Port from openclaw/openclaw#91950: normalize LLM-generated URLs like 'https:// docs.example' before web tool safety checks while preserving path and query encoding semantics.
* fix: cool down transient Telegram typing failures
Port from openclaw/openclaw#93020: add per-chat cooldown for transient sendChatAction failures so keep-typing refreshes do not hammer Telegram during network blips or rate limits.
* fix: support bare Telegram adapters in typing cooldown
* test: update typing backoff imports for relocated Telegram adapter
The Telegram adapter moved from gateway/platforms/telegram.py to
plugins/platforms/telegram/adapter.py since this branch was created;
point the test imports and monkeypatch targets at the new module.
--safe-mode promised to disable ALL customizations, but shell hooks
declared in config.yaml's hooks: block registered anyway —
register_from_config() runs independently of plugin discovery and
load_config() does not honor HERMES_IGNORE_USER_CONFIG. Gate it on
HERMES_SAFE_MODE at the single chokepoint so troubleshooting runs fire
zero user-configured code (plugins, MCP, and hooks).
Docs (en + zh) updated; positive + negative tests added.
The salvaged fix added a post-worker _interrupt_requested re-check to the
main OpenAI/Anthropic streaming poll loop. The Bedrock Converse poll loop
(interruptible_streaming_api_call, api_mode='bedrock_converse') has the same
bug class: its worker calls stream_converse_with_callbacks(on_interrupt_check=
...), which breaks out of the event loop on interrupt and returns a PARTIAL
response WITHOUT raising (bedrock_adapter.py). The worker sets result[
'response'] and exits with _interrupt_requested still True, so the in-loop
raise never fires and the poll loop returns the partial — silently swallowing
/stop on Bedrock exactly as it was on the paths the salvaged commit fixed.
Add the identical post-worker re-check before the Bedrock loop's return.
The non-streaming loop (interruptible_api_call) is structurally immune: its
worker's only early return fires off _request_cancelled, which is set by the
main loop immediately before it raises in-loop, so no swallow window exists.
Guard test flips _interrupt_requested True mid-stream (after the pre-flight
check) and asserts InterruptedError is raised; verified RED without the fix
(DID NOT RAISE) and GREEN with it.
The provider-mismatch guard now checks pool_provider and
current_provider != pool_provider. MagicMock.provider returns
a truthy child mock by default, which would trigger the guard
and skip the pool recovery tests. Set pool.provider='' explicitly.
Review findings: (a) an absolute path outside trusted roots passes
through unchanged and gets rejected downstream by skill_view — add a
debug log at the pass-through so the cron 'skill not found' symptom is
diagnosable next time; (b) test_relative_path_unchanged patched
get_skills_dir although the relative branch early-returns before any
root lookup — drop the misleading patch.
The extracted normalize_skill_lookup_name() resolved trusted roots via
agent.skill_utils.get_skills_dir(), but skill_view() enforces
tools.skills_tool.SKILLS_DIR — a separate module attribute that callers
and 60+ existing tests patch directly. With the helper reading a
different symbol than the enforcer, any SKILLS_DIR patch (or future
divergence between the two resolvers) makes normalization disagree with
enforcement and absolute-path loads regress silently. Read SKILLS_DIR at
call time (deferred import, cycle-safe) with get_skills_dir() as the
fallback, and align the new tests to patch the enforced symbol.
Follow-up to the salvage of #59829 by @HexLab98.
Add unit tests for normalize_skill_lookup_name and a cron scheduler
regression that absolute paths under the skills dir reach skill_view as
relative lookups.
Port from nearai/ironclaw#4547: treat a JSON null memory target as omitted so strict providers that fill optional fields with null use the documented default target instead of failing validation.
Deleting the matched user message breaks the strict role-alternation
invariant on the exact incident tail this fix targets — user(confirm) →
assistant('OK, restarting') becomes two consecutive assistant messages,
which strict providers reject and which the alternation-repair passes
upstream don't cover. Replace the message content with an explicit
'confirmation EXPIRED, re-confirm before any destructive action'
sentinel instead: the trigger text is still neutralized, the model gets
an affirmative instruction not to act, and the message sequence stays
valid. Adds an alternation-preservation regression test.
Follow-up to the salvage of #59640 by @knoal.
When a high-risk side effect (e.g. host restart via shutdown.exe) runs,
the user's plain-text confirmation phrase is persisted in the conversation
transcript. If the host restart killed the gateway process before the
assistant's tool result was written, the transcript tail ends on the
assistant's text response - and the dangerous confirmation text remains
in the user role.
On the next inbound message - possibly a casual 'are you there?' from
the user minutes later - the LLM sees the stale confirmation and may
interpret the new turn as a fresh re-confirmation, re-executing the
destructive action. This is the failure mode reported in #59607.
Fix:
- Add strip_stale_dangerous_confirmations() in agent/replay_cleanup.py
that removes user messages whose content matches a known dangerous
confirmation pattern AND whose timestamp is older than 60 seconds.
- Add is_dangerous_confirmation() helper with the matched patterns
(i18n-aware: covers 確認強制重開機 from the original incident).
- Wire the stripper into _build_gateway_agent_history() right after the
existing 75ed07ace strippers, so the strip chain is:
strip_interrupted_tool_tails -> strip_dangling_tool_call_tail ->
strip_stale_dangerous_confirmations.
- Update _build_replay_entry() to preserve the timestamp on user
messages (it was previously dropped), since the new stripper needs it.
Complements 75ed07ace (which strips the assistant side of the broken
tail) by handling the user side: a stale plain-text confirmation that
the assistant has not yet responded to in a way the resume logic
recognises.
Failing-test-first discipline: the bug-detection test
test_stale_confirmation_text_is_stripped_on_resume fails on unfixed
code (proves the test catches the bug) and passes after the fix.
Five additional safety tests confirm no regression on:
- fresh confirmations (within expiry) are preserved
- non-confirmation text is preserved
- non-matching histories are untouched
- dangerous-pattern detection works in all cases (case, i18n, None)
- direct unit test of the strip helper
Refs: #59607
The repair UPDATE ('SET active = 1 WHERE active IS NULL') was gated at
schema_version < 12, so already-v12+ databases — the exact population hit
by #51646, where the reconciler-added active column lacks its NOT NULL
DEFAULT 1 — never healed rows written as NULL by the pre-fix INSERTs.
Move the idempotent repair into unconditional startup so historical
gateway transcripts become visible again after upgrading.
Follow-up to the salvage of #59832 by @HexLab98.
* fix(gateway): use process-level HERMES_HOME for identity files
Gateway identity files (PID, lock, runtime status, takeover/stop markers)
were written via get_hermes_home() which honours the _HERMES_HOME_OVERRIDE
contextvar used for per-session profile dispatch. When a profile-context
task happened to be active at write time, files landed in the wrong profile
directory.
Add _get_process_hermes_home() that skips the contextvar and uses only the
HERMES_HOME env var or platform default, and route all gateway identity file
paths through it.
Fixes#56986
* chore(release): map liuhao1024 author email for PR #56993 salvage
---------
Co-authored-by: liuhao1024 <sunsky.lau@gmail.com>
Co-authored-by: Ben <ben@nousresearch.com>
* fix(dashboard): use loopback host for in-container WebSocket client (#58993)
Fixes#58993 - the in-container Dashboard's WebSocket client was dialing
the bind host (0.0.0.0) instead of 127.0.0.1, hijacking the host browser
when the container port was exposed.
* `hermes_cli/web_server.py::resolve_dashboard_ws_url()` now substitutes
127.0.0.1 for any 0.0.0.0 bind host discovered via the existing
`find_unused_port` / `get_listen_address` path. LAN IPs and explicit
`DASHBOARD_WS_HOST` overrides pass through unchanged.
* Existing tests preserved (no regression on the explicit-bind case).
Tests in `tests/dashboard/test_ws_client_host.py` cover:
- Bind host 0.0.0.0 → ws URL uses 127.0.0.1
- Bind host 127.0.0.1 → ws URL uses 127.0.0.1 (no regression)
- Bind host 192.168.1.5 → ws URL preserves the LAN IP
- DASHBOARD_WS_HOST env override wins over auto-detection
AI-assisted fix by https://github.com/SquabbyZ/peaks-loop
(cherry picked from commit 5501dd38d660619729ebcad85b75a8f46f8deb38)
* chore(release): map SquabbyZ email for AUTHOR_MAP attribution (#59682)
---------
Co-authored-by: SquabbyZ <601709253@qq.com>
* fix(auth): resolve Anthropic OAuth file per-profile + close port-binding platform gaps
Two focused pieces salvaged from PR #57563:
1. _HERMES_OAUTH_FILE was computed at module import time — frozen before
HERMES_HOME/profile overrides, so multiplexed profile turns read and
wrote the DEFAULT profile's .anthropic_oauth.json (OAuth path hijack).
Replaced with a lazy _get_hermes_oauth_file(); all web_server.py call
sites updated.
2. _PORT_BINDING_PLATFORM_VALUES was missing whatsapp_cloud and line —
both bind aiohttp TCP listeners, so a secondary multiplex profile
enabling them would collide with the primary's listener instead of
failing fast at startup.
Original work by @austinlaw076. The rest of #57563 was redundant on
main (adapter routing sweep superseded by #56854's salvage; cron secret
scope landed in fdab380a1; nested-config fallback in from_dict).
* chore(release): map austinlaw076 author email for PR #57563 salvage
* test(hermes_cli): patch _get_hermes_oauth_file instead of removed _HERMES_OAUTH_FILE constant
---------
Co-authored-by: Austin <austin@openvm067.space>
Co-authored-by: Ben <ben@nousresearch.com>
* fix(gateway): per-profile pairing whitelist isolation for multiplex gateways
Pairing approvals are stored per profile (profiles/<name>/pairing/) and
authz routes pairing checks through the serving profile's store, so one
profile's approved users no longer authorize against every other
profile's whitelist in multiplex mode.
The global store remains for the hermes pairing CLI and single-profile
gateways; unregistered/unstamped sources fall back to it, preserving
existing behavior.
Salvaged from PR #53045 (pairing half). The SOUL.md half was dropped:
the agent turn already runs inside _profile_runtime_scope on main, so
load_soul_md() resolves per-profile without changes.
Original work by @soddy022.
* ci: redispatch after arm64 docker dashboard-slot flake (unrelated to this PR)
---------
Co-authored-by: soddy022 <290613374+soddy022@users.noreply.github.com>
* fix(gateway): scope reset banners' session info to the serving profile
The auto-reset notice and the manual /reset //new banner both appended
_format_session_info() outside any profile scope, so a multiplexed
gateway advertised the base config's model/provider/context while the
session actually ran on the profile's.
Route both call sites through a new _reset_notice_session_info(source),
which enters _profile_runtime_scope for the source's profile when
gateway.multiplex_profiles is on (mirroring _run_agent's gating), so
_load_gateway_config()/_resolve_gateway_model() resolve the profile's
config.yaml via the existing context-local home override. Single-profile
gateways never enter the scope — behavior unchanged.
Both call sites invoke the helper via asyncio.to_thread: under the
scope, resolution can do blocking work (credential refresh,
context-length HTTP probes) that previously failed fast unscoped and
must not run on the event loop.
Fixes#59003
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* chore(release): map irresi author email for PR #59048 salvage
---------
Co-authored-by: irresi <blueirobin02@gmail.com>
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
A hosted agent whose Nous bootstrap session dies terminally (invalid_grant /
quarantine) looks HEALTHY to every liveness/connectivity probe — the machine,
relay ws, and dashboard all stay up — yet every inference turn hard-fails with
a provider-auth error until a human re-logs-in. Nothing currently surfaces that
condition to NAS.
Add get_nous_session_validity() (valid|terminal|unknown), classified from local
auth-store state (no working token required), and report it on the public
/api/status payload. NAS's 2-min health sweep reads it and re-mints the
bootstrap session in place on 'terminal'.
Anti-flap: only a terminal failure (relogin_required / persisted quarantine
marker with tokens cleared) maps to 'terminal'; transient/mid-rotation blips and
merely-expiring tokens report 'unknown' so a healthy box never triggers a
spurious re-mint.
Part of the hosted-agent bootstrap-session self-heal (NAS side reads this field).
The test_module_resolves_to_this_worktree guard asserted auth.__file__ contained
'worktrees/bootstrap-h2-logging' — a local dev crutch to defeat the editable-
install trap (venv points at the main checkout). In CI the code lives at
/home/runner/work/... so the assertion always fails. It never belonged in the
committed suite; the 5 behavioural tests are what matter.
A NAS-hosted Fly agent's Nous bootstrap session can take a terminal
invalid_grant and get quarantined in _quarantine_nous_oauth_state, which
clears the dead tokens from auth.json. Until now this quarantine was
completely silent: the only signal was a downstream "No access token found"
WARNING once the credential pool was already empty, which is too late to
root-cause. Because the Fly log drain is WARNING-only, nothing about the
terminal death reached centralized logging, and a real incident could not be
diagnosed because the evidence was never recorded.
Emit a WARNING+ forensic record AT the quarantine point, before the token
material is cleared. Fields: refresh_token hash prefix (12-char SHA-256 hex,
correlates to NAS's refreshTokenHash), client_id, agent_key_id, error code,
reason, auth.json path/size/mtime/exists, and whether the token was already
past its own expiry. WARNING level is deliberate — INFO never reaches the Fly
drain.
Redaction safety (load-bearing): the log dict is built only from computed
values (hash prefix, sizes, booleans). No raw refresh_token, access_token, or
agent_key bytes are ever passed into the log call, avoiding Hermes's known
credential-literal corruption bug class. A test asserts the raw refresh token
substring is absent from all emitted log output.
Note: no session_id field exists on Nous auth state; provenance is captured
via client_id + agent_key_id, which are non-secret routing identifiers.
The stage2-hook auth.json seed is first-boot-only ([ ! -f auth.json ]) to avoid
clobbering rotated refresh tokens on restart. That guard means a container whose
Nous bootstrap session took a terminal invalid_grant (tokens cleared,
providers.nous.last_auth_error.relogin_required stamped) cannot recover from a
restart — it stays unauthenticated until the credential is replaced.
Add a self-heal path: an orchestrator that manages the container supplies a
freshly-issued session via HERMES_AUTH_JSON_REBOOTSTRAP (distinct from the
create-only *_BOOTSTRAP var). On boot, scripts/docker_rebootstrap_nous_session.py
swaps ONLY the providers.nous entry, and ONLY when the on-disk entry is provably
terminal (quarantine marker + no usable tokens). Healthy/rotating/absent/
unparseable auth.json is always a no-op, so the env is safe to leave set across
restarts and never clobbers a good token. Pure stdlib, runs as its own
subprocess, always exits 0 so a re-seed error never fails the boot.
Reuses the same terminal predicate as get_nous_session_validity() so we re-seed
only a session that is genuinely dead.
Follow-up hardening on the cherry-picked pool-fallback fix. The original
_resolve_codex_usage_credentials wrapped BOTH resolve_codex_runtime_credentials()
and the separate _read_codex_tokens() account_id read in one broad
'except Exception: pass', which had three problems:
1. A transient refresh/network failure (non-AuthError) from the resolver was
silently swallowed and downgraded to pool.select(), which could report
/usage limits for a DIFFERENT pool account than the one actually running.
On main that error surfaced. This is a real behavior regression for the
multi-account/pool case.
2. If the resolver succeeded but only the account_id read raised, the whole
singleton tier was abandoned in favor of a pool token that carries no
ChatGPT-Account-Id header (PooledCredential has no account_id concept),
risking a wrong-account read or 401.
3. 'except Exception' masked genuine programming errors.
Fix: narrow the outer catch to AuthError (the documented 'no creds' failure
mode of both functions), and read account_id in a best-effort inner try so a
partial/missing singleton store can't sink an otherwise-usable credential.
Transient errors now propagate and fail open via the outer fetch_account_usage
guard rather than mis-routing to the wrong account. Adds debug breadcrumbs and
a comment characterizing when the tier-3 pool path actually fires.
Guard tests: a non-AuthError resolver failure must NOT swap to the pool
(fail-open, no snapshot); an account_id read failure keeps the singleton token.
Updated the existing pool-fallback test to use AuthError (the real failure
mode) instead of a generic RuntimeError.
Follow-up on the salvage of #59523. Two low-risk cleanups surfaced by review:
- Extract _ZAI_CODING_OVERLOAD_SHORT_ATTEMPTS as a module constant so
adaptive_rate_limit_backoff() and zai_coding_overload_retry_ceiling()
share one source of truth. Previously both hardcoded short_attempts=3
independently; tuning one without the other would silently desync the
retry ceiling from the backoff schedule.
- Replace the tautological formula-mirroring assert in
test_zai_overload_retry_ceiling_exceeds_short_attempts with a behavior
invariant (ceiling leaves headroom for every long-backoff entry), per the
repo's contracts-over-snapshots testing rule.
Assert the invariant that the Z.AI overload retry ceiling exceeds the
short-retry threshold (the original bug had them equal, so the long tier
was dead code), and walk the attempt range the retry loop actually
traverses to prove the full 30/60/90/120s long-backoff schedule now runs.
A fallback candidate can itself carry a stale credential (e.g. an
expired ANTHROPIC_TOKEN picked up by _try_anthropic). Its 401 previously
propagated out of the fallback call site and aborted the auxiliary task
— for compression: a 60s cooldown + context marker while the session
kept growing past the context cap. Live case: mattalachia debug dump
(Jul 2026), Codex timeout → Anthropic 401 x5 → 296K 'Cannot compress
further'.
Now each fallback candidate call is wrapped: on auth error, refresh the
candidate's provider credentials and retry once; if unrefreshable, mark
the provider unhealthy and walk the discovery chain again so the next
viable candidate serves. Sync + async paths. Non-auth errors still
raise unchanged.
Infer the concrete auxiliary auth provider from the selected client base
URL so provider:auto routes can refresh Copilot/Codex/Anthropic/Nous
credentials after auth errors, instead of skipping refresh because
resolved_provider stayed 'auto'. Adds the copilot branch to
_refresh_provider_credentials and evicts the stale auto-route cache
before retrying.
Fixes#20832. Salvaged from PR #20837, reapplied surgically onto current
main (branch predated the _retry_same_provider_sync/async extraction).
The -m flag seeds HERMES_MODEL/HERMES_INFERENCE_MODEL for the launched TUI
process only. But the per-turn config sync (_sync_agent_model_with_config)
computed its target via _config_model_target(), which fell back to those
env vars whenever config.yaml had no model.default — the normal state for
custom-provider-only setups. The sync then replayed the -m model as a
/model switch, and with model.persist_switch_by_default (default true)
_persist_model_switch wrote model.default/provider/base_url into
config.yaml. A one-shot CLI flag became the permanent global model,
visible in every new session and every model picker.
Two-sided fix:
- _config_model_target() no longer falls back to the env seed. Empty
model = config expresses no preference = sync is a no-op. The agent
keeps the session-scoped -m model; config.yaml edits still sync.
- _apply_model_switch() gains persist_override; all three internal
callers (config sync, /moa one-shot swap, /moa post-turn restore) pass
persist_override=False so session-mechanical switches can never write
config.yaml regardless of the persist-by-default setting. User-typed
/model keeps its existing flag/config behavior.
E2E-verified against an isolated HERMES_HOME with a custom-provider-only
config + -m env seed: sync no longer fires, config.yaml byte-identical,
_resolve_model() still returns the seed for the session's own agent.
When context.engine selects a plugin engine (e.g. LCM), the host
compression threshold — including the Codex gpt-5.5 50% -> 85%
autoraise — only configures the built-in ContextCompressor and never
reaches the plugin. The autoraise notice still fired, telling the user
auto-compaction was raised when nothing actually changed, and the
startup context-limit line printed the host percent next to the
engine's own threshold_tokens, contradicting itself.
- Clear _compression_threshold_autoraised when a plugin engine is
selected, suppressing both the CLI startup notice and the gateway
turn-1 replay via _compression_warning.
- Print the active engine's own threshold_percent in the startup
context-limit line so percent and token count agree.
- Built-in behavior is preserved, including the fallback path where a
configured engine fails to load and the built-in compressor takes
over.
Fixes#44439
The Codex gpt-5.5 compaction-threshold autoraise notice re-fired on every
agent init. Because the gateway rebuilds the agent per inbound message, the
notice spammed long-running Discord/Telegram/etc. sessions, and the only
documented remedy (`compression.codex_gpt55_autoraise false`) disables the
useful autoraise behavior itself.
Gate both emission surfaces — the CLI startup print and the gateway
`_compression_warning` replay — on a persisted per-profile marker under
`$HERMES_HOME` (`.codex_gpt55_autoraise_notice`), keyed on the from→to
percentages the notice displays. The notice now shows at most once per
profile; the autoraise still fires and `codex_gpt55_autoraise: false` still
disables it; and a later change to the raised threshold re-notifies once.
Docs updated to match.
The Codex gpt-5.5 compaction autoraise (#40957) overrode the effective
threshold unconditionally. If a user had set compression.threshold above
0.85, agent_init dropped them down to 0.85. That wastes usable window and
contradicts the feature's whole point: use more of the context, not less.
It happened silently too, since the one-time notice is suppressed when the
override doesn't raise.
The override is an autoraise. It must only raise. Pulled the apply logic
into a small pure helper that clamps the Codex case to never lower a
higher-or-equal user threshold, and emits the notice only when it actually
fires. Other overrides (Arcee Trinity) keep their existing unconditional
behavior.
Fixes the Codex gpt-5.5 compaction autoraise lowering a user's higher
configured threshold. A user on the Codex OAuth route with
compression.threshold > 0.85 was silently clamped to 0.85, compacting
earlier than they asked and using less of the 272K window the feature was
meant to unlock. The autoraise now only ever raises.
N/A
- [x] 🐛 Bug fix (non-breaking change that fixes an issue)
- [ ] ✨ New feature (non-breaking change that adds functionality)
- [ ] 🔒 Security fix
- [ ] 📝 Documentation update
- [ ] ✅ Tests (adding or improving test coverage)
- [ ] ♻️ Refactor (no behavior change)
- [ ] 🎯 New skill (bundled or hub)
- `agent/agent_init.py`: added `_resolve_compression_threshold()`, a pure
helper that combines the global threshold with a per-model override. The
Codex gpt-5.5 autoraise never lowers a higher-or-equal user threshold;
the notice is returned only when it actually raises. Rewired `init_agent`
to call it, replacing the unconditional `compression_threshold = _model_cthresh`.
- `tests/agent/test_arcee_trinity_overrides.py`: added 5 cases for the
helper — raise from default, never-lower regression, equal-is-noop,
no-override passthrough, and non-codex (Trinity) unconditional apply.
1. Set `compression.threshold: 0.90` and run gpt-5.5 on provider `openai-codex`.
2. Before: effective threshold drops to 0.85, no notice. After: stays 0.90.
3. Run `scripts/run_tests.sh tests/agent/test_arcee_trinity_overrides.py`.
Stash `agent/agent_init.py` and the new cases fail; restore and they pass.
- [x] I've read the [Contributing Guide](https://github.com/NousResearch/hermes-agent/blob/main/CONTRIBUTING.md)
- [x] My commit messages follow [Conventional Commits](https://www.conventionalcommits.org/) (`fix(scope):`, `feat(scope):`, etc.)
- [x] I searched for [existing PRs](https://github.com/NousResearch/hermes-agent/pulls) to make sure this isn't a duplicate
- [x] My PR contains **only** changes related to this fix/feature (no unrelated commits)
- [x] I've run `pytest tests/ -q` and all tests pass
- [x] I've added tests for my changes (required for bug fixes, strongly encouraged for features)
- [x] I've tested on my platform: macOS 15 (Darwin 25.5)
- [x] I've updated relevant documentation (README, `docs/`, docstrings) — or N/A
- [x] I've updated `cli-config.yaml.example` if I added/changed config keys — or N/A
- [x] I've updated `CONTRIBUTING.md` or `AGENTS.md` if I changed architecture or workflows — or N/A
- [x] I've considered cross-platform impact (Windows, macOS) per the [compatibility guide](https://github.com/NousResearch/hermes-agent/blob/main/CONTRIBUTING.md#cross-platform-compatibility) — or N/A
- [x] I've updated tool descriptions/schemas if I changed tool behavior — or N/A
gpt-5.3-codex-spark has a native 128K context window but the default
50% compaction trigger fires at ~64K, wasting half the usable window
before the session has accumulated enough turns to summarize
meaningfully. This raises the trigger to 70% (~90K) on the Codex OAuth
route only, leaving ~38K headroom for the summary and continued
conversation before the 128K hard limit.
The override is not gated by allow_codex_gpt55_autoraise because 128K
is the model's native window (unlike gpt-5.5's artificial 272K Codex
cap). Non-Codex routes are unaffected.
Also adds a boundary regression test verifying the short-session
scenario from the issue always yields a non-empty compressible window
(no silent context wipe).
The ChatGPT Codex OAuth backend caps both gpt-5.4 and gpt-5.5 at a 272K
context window, but the autoraise that lifts the compaction trigger to 85%
only matched gpt-5.5. On gpt-5.4 the global 50% threshold fired at ~136K —
half the usable window — compacting far earlier than necessary.
Rename _is_codex_gpt55 -> _is_codex_gpt54_or_gpt55 and match both families.
The one-time user notice is now model-aware (shows the actual slug). The
config key codex_gpt55_autoraise is kept as-is for backward compatibility.
Adds gpt-5.4 coverage to the autoraise tests.
Surfaces the usage_report()/provenance() data layer added in #36701 as a
user-facing CLI command. Unlike `hermes curator status` (scoped to
curator-managed agent-created candidates), `usage` lists every skill on disk
— bundled built-ins and hub-installed included — with per-skill use/view/patch
counts and an agent/bundled/hub provenance tag.
Flags: --sort {activity,recent,name}, --provenance {agent,bundled,hub} filter,
--json for machine-readable output.
* feat(oneshot): add --usage-file JSON usage report to hermes -z
Pipelines driving hermes -z (batch reviewers, cron scripts, eval
harnesses) had no way to account for per-invocation spend: the agent
computes estimated_cost_usd and full token counts internally, but
oneshot mode discards everything except the final response text.
- hermes -z PROMPT --usage-file PATH writes a JSON report after the
run: estimated_cost_usd, cost_status/source, input/output/cache/
reasoning/total tokens, api_calls, model, provider, session_id,
completed, failed.
- Written even when the run fails (with a failure field) so callers
can always account for spend; the write itself is best-effort and
never masks the run's own outcome.
- Flag registered in both the full parser and the Termux fast path;
added to both value-flag scan sets so profile detection stays
correct.
Validation: 6 unit tests + live E2E (real -z run produced a report
with real OpenRouter cost + token counts).
* test: include usage_file kwarg in oneshot dispatch assertions
The two dispatch tests assert the exact kwargs dict passed to
run_oneshot; the new usage_file kwarg must appear there.
A user-approved terminal/execute_code command could be SIGINT-killed
(exit 130 + "[Command interrupted]") by a stale interrupt bit that landed
on the execution thread during the blocking approval-wait, while the
result still carried the "...approved by the user." note. The terminal
tool runs sequentially inline on the execution thread, and nothing
cleared or re-checked the bit between approval-grant and env.execute.
Clear the current thread's interrupt bit once before an approved command
spawns its child (terminal foreground; execute_code local + remote), and
enrich the note to "...approved by the user, then interrupted." on a
genuine post-start interrupt instead of implying success. A genuine
interrupt arriving after execution starts (or during a retry backoff)
still SIGINTs the command; non-approved commands keep current behavior.
Adds regression tests covering stale-bit-clears, genuine-interrupt-still-
kills, the retry-backoff window, natural-exit-130 (not mislabeled), and
execute_code local + remote.
Follow-up on the cherry-picked #36896 commits, wiring 1Password into
the new registry as the reference *mapped* source:
- OnePasswordSource adapter (shape=mapped, scheme=op): fetch-only —
precedence, override semantics, conflict warnings, and env writes
move to the orchestrator; apply_onepassword_secrets kept as legacy
shim like Bitwarden's.
- Registered in _ensure_builtin_sources; mapped op:// bindings now
outrank bulk Bitwarden project dumps on contested vars.
- _cache.py FetchResult/is_valid_env_name re-exported from base so
there is exactly one canonical definition; bitwarden.py re-adapted
onto the contributor's DiskCache substrate.
- ErrorKind classification for op failures (auth/binary/empty/network).
- Registry + conformance coverage for OnePasswordSource, incl. the
headline multi-source test: both vaults claim the same var, mapped
1Password wins, conflict surfaced, provenance correct.
- env_loader tests migrated off the legacy apply_* mocks onto the
fetch layer; AUTHOR_MAP entry for @hwrdprkns.
The 1Password secret source resolves op:// references using
OP_SERVICE_ACCOUNT_TOKEN read from os.environ. Under systemd the gateway
gets that token via EnvironmentFile, but cron jobs, subprocesses, CLI
runs, macOS launchd, and Docker containers spawn fresh interpreters with
no inherited shell state — so they silently failed to resolve any
reference and fell back to empty strings.
Two patches close the gap, matching Bitwarden's reliability guarantees:
1. env_loader: auto-load ~/.hermes/.op.env after .env so the gitignored
bootstrap token is available everywhere. override=False plus an
explicit guard ensure it never clobbers a token already in env (e.g.
from a systemd EnvironmentFile, which keeps precedence).
2. credential_pool: _get_env_prefer_dotenv() now prefers the resolved
value in os.environ when .env still holds a raw op:// reference,
instead of handing a URL to provider auth. Non-op:// values keep the
existing .env-takes-precedence behaviour.
Also gitignore .op.env, document the three bootstrap-token options, and
add tests covering auto-load, no-override, and the resolved-vs-raw
precedence (plus regression guards).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Resolve provider credentials from 1Password op://vault/item/field references
at startup via the official `op` CLI, alongside the existing Bitwarden source.
Users map env-var names to references in secrets.onepassword.env; after .env
loads, each is resolved with `op read` and injected into os.environ. Auth is
whatever `op` already uses (service-account token or desktop/interactive
session) — Hermes never authenticates or installs `op` itself.
Startup-safe and fail-open: a missing binary, expired auth, a bad reference,
or an empty value each warn and fall back to existing credentials, never
blocking startup. Successful, complete pulls are cached in-process and on disk
(<hermes_home>/cache/op_cache.json, 0600) via the shared DiskCache; only
secret values are stored, never the token (auth is fingerprinted into the
key). Adds `hermes secrets onepassword {setup,status,set,remove,sync,disable}`
(aliases op/1password), config defaults, the cli-config example, docs, and
hermetic tests.
Hardening applied across both backends in env_loader: each source runs in its
own guard, config sections are coerced to dict, and cache_ttl_seconds is
coerced defensively — so a malformed secrets: section can't abort startup.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Follow-up to #59524. The one-shot running-claim stale-recovery window was a
fixed 30-min constant. Derive it from the cron inactivity timeout instead
(HERMES_CRON_TIMEOUT, the same limit the scheduler enforces per run) so the
safety valve tracks how long a run may actually go quiet:
- unset/invalid -> default 600s inactivity -> TTL 1800s (unchanged behaviour)
- positive N -> max(N * 3 headroom, 1800s floor)
- 0 (unlimited) -> no finite bound -> fall back to the 1800s constant
The fixed constant is kept as the floor + unlimited-case fallback. Resolved
once per due-scan. HERMES_CRON_TIMEOUT is a pre-existing internal env var
(already read by cron/scheduler.py); no new config surface.
E2E: with HERMES_CRON_TIMEOUT=1200 the claim now survives to 60min where the
old fixed 1800s constant wrongly expired it at 30min mid-run. +1 derivation
test; 640/640 cron tests pass.
When a bundled web provider (firecrawl, tavily, exa, ...) is listed in
plugins.disabled, its provider never registers and the web_search/
web_extract dispatchers emitted the misleading "No web extract provider
configured. Set web.extract_backend to ..." — even though the backend was
configured correctly. The real fix is to re-enable the plugin.
- web_tools.py + web_search_registry.py: when the configured backend names
a disabled bundled web plugin, both dispatchers now point the user at the
actual cause (re-enable the plugin) instead of a wrong config hint.
- plugins_cmd.py cmd_enable: enabling by canonical key now also clears the
manifest-name alias (web-firecrawl) from plugins.disabled, so the
suggested command actually re-enables the plugin ('explicit disable wins'
matches on the name too).
- plugins_cmd.py cmd_toggle / _run_composite_ui / _run_composite_fallback:
the interactive 'hermes plugins' menu now persists the canonical key
(web/firecrawl), never the bare manifest name — the drift that put the
offending entry in plugins.disabled in the first place.
Follow-up to #59518 (which fixed web credential resolution, a different
cause). Fixes the disabled-plugin symptom reported after that PR.
The PR predates #31884, which changed the non-interrupted api_calls==0
empty path from silence to a retry hint. Flip the contributed test to
assert the current (correct) behavior.
A /stop sets _interrupt_requested on the session's cached agent, but the
flag is only cleared by the turn finalizer. When the stopped run is hung
or still draining, the flag survives the forced lock release and the
session's NEXT user message is killed at the top of the tool loop
(conversation_loop.py interrupt check): the run completes with
interrupted=True, api_calls=0 and an empty response, which
_normalize_empty_agent_response passed through as pure silence — the
user's message was swallowed with no trace except a
'response ready: ... api_calls=0 response=0 chars' log line.
Two-layer fix:
- _interrupt_and_clear_session now evicts the cached agent whenever it
releases the running state. The next message rebuilds the agent from
session history (mirroring the /new and /model paths), while the old
agent object keeps its interrupt flag so a hung drain still dies when
it unblocks. This intentionally does NOT clear the flag in place:
turn_context deliberately preserves a pending interrupt across turn
start (it carries interrupt-message delivery), and clearing it could
revive a hung run the user just stopped.
- _normalize_empty_agent_response distinguishes a drain from a swallowed
turn: an interrupted run that did work (api_calls > 0) stays silent as
before (deliberate stop/steer; queued messages are delivered by the
recursive drain inside _run_agent), but an interrupted run with ZERO
api_calls never processed the user's message at all and now surfaces a
'send it again' notice instead of nothing.
Same silent-delivery class as a1f76ba7e (#29346), which covered the
extract-stripped case; regression tests added next to that coverage.
Fixes#44212
test_verification_status_outside_workspace_is_not_applicable passed tmp_path as
the cwd and asserted status == not_applicable, relying on tmp_path having no
project-marker ancestor. _marker_root() walks up to ~6 levels, so a stray marker
in a shared tmp-root ancestor (e.g. a /tmp/package.json left by another tool)
made project_facts_for() resolve tmp_path as a workspace and flip the status to
unverified. Green in clean CI, red on any dev box with a polluted /tmp.
Force the no-facts precondition by monkeypatching project_facts_for -> None so
the test deterministically exercises the not_applicable branch regardless of
ambient filesystem state. Test-only; no production change.
`summarize_background_review_actions` was structured on the assumption
that every parsed tool response is a fully-typed dict-of-fields. In
practice the memory/skill tools — and their wrappers over Mem0 OSS and
the skill_manage MCP server — sometimes serialize `_change` as a list
or scalar, and clamp `operations` to a single string when the field
came in via a partial JSON bridge.
The original code did the equivalent of:
change = data.get("_change", {})
change.get("description", "")
so when `_change` was a list the inner .get crashed with
`AttributeError: 'list' object has no attribute 'get'`, every ~10
turns the user saw the entire background review collapse.
Three defensive guards in summarize_background_review_actions:
- `call_details.get(tcid, {})` → `call_details.get(tcid) or {}` plus
`isinstance(detail, dict)` coercion. Catches stale scalar/None
values when a fork inherits partial state from a stale tool_call_id.
- `operations = detail.get("operations") or []` → `isinstance(ops_raw, list)`
coerce, then per-entry isinstance check before `.get()`. Skips
non-dict items without raising; an entire surrounding review no
longer goes down because one entry was malformed.
- `change = data.get("_change", {})` → `isinstance(change_raw, dict)`
coerce. The originally-reported crash class for skill_manage with
list-shaped _change now falls through to the generic summary path.
And the caller in `_run_review_in_thread` is wrapped in a try/except
that maps any residual summarize exception to `actions = []` and
emits a 'partial results' warning, so even an entirely unanticipated
shape won't take down the outer review — the user only sees
'Background memory/skill review failed' instead of the prior hard
crash that lost every successful action the fork had completed.
Tests: tests/test_background_review_list_shapes.py — standalone
pytest-free runner, 7/7 PASS:
a_change_as_list_does_not_crash (originally-reported shape)
a_change_as_int_does_not_crash (scalar fallback)
b_operations_as_string_treated_as_empty
b_operations_as_none_treated_as_empty
c_operations_contains_non_dict_entries (verbose-mode per-entry filter)
d_detail_non_dict_replaced_with_empty
e_call_defends_via_try_except (structural anchor)
Refs NousResearch/hermes-agent#59437
Server names with non-env-safe characters (dots, slashes, spaces)
produced invalid env-var keys like MCP_MY.SERVER_API_KEY or
MCP_GITHUB/MCP_API_KEY, breaking .env writes and ${VAR} header
substitution. _env_key_for_server now replaces any character outside
[A-Za-z0-9_] with an underscore.
Co-authored-by: Hermes Agent <agent@nousresearch.com>
Allow mainstream reverse-proxy path mounts to keep their X-Forwarded-Prefix when Home Assistant Supervisor ingress already consumes nearly the old 64-character budget. Keep validation bounded and keep rejected non-empty prefixes diagnosable with a deduplicated warning.
Constraint: HA Supervisor ingress prefixes are 63 chars before add-on subpaths, so the old 64-char cap dropped valid dashboard deployments.
Rejected: remove the length cap entirely | a bounded header budget is still a conservative validation guard.
Confidence: high
Scope-risk: narrow
Directive: Keep prefix validation centralized in hermes_cli.dashboard_auth.prefix so auth routes, cookies, and SPA asset rewriting agree.
Tested: python probe for the 73-char HA ingress prefix; scripts/run_tests.sh tests/hermes_cli/test_dashboard_auth_prefix.py -q; .venv/bin/python -m pytest tests/hermes_cli/test_web_server.py -k 'spa_assets_are_read_as_utf8' -q; python -m ruff check hermes_cli/dashboard_auth/prefix.py tests/hermes_cli/test_dashboard_auth_prefix.py; git diff --check
Not-tested: full test suite
The +60s next_run_at advance only delayed a duplicate one-shot dispatch by
one tick — a job that outlives the 60s tick interval (the reported 2.5-min
research prompt) still re-fired on the next tick after the window expired,
so the concurrent gateway+desktop double-delivery persisted.
Replace it with a durable run_claim (at+by, mirroring fire_claim) stamped
on the one-shot under the same jobs lock get_due_jobs holds, and checked at
the top of the due-scan: a fresh claim held by an in-flight run makes every
other scheduler process skip the job for its ENTIRE run, not one tick.
mark_job_run() clears the claim on completion; a ONESHOT_RUN_CLAIM_TTL
(30 min) safety valve re-dispatches a claim left by a tick that died mid-run
so a one-shot is never wedged.
E2E: long-running one-shot no longer double-fires at +28/+61/+120/+179s;
completion clears the claim + disables the job; crash recovery re-arms past
the TTL. +3 regression tests.
When two scheduler processes (gateway + desktop) run concurrently,
both could pick up the same one-shot job from get_due_jobs() because
its next_run_at was not advanced before execution started — only
recurring jobs were advanced (L3446). This caused duplicate deliveries
and wasted token spend (#59229).
Now _get_due_jobs_locked advances a one-shot's next_run_at by 60s
before returning it as due, persisted immediately under the same
file lock. mark_job_run re-anchors next_run_at on completion, so a
tick death between advance and execution only delays the job by one
tick window — it is never lost.
Closes#59229
Dashboard /chat for the default (launch) profile attaches to the
dashboard process's in-memory TUI gateway. The Node PTY child receives a
bridged TERMINAL_CWD env var, but the in-memory gateway process does not,
so cwd resolution fell through to os.getcwd() (wherever `hermes
dashboard` was launched) and ignored the configured terminal.cwd.
Read the launch profile's config.yaml directly in the in-memory cwd
resolution: a configured terminal.cwd now wins over a stale process env
and the launch directory. Widened to the resume/fallback session-cwd
sites (not just _completion_cwd) via a shared _default_session_cwd()
helper so fresh AND resumed sessions honor the config.
Co-authored-by: ygd58 <buraysandro9@gmail.com>
The CWE-22 traversal guard in SessionEntry.from_dict rejects any
interior '/' in session_key, but session_key is a logical routing
key (never used as a filesystem path) and Google Chat resource names
legitimately contain '/' (spaces/<id>, spaces/<id>/threads/<id>).
All Google Chat sessions were silently dropped on gateway start.
Split the validation: session_id keeps the strict _is_path_unsafe
guard (it's the value used as a filename); session_key now uses a
relaxed _is_session_key_unsafe helper that only blocks genuine
traversal vectors (parent-dir '..', leading '/', leading '\', leading
Windows drive-letter prefix) and allows interior '/'.
The CLI model-switch display (both picker and direct-switch paths)
omitted the custom_providers keyword when calling
resolve_display_context_length(). The function already supports it
(and the gateway correctly passes it), but the CLI call sites relied
on the fallthrough to probe-down default (256K) even when a
custom_providers entry specified a per-model context_length.
Fix: pass agent._custom_providers at both resolve_display_context_length
call sites in HermesCLI._apply_model_switch_result(), matching the
pattern already used for config_context_length.
build_preloaded_skills_prompt() (hermes -s <skill>, and tui_gateway's
HERMES_TUI_SKILLS deployment env var) loads skills via _load_skill_payload()
with a raw identifier, bypassing get_skill_commands()' scan-time disabled
filter entirely. Result: a skill an operator disabled via skills.disabled
still gets force-loaded and injected into every session — including every
session on a shared tui_gateway deployment where the operator set
HERMES_TUI_SKILLS.
The bundle-invocation path (#59156) already re-checks get_disabled_skill_names()
for exactly this reason; preloaded-skill loading was the other _load_skill_payload
call site still missing it.
Fix: check each resolved skill's name (and raw identifier) against
get_disabled_skill_names() before injecting it. A disabled skill is now
reported the same way an unknown one already is (skipped, listed in the
returned missing_identifiers) — no return-shape or caller changes needed.
No behavior change when no skill is disabled.
Same bug class as #40190: these providers read credentials via bare
os.getenv(), so keys stored in ~/.hermes/.env (hermes config layer)
were invisible in execution paths that never exported them into the
process environment. Add get_provider_env() on the WebSearchProvider
module as the shared config-aware lookup (get_env_value with os.getenv
fallback) and route all credential reads through it. SearXNG already
did this (#34290); Firecrawl fixed in the preceding cherry-picked
commit by @liuhao1024.
The Firecrawl provider used os.getenv() to read FIRECRAWL_API_KEY and
FIRECRAWL_API_URL, which only checks the process environment. When
values are supplied through Hermes's ~/.hermes/.env config mechanism
(via hermes_cli.config.get_env_value), they are not guaranteed to be
present in os.environ for every gateway/tool execution path.
Switch to get_env_value() which checks both os.environ and the .env
file, matching the pattern used by other providers (nous_subscription,
setup, discord adapter).
Fixes#40190
_classify_by_status() routes every other transient HTTP status to a retryable
reason (500/502 -> server_error, 503/529 -> overloaded, 429 -> rate_limit,
413 -> payload_too_large), but 408 Request Timeout fell through to the generic
`400 <= status < 500` branch and was classified as a non-retryable
format_error -- the same bucket as a 400 Bad Request.
A 408 is a transient timing failure the server itself flags as safe to retry
(RFC 9110 15.5.9), not a malformed request, so the retry loop aborted the turn
when a simple retry would recover. Common trigger: a reverse proxy in front of
a self-hosted backend (llama.cpp / Ollama / vLLM) returns 408 when a long
generation outruns the proxy's request-read window.
Route 408 to the existing FailoverReason.timeout (rebuild client + retry).
Add a regression test plus a boundary test asserting 400 stays non-retryable.
_sync_back_once defers a SIGINT that lands mid-sync, then re-delivers it once the
sync completes so the user's Ctrl+C isn't lost. It did so with
os.kill(os.getpid(), signal.SIGINT). That is not graceful on Windows: os.kill
only treats CTRL_C_EVENT(0)/CTRL_BREAK_EVENT(1) as console events; any other
value (SIGINT == 2) routes to TerminateProcess(sig), so a Ctrl+C during a
remote-backend (ssh/daytona/modal) sync-back hard-kills the whole CLI session
(exit code 2) on Windows instead of raising KeyboardInterrupt.
Use signal.raise_signal(signal.SIGINT) (3.8+), which invokes the restored
handler through C raise() on every platform. Verified on Windows: raise_signal
runs the handler (graceful) while os.kill(getpid, SIGINT) TerminateProcess-es
the process. Adds a cross-platform regression test that runs on Windows too (it
stubs the locked sync body, so unlike test_file_sync_back.py it needs no fcntl).
Completes the #59395 bug-class fix. create_job and update_job's
schedule-change path already reject past one-shots (via #59410/#59438);
this closes the two remaining doors that stored next_run_at=None for a
'once' schedule and re-created the silent ghost job:
1. update_job fallback-recompute (the safety-net that re-derives
next_run_at when it's missing on an enabled, non-paused job)
2. resume_job (resuming a paused one-shot whose time has already passed
— empirically confirmed to create a scheduled job that never fires)
The redundant update_job schedule-change hunk from the original PR was
dropped (already on main via #59438). Adds resume-reject + update-reject/
accept regression tests.
Salvaged from #59428 by isheng-eqi.
Follow-up to #59332 targeting the remaining PERCEIVED first-token latency
(the wire streaming was already per-token; these fix what the user sees):
1. display.show_reasoning default ON. On thinking models the reasoning
phase streams for tens of seconds; with the display off users stare
at a spinner the whole time and read it as a stall. Flipped in
DEFAULT_CONFIG, load_cli_config defaults, tui_gateway raw-YAML
fallbacks, and the hermes setup status line (all four read sites kept
in sync). Gateway per-platform defaults intentionally stay off —
messaging chats shouldn't fill with thinking text. /reasoning hide
still turns it off and persists.
2. Response box force-flushes long partial lines. _emit_stream_text only
painted on newline, so a response opening with a long paragraph
stayed invisible until the first \n — seconds of blank box. Now
partial lines wrap at terminal width and paint as tokens arrive
(mirrors the reasoning box's 80-char force-flush that existed since
day one). Table blocks remain batch-aligned; no content loss at wrap
boundaries (regression tests added).
3. hermes_time timezone resolution uses read_raw_config (mtime-cached +
libyaml C loader) instead of a raw yaml.safe_load of config.yaml
(~110-140ms measured) inside the FIRST system prompt build. First
build drops 320ms -> ~155ms on a 200-skill install.
4. Stale docs: configuration.md (en+zh) still documented the 70%/90%
[BUDGET WARNING] tool-result injections. Those were removed in April
2026 (c8aff7463) precisely because they hurt task completion; current
behavior is exhaustion-message + one grace call, no mid-loop
injection, no cache impact. Docs now describe reality.
Verified: token-count compression decisions already use API-reported
last_prompt_tokens (rough estimators are preflight-only and cost ~1.7ms
even on 1.7MB histories — not worth touching).
Widen the #59395 fix to the sibling site: update_job's schedule-change path
(cron/jobs.py) had the SAME unguarded compute_next_run -> next_run_at pattern,
so updating a job's schedule to a one-shot >ONESHOT_GRACE_SECONDS in the past
would re-create the ghost job (next_run_at=None, state='scheduled', never fires)
that create_job now rejects. Apply the identical guard on update (raise before
any disk write, so the original job is left intact), with regression tests for
the reject + future-accept cases.
Also surface ONESHOT_GRACE_SECONDS in the raised ValueError (not just the
warning log) so a caller knows how far in the past is too far. Message from the
competing PR #59410 by @isheng-eqi.
Co-authored-by: isheng-eqi <265044697+isheng-eqi@users.noreply.github.com>
Bare 'hermes sessions prune' keeps the historical 90-day default, but any
filter — now including --source — suppresses the implicit cutoff, so
'prune --source cron' targets ALL cron sessions instead of silently only
those older than 90 days (the surprise a user hit live: 'No sessions
match ... source cron' despite plenty of recent cron runs).
- CLI preview + confirmation now show the match count plus the oldest
and newest matching session start times before deleting.
- Dashboard /api/sessions/prune mirrors the semantics: attribute filters
without an explicit older_than_days match all ages (model_fields_set
distinguishes an explicit 90 from the Pydantic default); dry_run
responses gain oldest_started_at/newest_started_at.
- Docs + argparse help updated; tests for both surfaces.
The Skills Hub 'Browse Hub' landing page and index-backed search render
empty on fresh deployments (e.g. Fly.io VPS agents) with no stale cache.
Root cause: the centralized index at /docs/api/skills-index.json is a
large body (~34MB, tens of MB compressed) served with Content-Encoding:
br. httpx's streaming Brotli decoder — backed by brotlicffi 1.2.0.1,
which is pinned so aiohttp can decode Discord attachments — trips over
its own output_buffer_limit on a payload this size and raises:
DecodingError("brotli: decoder process called with data when
'can_accept_more_data()' is False")
_load_hermes_index() catches that (DecodingError is an httpx.HTTPError
subclass) and silently falls back to the on-disk cache. On a fresh box
that cache never existed, so HermesIndexSource.is_available is False,
the index contributes 0 skills, and the hub landing page — which is
built solely from an empty-query index search — is blank. Existing
installs only appear to work because they serve a (possibly weeks-)stale
cached index instead.
Fix: request 'gzip, deflate' on the index fetch so httpx never
negotiates the broken Brotli path, and retry once with 'identity' if a
DecodingError still occurs (defends against a proxy that ignores the
header). Falls through to the stale cache only when both attempts fail.
Verified on a live staging VPS agent: index_available flips False->True
and the featured landing list repopulates from 0 to 12.
Also un-freezes already-deployed images: skills added after an image was
built (e.g. the 'unbroker' optional skill) become reachable again via
the index, which is the whole point of the centralized catalog.
load_gateway_config() only surfaced the top-level `multiplex_profiles`
key into gw_data before calling GatewayConfig.from_dict(). A config.yaml
that pinned the flag under the nested `gateway:` section -- the form
written by `hermes config set gateway.multiplex_profiles true` -- was
silently ignored, so the gateway loaded with multiplex_profiles=False.
from_dict() already honors the nested fallback, but load_gateway_config()
builds gw_data from top-level keys first, so the nested value never
reached it.
Read gateway.multiplex_profiles into gw_data when the top-level key is
absent, mirroring the existing nested fallback for max_concurrent_sessions.
Adds a load_gateway_config() regression test that writes a config.yaml
with `gateway.multiplex_profiles: true` and asserts the loaded config has
multiplex_profiles=True (fails without the fix).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* feat(sessions): full filter surface for prune + new bulk archive subcommand
hermes sessions prune previously only supported --older-than N (integer
days) and --source — no way to target a window like 'the last 5 hours'
(e.g. a batch of CI smoke-test sessions), and no non-destructive option.
- SessionDB.prune_sessions gains keyword filters that AND together:
started_before/started_after epoch bounds, title_like, end_reason,
cwd_prefix, min/max_messages, archived tri-state. Default call is
byte-for-byte compatible (90-day cutoff, ended-only, source).
- New SessionDB.list_prune_candidates (backs --dry-run + confirmation
previews) and SessionDB.archive_sessions (bulk soft-hide via the
existing set_session_archived lineage-aware path; nothing deleted).
- CLI: prune gains --newer-than/--before/--after (durations like 5h/2d/1w,
bare days, or ISO timestamps), --title, --end-reason, --cwd,
--min/--max-messages, --include-archived, --dry-run. New
'hermes sessions archive' takes the same filters, requires at least one,
and is idempotent. Both show a preview before confirming.
- Dashboard /api/sessions/prune accepts the same filters + dry_run.
- Docs: sessions.md + cli-commands.md updated.
Filter parsing lives in hermes_cli/session_filters.py with unit tests;
DB filters covered in tests/test_hermes_state.py.
* feat(sessions): prune/archive filters for model, provider, user, chat, branch, tokens, cost, tool calls
Extends the prune/archive filter surface to everything identifiable in
the sessions table:
- --model (substring on model slug), --provider (exact on
billing_provider, case-insensitive), --user, --chat-id, --chat-type
(exact), --branch (substring on git_branch), --min/--max-tokens
(input+output), --min/--max-cost (USD, actual_cost_usd falling back to
estimated_cost_usd), --min/--max-tool-calls.
- SessionDB prune/archive/list_prune_candidates now share the filter
kwargs via **filters into _prune_filter_where (unknown names raise
TypeError); candidates listing + CLI preview now include the model.
- Any attribute filter (except legacy --source) suppresses the implicit
90-day default so 'prune --model X' matches all ages.
- Dashboard /api/sessions/prune passes the new fields through.
- Docs + tests updated (7 new DB tests, 3 new parser tests).
Fixes#50051 by preserving nested gateway.multiplex_profiles and routing gateway config env reads through the active profile secret scope when present.
This keeps secondary profile adapter startup from inheriting default-profile platform tokens or port-binding enables while preserving legacy single-profile behavior outside a scope.
Constraint: latest upstream main f57ff7aef1 still reproduced both nested-config loss and cross-profile env leakage
Rejected: special-casing API_SERVER_* only | left other profile-scoped tokens vulnerable to the same leak
Confidence: high
Scope-risk: moderate
Directive: keep future gateway/config env reads on the scoped helper path unless a variable is explicitly process-global
Tested: pytest -q tests/gateway/test_multiplex_phase0.py tests/gateway/test_multiplex_credential_isolation.py tests/gateway/test_config.py -k 'multiplex or scope or getenv or api_server or relay'
Not-tested: full gateway startup across live platform adapters
test_group_new_keeps_existing_reset_semantics_when_dm_topic_mode_enabled
asserts 'parallel work' not in the /new reply — but /new appends a
random tip from hermes_cli.tips (380 entries), and one tip's text
contains exactly that phrase (the delegate_task concurrency tip). CI
failed on PR #59331 slice 2 when the dice landed on it. Pin
get_random_tip in the test.
The routing sweep sends these paths through _adapter_for_source, which
reads source.profile. A bare MagicMock auto-attribute is truthy, so the
fixtures looked like stamped secondary profiles and hit the new
fail-closed branch. Real SessionSource.profile is None or str
(AGENTS.md pitfall #17).
register_mcp_servers now nudges cached entries whose session is None
via _signal_reconnect, so a new agent session recovers a parked server
immediately instead of waiting up to _PARKED_RETRY_INTERVAL for the
next self-probe (#50170). Gate-check idea credit: @izumi0uu (#50184),
@LeonSGP43 (#37772), @Tranquil-Flow (#37899).
The dead-session half-open test drives _signal_reconnect with
session=None; the salvaged _ReconnectAdapter assumed a live old
session. Also count set() calls explicitly instead of relying on
MagicMock introspection.
Four independent pre-request stalls sat on the critical path between
prompt submission and the first streamed token, measured with cProfile
against a live process:
1. Discord capability detection (~2.0s, worst 5s): get_tool_definitions
-> _get_dynamic_schema made a BLOCKING https call to discord.com
inside AIAgent.__init__ for any user with DISCORD_BOT_TOKEN set, on
every platform, every cold process. Now non-blocking: memory cache ->
24h disk cache -> permissive default + one background detection that
seeds the disk cache for the next process. The permissive default is
pinned per-process so tool schemas never flip mid-conversation
(prompt-cache safety); it mirrors the existing detection-failure
fallback (all actions exposed, 403s enriched at call time).
2. Ollama /api/show probe (~0.3s): get_model_context_length step 5e
POSTed to <base_url>/api/show for KNOWN providers (openrouter etc.),
got a 404, and never cached the miss - so every fresh process paid a
full HTTP round-trip. Known non-Ollama providers now skip the probe;
local/custom/unknown endpoints keep the exact previous behavior.
3. env_probe subprocess sweep (~0.5s): the Python-toolchain probe ran
4-8 subprocess calls inside the FIRST system prompt build. Now warmed
off-thread during agent init; the prompt build hits the cache (same
lock, so a mid-flight warm just joins instead of recomputing).
4. tools.mcp_tool import (~0.4s): the between-turns MCP refresh in
build_turn_context imported the whole mcp package even with zero MCP
servers configured. MCP tools can only exist if tools.mcp_tool was
already imported (discovery/reload paths), so gate the import on
sys.modules membership - no behavior change for MCP users.
CLI additionally pre-imports run_agent + openai off-thread during the
idle banner window (same pattern as the /model picker prewarm), hiding
the remaining ~1.5s of module imports while the user types. Fixes 1-4
apply to every interaction layer (CLI, gateway, TUI, desktop, cron).
Measured cold first turn (submit -> request dispatched, openrouter,
discord token set): 4.3s before -> 0.9s after CLI prewarm (~80%); the
agent-side non-import cost drops 2.9s -> 0.36s (init) + 0.27s (turn
prologue).
Some MCP servers (e.g. DocuSeal) serve their web UI on HEAD/GET but
speak Streamable HTTP only via POST. The preflight probe now tries a
lightweight JSON-RPC `initialize` POST before rejecting endpoints
whose HEAD/GET returns a non-MCP content type (e.g. `text/html`).
If the POST returns `application/json` or `text/event-stream` with a
2xx status, the endpoint is accepted. Otherwise the original rejection
behaviour is preserved.
Adds 5 new test cases covering the POST probe path:
- POST rescues HTML HEAD with JSON response
- POST rescues HTML HEAD with event-stream response
- POST still rejects when it also returns HTML
- POST still rejects on non-2xx status
- POST not attempted when HEAD already returns valid MCP content type
Follow-up to #9006/#58899. The gateway routing index (session_key ->
SessionEntry) now lives in a new gateway_routing table in state.db as the
primary store; sessions.json is demoted to an optional legacy mirror.
- hermes_state.py: schema v19 — gateway_routing table (scope + session_key
PK; scope = resolved sessions_dir so multiple stores sharing one state.db
never cross-contaminate) with save/replace/load/delete methods
- gateway/session.py: _save() writes the whole index atomically to the DB
(mirrors the old full-file JSON rewrite semantics) and only falls back to
JSON when the DB write fails; _ensure_loaded reads the DB first and folds
in legacy sessions.json entries for keys the DB lacks (pre-migration
import; DB entries win over stale JSON)
- gateway/config.py + hermes_cli/config.py: new write_sessions_json flag
(default true for compat/downgrade safety); gateway.write_sessions_json:
false stops producing the file entirely
- sessions.json _README updated to say it's a legacy mirror + how to
disable it
Rehydration is now lossless across restarts even with sessions.json deleted:
suspended/resume_pending/model_override/token state all round-trip through
the DB (the old sessions-table recovery only rebuilt the bare key mapping).
When the main agent uses a named custom provider (custom:<name>),
resolve_runtime_provider correctly resolves the base_url and api_key.
But the auxiliary client re-resolves from the bare 'custom' provider
name, losing the provider identity. The bare 'custom' falls back to
OpenRouter, which _resolve_custom_runtime() then rejects — leaving all
auxiliary tasks (title gen, compression, vision, session search, etc.)
with no credentials.
Fix: when resolve_provider_client receives a main_runtime dict
containing concrete base_url + api_key, use it directly instead of
re-resolving. The main agent already solved provider resolution;
the auxiliary client just needs to reuse its answer.
Closes#45472
Regression tests from PR #51586: the inspection agent must receive the
platform-resolved enabled_toolsets and agent.disabled_toolsets, and a
Blank Slate profile's prompt-size must count exactly the 6 file/terminal
tool schemas.
Two pre-existing tests awaited run() to return after initial-connect
retry exhaustion; with #57477's parking that await hangs (CI: 300s
SIGKILL on slices 4 and 6). Assert the new contract instead: the task
stays alive (parked) and exits on shutdown.
The local retries variable in MCPServerTask.run() accumulated across
transient disconnections — each transport exception incremented it, but
only clean transport returns (auth recovery / manual refresh) or
park-wake reset it. Five transient blips over a long-uptime gateway
would permanently park the MCP server.
Promote retries to instance attribute _reconnect_retries and reset it
at all 4 session-establishment sites in _run_stdio / _run_http, so only
consecutive failures without successful reconnection count toward the
parking budget.
Fixes#57604
Persists as the server's connect_timeout in config, which the probe
now honors. CLI-flag portion of PR #54494; the probe-wrapper portion
was superseded by resolving connect_timeout inside _probe_single_server.
_reauth_oauth_server (hermes mcp login / reauth) called
_probe_single_server without a timeout, so it always used the 30s
probe default — far too short for a human browser OAuth round-trip
(open → sign in → consent → loopback redirect). The server-level
connect_timeout in config.yaml was silently ignored, so login timed
out at ~40s no matter what the user configured.
Pass the server's configured connect_timeout through, with a 180s
floor for the interactive login path. Update the two TestMcpLogin
probe mocks for the new kwarg and assert the login path propagates a
>=180s timeout.
Follow-up to the salvaged #54938: the bounded reader gives a proper 413 +
anomaly telemetry for oversized chunked bodies; client_max_size makes
aiohttp enforce the same 1 MiB cap on every other read path
(#58536/#58902/#59180 pattern). Test fixture's fake Application now
accepts kwargs.
The salvaged #25296 fixture's _FakeRequest.read() calls json.dumps but the
test module never imported json — the NameError was swallowed by the
handler's generic except → 400, failing 10 payload tests.
_handle_webhook() called request.read() with no size guard. Since the
endpoint is publicly reachable, an attacker can send an arbitrarily large
POST body to exhaust gateway memory.
Add _TWILIO_WEBHOOK_MAX_BODY_BYTES (64 KiB — well above any real Twilio
payload) and gate on both Content-Length and actual read size, returning
HTTP 413 with an empty TwiML Response on oversized requests. Mirrors the
guard already present in the Raft adapter.
A `hermes update` that bumps the spectrum-ts pin rewrites the Photon
sidecar's package-lock.json but never reinstalls node_modules. The sidecar
then spawns against the old install and the v8 postinstall patch throws
"@spectrum-ts/imessage dist not found", so the gateway retries the photon
platform every 300s forever without ever repairing the deps. Observed in
the wild: a June pin bump to spectrum-ts 8.0.0 left node_modules at 3.1.0,
and inbound/outbound iMessage stayed dead for days with the reconnect loop
faithfully restarting into the identical broken state.
_start_sidecar only checked that node_modules exists, not that it matches
the lockfile, so restart never became repair. Detect the skew with the same
signal npm ci uses: the top-level package-lock.json being newer than npm's
node_modules/.package-lock.json install marker. When stale, reinstall
(npm ci, falling back to npm install) before spawning. The reinstall runs
via asyncio.to_thread so a cold install can't block the event loop and stall
every other platform's traffic; worst case it heals on the next reconnect
tick instead. First-run "deps not installed" behavior is unchanged, and a
missing/unreadable marker fails safe to "not stale" so start is never
blocked.
Reuses the existing npm ci -> npm install fallback from
`hermes photon install-sidecar`. Adds unit tests for the staleness signal
(stale / fresh / missing-marker).
Follow-up to the #55911 salvage: inherit model.api_key only when the aux
base_url resolves to the same hostname as the main model's base_url
(runtime override or config). A misconfigured aux endpoint on a different
host keeps the fail-safe no-key-required placeholder instead of leaking
the main credential cross-host.
When an auxiliary task is configured with provider=custom and an explicit
base_url but an empty api_key, the custom_key fallback chain in
resolve_provider_client() jumped straight to the no-key-required
placeholder without consulting model.api_key from config.yaml. Users
on self-hosted gateways who share the same endpoint and credentials for
both the main model and auxiliary tasks got 401 auth errors.
Add _read_main_api_key() following the same pattern as _read_main_model()
and _read_main_provider(): checks _RUNTIME_MAIN_API_KEY (runtime override)
first, then config.yaml model.api_key. Insert it into the fallback chain
before no-key-required so real credentials are used when available, while
local servers without auth still get the placeholder.
Overlap-invariant regression test from PR #58686 — no toolset in the
blank-slate disabled_toolsets may share a tool with a kept toolset,
since the subtraction happens at tool granularity (#57315, #58281).
The disabled_toolsets subtraction loop in _compute_tool_definitions
preserved shared core tools only for hermes-* platform bundles (#33924),
subtracting bundle_non_core_tools(); every other name took the else
branch and got a full resolve_toolset() subtraction. The `coding`
toolset is a posture toolset (posture: True) that re-lists the shared
_HERMES_CORE_TOOLS it does not own, so disabled_toolsets=["coding"]
stripped those core tools from the whole schema (34 tools collapsed to a
handful; terminal/read_file/write_file/web_search/execute_code gone).
Extend the core-preserving branch to also match posture toolsets, so
they subtract only the non-core delta. Only `coding` carries
posture: True, so atomic toolsets stay fully removable. The
bundle-misconfiguration info log is gated to hermes-* names, since its
wording is bundle-specific and disabled_toolsets=["coding"] is a
legitimate config written by older `hermes setup` runs.
Adds a regression test (TestDisabledToolsetsPostureToolset) alongside
the existing #33924 bundle tests.
Blank Slate's _blank_slate_minimal_toolsets() adds every TOOLSETS entry
to agent.disabled_toolsets except file and terminal. The coding
posture toolset (session-level, selected by agent/coding_context.py)
slips through because the loop only skips hermes-* composites and
includes-only groups.
At runtime, model_tools.get_tool_definitions() resolves coding and
subtracts its tools — terminal, read_file, write_file, patch,
search_files, process — erasing the entire Blank Slate minimal surface.
The agent ends up with only cronjob.
Skip posture toolsets in the disabled-list computation. Posture
toolsets are not user-facing capabilities to disable; they are
per-session selections that should never appear in agent.disabled_toolsets.
Fixes#57315
Assert the "Test server" probe skips prompts/list when tools.prompts is false,
skips both families when the server advertises neither capability (the Unreal
MCP server case), probes both when advertised and enabled, and falls back to
the legacy always-try behaviour when no capability info was captured.
* fix(docker): heal pairing-dir ownership after `docker exec` writes (#10270)
The official Docker image runs the gateway as the unprivileged `hermes`
user (uid 10000) via `gosu`, but `docker exec` defaults to root. Approval
files written by `docker exec <container> hermes pairing approve <code>`
end up as `-rw------- root:root`, and the post-gosu gateway process
cannot read them. The approval is silently ignored — the user keeps
hitting 'Unauthorized user' on every message.
The entrypoint's existing top-level chown is gated on the top-level
$HERMES_HOME being mis-owned, so on warm boots (where /opt/data is
already hermes:hermes) the recursive chown is skipped — meaning a
container restart does NOT self-heal the bug either.
Three-part fix:
1. docker/entrypoint.sh: chown the platforms/pairing/ (and legacy
pairing/) subtree on every container start, regardless of the
top-level decision. The directory is tiny (a few JSON files), so
the unconditional chown is effectively free. Container restart
now self-heals.
2. gateway/pairing.py: PairingStore._load_json was swallowing
PermissionError under its bare 'except OSError' branch, which is
what made this a silent failure. Split it out: log a WARNING that
names the file, the gateway's uid, the file's owner/mode, and the
exact docker exec -u hermes workaround. Still falls back to {} so
the gateway stays up.
3. website/docs/user-guide/security.md: add a Docker tip to the
pairing-CLI section pointing users at `docker exec -u hermes …`
up front.
Reproduced end-to-end in a containerized harness — before the fix
the gateway sees 0 approved users after `docker exec` + restart;
after the fix it sees the expected 1, and the file on disk goes
from `root:root 600` back to `hermes:hermes 600` on next start.
Fixes#10270
* fix(pairing): gate os.geteuid for Windows in PermissionError warning
Adds approvals.deny to config.yaml — a list of fnmatch globs matched
against terminal commands. A match blocks unconditionally, BEFORE the
--yolo / /yolo / approvals.mode=off bypass, making it the user-editable
counterpart to the code-shipped hardline blocklist.
- Checked in both command gates (check_dangerous_command and
check_all_command_guards), after the hardline floor and sudo-stdin
guard, before the yolo bypass and permanent allowlist.
- Matching runs over the same normalized/deobfuscated command variants
as the dangerous-pattern detector, case-insensitive.
- Opt-in: empty/absent list is a no-op; behavior unchanged.
Supersedes the trust-engine approach from #21500 with a minimal
config-native design: the only capability the existing stack lacked
was deny-that-beats-yolo. Allow already exists (command_allowlist),
ask already exists (session approvals).
Sibling sweep from the #58902 raft review found aiohttp servers still
running on the implicit 1 MiB default with no explicit body cap:
- bluebubbles webhook (127.0.0.1): 1 MiB explicit cap — events are small
JSON/form payloads; attachments arrive via the REST API
- teams Bot Framework listener (0.0.0.0 bind — most exposed): 1 MiB cap;
activities are JSON well under that
- hermes proxy server: 10 MB cap mirroring api_server's MAX_REQUEST_BYTES
(chat-completion payloads can be large, but must stay bounded)
client_max_size bounds every read path including chunked transfer-encoding
requests that carry no Content-Length (#58536/#58902 pattern).
Deliberately excluded: feishu, whatsapp_cloud, sms, line, wecom, msgraph —
open contributor PRs (#54938, #54944, #54620, #54931, #54934, #25296)
already cover those; reviewing them separately preserves their credit.
3 regression tests pin the wiring.
PR #58889 fixed the CLI-fallback transport; review of that fix found the
same leak class at four sibling spawn sites of the third-party cua-driver
binary:
- _resolve_mcp_invocation (cua-driver manifest): no env= at all — full
parent environment inherited
- cua_driver_update_check (check-update --json): telemetry env but no
secret sanitization
- doctor._drive_health_report (<binary> mcp Popen): telemetry env only
- permissions._run (every macOS/Linux permission probe): telemetry env only
All now route through _sanitize_subprocess_env(cua_driver_child_env()),
matching the sanctioned MCP spawn and the #53503/#55709/#58889 strip-by-
default policy for non-terminal spawns. Sanitization degrades gracefully
(falls back to the telemetry env) so doctor/permission probes never break
on an import error.
4 regression tests covering each site.
Skill bundles load their member skills via _load_skill_payload directly,
bypassing the scan-time disabled filter in get_skill_commands(). PR #58888
closed this gap for stacked slash-skill invocations, but /<bundle> dispatch
in the gateway had the same class of bypass: a skill an operator disabled
for a platform via skills.platform_disabled still got its full content
injected when referenced by a bundle.
build_bundle_invocation_message() now accepts a platform kwarg, filters
members against get_disabled_skill_names(platform=...), and reports skipped
skills in the bundle header. Gateway dispatch passes the event's platform
explicitly (env-var resolution can't be trusted in the multi-platform
gateway process, same reasoning as the #58888 gate).
When a single line exceeds the entire char budget, its tail is
unreachable via offset pagination (offsets are line-granular). Tell
the model so it doesn't assume it saw the full line.
read_file previously hard-rejected any read whose formatted output exceeded
the ~100K char safety limit, returning an error with zero content. A file
with few but very long lines (logs, wide CSV rows, minified data) sails past
the line-count limit and then trips the char guard, so the model gets nothing
and must guess a smaller limit — wasting a full round-trip.
Now the read is trimmed to the last complete line that fits the budget and
returns the partial content plus truncated_by="bytes" and a next_offset, so
the model paginates forward instead of starting over. A single line larger
than the whole budget is clamped on a code-point boundary (never empty) and
the cursor still advances. Applies at both read paths (normal + extracted
documents).
Adapted from IronClaw's Rust dual line/byte cap to hermes's Python tool-layer
char guard, which is the single uniform chokepoint over the gutter-rendered
content for every backend.
Follow-up to the salvaged toggle commit:
- file_tools.py / code_execution_tool.py: carry docker_network in their
container_config dicts so those environment-creation paths honor the
lockdown instead of silently defaulting back to bridge (the probe/exec
asymmetry class reported on #46358).
- docker.py: cross-process reuse now inspects HostConfig.NetworkMode when
docker_network=false and removes a mismatched (networked) container
before starting a fresh air-gapped one. Fails closed when inspect fails.
Default-network config never churns containers, so operators using
docker_extra_args --network=none are unaffected.
- tests: AST invariant that every container_config site carrying
docker_run_as_host_user also carries docker_network, plus three reuse
guard tests (reject bridge under lockdown / keep matching none /
no inspect when network enabled).
- docs: configuration.md gains terminal.docker_network + env var row.
Port from qwibitai/nanoclaw#2713: expose Hermes' existing Docker network isolation primitive through terminal config so operators can opt out of container egress.
_resolve_task_provider_model returns early on an explicit provider arg,
which skips the config block that consults auxiliary.<task>.base_url /
api_key. Any caller passing provider explicitly (e.g.
resolve_vision_provider_client(provider="custom", ...)) bypasses the
configured custom endpoint and falls through to main-runtime resolution,
silently routing the task to the wrong backend.
Adopt the task's configured base_url/api_key before the early returns,
but only when no explicit base_url was given and the config targets the
same provider (or names none) — a caller forcing a *different* provider
keeps full explicit-arg priority, and an explicit base_url still wins
over config.
Fixes#58515
The load_config() cache is keyed on config file mtime/size only, so a
load_config() that runs before load_hermes_dotenv() populates the process
environment caches the unexpanded ${VAR} literal and serves it for the
life of the process — auxiliary.<task>.api_key/base_url env refs reach the
provider client verbatim (auth failure / silent fallback), while
providers.* appear to work because provider credential resolution re-reads
the environment at call time.
Record a snapshot of every ${VAR} name referenced in the raw config
(user + managed) with its os.environ value at expansion time, and treat
the cache as stale when any of those values change. Covers both the late
.env load and in-process key rotation; an unchanged environment still
takes the cache-hit path.
Fixes#58514
Moves gateway routing metadata (display_name, origin_json, expiry_finalized)
into state.db, making SQLite the single source of truth for gateway session
discovery. Eliminates the dual-file (sessions.json + state.db) polling
dependency that caused the mcp_serve new-conversation race (#8925).
- hermes_state.py: schema v18 (3 new sessions columns + sessions.json
backfill migration), record_gateway_session_peer gains
display_name/origin_json, new set_expiry_finalized(),
list_gateway_sessions(), find_session_by_origin()
- gateway/session.py: peer recorder persists display_name + full origin
JSON; new SessionStore.set_expiry_finalized() single write-path
- gateway/run.py: expiry watcher success + give-up paths use the store
helper so the flag lands in both sessions.json and state.db
- mcp_serve.py: routing index reads state.db first (sessions.json fallback
for pre-migration DBs); _poll_once collapses to a single state.db mtime
check — the #8925 race is structurally impossible now
- gateway/mirror.py, gateway/channel_directory.py, hermes_cli/status.py:
query state.db first, sessions.json fallback
Closes#9006
Port from openclaw/openclaw#95108: an unbounded response.read() on a
non-OK *streaming* response can balloon memory (huge body) or hang the
agent forever (body opens then stalls with no further bytes). The
diagnostic body is only ever shown truncated, so reading megabytes or
blocking indefinitely buys nothing.
Add agent/bounded_response.read_streaming_error_body() which caps the
read at a byte limit and enforces a hard wall-clock deadline (run on a
worker thread so it can interrupt a socket read that stalls mid-chunk,
which a between-chunk wall-clock check cannot). Wire it into all three
streaming error-body sites that previously did a bare response.read():
native Gemini, Gemini Cloud Code, and Antigravity Cloud Code. The
existing error builders now accept an optional pre-read body_text so
classification (status code, RESOURCE_EXHAUSTED, free-tier guidance,
Retry-After) is preserved unchanged.
Tests use a real in-process socket server (no mocks): oversize body is
capped, stalled body hits the deadline with partial text preserved,
normal error envelope reads intact and parses.
_redact_env already skips redaction when a KEY=value assignment's value is
a programmatic env lookup (os.getenv(...), os.environ[...], process.env.X)
per issue #2852 — masking it would corrupt a code snippet, not redact a
secret. _redact_json (JSON "key": "value" syntax) and _redact_yaml
(unquoted key: value syntax) are separate closures in the same function
and never got the same check, so the identical code-snippet-in-config-
syntax case still gets mangled:
{"apiKey": "os.getenv('OPENAI_API_KEY')"} -> {"apiKey": "os.get...EY')"}
api_key: os.getenv("OPENAI_API_KEY") -> api_key: os.get...EY")
Fix: apply the same _ENV_LOOKUP_VALUE_RE.match(value) check in both
closures before masking, mirroring _redact_env exactly. Real secret
values in JSON/YAML syntax are still redacted (verified live and via new
tests) — this only skips the case where the "value" already look like a
code snippet.
a0a3c716f fixed the exact same failure mode for Telegram (#58563):
post-#48648, oversized mid-stream edits truncate to a one-message preview
instead of splitting. Once a long streamed reply grows past that cap, every
subsequent progressive edit truncates to the SAME preview text — re-sending
an identical edit every tick still counts against the platform's edit rate
limit for the rest of the stream.
Discord's edit_message() has the identical architecture (mid-stream
truncate-in-place, both pre-flight and reactive-after-50035 truncation
paths) and this file's own docstring already calls out "the Telegram #48648
lesson" it's built on — but the saturated-preview dedup fix itself was never
ported over.
Fix: track the last truncated preview per (chat_id, message_id), mirroring
a0a3c716f exactly. Skip the edit call when the new truncation is identical;
still deliver when the visible content actually changes (e.g. the
chunk-count marker crosses (1/2) -> (1/3) as the stream grows). State
clears on finalize and when content shrinks back under the cap, so dedup
can never mask a real edit.
11b4a21a5 cleared the per-session _last_resolved_model cache on /new and
the compression-exhausted auto-reset, so a resumed/reset conversation
resolves the model from current config instead of a stale cached value
(#58403). Three other sites documented as the same "full conversation
boundary" treatment — pop _session_model_overrides, clear the reasoning
override, pop _pending_model_notes — still missed _last_resolved_model:
- _session_expiry_watcher's permanent finalization block (gateway/run.py):
a session that goes idle and is finalized, then resumed, could serve a
model cached before it went idle on a transient config-cache miss.
- The daily/idle/suspended auto-reset cleanup (_was_auto_reset handling,
gateway/run.py): same failure mode, different trigger.
- /resume (gateway/slash_commands.py), whose own comment already says
"conversation boundary just like /new" for the sibling dicts it clears.
Fix: pop the session's _last_resolved_model entry in all three, mirroring
the exact pattern 11b4a21a5 established.
_handle_wake() and _handle_activity() enforced max_body_bytes only via
the Content-Length header. A Transfer-Encoding: chunked request
(content_length=None) or a spoofed small Content-Length bypassed the
cap entirely, letting the actual read be bounded only by aiohttp's
implicit 1 MiB client_max_size default (64x the 16 KB default) — the
same pattern ec29590a0 just fixed for gateway/platforms/webhook.py.
Fix: web.Application(client_max_size=self._max_body_bytes) so aiohttp
enforces the cap on every read path including chunked bodies, catch
HTTPRequestEntityTooLarge -> 413 on both endpoints (was swallowed into
a generic 400), and re-check the actual bytes read as defense in depth.
Exposure here is narrower than the webhook adapter (binds to 127.0.0.1
by default and requires the bridge token), but the bypass is otherwise
identical.
_redact_telegram_error_text() strips bot tokens from api.telegram.org
URLs embedded in transport-error text, and is already applied across the
send/edit transient-error paths. Four sites still built their message
from the raw exception:
- connect()'s fatal-error handler is the most severe: the raw text is
passed to _set_fatal_error(), which persists it via
write_runtime_status() to a dashboard/admin-facing runtime status
file, not just a log line. A transient network error during startup
commonly embeds the request URL
(https://api.telegram.org/bot<TOKEN>/getMe), so this could leak the
live bot token into that surface.
- disconnect(), send_document(), send_video() build the same unredacted
pattern into a warning log line (lower blast radius, but the same
leak class).
Fix: route all four through the existing _redact_telegram_error_text()
helper before building the message/log line, mirroring the send/edit
paths exactly. Also drops exc_info=True from the two logger.error/
logger.warning calls that had it — exc_info prints the exception's own
traceback (including its unredacted message) separately from the format
string, which would otherwise defeat the redaction; the already-redacted
sibling call sites in this file follow the same convention.
_CuaDriverSession._call_tool_via_cli() (the EAGAIN/silent-empty MCP
fallback transport) invokes `cua-driver call <tool> <json>` via
subprocess.run() with no env= argument, so the third-party cua-driver
binary inherits the full, unsanitized parent environment. The primary
MCP spawn site (_lifecycle_coro) already applies
_sanitize_subprocess_env(cua_driver_child_env()) before opening the
stdio client, per the same policy #53503/#55709 established for other
subprocess spawn points — this fallback path, added alongside the
EAGAIN/silent-empty-capture hardening, missed it.
Fix: apply the same env=_sanitize_subprocess_env(cua_driver_child_env())
to the subprocess.run() call in _call_tool_via_cli(), mirroring the
sanctioned spawn site exactly (telemetry policy applied first, then
Hermes-managed secrets filtered).
_handle_message() re-checks a slash-skill command's per-platform disabled
status before dispatch, because get_skill_commands() only applies the
global disabled list at scan time. That check only covered the leading
skill: split_stacked_skill_commands() resolves additional /skill tokens
that follow it (stacked invocations, up to 5 skills, #57987), and
build_stacked_skill_invocation_message() loads every one of them via
_load_skill_payload() with no disabled-status check of any kind.
A message on a platform with skills.platform_disabled configured for a
given skill could still get that skill's full SKILL.md content injected
into the agent's context for the turn, as long as it was typed after an
allowed skill: `/allowed-skill /disabled-skill do X`.
Fix: after computing the stacked extra_keys, look up each one's skill
name and re-check it against the same get_disabled_skill_names(platform=)
set already used for the leading skill. If any stacked skill is disabled
for the platform, reject the whole invocation with the same style of
message the leading-skill check already returns, instead of partially
loading it.
Port from openai/codex#21069 ("Spill large hook outputs from context").
Both shell hooks and Python plugins can return {"context": "..."} from
pre_llm_call, which gets appended to the current turn's user message on
every subsequent API call. A plugin that emits a large blob inflates
every turn and blows out the prompt cache prefix.
- tools/hook_output_spill.py: shared helper that writes oversized
context to $HERMES_HOME/hook_outputs/<session_id>/<uuid>.txt and
returns a head/tail preview plus the saved path. Never raises.
- agent/turn_context.py: apply the cap at the pre_llm_call aggregation
site (moved here from run_agent.py since the original PR), covering
both Python plugins and shell hooks.
- agent/shell_hooks.py: reserve output_spill as a sub-key under hooks:
so the config block doesn't emit unknown-hook-event warnings.
- Docs: document the cap + config in build-a-hermes-plugin.md.
Config (behaviour-preserving when absent):
hooks.output_spill: enabled/max_chars/preview_head/preview_tail/directory
Tests: 14 unit tests; shell_hooks (56) and plugins (100) suites green.
E2E validated with isolated HERMES_HOME (spill, passthrough, traversal
sanitisation, reserved-key skip).
Feishu adapter's disconnect() cancelled WSS-thread tasks but never
called the lark_oapi client's _disconnect() coroutine, so no
WebSocket CLOSE frame was sent. Feishu's server kept routing
messages to the stale endpoint for minutes (CLOSE-WAIT timeout),
silencing the channel across every shutdown path — systemd restart,
hermes update, hermes gateway restart, and the --replace takeover
during 'hermes dashboard' invocations.
Schedule ws_client._disconnect() on the WSS thread loop via
run_coroutine_threadsafe with a 5s timeout before the existing
task-cancel + loop-stop sequence. Defensive hasattr guard + broad
except keeps disconnect() resilient if lark_oapi's internals shift.
Fixes#10202
Port from Kilo-Org/kilocode#11555: GLM-5.2 exposes a native
reasoning_effort knob with two enabled levels (high / max) on its
OpenAI-compatible endpoints. Previously the zai profile (direct Z.AI
/api/paas/v4) used the base ProviderProfile and emitted nothing, and the
OpenCode Go profile only handled Kimi K2 / DeepSeek — so a user's effort
preference for GLM-5.2 was silently dropped on both routes.
- zai: ZaiProfile maps effort onto high/max (xhigh/max -> max, lower -> high)
- opencode-go: same mapping for GLM-5.2, alongside existing Kimi/DeepSeek
- alias spellings recognized (glm-5.2 / glm-5-2 / glm-5p2, vendor-prefixed)
- disabled / no effort leaves the server default untouched
The google-antigravity and google-gemini-cli OAuth providers were removed
in #50492. They were the only producers of a cloudcode-pa:// base_url, so
the account-level-quota early-returns in _pool_may_recover_from_rate_limit
and _credential_pool_may_recover_rate_limit are now unreachable.
- Drop the dead cloudcode-pa:// checks and the now-unused provider/base_url
params on _pool_may_recover_from_rate_limit (only caller updated).
- Prune the obsolete CloudCode-specific regression tests; keep the live
single/multi-entry pool-rotation invariants (#11314).
ruff check --fix --select F541 . on current main. Pure prefix removals;
adjacent-string concatenations keep the f only on interpolating fragments.
No string content or live placeholder altered.
TestMcpParallelToolBatch seeded provenance under old-style
mcp_<server>_<tool> names, which no longer pass the
is_mcp_tool_parallel_safe() prefix gate after the naming change.
Port from anomalyco/opencode#33533. Native MCP tools now register as
mcp__<server>__<tool> (double-underscore delimiter) instead of
mcp_<server>_<tool>, aligning with the convention used by Claude Code,
Codex, and OpenCode.
The double-underscore delimiter disambiguates the server/tool boundary
even when either component contains underscores (the single-underscore
form was ambiguous, which is why is_mcp_tool_parallel_safe already had to
track provenance in a side-map). It also unifies native registration with
the Anthropic-OAuth wire form (_MCP_TOOL_PREFIX = 'mcp__'), so the
single->double promotion that path performed is now a no-op for native
tools while still handling legacy replayed names.
- tools/mcp_tool.py: add MCP_TOOL_NAME_PREFIX + mcp_prefixed_tool_name()
helper; route _convert_mcp_schema, utility schemas, refresh stale-set,
and the parallel-safe prefix gate through it
- agent/transports/codex_event_projector.py: mirror convention in the
deterministic call_id input for MCP server-executed tool calls
- tests: update produced-name assertions to the new convention
Defense-in-depth alongside the interpreter-shutdown guard: run_job closed
the cron agent's async resources (agent.close + cleanup_stale_async_clients)
in its finally block BEFORE run_one_job called _deliver_result, so a live
delivery could race a torn-down async client. run_job now accepts an optional
defer_agent_teardown holder; when set it hands the live agent back instead of
closing it, and run_one_job tears it down (via the extracted _teardown_cron_agent
helper) only AFTER delivery — in a finally so a failed run never leaks. Default
path (holder=None) is unchanged, so every existing caller keeps inline teardown.
Reorder approach based on #58777 by @LavyaTandel; reworked to keep a single
delivery site in run_one_job and add regression coverage.
Co-authored-by: LavyaTandel <lavya@loom.local>
Pins `_interpreter_shutting_down()` (finalizing flag + shutdown-error-text
fallback) and asserts the standalone delivery path skips gracefully without
scheduling a send when the interpreter is finalizing, while the normal
non-finalizing path still delivers. Source guardrails keep the guard wired
into both the dispatch (`_submit_with_guard`) and standalone-delivery sites.
Assert _await_thread_exit lets a coroutine scheduled onto the running loop by a
blocked worker thread complete (the #58818 deadlock a synchronous join caused),
returns False when the thread outlives the timeout, and handles None/dead
threads.
Phase-2 review follow-ups on the unreadable-config chokepoint work:
- hermes_cli/xai_retirement.py apply_migration() is a full-file config.yaml
rewriter (ruamel round-trip + plain open("w")) that lives outside the
atomic_yaml_write path, so the chokepoint didn't cover it. It reads the
file first (which already fails closed on an unreadable file), but add
require_readable_config_before_write() right before the write as a
backstop for the read-then-write window, and a regression test asserting
the original bytes survive an unreadable config.
- Drop the unnecessary "Path" string quotes on atomic_config_write's
annotation — Path is imported eagerly at module top, no forward ref needed.
auth.py _update_config_for_provider / _reset_config_provider intentionally
keep their standalone require_readable_config_before_write guard + bare
atomic_yaml_write: the guard must fire BEFORE the read (fail-fast) at those
read-then-write sites, and a test pins the atomic_yaml_write call. Both are
already fully guarded against the bug; routing them through the wrapper
would move the check to write time for no benefit.
The unreadable-config-overwrite bug (an existing config.yaml that reads as
{} on a permission/IO error gets replaced with only defaults or the edited
section) is not limited to save_config / config set / auth. The same
read-then-atomic_yaml_write pattern lives at ~7 other independent write
sites that don't route through those functions:
- gateway/slash_commands.py: _save_config_key, memory/skills write_approval
toggles, tool_progress toggle, runtime_footer toggle, personality set
- hermes_cli/doctor.py --fix (stale root-key migration)
- gateway/platforms/yuanbao.py auto-sethome
- plugins/platforms/telegram/adapter.py topic thread_id persistence
- tui_gateway/server.py _save_cfg
- agent/onboarding.py mark_seen
Rather than sprinkle require_readable_config_before_write() at each site,
add a single fail-closed chokepoint, atomic_config_write(), that runs the
guard then delegates to atomic_yaml_write, and route every config.yaml
write through it. Root cause remains that read_raw_config() can't tell an
absent file from an unreadable one (returns {} for both) — read-only
callers correctly stay fail-open, but any full-file replacement now fails
closed in one enforced place instead of relying on each caller to remember
the guard.
save_config / set_config_value / auth keep the contributor's original
guard calls (their commit); this commit widens the fix to the sibling
call paths and adds a regression test on the chokepoint (fails closed on
unreadable existing file + still creates a genuinely absent file).
httpx ignores the client-level `limits` kwarg when a custom `transport`
is supplied. The #31599 keepalive fix injected limits via
`httpx_kwargs[limits]`, but the fallback-IP branch also passes a
custom `TelegramFallbackTransport` — so the limits were silently
discarded and the inner AsyncHTTPTransport instances ran with httpx
defaults (keepalive_expiry=5.0), leaking CLOSE_WAIT fds.
Pass the tuned limits directly into `TelegramFallbackTransport`
via `transport_kwargs` so its inner transports honour keepalive_expiry.
Only affects the fallback-IP branch; proxy and direct-DNS branches
continue to use `_with_limits()` as before.
Fixes#58790
The two TestSourceGuardrail tests asserted the presence of literal
strings ("#58753", "_user_survives") in context_compressor.py. Those
are change-detector tests that break on any refactor without catching a
real regression. The four behavioral tests in
TestCompressAlwaysKeepsAUserTurn already exercise the real compress()
path and fully cover the invariant (user turn survives, summary pinned
to user, no consecutive user roles, surviving tail user untouched).
Regression coverage for the kanban-worker crash where compression left a
transcript with no user-role messages, triggering a non-retryable
`400 No user query found in messages` from vLLM/Qwen.
Exercises the real `compress()` path with the reporter's shape (no system
prompt in the list, a re-compaction with the only user turn in the
compressed middle) and asserts the output always keeps >=1 user turn,
never introduces consecutive user roles, and leaves a surviving tail user
message untouched. A source guardrail pins the guard so a future refactor
cannot silently drop it.
Add an opt-in toggle (require_admin_for_exec_approval, default false) that
restricts who can click Approve/Deny on a dangerous-command prompt to admins
listed in allow_admin_from. Off by default, so the v0.16-restored user-scope
behavior is unchanged. When on, the clicker must pass the normal admission
check AND be an admin; fails closed (logged) when no admins are configured.
Only ExecApprovalView is gated — model picker / clarify / update-prompt stay
user-scope.
coerce_tool_args only repaired the outermost value, so JSON-encoded
*elements* of array properties (and nested object sub-fields) were left
as strings. Three core tools have array<object> schemas — todo.todos,
delegate_task.tasks, memory.operations — so a model emitting
{"todos": ["{...}"]} would pass raw JSON strings into the tool and fail
downstream on item["id"]/item["goal"] access.
Adds a schema-guided recursive pass (_normalize_json_strings_for_schema)
that parses JSON-string array items and nested object fields only when
the matching schema position expects an array/object, preserving
legitimate JSON-looking string fields (type: string).
Adapted from cline/cline#11803 to hermes-agent's existing coercion layer.
Follow-up to the salvaged #58696 (devatnull) + #41343 (annguyenNous)
commits: instead of fully suppressing commentary/analysis-phase stream
deltas, fire on_reasoning_delta so the CLI/gateway display them like
thinking text. Matches Codex CLI semantics where commentary is never
the turn's final answer, while keeping the narration visible in the
reasoning display. Adds devatnull to AUTHOR_MAP.
- bridge: only enqueue poll_update events for polls Hermes itself created
(tracked via recentlySentIds when /send-poll returns) so arbitrary human
polls in group chats don't inject agent-visible messages on every vote
- update test_already_whatsapp_italic for the new markdown-italic mapping
- AUTHOR_MAP entry for @devatnull (PR #58704 salvage)
The SOM/AX element list dropped labels for two extremely common cua-driver
render forms, leaving the model unable to target elements by name:
- [79] AXButton (Dark) -> parenthesised label
- [4] AXStaticText = "Wi-Fi" -> = "value" form
- [92] AXPopUpButton = "Automatic" -> = "value" form
The old regex only matched quoted "label" and id=Label, so System Settings
buttons/text/popups all surfaced with empty labels. That's why selecting the
macOS Appearance 'Dark' button by element index required guessing — the
labels weren't available to aim with.
Fix: extend _ELEMENT_LINE_RE to capture all four label forms (= "value",
"quoted", (parenthesised), id=Label), skipping a pure-digit (N) order number
in favour of the id= label. Verified live against System Settings: the
Appearance buttons now surface as Auto/Light/Dark.
Adds a regression test covering all label forms. Full suite: 84 passed.
The first fix handled the EAGAIN McpError path. But the persistent MCP
session (long-running gateway/desktop worker) has a second failure mode:
list_windows or get_window_state 'succeed' over MCP yet return a
degenerate/empty payload (no windows, or no screenshot + blank tree)
WITHOUT raising — typically when the bridge reconnected mid-call and
dropped the heavy response. That surfaced to the model as a silent 0x0
capture with no error and no fallback firing (0.00s empty return).
Fix: detect empty results in capture() and re-fetch over the CLI
transport before giving up:
- empty list_windows -> CLI re-fetch the window list
- empty get_window_state (som/ax) -> CLI re-fetch the AX tree + screenshot
- empty screenshot (vision) -> CLI re-fetch get_window_state for the PNG
Adds 2 regression tests. Full suite: 83 passed.
The cua-driver MCP stdio bridge intermittently (and on some machines
persistently) fails to forward heavier calls like get_window_state to
the daemon with POSIX EAGAIN — 'daemon transport error forwarding
get_window_state: Resource temporarily unavailable (os error 35)'.
The wrapper surfaced this as an empty 0x0 capture, so computer_use
returned blank screenshots even though the display, permissions, and
the daemon were all healthy (the direct 'cua-driver call' CLI path
worked fine throughout).
Fix: when the MCP path raises the transient/transport error, fall back
to the 'cua-driver call' subprocess transport, which talks to the
daemon over a different socket. The CLI fallback routes get_window_state
screenshots to a temp file via screenshot_out_file (tiny JSON response
instead of a multi-MB base64 blob that congests the socket), reads the
PNG back, retries with backoff, and remaps the JSON into the same
{data, images, structuredContent, isError} shape the MCP path produces
so capture()/_action() are transport-agnostic.
Adds _is_transient_daemon_error() classifier and 3 regression tests.
Verified live: captures that returned 0x0 now return full
1567x905 screenshots with the AX element tree.
The 'never reached ready' error (issue #57025) was undiagnosable — doctor
and MCP test pass while the wrapper times out, with no hint where startup
stalled. Track a phase marker through _lifecycle_coro (binary-check →
manifest-discovery → mcp-initialize → capability-discovery → ready) and
include it in the timeout RuntimeError plus a pointer to doctor and the
agent.log phase timings.
Complements the 15s→30s bump + success-path phase timing log from #58760.
Replaces the POSIX `/bin/bash -c "$(curl …)"` invocation with a
download-then-exec flow: curl the upstream install.sh into a mkstemp
temp file (unpredictable name, 0600) and run it as a plain argv list.
No shell=True, no command substitution. The temp script is removed in
a finally block; download failures return cleanly without exec.
Salvages the intent of #34974 by @ErnestHysa. His original patch
targeted a fixed /tmp/cua-driver-install.sh path (symlink/TOCTOU-prone
on multi-user hosts) and predates Windows/Linux installer support;
this version uses mkstemp and keeps the powershell path untouched.
Co-authored-by: ErnestHysa <takis312@hotmail.com>
Follow-up to #57987: after /skill-a the completer previously went silent
for a second /skill token. Now, while the leading tokens form an unbroken
skill chain (each token a distinct installed skill, under the 5-cap) and
the word under the cursor starts with '/', the completer keeps offering
the remaining skill commands, and SlashCommandAutoSuggest ghost-suggests
the rest of the next skill name. Instruction text, path-like tokens, and
broken chains get no suggestions. The TUI's complete.slash RPC reuses
SlashCommandCompleter, so it inherits the behavior with no changes.
* fix(cli): unwedge cua-driver installer timeouts — group-kill on timeout, stale-lock pre-clear, 660s ceiling
The cua-driver refresh in hermes update could wedge permanently:
subprocess timeout (300s) killed only the outer shell, orphaning the
curl|bash grandchildren and the upstream installer's concurrent-install
lock (~/.cua-driver/packages/.install.lock.d). The installer only
reclaims a stale lock after 600s of waiting — longer than our old
ceiling — so every subsequent run was killed before recovery could
fire: 'always times out'.
- Run the installer in its own process group (start_new_session) and
SIGKILL the whole group on timeout, so no lock-holding orphans survive.
- Pre-clear a provably-stale lock (dead holder pid, or pid-less and
older than the upstream 600s window) before invoking the installer.
- Raise the ceiling to 660s (> upstream LOCK_STALE_AFTER_SECONDS=600).
- Timeout message now names the lock path and the manual re-run command.
Fixes#58762
* chore: suppress windows-footgun lint on platform-gated kill calls
Both sites are POSIX-only: _clear_stale_cua_install_lock early-returns
on win32, and os.killpg sits in the 'not is_windows' branch.
The custom ``httpx.HTTPTransport(socket_options=[SO_KEEPALIVE, ...])``
in ``_build_keepalive_http_client()`` was introduced to fix CLOSE-WAIT
socket accumulation on long-lived connections (#10324).
That approach broke streaming for providers behind reverse proxies
(OpenResty, Cloudflare, etc.) because the custom socket options
conflict with the proxy's chunked-transfer handling (#54049, #12952).
It also stripped TCP_NODELAY, stalling TLS handshakes and SSE encoding.
Narrow per-provider bypasses were added for Copilot (#50298), Codex
(#36623, #12953), but the root cause remained.
The fix moves connection lifecycle management from the socket layer to
the HTTP pool layer:
- ``httpx.Limits(keepalive_expiry=20.0)`` tells httpx to close idle
pooled connections at 20 s, before a reverse proxy's typical 30-60 s
timeout drops them and causes CLOSE-WAIT accumulation.
- The default httpx transport preserves OS TCP defaults (including
TCP_NODELAY), so TLS handshakes and SSE chunked encoding work
correctly.
- ``trust_env=False`` prevents httpx from double-dipping on env vars
(we handle proxy detection ourselves via ``_get_proxy_for_base_url``
which respects NO_PROXY).
- The Copilot host bypass (line 3632) is no longer needed since all
providers now use the same standard httpx.Client.
Closes#54049. Supersedes #12010, #36623, #12953, #50298.
Follow-up to #56236: the broadened root token /[/.]*\** treats any run of
dots after the root slash as a collapse spelling, so a literal root-level
directory named '...' (rm -rf /...) was unconditionally hardline-blocked
with no approval path. Tighten the token to /(?:(?:\.\.?)?/)*(?:\.\.?)?\**
so each inter-slash segment must be exactly '.' or '..' — all real collapse
spellings (//, /., /./, /.., //*, ///, /../..) stay on the hardline floor
while literal dot-run dirs fall through to the softer DANGEROUS_PATTERNS
rules like every other real path.
* feat(approvals): /deny <reason> relays denial reason to the agent
Port from qwibitai/nanoclaw#2832 (reject with reason).
Gateway /deny now accepts an optional trailing reason (/deny <reason>
or /deny all <reason>). The reason rides on the per-session approval
entry through resolve_gateway_approval -> _await_gateway_decision and is
appended to the BLOCKED tool result the agent receives, so a declined
agent can adapt instead of only hearing 'denied'.
Adapted to hermes-agent's synchronous single-command /deny model: no DB
state, no second-message capture step, no migration. Reason is capped at
280 chars and threaded through both the terminal-command guard and the
execute_code guard. Plain /deny and the approve paths are unchanged.
- tools/approval.py: _ApprovalEntry.reason; resolve_gateway_approval gains
optional reason; _await_gateway_decision returns it; both gateway BLOCKED
messages include it
- gateway/slash_commands.py: parse leading 'all' + trailing reason
- locales/en.yaml: deny.denied_reason_{singular,plural}
- hermes_cli/commands.py: /deny args_hint '[all] [reason]'
- tests: 3 new (with-reason, all+reason, plain-deny regression)
* fix(ci): localize deny-reason keys across all locales + update interrupt-path assertions
CI surfaced two enforced invariants broken by the deny-with-reason change:
- test_i18n catalog-parity requires every locale to carry the same keys as
en.yaml with matching placeholders. Added deny.denied_reason_singular/plural
(with {count}/{reason}) to all 15 non-English locales.
- test_approval_interrupt asserts the exact dict from _await_gateway_decision,
which now carries a 'reason' key (None on the interrupt/timeout paths).
Inspired by Claude Code v2.1.199 (July 2, 2026): stacked slash-skill
invocations load all leading skills (up to 5), not just the first.
- agent/skill_commands.py: split_stacked_skill_commands() consumes leading
/skill tokens (stops at the first non-skill token so slash-path arguments
are never swallowed); build_stacked_skill_invocation_message() composes
the multi-skill turn reusing the existing bundle scaffolding markers so
extract_user_instruction_from_skill_message() keeps memory providers
storing the user's instruction, not N skill bodies.
- cli.py + gateway/run.py: dispatch the stacked path on both surfaces.
- 11 new tests + docs section in skills.md.
Port from anomalyco/opencode#34529: MCP servers can emit
notifications/message logging notifications (RFC 5424 levels), but the
MCP SDK's default logging_callback silently discards them — server-side
warnings/errors during tool calls were invisible.
- tools/mcp_tool.py: pass a logging_callback to every ClientSession
(stdio, SSE, streamable HTTP old+new API paths via the shared
sampling_kwargs sites), mapping the 8 MCP log levels onto Python
logging levels and tagging entries with [server/logger] origin.
- JSON-serialize non-string payloads, cap at 2000 chars so a chatty
server can't flood agent.log, never raise from the handler.
- Gated on SDK support (_check_logging_callback_support) mirroring the
existing message_handler gate for old SDK versions.
- tests/tools/test_mcp_server_log_notifications.py: 10 tests covering
level mapping, origin tagging, JSON payloads, truncation, and the
never-raise contract.
Inspired by Claude Code v2.1.199 (July 2, 2026): SSL certificate errors
(TLS-inspecting proxies, missing CA bundles, expired certs) no longer
burn retries before showing actionable guidance — they fail immediately
with the fix hint.
- agent/error_classifier.py: new FailoverReason.ssl_cert_verification +
_SSL_CERT_VERIFY_PATTERNS, checked BEFORE the transient-SSL patterns
(cert-verify messages also contain '[SSL:' and previously retried
forever as timeout). Non-retryable, no compression, no fallback churn.
- agent/conversation_loop.py: dedicated status line + per-cause fix
hints (corporate proxy CA bundle, certifi refresh, self-signed local
endpoints) on the non-retryable abort path.
- 7 new tests incl. regression guards (transient alerts still retry,
large-session cert failure doesn't trigger compression).
`hermes profile export default` crashed with `shutil.Error` when
HERMES_HOME pointed outside ~/.hermes (common in Docker deployments)
and the workspace contained broken symlinks. Two root causes:
1. `copytree` defaults to `symlinks=False` and follows link targets;
broken ones crash. #58397 (liuhao1024) drafted a minimal
`symlinks=True` flag fix; this PR adopts that change.
2. `copytree` was invoked against the entire HERMES_HOME root (which
doubles as cwd in Docker layouts). The post-hoc blacklist at
`_DEFAULT_EXPORT_EXCLUDE_ROOT` is a fixed-length enumerate-and-pray
list that can't anticipate every unrelated sibling directory
(`x11-dev/`, etc.). Replaced with a positive allow-list at
`_DEFAULT_EXPORT_INCLUDE_ROOT` enumerating the known Hermes profile
artifacts (config, persona, skills, cron, scripts, sessions,
plugins, memories, knowledge, preferences). Sensitive runtime
surfaces (`state.db`, `logs/`, auth files, other profiles) are
intentionally not in the allow-list so the export stays a
portable, credential-free snapshot of the user-facing surface —
which means the existing `test_export_default_excludes_infrastructure`
regressions remain green.
Adds two regression tests:
* test_export_default_uses_allowlist_for_unrelated_dirs — >x11-dev<
sibling directories must not leak into the archive.
* test_export_default_handles_broken_symlinks — symlinks inside
allowed artifacts survive instead of crashing the export.
closing that PR as superseded once this lands.
Closes#58394
shutil.copytree() defaults to symlinks=False which follows symlinks and
crashes on broken ones. In Docker/custom HERMES_HOME deployments,
unrelated directories may contain stale symlinks that break export.
Add symlinks=True to both copytree() calls in export_profile() so
broken symlinks are preserved as symlink entries in the archive.
Fixes#58394
video_analyze_tool's local-path branch read raw bytes via
_detect_video_mime_type (extension-only, no magic-byte check) with no
call to agent.file_safety.raise_if_read_blocked, unlike the image-gen
and video-gen provider plugins that already route local inputs through
that shared chokepoint (#57698). A model could point video_url at a
credential store (e.g. .env, auth.json) renamed or symlinked to a
video-like extension and have its raw bytes base64-encoded and sent to
the vision provider.
vision_analyze_tool and its native fast path (_vision_analyze_native)
had the same gap in their local-file branches; they were only
incidentally protected by the image magic-byte sniff rejecting
non-image content, not by the intended read guard.
Add raise_if_read_blocked() to all three local-file branches, mirroring
the existing plugins/image_gen and plugins/video_gen call sites.
* fix(cli): set correct x-initiator header per Copilot turn
copilot_default_headers() always hardcoded x-initiator: agent, but
GitHub Copilot billing requires "user" for user-initiated prompts and
"agent" for tool/follow-up calls. This caused premium requests to never
be consumed correctly, risking billing issues or account bans.
Adds is_agent_turn param to copilot_default_headers() and injects
extra_headers={"x-initiator": "user"} on the first API call of each
user turn when targeting Copilot URLs. The flag flips to False after
injection so subsequent calls (tool use, streaming fallback) default
back to "agent".
Fixes#3040
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* chore(release): add AUTHOR_MAP entry for @tjp2021 (PR #4097 salvage)
---------
Co-authored-by: Tim <tim@iteachyouai.com>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Both gateway compression entry points (session-hygiene auto-compress in
run.py; manual /compress in slash_commands.py) filtered the transcript
to user/assistant-only, content-bearing messages before calling
_compress_context. That starved the compressor:
- tool results are usually the bulk of the context, and
_prune_old_tool_results never saw them
- short filtered histories tripped the protect-first/last early-return,
so compression became a no-op even on huge sessions
- assistant tool_calls stubs (content=None) were dropped, so even the
summary lost the tool activity
Pass user/assistant/tool messages through intact, matching what the
agent loop itself feeds _compress_context.
Port of PR #3854 onto current main (the manual-compress handler moved
from run.py to slash_commands.py since the PR branched); regression test
asserts tool messages reach the compressor.
Authored-by: David Zhang <david.d.zhang@gmail.com> (@Git-on-my-level)
Co-authored-by: David Zhang <david.d.zhang@gmail.com>
_do_reconnect() succeeded but never called
YuanbaoAdapter.set_active(adapter), leaving get_active()
permanently returning None after any WS disconnect/reconnect
cycle. This caused cron delivery to silently fail because
_send_yuanbao() checks get_active_adapter() and gives up
immediately when it returns None.
Fix: call set_active(adapter) after successful reconnect,
matching the pattern in connect().
Fixes#58363
Post-#48648, oversized mid-stream edits truncate to a 4096-char preview
instead of splitting. But when rich messages raise the consumer's overflow
budget to 32k, the consumer keeps accumulating past 4096 and keeps issuing
progressive edits every edit_interval — each one truncating to the SAME
preview text. Telegram counts every one of those no-op requests against the
flood budget: a long streamed reply fires ~1 identical edit per 0.8s for
the rest of the stream, trips flood control (200s+ penalties), and the
final delivery hangs behind inline flood sleeps. Users see the bot stuck
'streaming' and the chat unresponsive.
Fix at the chokepoint: track the last truncated preview per
(chat_id, message_id) and skip the API call when the new truncation is
identical. Previews still update when the visible prefix actually changes
(e.g. chunk-count marker 1/2 → 1/3). State clears on finalize and when
content shrinks back under the cap, so dedup can never mask a real edit.
Live repro: 19,956-char streamed reply, transport=edit, rich available —
4x flood-control hits within ~700ms, 250s penalties, hung final delivery.
E2E harness on the same stream: 14 edit calls on main vs 7 with the fix
(the delta is pure no-op duplicates; scales with stream length).
Extend the pre_tool_call plugin hook return contract with a new directive:
{"action": "approve", "message": "why this needs human confirmation"}
Previously a pre_tool_call hook could only veto a tool call (action: block)
or allow it silently. It could not escalate to the existing human-approval
flow. This unlocks user-defined runtime approval rules on ANY tool (HTTP
writes, file writes to sensitive paths, email sends), enforced at runtime —
resolving #51221 as a pure plugin, with no core approval.py rule schema.
Mechanism:
- get_pre_tool_call_directive() returns (action, message) for block|approve;
get_pre_tool_call_block_message() kept as a block-only back-compat shim.
- resolve_pre_tool_block() is the single dispatch-site chokepoint: fetches
the directive and, for approve, invokes the human gate; fail-closed to a
block on denial, timeout, or gate exception. ALL FOUR tool-dispatch sites
now call it: tool_executor (concurrent + sequential), agent_runtime_helpers,
and model_tools.handle_function_call.
- request_tool_approval() escalates via the SAME machinery as Tier-2
dangerous commands: session/permanent allowlist, prompt_dangerous_approval
(CLI) / submit_pending (gateway), [o]nce/[s]ession/[a]lways/[d]eny,
timeout fail-closed, approvals.cron_mode for cron contexts.
Architecture: extracted the shared decision core into _run_approval_gate(),
called by BOTH check_dangerous_command() and request_tool_approval() so the
fail-closed / cron / gateway / yolo / persist policy lives in ONE place and
cannot drift. Fixed a latent divergence — the plugin path now honors --yolo.
Approval grain: [a]lways is keyed on tool_name + a hash of the reason (an
explicit plugin rule_key overrides), so distinct reasons on the same tool
persist independently instead of one 'always' blanketing the whole tool.
Non-interactive: cron honors approvals.cron_mode (parity with commands); any
other non-interactive non-gateway context fails CLOSED for the plugin path
(the command path keeps its historical fail-open default, unchanged).
No new config schema, no new env vars, no new hook events.
On Windows every Hermes process (gateway, serve, TUI/slash workers, MCP
servers, CLI commands) writes the shared rotating logs through
concurrent-log-handler's cross-process rotation lock. When the emitting
thread is an asyncio event loop, a lock wait blocks the loop — stalling it
for seconds and dropping WebSocket clients (the 'gateway keeps going down'
symptom seen in #58265).
Route every file handler through a single QueueListener on a dedicated
thread: loggers only enqueue (non-blocking); the listener does the file I/O
and rotation-lock wait off the hot path. The QueueHandler funnels via the
root logger; per-handler levels and component filters are preserved by
respect_handler_level + handler.handle on the listener thread. An atexit
hook stops the listener before logging.shutdown closes the file handlers.
- _NonFormattingQueueHandler passes the raw record (in-process queue) so
target handlers apply their own RedactingFormatter/filters.
- flush_log_queue() drains synchronously (shutdown + tests).
- rotating_file_handlers() exposes the handlers now behind the listener;
tests updated to use it.
Extends the #58265 fix: the provider-key warn-storm was one amplifier of
this contention; this takes the contention off the event loop entirely.
_normalize_custom_provider_entry() runs on every load_picker_context()
call (per picker/inventory request) and warned each time for (a) the
redundant `provider` key that Hermes' own config writer emits into
provider entries and (b) any other unknown key. On Windows the serve
launcher+worker pair share one rotating log via concurrent-log-handler's
cross-process lock, so that per-load warning volume drove 'Cannot acquire
lock after 20 attempts' retries that pegged a core, stalled the event
loop ~14s, and dropped every desktop/TUI WebSocket while /health stayed
green (gateway looked down; dashboard looked fine).
- Accept `provider` as a known key (silently ignored) so self-written
legacy configs don't warn.
- Deduplicate the normalizer's warnings per (provider, signature) so a
static config quirk is surfaced once, not on every inventory load.
Adds regression tests for both.
Fixes#58265
After a config change (e.g. switching model provider), the /new command
must clear the per-session _last_resolved_model cache so the next turn
resolves the model from the updated config instead of falling back to
the stale cached value.
Without this fix, if a transient config-cache miss occurs on the first
post-/new turn, the #35314 recovery path serves the old model from the
cache — the user sees the old model being used even though they changed
config.yaml and explicitly ran /new.
Fix applies to both call sites that reset session model state:
- GatewaySlashCommandsMixin._handle_reset_command (slash_commands.py)
- GatewayRunner compression-exhausted auto-reset (run.py)
Fixes#58403
Per-session /model overrides supplied api_key and provider but omitted
credential_pool, so billing rotation never ran on HTTP 402. Wire the pool
on fast override, rehydrate, and apply paths; backfill from provider for
legacy persisted overrides. Regression tests in tests/gateway/.
_try_anthropic() hard-failed (return None, None) when the anthropic
credential pool was present but had no selectable entry — e.g. the pooled
OAuth token expired and its refresh_token had gone stale, so
_select_pool_entry("anthropic") returned (True, None). This wedged every
auxiliary task routed to Anthropic (goal judge surfaced "no auxiliary
client configured") even when a perfectly valid ANTHROPIC_TOKEN /
credentials-file token was available. The main session stayed healthy
because it resolves the env token directly.
The openrouter path (_try_openrouter) and codex path already fall through
to their standalone credential on (True, None); anthropic was the only
provider that hard-failed. Make _try_anthropic fall through to
resolve_anthropic_token() on that branch so the three paths are symmetric:
a temporarily dead pool entry must not block auxiliary tasks when a valid
standalone credential exists.
Adds a regression test covering: (1) pool present + no entry + valid env
token -> client built from the env token, (2) pool present + no entry + no
resolvable token -> clean (None, None), (3) base_url defaults correctly
when falling through with pool_present=True.
The webhook adapter enforced max_body_bytes only via the Content-Length
header; a Transfer-Encoding: chunked request (content_length=None) or a
spoofed small Content-Length bypassed the cap entirely and read the full
body (bounded only by aiohttp's implicit 1 MiB default, above any
operator-configured smaller limit).
- web.Application(client_max_size=max_body_bytes): aiohttp enforces the
cap on every read path, chunked included
- catch HTTPRequestEntityTooLarge -> 413 (was swallowed into generic 400)
- post-read length re-check as defense in depth
- chunked-upload regression test
Manual port of PR #3955 by @Gutslabs onto current main (handler had
been restructured since); authorship preserved.
The container exec-read piped the whole file through base64 with no size
guard — the 50MB cap was only enforced host-side AFTER the full payload
had already streamed into host memory. A prompt-injected read of a huge
container file (or /dev/zero) could balloon the gateway process.
head -c (cap+1) bounds the read inside the sandbox; the +1 byte lets the
host distinguish at-cap from over-cap and reject with SourceTooLarge.
Input redirect replaces 'base64 --' (no argv exposure at all for
leading-dash paths). Docker integration tests re-verified live.
- _normalize_to_supported_image: ensure cache/vision exists before writing
the converted PNG (fresh HERMES_HOME had no dir -> FileNotFoundError).
- resolver: SVG passes through as image/svg+xml instead of erroring; the
call sites rasterize to PNG via the salvaged normalize step (cairosvg /
svglib / rsvg-convert / inkscape, best-effort with actionable error).
- normalization offloaded via asyncio.to_thread at both call sites.
- tests: resolver pass-through + rasterization + no-converter error paths.
- AUTHOR_MAP: jonathan@mintrx.com -> JAlmanzarMint (PR #52688 salvage).
vision_analyze embedded SVG (and BMP/TIFF) tool-results into conversation
history with media_type image/svg+xml. Anthropic only accepts jpeg/png/
gif/webp, so the request fails with a non-retryable 400. Because the image
is baked into immutable history and re-sent every turn, the session is
permanently wedged on resume — retries re-send the same bad bytes.
Add _normalize_to_supported_image(): SVG is rasterized to PNG (best-effort
via cairosvg/svglib/rsvg-convert/inkscape), other non-supported raster
formats are re-encoded to PNG via Pillow, and if conversion is impossible
the tool returns an actionable error instead of a session-wedging payload.
Wired into both the native-vision fast path and the auxiliary-API path so
the whole bug class is covered, not just the one call site.
All 99 existing vision tests pass.
- resolve_image_source: bare cwd-relative filenames ('pic.png') resolve
again (main accepted them; the path-shape gate regressed them — review
by egilewski). Unknown explicit schemes (ftp://, s3://) still rejected.
- Local backend: nonexistent path now raises a clean 'image file not
found' instead of a misleading sandbox-fallback message.
- W4: remove the now-dead path-based _detect_image_mime_type (suffix-trust
SVG acceptance) so future callers can't reintroduce it.
- W3: SVG sources rejected with a dedicated actionable message + test.
- Polish: host read_bytes / temp-file write_bytes offloaded via
asyncio.to_thread (matches the container exec-read); unused
ResolveContext.cfg/extra_roots fields dropped; duplicate policy check
documented as intentional pre-flight short-circuit.
Salvage of #35362, evolved to also close the vision sandbox-escape
(GHSA-gpxw-6wxv-w3qq). The two were the same root cause — vision read image
bytes host-side while every other tool reads through the terminal backend —
so one resolver fixes both the delivery gaps and the escape.
Delivery (from #35362, re-authored against current main since the branch was
4140 commits stale and vision_tools.py had been rewritten on both sides):
- tools/image_source.py: one resolver for data:/http(s)/file/local/container
image sources, returning raw bytes through a single magic-byte-sniff +
50MB-ingest chokepoint. Fixes 'no image attached' / 'Invalid image source'
for every source type (#7571, #25118, #29643, #22328, #32709, #9077).
- tools/credential_files.py: from_agent_visible_cache_path, the container->host
cache reverse-map (inverse of the existing forward twin).
- tools/vision_tools.py: both vision sites route through the resolver with
task_id threaded from the handler; resolved bytes are materialized to a temp
file so main's evolved encode/resize/embed-cap pipeline is reused verbatim
(kept over the PR's older bytes-core resize to avoid touching browser_tool /
conversation_compression callers).
Security (fills #35362's deliberately-stubbed _within_allowed_roots seam):
- Under a non-local terminal backend the file tools are confined to the sandbox
(SECURITY.md 2.2), but vision read host-side — a prompt-injected
vision_analyze('/etc/passwd') exfiltrated host secrets, and read_file even
redirects the model to vision_analyze for image paths. The resolver now
enforces the same boundary: local backend reads any host path (chosen
posture); non-local backend host-reads ONLY the media caches under
HERMES_HOME (where the gateway/download media lives) and routes every other
path to an in-sandbox base64 exec-read — which reads the CONTAINER's file,
the same one 'cat' would, never the host's. Paths are resolve()-d so a
symlink can't escape a cache; fail-closed when no sandbox env exists.
This closes the escape AND delivers container-only images (#32709) with the
same mechanism.
Tests: unified resolver + confinement model (tests/tools/test_image_source.py,
incl. proof a non-cache host path under Docker yields container bytes not the
host secret); existing vision tests updated to the resolver boundary; Docker
integration test verified green against a real daemon (exec-read of a tmpfs
/workspace file, a root-owned mode-600 file, and the host-secret invariant).
Fixes GHSA-gpxw-6wxv-w3qq.
Co-authored-by: banditburai <promptsiren@gmail.com>
'KEY=os.getenv(...)' / 'os.environ[...]' / 'process.env.X' values are
variable-name references in code snippets, not leaked secrets. Masking
them corrupted pasted code in prose/log contexts (issue #2852):
ha_token=os.getenv('HOMEASSISTANT_TOKEN') -> ha_token=os.get...EN').
Skip these values inside _redact_env, which covers all three passes that
share the closure (_ENV_ASSIGN_RE, _CFG_DOTTED_RE, _CFG_ANCHORED_RE).
Real secret values are still masked.
Salvage of PR #2852-fix #2854 — the PR's own placement (an unconditional
pass before the code_file gate) would have reintroduced the code-file
false-positive class; the skip is applied inside the existing gated pass
instead. Tests adapted from the PR.
Co-authored-by: crazywriter1 <sampiyonyus@gmail.com>
Add unit tests for resolve_placeholder_terminal_cwd and extend the config
bridge simulation for docker mount-on vs mount-off vs local fallback.
Co-authored-by: Cursor <cursoragent@cursor.com>
The conflict-retry ladder schedules a background recovery task via
loop.create_task(self._handle_polling_conflict(...)) on each failed
start_polling. test_polling_conflict_becomes_fatal_after_retries never
cancelled the last one, so under a loaded scheduler a leaked task could
get a turn, re-drive the counter into the fatal branch, and fire
_notify_fatal_error a second time — breaking assert_awaited_once()
non-deterministically. The bounded updater.stop() guard added in this
salvage introduced an extra await/scheduling yield that surfaced the
latent leak in CI slice 3. Cancel the leaked task before the fatal
assertions so the test is deterministic regardless of scheduler timing.
The salvaged fix (#58272) guarded the primary network-error reconnect path.
Issue #58270's Scope section names three more unguarded await updater.stop()
sites that can hang identically on a CLOSE-WAIT socket:
- conflict handler (before the retry back-off sleep)
- conflict-retries-exhausted teardown (before the fatal notify)
- disconnect() teardown (would hang gateway shutdown/restart)
Each is now wrapped in asyncio.wait_for(..., _UPDATER_STOP_TIMEOUT) with a
warning on timeout, matching the primary path, so no reconnect/teardown ladder
can wedge on a dead socket.
Also hoist the shared 15.0s bound to a single module constant
_UPDATER_STOP_TIMEOUT (self-documenting + DRY across all 4 sites), and update
the CLOSE-WAIT regression test to patch that constant instead of monkeypatching
asyncio.wait_for process-wide.
Fatal-notify idempotency: bounding the conflict-exhausted teardown stop() adds
an await AFTER _set_fatal_error, which yields the loop and lets a concurrent
retry task (scheduled by an earlier conflict, already suspended past the entry
guard) reach the fatal branch too — double-firing the fatal handler (surfaced
as a Python 3.11 CI failure in test_polling_conflict_becomes_fatal_after_retries).
Snapshot the pre-transition fatal state and only notify on the first transition.
When the TCP connection enters CLOSE-WAIT the PTB polling task is blocked
on epoll on a dead socket and never wakes. updater.stop() awaits that task
and therefore hangs indefinitely.
Consequence: _polling_error_task stays alive-but-blocked forever; every
subsequent heartbeat probe sees it as "in-flight" and skips triggering a
new reconnect; the gateway silently drops messages for hours until a manual
restart. Field incident: 11-hour outage on 2026-07-04 UTC despite the
heartbeat loop firing a reconnect at 01:11 — stop() blocked the entire
ladder.
Fix: wrap the updater.stop() call inside asyncio.wait_for(timeout=15).
On TimeoutError log a warning and continue to _drain_polling_connections()
+ start_polling() — same recovery path, just unblocked.
The heartbeat loop (PR #48496) correctly detects the dead socket and fires
_handle_polling_network_error. This commit is the missing second half:
ensuring the reconnect itself always completes.
Test: test_handle_polling_network_error_updater_stop_timeout() simulates
a hang by making stop() sleep forever and verifies that drain + start_polling
are still reached after the timeout.
Fixes#58270
Strict providers (DeepSeek) reject a payload where the same tool_call_id
appears more than once with HTTP 400 'Duplicate value for tool_call_id'.
The issue was filed as an 'orphaned tool message' compression bug, but the
pasted error is a DUPLICATE tool_call_id — orphans are already handled on
main; duplicates were not. Reproduced live on main: both shapes leaked
through repair_message_sequence and sanitize_api_messages.
Two chokepoints, two shapes:
- repair_message_sequence: consume the id from known_tool_ids on first
match so a SECOND tool result reusing it falls into the drop branch
(duplicate tool-result shape). This is @Robinlovelace's kernel from
#55436 (applied manually — that PR was ~800 commits stale and bundled
an unrelated duplicate-DB-write change for #860, which is dropped here).
- sanitize_api_messages (final pre-API pass): add a dedup pass covering
BOTH (a) duplicate tool_calls sharing an id WITHIN one assistant message
(the message[6] shape) and (b) later tool result messages reusing an
already-seen id. #55436 covered neither of these at this chokepoint.
Tests: duplicate-tool-result dedup at both functions, duplicate-assistant-
tool_call-id collapse, and a negative control proving distinct ids are
never dropped (no over-dedup).
Credit: @Robinlovelace (#55436) for the repair_message_sequence dedup kernel.
Closes#58327.
22c5048d9 restored Anthropic-style cache_control for two of MoA's three
call paths: the acting aggregator (MoAChatCompletions.create, the
persistent `provider: moa` model) and the advisor fan-out (_run_reference).
aggregate_moa_context() -- the /moa <prompt> one-shot command's synthesis
call -- is the third, independent call path and was never covered: its
call_llm(task="moa_aggregator", ...) sent a single undecorated user message
containing the full joined reference output, re-billing the entire input on
every invocation even when the resolved aggregator slot is a cache-honoring
route (Claude on OpenRouter/native Anthropic, MiniMax, Qwen/DashScope).
- Generalize _maybe_apply_advisor_cache_control to
_maybe_apply_moa_cache_control (it never had advisor-specific logic --
same policy function, same breakpoint layout as the main loop, judged
purely on the passed-in runtime) and reuse it in aggregate_moa_context
the same way _run_reference already does.
- Compute _slot_runtime(aggregator) once and reuse it for both the
decoration call and the call_llm kwargs, instead of calling it twice.
Mutation-verified: reverting the moa_loop.py change makes the new
regression test fail by asserting a plain string aggregator-message
content where the cache-honoring case expects native cache_control
content blocks.
Follow-up to @msh01's wall-deadline init-timeout fix.
- Resource leak: on timeout the initialize() task is abandoned without
awaiting its (shielded, possibly-never-completing) cancellation, so the
half-built PTB app's httpx client / connection pool was never closed —
up to 8x across the retry ladder. Add an optional on_abandon cleanup to
_await_with_thread_deadline that best-effort app.shutdown()s the abandoned
app, run detached + exception-swallowed so it can never re-block or re-hang
the ladder (mirrors _close_client_on_timeout in agent/auxiliary_client.py).
- Cover the helper itself: the salvaged test monkeypatched out the real
_await_with_thread_deadline, so its abandonment/cleanup path was untested.
Add direct tests for happy-path return, prompt-timeout-with-cleanup, and
cleanup-error-swallowed; the wedged coroutines swallow cancellation for a
bounded window (proving the helper returns before cancellation completes,
the #58236 shielded-scope behavior) without leaving an immortal task that
would wedge pytest teardown. Widen the salvaged stub to accept on_abandon.
- Attribution: add yingwaizhiying@gmail.com -> msh01 to AUTHOR_MAP (bare
gmail does not auto-resolve the check-attribution gate).
Known follow-up (not addressed here): the retry ladder reuses the same
self._app across all 8 attempts; a fresh app per attempt would fully close
the coherence risk if an abandoned initialize() completes in the background.
That is a larger restructure of the ~130-line builder+handler setup, left
for a separate change.
Follow-up to @srojk34's basename-denylist widening. Two gaps the
basename-only guard left, both covered by the two canonical guards it
mirrors:
- Directory-tree stores mcp-tokens/ (live MCP OAuth tokens) and pairing/
are denied as whole trees by gateway.platforms.base._ROOT_CREDENTIAL_DIRS
and agent.file_safety, but the dashboard files API descends into subdirs,
so mcp-tokens/<server>.json (non-canonical basename) stayed
listable/readable/downloadable. Add _is_sensitive_path(), a path-aware
check that blocks any path with a credential-directory component, and
route all three call sites (list/read/download) through it.
- Add .git-credentials to the basename set (agent.file_safety blocks it too).
- Correct the docstring: it now says it mirrors the credential-FILE basenames
of the canonical guards, with the directory trees handled by the new
path-aware helper (the prior wording overstated parity).
Scope stays on the read/list/download exfil surface (#57505); the write
endpoints (upload/mkdir/delete) are a separate threat and out of scope.
Tests: dir-tree descent blocked (mcp-tokens/pairing per-server files),
.git-credentials blocked, plus a positive control that a benign subdir file
stays browsable. Mutation-checked (neuter _is_sensitive_path -> new tests
fail). 39 web_server_files + fs tests pass, ruff clean.
_is_sensitive_filename() only blocked .env / .env.<suffix>, but the
dashboard Files tab's managed root is operator-configurable and, per the
docker-mount scenario #57505 was filed against, can point directly at
HERMES_HOME — where the canonical credential stores enforced elsewhere
in the codebase (gateway.platforms.base._ROOT_CREDENTIAL_FILES,
agent.file_safety.get_read_block_error) all live: auth.json, OAuth
token stores, webhook HMAC secrets, the Bitwarden disk cache. None of
those basenames were blocked, so the Files tab could still list, read,
and download them. .envrc (direnv) also slipped past the old check
since it doesn't equal ".env" or start with ".env.".
Widen the basename set to mirror both existing guards so the dashboard
doesn't lag behind them.
Both wired against features the iron-proxy author (@mslipper) confirmed on
PR #30179 — and both verified present in the pinned v0.39.0 source.
Header-auth providers (match_headers):
- New _HEADER_AUTH_PROVIDERS: Anthropic native (x-api-key), Azure OpenAI
(api-key on *.openai.azure.com / *.cognitiveservices / *.services.ai),
Gemini (x-goog-api-key + ?key= query param via match_query).
- TokenMapping grows match_headers + alias_env_names; per-provider header
sets flow into the secrets rules; mappings.json roundtrips them
(legacy files load with the Authorization default).
- GEMINI_API_KEY / GOOGLE_API_KEY collapse into ONE mapping (two
require-rules on the same host would reject each other); the sandbox
gets the token under both names, and the proxy child env mirrors the
alias into the canonical name when only the alias is set.
- Docker backend injects alias env names alongside canonical ones.
- The fail-closed tier is now empty, so fail_on_uncovered_providers and
discover_blocked_providers are deleted (dead toggle otherwise);
_NON_BEARER_PROVIDERS shrinks to genuinely-unswappable signature auth
(AWS SigV4, GCP service-account OAuth) — warn-only, as before.
Management API (hot reload):
- Generated proxy.yaml enables the v0.39 management listener: loopback
only at tunnel_port+2, bearer key from HERMES_IRON_PROXY_MGMT_KEY.
- Key minted at setup (management.token, 0600); start_proxy injects it
(v0.39 refuses to start when api_key_env is empty).
- hermes egress reload -> POST /v1/reload: re-reads proxy.yaml and
atomically swaps the pipeline; 422 leaves the running ruleset
untouched; actionable errors for not-running / pre-management config /
key mismatch. Secrets changes still require restart (daemon env is
read at spawn) — the CLI says so.
Validation: 218/218 unit+CLI+docker tests; 3/3 gated live E2E against the
real v0.39.0 binary (Authorization swap, x-api-key swap, live reload with
token rotation on the same pid). Docs updated.
Two review findings on the #57922 salvage:
1. Stale inline comment at the login-exchange site still claimed the token
endpoint uses the claude-code/ UA prefix and 404s claude-cli/ — now
contradicts the axios/ fix. Repointed it at _OAUTH_TOKEN_USER_AGENT.
2. The inherited Path.home test isolation on the three TestRefreshOauthToken
tests only stubbed the ~/.claude *file* source, not the macOS Keychain.
_refresh_oauth_token re-reads read_claude_code_credentials() (keychain
first) in its adopt-already-refreshed branch, so on any macOS dev/CI runner
with real Claude Code creds the branch short-circuits and the 3 tests fail.
Stub read_claude_code_credentials -> None so the tests are hermetic.
(The remaining TestResolveAnthropicToken/TestResolveWithRefresh/TestRunOauthSetupToken
failures on macOS are the same pre-existing keychain-leak class on origin/main,
unrelated to this OAuth-UA fix, and pass in CI — left out of scope.)
hermes auth add anthropic fails 100% at token exchange with HTTP 429 while
Claude Code /login succeeds through the same client_id/redirect/scope. The
discriminator is the User-Agent on the /v1/oauth/token request.
Verified live against platform.claude.com (throwaway code, nothing burned):
claude-code/2.1.200 (external, cli) -> 429 rate_limit (Hermes, blocked)
Mozilla/5.0 -> 429 rate_limit
axios/1.7.9 -> 400 invalid_grant (reached validation)
node / empty / SDK-style UAs -> 400 invalid_grant
Anthropic now rate-limits token-endpoint requests whose UA starts with
claude-code/ (the anti-abuse net for Max-sub-as-API-key). This is the same
prefix-block shape that #48534 first hit on claude-cli/, then #56263 dodged by
switching to claude-code/ — which held ~2 weeks and is now blocked too. Bumping
_CLAUDE_CODE_VERSION_FALLBACK cannot help; the gate is prefix-based.
Fix: shared _OAUTH_TOKEN_USER_AGENT (axios/) on the token endpoint only — the
two refresh POSTs (refresh_anthropic_oauth_pure) and the login exchange POST
(run_hermes_oauth_login_pure). The real Claude Code CLI exchanges the auth code
with a bare axios client, NOT its claude-code/ inference UA.
The INFERENCE client (build_anthropic_kwargs, /v1/messages) is deliberately left
on claude-code/ + x-app: cli — that fingerprint is required there and is NOT
throttled on the messages API. Two endpoints, opposite UA requirements.
Also isolate two _refresh_oauth_token tests from live ~/.claude creds and update
the UA regression tests to assert the split (token endpoint uses a
non-claude-code UA while inference keeps claude-code/).
Verified E2E: Hermes' own login path now returns 400 (past the 429 wall)
instead of 429, using the real _OAUTH_TOKEN_USER_AGENT constant against the live
platform.claude.com token endpoint.
Salvaged from #57922 (authorize-host + scope changes dropped as non-load-bearing;
they only add a redirect hop back to claude.ai and the UA fix alone clears 429).
repair_message_sequence Pass 1 registered only tc.get("id") when building
the set of known assistant tool_call ids, then matched tool results against
it by tool_call_id. In the Codex Responses format an assistant tool_call
carries both id (fc_...) and a distinct call_id (call_...); a tool result's
tool_call_id may be keyed on either depending on which builder produced it.
Registering only id made a valid tool result whose tool_call_id matched
call_id look orphaned, so the pass dropped it and left the assistant
tool_call unanswered -- producing HTTP 400 on strict providers (DeepSeek,
Kimi): 'Messages with role tool must be a response to a preceding message
with tool_calls'. Long-running sessions that persisted such a sequence were
permanently broken, re-sending the orphan every turn.
Register both id and call_id for each assistant tool_call so a result
matching either key is recognized, consistent with
AIAgent._get_tool_call_id_static and the compressor's _sanitize_tool_pairs.
Apply the same call_id||id precedence to the corrupted-args sanitizer's
existing-result scan / stub insertion, which had the identical mismatch.
Adds 3 regression tests covering the codex id!=call_id case (match on
call_id, match on only call_id, match on id when both present).
On X11 a window's PID comes from the optional _NET_WM_PID property, so
cua-driver's list_windows legitimately returns pid: null for windows that
don't set it (desktop root, panels, override-redirect popups). capture()
and focus_app() coerced every entry via int(w["pid"]) inside a list
comprehension, so a single null-pid window raised TypeError and aborted
the whole enumeration before any screenshot — capture was impossible on
any X11 desktop with even one such window.
Route both ingestion sites through a new _ingest_windows() helper that
skips entries lacking a usable pid/window_id (uncapturable anyway) and
coerces the rest.
Adds tests/tools/test_computer_use_null_pid_windows.py covering the
helper's filtering/coercion and an end-to-end capture() regression that
reproduced the crash.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The #22773 heuristic classified any telegram:<positive_chat_id>:<numeric_thread_id>
cron target as a Bot API channel Direct-Messages topic and routed it via
direct_messages_topic_id, which nulls message_thread_id. A normal forum-style
topic inside a private chat has the identical shape, so every such cron
delivery landed in General instead of the target thread (#52060). It also
means the only way to address a genuine channel DM topic was this same
ambiguous guess.
Disambiguate with the real runtime signal instead: probe the live adapter's
get_chat_info once and route via direct_messages_topic_id only when the chat
is actually a channel; everything else (private forum topic, forum supergroup,
group) and any probe failure fall back to message_thread_id — parity with
0.16.0 and with live reply routing. This fixes forum-topic delivery AND makes
genuine channel DM-topic cron delivery work correctly.
PR #57601's original branch added a top-level reasoning_effort emit to the
LEGACY build_kwargs path (agent/transports/chat_completions.py), but
provider=custom resolves to CustomProfile (plugins/model-providers/custom/),
so chat_completion_helpers takes the profile path and returns early — the
added branch was unreachable dead code for every custom endpoint.
Move the fix to its real site, CustomProfile.build_api_kwargs_extras(), and
follow the DeepSeek/Zai profile precedent:
- disabled -> extra_body.think = False (unchanged)
- enabled + effort -> TOP-LEVEL reasoning_effort (the OpenAI-compatible
format GLM-5.2/ARK expect), passed through verbatim
incl. max/xhigh
- enabled + no effort -> omit, so the endpoint's server default applies
(avoids silently forcing 'medium' as the original
branch did)
Deliberately does NOT force think=True on enable — that flag is Ollama-only
and risks a 400 on GLM/vLLM endpoints that don't recognize it; thinking is
already server-default-on for these backends.
Verified end-to-end through the real profile dispatch (temp HERMES_HOME):
custom+high -> reasoning_effort=high; custom+max -> reasoning_effort=max;
custom+none -> think=False; custom+unset -> nothing; num_ctx composes.
Adds tests/plugins/model_providers/test_custom_profile.py (13 cases).
Addresses the custom-provider half of #55276.
Co-authored-by: huanshan5195 <huanshan5195@users.noreply.github.com>
Follow-up to salvaged #57601. Adding "max" to VALID_REASONING_EFFORTS
made parse_reasoning_effort("max") valid, so:
- test_unknown_levels_return_none no longer lists "max" (it is now valid;
auto-covered by test_each_valid_level which iterates the tuple).
- test_known_supported_levels_are_documented and the parse_reasoning_effort
docstring now include "max" so the doc-sync guard actually protects it.
Follow-up to the salvaged SSH-tilde-cwd fix. The predicate
"backend == ssh and (cwd == ~ or cwd.startswith(~/))" was inlined at
each expanduser guard site, which is how the test simulator drifted from
production (it grew an SSH guard on a top-level-alias branch that has no
production counterpart).
- Add tools/terminal_tool._is_ssh_remote_tilde_cwd(backend, cwd) as the
single source of truth (case/whitespace-tolerant).
- Use it in _get_env_config and the gateway config bridge.
- Test simulator imports the real helper instead of re-implementing the
predicate; revert the phantom SSH guard on the top-level-alias branch
(production maps top-level cwd: to a plain env var, not TERMINAL_CWD via
an SSH-guarded path — that branch tested nothing real).
Follow-up to the salvaged #58000 fix.
- Extract _raise_if_non_interactive(lead) so the shared 'hermes mcp login'
next-step wording lives in one place across both OAuth boundaries
(_redirect_handler, _wait_for_callback), rather than two copy-pasted
inline raises. Boundary-specific lead sentences preserved verbatim, so
existing message-match tests stay green.
- Add a positive-control test asserting the guard does NOT over-fire on the
interactive path (valid/refreshable tokens keep working), satisfying an
explicit regression-coverage line from #57836.
Add TestNonInteractiveFailFastAtCallbackBoundary: the callback boundary must
reject before binding a listener and without entering the poll loop, the guard
must hold even when a (stale) token file exists on disk, the redirect handler
must not print a URL or open a browser, and both boundaries must point users at
`hermes mcp login`.
Mark the existing timeout test and the SSH-hint redirect tests interactive so
they exercise their intended paths rather than short-circuiting on the new
non-interactive guard.
The salvaged max-output/pressure fix set conversation_history=None after
the new pre-API compaction. That is only correct for legacy session-
rotation. Under the default in-place compaction (compression.in_place:
True), archive_and_compact inserts the compacted rows into the session DB
directly without stamping them with the intrinsic persisted-marker, so a
subsequent flush with conversation_history=None re-appends them — doubling
the active context and retriggering compression (the early-persist
duplicate-row trap).
Use conversation_history_after_compression(agent, messages), matching the
two existing compaction sites (post-response should_compress and the
turn-prologue preflight), which returns None for rotation and
list(messages) for in-place so the compacted dicts are skipped by identity.
Adds a regression test with a real SessionDB + real archive_and_compact
that asserts the compacted summary row is persisted exactly once (fails
with 2 copies on the None variant).
Follow-up to the salvaged #57845 fix. _can_carry_marker used
any(isinstance(part, dict)) but _apply_cache_marker only marks the LAST
content part, so a list whose last element is a non-dict passed the carrier
gate yet received no marker — wasting one of the four breakpoints. Tighten
the predicate to require content[-1] to be a dict (mirroring the apply
logic) and add a regression test. Flagged by a 3-agent review.
Gateway users can now search resumable sessions from messaging surfaces:
/sessions search <query> (alias: find) matches titles and session ids —
including every title/id in a row's forward compression chain, so a
compressed-away title still surfaces its live tip — plus a
punctuation-normalized variant so 'an94' matches 'AN-94'.
Implemented by generalizing the existing id_query chain-filter in
SessionDB.list_sessions_rich into a combined SQL-level filter (search
stays ORDER BY last-active + LIMIT at SQL level), threading a
search_query through the shared query_session_listing helper, and
teaching parse_session_listing_args to split off a search query.
Search results pass through the existing _resume_row_visible guard
unchanged: origin scoping, admin-only 'all', and the fail-closed
legacy-row posture from the July 1 hardening are preserved exactly.
Over-fetch (50) before the visibility cut so origin-invisible matches
can't starve the page.
Salvages the feature direction of PR #57595 by @GodsBoy with a minimal
implementation that keeps the resume authorization surface untouched.
Post-merge follow-ups + several review rounds + a hub-search rework, folded together.
Merge-scuff restores (a stale-base refactor had reverted two live-on-main fixes):
- gateway: SessionStore compression-tip healing + its regression test.
- desktop: messaging session/transcript polling in desktop-controller
(MESSAGING_POLL / ACTIVE_MESSAGING_SESSION_POLL, refreshMessagingSessions,
refreshActiveMessagingTranscript, the richer sameCronSignature) so inbound
platform traffic updates live again instead of freezing until manual refresh.
Profile-switch isolation (epoch/close/guard on every profile-scoped async):
- Hub store clears + in-flight runHubAction bails (and swallows the post-switch
404 instead of a phantom toast); hub preview/scan/search/sources profile-scoped.
- MCP: probe/auth epoch guards, dirty-draft reset, sidebar mutations blocked
until config resettles AND every persist re-checks the epoch post-await;
profilePending clears on config settle incl. error; logs re-key on profile.
- Model settings reload on switch and epoch-guard setModelAssignment /
saveMoaModels / API-key activation.
- Config draft resets + cancels its autosave on switch; skill editor/archive and
star-map node dialogs close on switch; openSkillEditor / star-map openEdit
discard stale fetches; tool-usage analytics loads are profile-guarded/keyed.
Correctness + UX:
- Unique per-skill action names for hub install AND uninstall; hub/catalog rows
flip only on a clean exit_code; catalog install polls the background bootstrap
to completion, reconciles the mcp.json draft (no dropped server), and fails
loudly on non-zero exit; MCP catalog query keyed by profile.
- /test reports needs-auth for anonymous auth:oauth servers; /auth snapshots +
restores tokens on a failed re-auth and clears the full 300s callback window.
- config-settings shows a retry on load failure; CodeEditor/JsonDocumentEditor
go read-only while saving so edits typed mid-save aren't dropped.
- Deep-link highlighter deletes its param only after a successful scroll.
- Restored the PageSearchShell trailing slot → Artifacts refresh button/spinner.
- /settings?tab=mcp redirect keeps server=.
Progressive hub search: fan out one query per backend-searchable source
(index-covered API sources stay unsearchable → no ~70-call GitHub re-hammer),
merge/dedupe by trust as each lands, per-source spinner overlaid on the dimmed
chip — results stream in without blocking on the slowest, no layout shift.
test(web): /api/skills list carries usage + provenance (CI contract).
from a live blocked-sites pass (no PII):
- posture shift: blind opt-out is the DEFAULT, not a fallback -- submit on every site with an
accessible removal channel even without first confirming a listing (own identifiers to the broker's
own official channel = still least-disclosure). guided flows double as the authoritative search.
- blocked-form rule: when a form is automation-hostile (hard captcha / cloudflare / datadome /
slide-to-verify), default to the broker's CITED rights-email rather than recording blocked.
- captcha policy clarified: never defeat behavioral/token/slider challenges; ok to read a static
distorted-text or plain-arithmetic captcha on the subject's own opt-out; stop if the whole
submission is rejected after a correct answer (fingerprinting the automation, not grading it).
- intelius/peopleconnect: delete-wipes-suppression is field-confirmed -- a deletion-complete email
means the suppression is gone and the subject re-lists cluster-wide; re-run suppression and verify
the Control step reads "suppressed". guided-mode session persists; DOB is an <input type=date>.
- new records: addresses.json (intelius front-end, cluster-covered) and socialcatfish.json
(cited rights-email lane + automation-hostile form).
- new references/site-playbooks.md: per-site game-plan matrix (8 blocked-tail sites), the meta-search
no-op skip-list (idcrawl/lullar/yasni/webmii/namesdir/itools/skipease), and the infopay /
peopleconnect backend clusters. OSINT-list triage taxonomy added to methods.md.
- state-machine.md: fixed doc drift + documented submitted->not_found illegal (resolves as
awaiting_processing), blocked->submitted via action_selected, operator_manual_check, --evidence & pitfall.
tests: standalone 99, PR 97 (+1 cluster-coverage regression); ruff + windows-footguns clean.
Route the app off its hand-rolled helpers onto lib/{text,time,format,json-format}
and the new primitives, plus assorted small tidy-ups:
- compactNumber for counts/tokens; normalize/capitalize/asText at the many
filter/label sites; shared Intl date/time formatters; row-hover + framed
editor adoption; scrollbar-gutter + padding parity on list surfaces.
- Messaging/Artifacts/Cron search hints + narrow-viewport tab dropdown;
floating-pet adopts useOnProfileSwitch; number formatting in statusbar,
command-center, agents.
- Electron: native overlay width + backend spawn tidy.
- Settings > Keys: credential fields read as plain subtext (all-unset) until
the group is focused or expanded, then take full input chrome with no
horizontal/vertical shift; inline Remove (trash) + Save mirror SearchField's
trailing-clear pattern instead of a floating hint that overlapped the card;
Esc still cancels. Drops the now-dead or/escToCancel i18n keys.
- Shared TabDropdown/ResponsiveTabs (components/ui): PageSearchShell and the
Command Center log file/level filters reuse the one narrow-width collapse.
- OverlayNav: data-driven pane nav — persistent rail on wide, a single dropdown
riding the titlebar strip on narrow; Settings and Command Center adopt it, and
the mobile dropdown carries the same section icons as the rail. Fixes narrow
vertical centering, redundant mobile section titles, gateway-status wrap, and
Panel master/detail stacking.
- OverlayIconButton is now the titlebar ghost button, matching the close X at
every size. Settings sub-view nav opens section + sub-view in one navigate so
API-keys/accounts actually open on narrow.
- Settings > Model: cube icon (was the {} namespace glyph) and a DOM-shaped
skeleton in place of the centered spinner.
- Command palette / session switcher clear the macOS traffic lights on small
screens.
- Prettier/eslint sweep across the touched files.
phase-2 work (sending webmail, clearing session-bound gates like peopleconnect guided-mode) needs
the operator's own logged-in browser, not a cloud browser. new `pdd.py cdp`:
- finds chrome/chromium/brave/edge (macos/linux/windows), launches it detached on a dedicated debug
profile ($HERMES_HOME/chrome-debug) with --remote-debugging-port, waits for the port, prints the
CDP endpoint (webSocketDebuggerUrl)
- `--check`: report whether a debug browser is already live (never double-launches)
- `--print`: emit the exact command for the operator to run themselves
- doctor, SKILL.md, and methods.md all point at it
- windows-safe detach (start_new_session on posix, DETACHED_PROCESS on windows); stdlib only
tests: standalone 98, PR 96 (+6 cdp); ruff + windows-footguns clean.
from a live run (NY subject, 43 brokers):
- fanout default 8->5 (8+ batches time out)
- setup/doctor read $HERMES_HOME/.env so creds hermes already loads are detected
- new `show <subject> <broker>`: reads back case state+evidence for cheap parent re-verify
- intelius: requires.dob + 5-step guided-mode gate; planner pre-warns when dob is missing
- rehold.json: property-record != PII (an address-only match is not_found, not removable)
- tps/fps: match_signal_notes tell the scanner to ignore SEO-templated titles
- methods.md: browser backends (scan vs execute + operator chrome over CDP), property/SEO callouts
- doctor: warn when browser email-mode pairs with a cloud scan backend (needs operator chrome/CDP)
- ledger: found->not_found retract (false-positive), blocked->human_task_queued
- autopilot: indirect-exposure web-form fallback; drop a stray f-string
tests: standalone 92 pass; ruff clean.
A plugin-registered WebSearchProvider with no built-in provider credentials
must light up web_search / web_extract and be discoverable by the backend
selectors. Covers check_web_api_key(), _get_backend(), _is_backend_available()
registry delegation, per-capability extract selection (#32698), and that the
web_search / web_extract tool registry entries are not filtered out.
Tests contributed by @m0n5t3r (PR #28652, issue #28651).
Asserts the behavior contract that run_one_job installs a profile secret
scope around run_job under multiplexing (so resolve_runtime_provider's
get_secret does not fail-close with UnscopedSecretError) and tears it
down afterward. Mutation-verified: fails on unmodified main with the
exact UnscopedSecretError, passes with the fix.
Fold the xAI video credential-read guard into the same shared
agent.file_safety.raise_if_read_blocked chokepoint this PR introduces for
the image providers, so the whole image+video bug class is covered by one
enforced boundary. Consolidates the parallel salvage of #57695 (xAI
image+video) into this PR; #57727 is now redundant and will be closed.
- video_gen/xai: guard _image_ref_to_xai_url and _video_ref_to_xai_url
(the video image + video byte-read chokepoints) via the shared helper.
- Regression tests: symlinked auth.json with .png/.mp4 names are blocked
across both video read paths (mutation-checked).
Follow-up to the per-provider guards. Three improvements from review:
1. Extract agent.file_safety.raise_if_read_blocked() as a single shared
chokepoint and route the OpenAI, OpenRouter, and (newly) xAI image
providers through it, replacing the 3x-duplicated inline try/except.
Fixes the whole bug class: xai/_xai_image_field read a model-supplied
local path via open() with no guard — the same vulnerability the PR
fixed for OpenAI/OpenRouter, in a sibling provider it missed.
2. Strengthen the regression tests from pass-on-any-ValueError to true
security invariants: spy open()/read_bytes() and assert the blocked
credential is NEVER read; add negative controls (legit local image
still loads; remote/data: URIs pass through unguarded) so a
block-everything regression can't pass.
3. Guard is best-effort by design (defense-in-depth, not a security
boundary) — documented on the shared helper.
- agent/file_safety.py: raise_if_read_blocked()
- plugins/image_gen/{openai,openrouter,xai}: route through helper
- tests: no-read spies + negative controls across all three providers
Five follow-ups to #57659 from post-merge review:
1. install.ps1: gateway scheduled-task re-enable now runs in a finally
(a thrown Remove-Item/uv venv failure previously stranded the user's
gateway autostart disabled), and tasks that were already disabled
before the install are no longer blindly re-enabled.
2. The venv-python holder guard is no longer bypassed by plain --force
(which the desktop bootstrap passes on every update while its lock
probe only checks hermes.exe/app.asar). New explicit --force-venv is
the escape hatch; --force keeps bypassing only the hermes.exe shim
guard.
3. _detect_venv_python_processes now also catches uv/base-interpreter
trampolines whose exe is outside the venv, via cmdline (venv path or
'-m hermes_cli.main' tied to this install root) and cwd.
4. Missing venv python is now UNHEALTHY on managed installs
(.hermes-bootstrap-complete / .update-incomplete markers) so the
repair lane runs instead of 'Already up to date!'; the repair branch
recreates the venv first when it's gone entirely. Dev checkouts keep
reporting healthy.
5. install.ps1 comment no longer claims a Startup-folder disarm the
code doesn't perform (logon-only, not a mid-install respawner).
The salvaged tests from #53754 predate _handle_vision_analyze becoming
async and the native fast path; await the handler and force the legacy
aux path so the model-resolution assertion is actually exercised.
_handlers for vision_analyze and video_analyze read model name from
config.yaml (auxiliary.vision.model / auxiliary.video.model) before
falling back to AUXILIARY_VISION_MODEL / AUXILIARY_VIDEO_MODEL env
vars. Matches the existing config-first pattern for timeout and
temperature in the same file.
Fixes#53749
A custom:<name> main provider resolves at runtime to the bare provider id
"custom". In the vision auto-detect chain, the main-provider branch called
resolve_provider_client("custom", ...) WITHOUT explicit_base_url/api_key,
so it returned (None, None) ("no endpoint credentials found") and the whole
chain fell through to OpenRouter/Nous. A user on a custom endpoint with no
aggregator configured then got "No LLM provider configured for task=vision
provider=auto" on every image, even though their main model fully supports
vision.
Recover the live endpoint that set_runtime_main() records each turn
(_RUNTIME_MAIN_BASE_URL/_API_KEY/_API_MODE) and forward it to Step 1, with
a fallback to _resolve_custom_runtime() for non-gateway callers. Mirrors the
existing explicit-base_url branch directly above.
Adds TestResolveVisionCustomProvider covering custom, custom:<name>, and the
no-runtime fallback path.
When model.provider is set to custom:<name>, _supports_vision_override()
previously tried only the runtime provider key ('custom') and the raw
config value ('custom:my-proxy'). It did not try the stripped name
('my-proxy'), which is the actual key under providers: in config.yaml.
This caused native image routing to fall back to text mode even when the
user explicitly declared supports_vision: true on the named provider's
model entry.
Fixes#39963
A user with tts.openai.model set to a direct-OpenAI model (e.g. tts-1-hd)
but no VOICE_TOOLS_OPENAI_KEY/OPENAI_API_KEY (or with tts.use_gateway)
routes TTS through the managed Nous audio gateway, which only proxies
gpt-4o-mini-tts. The request 400s with:
VALIDATION_ERROR: Unsupported managed OpenAI speech model
{'model': 'tts-1-hd', 'supportedModels': ['gpt-4o-mini-tts']}
_resolve_openai_audio_client_config now reports whether it resolved the
managed gateway; _generate_openai_tts coerces the model to a
managed-supported one (logging a warning that points at the direct-key
escape hatch) unless the user redirected base_url to their own endpoint.
Direct-key users keep their tts-1/tts-1-hd preference unchanged.
Follow-up to #57507: .ENV / .Env.local on case-insensitive filesystem
mounts slipped past the guard. Lowercase the name before matching and
add a regression test. Addresses egilewski's open review note.
`_dispatch_sync` gathers the mautrix per-event handler tasks with a bare
`asyncio.gather(*tasks)`. Without `return_exceptions=True`, the first handler
that raises aborts the gather, so the sibling events in the same sync response
are dropped unprocessed — the exception propagates up to the sync loop, which
logs a single "sync error" and moves on. The invite/redaction gathers a few
lines above already use `return_exceptions=True`.
Use `return_exceptions=True` and log each failing handler, so one bad event no
longer takes out the rest of its batch and per-event failures stay visible.
Regression test: a batch with one failing and one succeeding handler no longer
raises, the good handler still runs, and the failure is logged (mutation-
verified — reverting re-raises RuntimeError out of _dispatch_sync).
Source: https://github.com/NousResearch/hermes-agent/pull/52346
Related prior work: https://github.com/NousResearch/hermes-agent/pull/39462
Related prior work: https://github.com/NousResearch/hermes-agent/pull/27426
Maintainer direction: https://github.com/NousResearch/hermes-agent/pull/52346#issuecomment-4854881612
Remove acp_command and acp_args from the model-facing delegate_task schema and
dispatch paths. Child agents can still use ACP subprocess transport when it
comes from trusted delegation config or parent inheritance, but a model tool
call can no longer choose the command or arguments that reach child
construction.
This is salvageable because the risky boundary is model control over child ACP
transport, not ACP itself. The patch follows the maintainer direction from the
source discussion by preserving trusted ACP configuration and prior integration
work while removing the untrusted tool-call fields from both top-level and
per-task delegate inputs.
Reproduced on main by passing acp_command through delegate_task and observing it
reach _build_child_agent. Verified after the fix that model dispatch strips the
hidden top-level fields and per-task hidden fields are ignored before child
construction.
Co-authored-by: Carlosian <claudlos@agentmail.to>
Co-authored-by: ssiweifnag <120658181+ssiweifnag@users.noreply.github.com>
Co-authored-by: nikshepsvn <23241247+nikshepsvn@users.noreply.github.com>
Replace the exact-filename frozenset with _is_sensitive_filename()
that matches .env plus any .env.<suffix> variant. This covers
shorthand suffixes like .env.prod that the previous enumeration
missed.
Add test_sensitive_env_suffix_variants_blocked regression test
covering .env.prod, .env.dev, .env.staging.local, and .env.ci.
Addresses review feedback from egilewski on PR #57507.
The dashboard Files tab could list, read, and download .env files
containing API keys when running with a bind-mounted Hermes home
directory (e.g. docker run -v ~/.hermes:/opt/data).
Add _SENSITIVE_FILENAMES frozenset and filter these from
list_managed_files(), read_managed_file(), and download_managed_file().
Return 403 for direct read/download attempts on sensitive files.
Fixes#57505
_resolve_media_to_data_urls's ad-hoc _MEDIA_TAG_RE matched any bare
token after MEDIA: (no absolute-path anchor) and read the resolved
path directly with no denylist. A relative/traversal path like
MEDIA:../../../../etc/passwd.png slipped through, and any image-
suffixed file the process could read (including under ~/.ssh, ~/.aws,
etc.) was base64-inlined into the API response if its path merely
appeared in the model's own final reply text.
Every other platform adapter's MEDIA: handling already goes through
two shared primitives in gateway/platforms/base.py:
- MEDIA_TAG_CLEANUP_RE, which anchors the path to ~/, /, or a
Windows drive letter plus a known deliverable extension.
- validate_media_delivery_path, which resolves symlinks and rejects
paths under the credential/system-path denylist.
Reuse both here instead of the local unanchored pattern and naive
Path().expanduser() resolution.
browser_cdp's frame_id (OOPIF) path returned early via
_browser_cdp_via_supervisor before _browser_cdp_private_guard ever ran,
unlike the stateless path a few lines below. A model that navigated a
cloud browser to a private/internal URL could still read page content
by passing frame_id, bypassing the same SSRF/private-page boundary
already enforced on Runtime.evaluate, Page.navigate, and other raw CDP
calls.
Apply the same guard call used by the stateless path before dispatching
to the supervisor, so both routing modes share one boundary.
Root-causes the July 2026 Windows incident chain (locked _brotlicffi.pyd /
_sodium.pyd during install, then 'No module named annotated_doc' with
'hermes update' insisting 'Already up to date!'):
- hermes update: probe venv core imports even when the checkout is current;
a half-updated venv (dep sync killed mid-flight by a locked .pyd) is now
detected and repaired instead of being reported as up to date
- hermes update (Windows): after pausing gateways, refuse to mutate the venv
while other processes run from the venv interpreter (the Desktop backend
runs as python.exe so the hermes.exe shim guard never saw it); --force
keeps the old behavior
- install.ps1 venv stage: disarm gateway autostart Scheduled Tasks before
the kill sweep (they respawn the gateway inside the kill->delete window),
make the sweep a bounded loop requiring 3 clean passes, and rename-then-
delete the old venv (a rename succeeds even with mapped DLLs) with stale-
dir cleanup on the next run
- desktop updater: 'venv shim still locked after 15s' now ABORTS the update
hand-off (restarting our backend, surfacing the holder to the user)
instead of 'proceeding anyway (force)' into guaranteed venv corruption;
the unlock wait also re-kills respawned backends each poll tick
A Cursor-style MCP manager inside Capabilities, plus the backend it needs.
- Server list with brand/favicon avatars + live status dot and a capability
summary (N tools, M prompts, K resources); Servers | Catalog views.
- Catalog: one-click install of Nous-approved servers with required-env prompts.
- GUI OAuth: Authenticate opens the system browser from the TTY-less backend and
verifies a token actually lands; header/API-key servers are never pushed down
OAuth; a dirty mcp.json can't drop a freshly-persisted auth field.
- Full-width mcp.json editor (ecosystem document format) + pinned stdio/agent
LogTail; probes cached 5m and keyed by (profile, config) so revisiting never
respawns the fleet or shows a stale probe.
- Whole-map persistence (PUT /api/mcp/servers) so deletes/toggles actually stick
(the generic /api/config deep-merge could not remove keys).
- perf: MCP probe/auth no longer hold the global skills lock, so a slow stdio
spawn can't stall every other request into a 15s timeout.
- per-tool include/exclude gating (lib/mcp-tool-filter) mirroring the CLI loader.
The routing-heal added to get_or_create_session calls
SessionDB.get_compression_tip; the stale-guard suite's bare MagicMock db
returned a Mock the heal then assigned as session_id, failing JSON
serialization. Model the real contract (a non-compressed session's tip is
itself) so the heal is a correct no-op.
Inbound Telegram/WeChat/Discord messages are written by the background
gateway, not the desktop websocket that drives local chats. Without
explicit polling the messaging sidebar and the open transcript stay
frozen until the user manually refreshes.
Desktop:
- MESSAGING_POLL_INTERVAL_MS (10 s): interval poll of the messaging
session list so new platform sessions surface automatically.
- ACTIVE_MESSAGING_SESSION_POLL_INTERVAL_MS (5 s): poll the currently-
viewed messaging transcript and re-hydrate the chat state when the
FNV-1a signature changes (hash covers role + timestamp + content).
- sameCronSignature now compares lineage_root_id / source / profile /
preview / message_count / last_active / ended_at so stale previews
and activity times are no longer silently ignored.
- sessionMatchesStoredId helper de-dups the id / _lineage_root_id check.
- refreshMessagingSessions exposed from useSessionListActions so the
controller can use it in the poll effect.
Gateway:
- SessionStore._compression_tip_for_session_id: look up the latest
compression continuation for a session id.
- SessionStore._heal_compression_tip_locked: rewrite a stale entry to
the compression child before returning it, so a restart or failed send
no longer leaves the store pinned to the compressed parent.
Co-authored-by: lawyer112 <lawyer112@users.noreply.github.com>
Follow-up to the salvaged early-exit retry fix (#35617): the debug-browser
launch path was fire-and-forget (stderr to DEVNULL, no logging), so every
platform failure — Windows singleton forward to an existing instance, bad
profile dir, missing shared libraries, policy blocks — collapsed into the
same unactionable 'port 9222 isn't responding yet' message and debug
reports contained nothing.
- launch_chrome_debug() returns a structured ChromeDebugLaunch with
per-candidate attempts (state, exit code, stderr tail)
- browser stderr is captured to <hermes_home>/chrome-debug/launch-stderr.log
- clean exit (code 0) without the port opening is detected as Chromium's
single-instance forward and produces a targeted user hint to close all
running instances of that browser
- crash exits surface the stderr tail (e.g. missing libnspr4.so)
- every spawn/exit is logged to agent.log so hermes debug share captures it
- CLI (/browser connect) and TUI/desktop (browser.manage) both print the hint
* feat(desktop): CLI/dashboard parity — skills hub browser, MCP test/toggle/catalog, maintenance ops, log filters
Brings desktop GUI to parity with hermes skills/mcp/doctor/backup/debug-share/
curator/memory CLI commands and the dashboard's System + Skills-hub pages:
- Skills page: new Browse Hub tab (search official/GitHub/community sources,
preview SKILL.md, security scan verdicts, install/update with live action log)
- MCP settings: connection test (tool listing), per-server enable/disable
toggle, and a Catalog tab installing Nous-approved MCP servers with env prompts
- Command Center: new Maintenance section (doctor, security audit, backup,
debug share links, curator status/pause/run, memory file status + reset)
- Command Center system logs: file (agent/errors/gateway/desktop), level, and
substring filters instead of a fixed agent.log tail
- hermes.ts API client + types for all the above; en/zh locale strings (ja and
zh-hant inherit via defineLocale)
* feat(desktop): backend model catalogs in toolset config — hermes tools parity
Completes the `hermes tools` parity gap: after picking an image/video
generation backend the CLI runs a model picker (e.g. FAL's multi-model
catalog with speed/strengths/price); the desktop toolset drawer now has the
same flow as a radio-card list.
- web_server: GET /api/tools/toolsets/{name}/models (catalog + current +
default for the active or named provider row) and PUT .../model
(validated write to image_gen.model / video_gen.model), reusing the CLI's
plugin catalog helpers so GUI and `hermes tools` stay in lockstep
- desktop: ModelCatalogPicker in ToolsetConfigPanel — per-model cards with
speed/strengths/price, in-use + default badges, disabled until the
backend is the active one; provider selection now mirrors is_active
locally so the catalog unlocks without a refetch
- tests: 3 backend endpoint tests (catalog shape invariants, persist +
validation), 2 component tests, 2 API-contract tests; en/zh strings
OpenCode Go serves minimax/qwen via Anthropic Messages (base URL without
/v1 — the SDK appends /v1/messages) and glm/kimi/deepseek/mimo via OpenAI
chat completions (base URL WITH /v1). The runtime stripped /v1 for
anthropic-routed models, and the TUI/desktop + gateway persisted that
stripped URL to model.base_url. Every later chat_completions model then
POSTed to https://opencode.ai/zen/go/chat/completions — a 404 (the
marketing site). Result: only minimax worked; glm/deepseek/kimi all 404ed.
- New normalize_opencode_base_url(): symmetric /v1 normalization —
strip for anthropic_messages, re-append for chat_completions /
codex_responses on opencode.ai hosts (heals persisted stripped URLs;
custom proxy overrides untouched)
- Applied at all three former one-way strip sites (resolve_runtime_provider
x2, switch_model)
- opencode_model_api_mode: all Qwen models on Go AND Zen now route via
/v1/messages per current published endpoint tables (previously only
qwen3.7-max on Go — qwen3.6-plus etc. would 404 the same way)
- Catalog refresh: Go gains deepseek-v4-pro/flash, glm-5.2,
kimi-k2.7-code, minimax-m3, qwen3.7-plus; Zen gains glm-5.2,
kimi-k2.7-code, minimax-m3, qwen3.7-plus
Reported by IndieSuperhuman on X: opencode-go 404s for any model other
than minimax.
A single-model Hermes agent never sends temperature; the provider default
applies. MoA hardcoded reference_temperature=0.6 / aggregator_temperature=0.4,
and the coercion float(preset.get(key, 0.6) or 0.6) made unset IMPOSSIBLE to
express: absent, null, empty, and even an explicit 0 all collapsed to the
baked-in default. Every MoA advisor and aggregator therefore ran at 0.6/0.4
while the same model running solo used the provider default — silently
skewing solo-vs-MoA comparisons and overriding provider-tuned defaults.
- moa_config normalization: temperatures coerce to None when absent/blank/
invalid (new _coerce_float_or_none); explicit values incl. 0 honored.
- moa_loop: _preset_temperature() resolves preset values; None flows to
call_llm, which already omits the parameter when None (same contract as
max_tokens). Aggregator still inherits the acting agent's own configured
temperature when the preset doesn't pin one.
- conversation_loop (context-mode MoA): same resolution, no more hardcoded
0.6/0.4 at the call site.
- DEFAULT_CONFIG preset + web_server payload models + docs updated: unset
is the default, pinning stays available.
Follow-up to the per-site strips from the review gate. The two copy-site
strips are correct but positional — a copy site added after the assembly
loops would re-leak _db_persisted into the child-session flush. Add a single
terminal sweep (_strip_persistence_markers) run once on the fully-assembled
compressed list so the invariant 'no compacted message leaves compress()
carrying a persistence marker' is structural, not dependent on copy-site order.
- agent/context_compressor.py: _strip_persistence_markers() called before
compress() returns; helper docstring notes the sweep is the authoritative guard
- tests/agent/test_context_compressor.py: structural regression — neuter the
per-site helper to a leaking copy, assert the terminal sweep still strips
- tests/run_agent/test_compression_persistence.py: pin the fixture assumption
behind the exact-equality row-count assertion
Shallow messages[i].copy() during context compression propagated the
_db_persisted marker from cached gateway incremental flushes into the
post-rotation compressed list. _flush_messages_to_session_db then skipped
every row when writing to the new child session, so gateway restarts
lost the compacted transcript (severe amnesia).
Strip the marker in _fresh_compaction_message_copy() and add regression
tests for rotation flush + compressor assembly.
Fixes#57491
The reaction-guard regression test defined a local _should_react lambda and
asserted it against itself — a tautology that would stay green even if the
production guard at _handle_slack_message reverted to (is_dm or is_mentioned),
re-introducing the unmentioned-MPIM reaction spam this PR fixes.
Replace it with a shared _reaction_guard helper plus a source-introspection
test that pins the production expression: asserts (is_one_to_one_dm or
is_mentioned) is present and (is_dm or is_mentioned) is absent. Mutation-checked
— reverting the adapter guard now fails the test.
Follow-up self-review finding on the salvage of #57339.
Group DMs (MPIMs) were classified as DMs and thereby exempted from every
operator control that shared surfaces are supposed to honor: allowed_channels,
require_mention, strict_mention, free_response_channels, and the reaction
guard. Symptom: the bot added 👀/✅ to unmentioned MPIM
messages and still invoked the agent (which then returned NO_REPLY) instead of
the gateway dropping the event before model execution. Removing an MPIM from
allowed_channels did not disable it.
Root cause is the DM classification at adapter.py:
is_dm = channel_type in {"im", "mpim"}
used for BOTH routing exemptions and reaction gating. An MPIM is a shared
surface (multiple humans can see and trigger the bot), not a private 1:1 DM,
so it must be gated like a channel.
This behavior was introduced/reinforced by a trail of Slack group-DM PRs:
- #4633 fix(slack): treat group DMs (mpim) like DMs + reaction guard
- #54632 fix(slack): subscribe to message.mpim + mpim scopes so group DMs work
- #54663 fix(slack): group DMs work OOTB + reinstall nudge
#54632/#54663 correctly made MPIM messages *reachable*; #4633 over-reached by
giving them the DM mention/reaction *exemptions*. This corrects only that
over-reach.
Fix (minimal): introduce `is_one_to_one_dm = channel_type == "im"` and key the
two EXEMPTION sites off it instead of `is_dm`:
- mention/allowlist gating block (`if not is_one_to_one_dm and bot_uid:`)
- reaction guard (`(is_one_to_one_dm or is_mentioned)`)
`is_dm` is intentionally retained for session/thread scoping and chat_type
labeling, where treating an MPIM as a persistent multi-party conversation is
correct — only the mention/reaction exemptions were wrong.
Docs: slack.md now distinguishes 1:1 DMs (mention-exempt) from group DMs
(shared surface; obey require_mention/strict_mention/allowed_channels/
free_response_channels; reactions only when @mentioned).
Tests: +7 in test_slack_mention.py (MPIM unmentioned dropped under
require_mention and strict_mention; MPIM mentioned processed; MPIM off
allowed_channels dropped; MPIM in free_response opted in; 1:1 IM still exempt;
reaction guard drops unmentioned MPIM). Updated _would_process to model the
is_one_to_one_dm gating + strict_mention. 72 passed.
PeopleConnect is the exception to deletion-beats-suppression: "DELETE MY
USER DATA" also deletes suppressions on file, and deletion does not stop the
people-search sites from showing you (public records re-list). Suppression
is the effective lever and must be maintained.
- intelius.json: deletion.prefer=false; playbook/quirks/notes rewritten with
the verbatim privacy-center language; delete is the data-purge-only path.
- autopilot: honor deletion.prefer -> prefer_suppression when false.
- methods.md / SKILL.md / README: exception called out.
- tests updated + prefer-flag routing test (86 tests).
unbroker finds where a consenting person's info is exposed across data
brokers and people-search sites and files the removals, running as far as
each site allows and handing only genuinely human-only steps (hard CAPTCHA,
gov-ID, phone, fax) back as an end-of-run digest.
- Deterministic stdlib CLI (scripts/pdd.py) owns config, dossiers+consent,
the broker DB, tier planning, the ledger, email, and the autonomous
action queue; the agent scans/submits with native tools (web_extract,
browser_*, delegate_task, cronjob, terminal).
- Verify-before-disclose, least-disclosure (never volunteers SSN), consent
gate, opaque ids, optional age-at-rest encryption, file-locked ledger.
- Jurisdiction-aware (CCPA/CPRA, GDPR, generic); CA DROP one-shot covers
the state registry (~545) in a single request; BADBOOL + curated
people-search coverage; scheduled re-scan for re-listing.
- No CAPTCHA-solving services or anti-bot bypass; browser email mode needs
no stored password.
- 85 hermetic tests (tests/skills/test_unbroker_skill.py; SMTP/IMAP via
injected fakes, registry via CSV fixtures). Ships placeholder data only.
Broker dataset adapted from BADBOOL (Yael Grauer, CC BY-NC-SA 4.0).
hermes debug share reads os.getenv — the invoking terminal's environment — but
launchd/systemd and the desktop-spawned `serve` backend load credentials from
~/.hermes/.env, not the login shell. A key exported in the shell but absent
from .env is invisible to the backend, yet the dump printed a bare "set",
sending support down a phantom "the key is configured" path.
This was the actual trap behind a "Desktop has no web_search / no tools"
report: FIRECRAWL_API_KEY was a shell export (so `debug share` in a terminal
read "firecrawl set") but not in .env, so the launchd backend's
check_web_api_key returned False and web_search was gated off — which a
contributor then misdiagnosed as a missing `desktop` platform registration.
The dump now annotates any key set in-process but missing from ~/.hermes/.env
with "(shell only — not in .env; managed/desktop backend may not see it)" so
the mismatch is obvious instead of hidden behind "set".
The merged webhook session-close fix (#57370, salvaging #57322) wrapped
handle_message in a try/finally — but BasePlatformAdapter.handle_message
is fire-and-forget: it spawns _process_message_background and returns
before the agent run starts. The finally-close therefore ran BEFORE
get_or_create_session created the session row, found no session_id, and
silently no-op'd — the ghost-session leak persisted on the real path.
(The shipped test masked this by stubbing handle_message with a fake
that created the row synchronously.)
Move the close to an on_processing_complete override — the lifecycle
hook the base class fires at the TRUE end of the run, on the success,
failure, and cancellation paths alike. Empirically verified through the
real fire-and-forget pipeline: before, ended_at stayed NULL; after,
ended_at is set with end_reason=webhook_complete and the row is
prunable.
Tests now stub only the runner-side _message_handler (the seam the live
gateway injects) so handle_message / _process_message_background /
on_processing_complete all run for real; adds an AsyncSessionDB-facade
coverage test for the coroutine-await branch.
Follow-up to @helix4u's #57336 salvage. Two review findings:
- W1: model-picker grouped custom-provider rows by
(api_url, credential, api_mode) but NOT extra_headers. Entries sharing a
URL+credential+api_mode yet declaring different headers (e.g. per-tenant
routing behind one proxy) collapsed into one row and probed /models with
whichever header set was seen first (order-dependent). Fold a canonical
header identity into group_key so distinct header-authed endpoints stay
separate; drops the now-dead first-non-empty merge branch.
- W2: the extra_headers stringify+None-filter comprehension existed in 5
copies (config.py x2, runtime_provider.py, model_switch.py, models.py).
Extract one shared hermes_cli.config.normalize_extra_headers primitive;
all sites now call it.
Tests: +normalize_extra_headers unit tests, +regression test proving two
same-endpoint entries with different headers stay distinct and each probes
with its own headers. 223 targeted tests pass; ruff clean.
Follow-up to the cherry-picked #31856 fix. The contributor's guard defers
idle-TTL eviction until the session store reports the session expired, so the
expiry watcher can tear the agent down and fire MemoryProvider.on_session_end()
with the live transcript. Two gaps remained:
1. Memory-leak regression for mode='none' sessions. _is_session_expired()
returns False forever for the 'none' reset policy, so the naive guard would
never idle-evict those agents — reopening the unbounded-cache leak the idle
sweep (#11565) exists to relieve. Added SessionStore.is_session_finalizable()
(a public predicate: will the expiry watcher EVER finalize this session?) and
gate the deferral on it. mode='none' agents fall through to soft eviction as
before.
2. on_session_end still dropped on the LRU-cap path. Both cache-pressure paths
(_enforce_agent_cache_cap and _sweep_idle_cached_agents) soft-evict via
_release_evicted_agent_soft, which by design does NOT fire on_session_end.
If cache pressure evicts a finalizable-but-not-yet-expired agent before it
expires, the watcher later finds no cached agent and the hook is skipped.
Added _commit_memory_before_soft_evict(): at LRU eviction, if the session is
finalizable and not yet expired, commit end-of-session extraction via the
live agent's own (fully-scoped) memory manager using commit_memory_session()
— extraction WITHOUT provider teardown, so the eviction stays soft and a
resumed turn keeps working. Skipped for mode='none' (no missed boundary to
compensate) and expired sessions (the watcher tears those down directly).
This closes#11205 for ALL eviction paths and reset policies, not just the
idle-sweep + finite-policy case, while preserving the soft-eviction
resumability contract (never calls close() on a live session).
Tests: 5 new cases in test_agent_cache.py (mode='none' still reaped, LRU-cap
commits for finalizable / skips for none, real is_session_finalizable
predicate); all mutation-checked. Contributor's original 2 tests updated to
assert the finalizable path explicitly.
The idle-TTL sweep (_sweep_idle_cached_agents) was evicting agents
as soon as they passed _AGENT_CACHE_IDLE_TTL_SECS, even when the
session hadn't expired yet. In daily-reset mode the reset can fire
hours after the last user message — evicting the agent early means
the session-expiry watcher has no agent in cache to call
on_session_end() with, so memory providers miss the live transcript.
Now the sweep checks the session store before evicting: if the
session still exists and hasn't expired, the agent stays in cache
so the expiry watcher can tear it down properly later.
When the session store is unavailable or throws, falls back to the
original eviction behavior (safe default).
Fixes: #11205
Self-review (ruff+ty lint diff = 0 net-new; 2-agent deep review) surfaced one
Warning + comment-accuracy nits; no Critical:
- W1: the local-probe TTL cache memoized None (probe failure) for 30s, so a
probe that failed during a startup race would suppress a legit retry once
the server came up. Cache only positive results — still fully bounds the
hot-path probe rate (reachable servers cache their value) while an
unreachable one re-probes on the next call. Add a regression test asserting
a None result is NOT cached (retry re-probes); mutation-verified.
- Tighten the platform-guard comment: gateway/TUI/cron already construct with
quiet_mode=True (gated by `not agent.quiet_mode`), so the guard's active job
is CLI dedup vs show_banner, not "filling the gateway/TUI gap" as originally
worded.
Verified not-issues (per review): positive-value 30s cache does not break the
reconcile-after-restart freshness contract (restart = fresh process, empty
cache); cache key is collision-safe; platform guard is correct in both
directions (no runtime path leaves platform None on a non-CLI surface).
Tests: 149 passed. ruff clean; ty 0 net-new vs base.
Desktop/dashboard WebSocket connections drop during long agent operations
(delegate_task subagents, large model outputs) when the uvicorn event loop is
GIL-starved for minutes. Root cause: uvicorn's ws keepalive ping runs on the
SAME event loop as agent turns. A single synchronous GIL-holding call on a
worker thread (a regex/scrub over a large output, or a long subagent turn)
freezes the loop, so it cannot process the incoming pong within ws_ping_timeout
and uvicorn closes an otherwise-healthy connection (#53773: 'event loop stalled
226.3s'; #48445/#50005). Loosening the timeout only raises the threshold — a
multi-minute stall sails past any finite window.
The keepalive ping exists to detect half-open connections (reverse-proxy 524,
dropped tunnels), which cannot happen on loopback: there is no network or proxy
in the path, and a dead local client tears the socket down with a real FIN/RST
that starlette surfaces as WebSocketDisconnect regardless of the ping. So on
loopback the ping provides ~no liveness value while actively killing
recoverable stalls — disable it entirely (ws_ping_interval/timeout=None).
Non-loopback (public) binds sit behind a Cloudflare Tunnel where half-open IS a
real failure mode, so the ping stays at 20/20 to detect it.
Empirically verified (real uvicorn + websockets peer): with ws_ping=None the
server never closes a silent peer during an 8s window; with the pre-fix 2s/2s
window uvicorn closes it. A genuinely-dead client still fires the
WebSocketDisconnect reap path regardless of the ping.
Note: this fixes the local Desktop case (the OP's scenario). A remote Desktop
over an authenticated public dashboard route (McCalebTheSecond's comment) keeps
the ping and needs the deeper GIL-hotspot fix — tracked separately.
Closes#53773
Salvage review of #56431 surfaced one Critical + two Warning issues; fix
them on top of the contributor's cherry-picked commits:
1. Critical — duplicate non-agentic warning on the interactive CLI. The new
agent_init warning fires on every platform, but cli.py show_banner()
already warns on CLI (richer output + /model hint), so a CLI user saw the
warning twice per startup. Guard the agent_init emit to skip platform=="cli"
— it now fills exactly the gateway/TUI gap the PR intended, no duplication.
2. Warning — vLLM error-parse regex under-matched. The patterns required a
literal space before the number, so "max_model_len: 32768", "=32768",
"(32768)", and "... is 32768" all returned None. Broaden both patterns to
accept :/=/(/ 'is' delimiters. Add a parametrized test over all delimiter
variants.
3. Warning — per-call live probe latency on local endpoints. The new
reconcile-on-hit + pre-defaults step-7 probe made every local resolution
fire a synchronous network probe (banner + /model switch + compressor
update_model each within one startup). Add a 30s in-process TTL cache
keyed by (model, base_url) around _query_local_context_length so back-to-
back resolutions reuse one round-trip; not persisted to disk, so the
reconcile freshness contract (re-probe after restart) is preserved. Add an
autouse fixture clearing the cache between tests + TTL coverage.
Tests: 148 passed (was 138). ruff clean.
Replace the webhook delivery-close path's direct reach into private
SessionStore._entries (which also bypassed the store lock) with a public,
lock-held peek_session_id(session_key) accessor. Mirrors the existing
lookup_by_session_id inverse helper. Keeps a getattr fallback for older
stores / test doubles. Adds a unit test for the accessor.
Webhook deliveries created a unique one-shot session (delivery_id baked into
the session key at gateway/platforms/webhook.py:668) but the adapter fired
handle_message via asyncio.create_task WITHOUT ever ending the session
(webhook.py:713, pre-fix). Nothing else closes it: the gateway caches/expires
the agent per session_key but never calls end_session for the webhook path,
and _end_session_on_close teardown doesn't run for these fire-and-forget tasks.
SessionDB.prune_sessions (hermes_state.py:4965) only deletes rows WHERE
ended_at IS NOT NULL. So every webhook session stayed with ended_at NULL ->
unprunable -> unbounded state.db growth. This was the primary driver of the
SQLite lock-contention gateway outage.
Fix: wrap the delivery in _run_delivery_and_close, which awaits
handle_message and then (in finally, so failures still reap) calls
_end_webhook_session -> SessionDB.end_session(session_id, 'webhook_complete').
This mirrors how cron closes its session with 'cron_complete'
(cron/scheduler.py:3065). end_session is first-reason-wins and no-ops on an
already-ended row, so it never clobbers a compression/agent_close reason.
Adds tests/gateway/test_webhook_session_close.py asserting the invariant
(a completed webhook session has ended_at set + is prunable), including the
error-path case, against a real SessionStore + SessionDB.
Add regression tests for vLLM max_model_len error parsing, stale local
cache reconciliation, live probes over llama defaults, and the 64K minimum
guard on persistent cache writes.
(cherry picked from commit 1cb47ef437de7ce289cb358e8d6b89e9194b43ed)
Refine the blank-line handling so a blank line only continues a list run
when the next non-blank line is another list item. This keeps a list ->
paragraph -> list sequence as three separate blocks and matches the
contiguous-list layout for mixed/nested lists (one rich_text block, split
into sub-lists by (indent, ordered)), rather than emitting a separate
block per item.
Adds regression tests for the mixed blank-separated layout and the
list->paragraph->list boundary.
When a Markdown ordered list has blank lines between items (common in
LLM-authored content), the list run loop breaks on each blank line.
Slack numbers each rich_text_list independently, so N items produce N
lists each starting at 1.
Skip blank lines inside the list run as soft separators instead of
breaking, so ordered items stay in one rich_text_list and Slack renders
the correct numbering.
Fixes#57076
setup.runtime_check and setup.status are polled by the Desktop frontend on
connect and periodically (use-status-snapshot → evaluateRuntimeReadiness), but
neither was in _LONG_HANDLERS — so dispatch() ran both inline on the WS reader
thread. Under GIL pressure from concurrent agent turns (terminal I/O, large
output, background-process completions) either can block for seconds:
- setup.runtime_check → resolve_runtime_provider() (config read, auth check,
may probe the provider endpoint)
- setup.status → _has_any_provider_configured() (provider config + credential
scan)
While either blocks the reader thread the WS read loop can't service later
requests; the frontend RPC timeout fires, the client drops the socket, and the
lost setup.runtime_check response reads as ready=false — a false "needs setup"
/ "Settings failed to load" even though the provider is configured.
Route both to the RPC pool (same precedent as #55545's session.list/pet.info/
process.list). The handlers are read-only and pool writes go through the
lock-guarded write_json, so there's no ordering or safety concern.
Test asserts all 5 frontend-polled RPCs are pool-routed.
Co-authored-by: izumi0uu <izumi0uu@gmail.com>
delete_profile stopped only the process named in gateway.pid, but a Desktop
app spawns a headless `serve`/`dashboard` backend per profile that holds the
profile's SQLite connection open and keeps writing sessions/WAL/sandbox files.
That backend is never in gateway.pid, so a CLI `hermes profile delete` run
while the Desktop app is up left it writing into the tree — rmtree's final
rmdir then failed with ENOTEMPTY (#47368 "Bug 2"), and pre-guard it also
resurrected the directory.
- _profile_bound_backend_pids(): find running Hermes backends bound to this
profile via a `--profile <name>` selector or a HERMES_HOME env resolving to
the profile dir. Tightly scoped — current-user only, backend subcommands
(serve/dashboard/gateway) only so an interactive chat is never killed, and
never this process or its ancestors.
- _stop_profile_backends(): terminate them (graceful, then force), best-effort
so it can never make delete worse.
- _rmtree_with_retry(): a few spaced retries absorb the ENOTEMPTY / Windows
file-lock race from a just-terminated writer's in-flight -wal/-shm/sandbox
writes instead of failing the whole delete on a race the next attempt wins.
Complements the recreation guard (deleted profiles no longer reappear) and the
Desktop teardown-before-delete flow; this is the CLI-side convergence fix for a
delete run while a Desktop-managed backend is live.
Part of #47368.
A Z.ai desktop user reported thinking reverting to medium after one turn,
burning ~200% of a week's credits in 4 days despite reasoning_effort: false
in config.yaml. Four compounding bugs:
- _session_info reported reasoning_effort "" for disabled reasoning,
indistinguishable from unset — the desktop adopted it after the first
turn, wiping its sticky "thinking off" pick so every later chat
reverted to the default effort.
- config.set key=reasoning always wrote agent.reasoning_effort to global
config.yaml, so every desktop model-menu selection (preset.effort ??
'medium') clobbered the user's configured value. Now session-scoped
like the messaging gateway's /reasoning, landing on
create_reasoning_override so lazily-built sessions keep it too.
- YAML `reasoning_effort: false`/`off`/`no` (boolean False) was coerced
to "" by every loader's `str(x or "")`, silently re-enabling thinking.
parse_reasoning_effort now treats False/"false"/"disabled" as
{"enabled": False}; loaders (tui gateway, gateway, cli, cron,
delegate) pass the raw value through. The desktop config reader also
crashed on the boolean (false.trim()), aborting voice/STT settings.
- The zai provider profile never sent thinking on the wire, and GLM-4.5+
defaults to thinking ON server-side — so disabling reasoning was a
silent no-op on direct Z.ai, the actual token burner. The profile now
emits extra_body.thinking {"type": "enabled"|"disabled"} for
thinking-capable GLM models, mirroring the DeepSeek profile.
Also: /new (session reset) now carries reasoning_config across the
rebuild like model_override; config.get reasoning prefers the session's
live value and maps a config False to "none"; Settings shows "Off"
instead of a blank select for hand-written false.
The terminal-refresh quarantine filtered in-memory entries on
source == "device_code" but built removed_ids from the deleted
"loopback_pkce" source name, so the revoked device-code entry was
never pruned from the persisted pool in auth.json. Also restores the
_print_loopback_ssh_hint test suite scoped to Spotify (the helper's
remaining caller) instead of deleting it wholesale.
Replace the loopback/PKCE-callback server and manual-paste fallback with
the RFC 8628 device-code flow as the only xAI Grok OAuth login path. The
flow works in headless/SSH/container sessions with no 127.0.0.1 listener,
shrinking the local attack surface.
- Poll the token endpoint with server-provided interval, honoring
slow_down and expires_in; store tokens with auth_mode
oauth_device_code.
- Adaptive proactive refresh skew for short-lived device-code JWTs;
rotated tokens sync back to auth.json, the global root store, and the
credential pool (no refresh-token replay).
- Clear source suppression on successful re-login (CLI + dashboard) and
drop the duplicate dashboard pool entry so exactly one seeded
device_code entry exists.
- Use the shared device_code source name for consistency with the
nous/codex device-code providers.
- Desktop: remove the loopback OAuth flow states and dead type variants;
pkce providers' sign-in URL selection is unchanged.
- Docs (EN + zh-Hans) rewritten for device-code login; drop the deleted
--manual-paste flag from documented commands.
MSYS_NO_PATHCONV is honored by Git for Windows bash only. _find_bash's
final shutil.which fallback can return MSYS2-proper or Cygwin bash,
which ignore it and honor MSYS2_ARG_CONV_EXCL instead. Set both so argv
path conversion stays disabled regardless of which bash flavor spawns.
Also subsumes the cmd /c mangling in #56147.
Per-session /model overrides (_session_model_overrides) were in-memory only,
so a gateway restart silently reverted every session to the global default
model. Persist the non-secret parts (model/provider/base_url ONLY — never
api_key) into the session entry in sessions.json and lazily rehydrate them
on first use after a restart, re-resolving credentials through the normal
runtime provider resolution.
- gateway/session.py: SessionEntry.model_override field with
sanitize_model_override() (allowlist: model/provider/base_url) applied on
both serialization and deserialization; SessionStore.set_model_override /
get_model_override accessors. reset_session() already creates a fresh entry,
so /new keeps its clear-on-reset semantics — a restart cannot resurrect an
override the user reset away.
- gateway/slash_commands.py: write-through at both /model set sites (text
command + picker) after storing the in-memory override.
- gateway/run.py: _rehydrate_session_model_override() called from
_resolve_session_agent_runtime(); in-memory state always wins, credentials
are re-resolved per provider (credential-less fallback on failure). Session
expiry finalization also drops the persisted override.
- tests/gateway/test_session_model_override_persistence.py: restart
round-trip, /new clearing, api_key-never-serialized (including tampered
sessions.json), rehydration + live-state precedence + credential-failure
degradation.
Salvaged from #3659 by @Git-on-my-level, narrowed to the restart-persistence
gap confirmed in triage.
Named providers / custom_providers entries in config.yaml now accept an
extra_headers dict scoped to that endpoint — for reverse proxies, API
gateways, and custom auth schemes (e.g. Cloudflare Access service tokens).
- hermes_cli/config.py: normalize extra_headers on provider entries
(_normalize_custom_provider_entry + providers-dict translation), add
get_custom_provider_extra_headers /
apply_custom_provider_extra_headers_to_client_kwargs helpers keyed on
base_url (case/trailing-slash insensitive, no substring bypass —
mirrors the TLS helpers)
- hermes_cli/runtime_provider.py: surface extra_headers in the resolved
runtime for named custom providers (providers dict, legacy
custom_providers list, and the credential-pool path)
- run_agent.py / agent/agent_init.py: merge per-provider extra_headers
onto the OpenAI client default_headers at construction and on every
_apply_client_headers_for_base_url re-application (credential swaps,
rebuilds), most-specific level wins; OpenAI-wire only (native
Anthropic/Bedrock scoped out)
- agent/auxiliary_client.py: accept model.extra_headers as an alias of
model.default_headers for the global variant
- cli-config.yaml.example: documented commented example
- Header values are treated as secrets and never logged
Salvaged from PR #3526 by @jneeee, reimplemented against current main.
Co-authored-by: Teknium <127238744+teknium1@users.noreply.github.com>
Adds a no-code routing layer to the OpenAI-compatible API server so one
Hermes deployment can map different API clients to different
model/provider backends. Clients pick a backend by sending a configured
alias as the OpenAI 'model' field; unmatched values fall back to the
global model. Configured aliases are listed by GET /v1/models.
Precedence (highest first): session /model override > model_routes
route > global config. Route provider credentials resolve through
_resolve_runtime_agent_kwargs_for_provider (same seam as
channel_overrides); per-route api_key/base_url are upstream provider
credential overrides — never caller auth, never logged.
Salvaged and rebased from PR #3176 by @Mibayy onto current main.
Salvaged from PR #3243 by @Mibayy, reimplemented against current main
(the original diff targeted a removed gateway/run.py handler).
- /compact is now a first-class alias of /compress (CLI, gateway,
Telegram/Slack/Discord command lists, autocomplete) — also fixes the
dangling '/compact' references in gateway error messages
(gateway/run.py context-exhausted banners).
- --preview / --dry-run: report what WOULD be compressed (message
counts, token estimate, 'here [N]' boundary) without touching the
transcript. Flags coexist with the existing 'here [N]' / focus-topic
args on both the CLI and gateway surfaces via shared pure helpers in
hermes_cli/partial_compress.py.
- --aggressive (LLM-free hard truncation) is intentionally NOT
implemented: it would need its own transcript-persistence branch
outside the guarded _compress_context rotation machinery (#44794
data-loss class). The flag is recognized and returns an explanatory
message pointing at '/compress here [N]' and /undo instead of being
mis-parsed as a focus topic.
- locales: gateway.compress.aggressive_unsupported added to all 16
catalogs (parity test enforced).
- release.py: AUTHOR_MAP entry for contributor credit.
Salvage of #3459 by @keslerm, reimplemented against the restructured
progress-callback block in gateway/run.py (resolve_display_setting,
needs_progress_queue, thinking-relay). Duplicate PR #3458 by @dlkakbs was
submitted 4 minutes earlier with the same feature — both credited.
Co-authored-by: Dilee <uzmpsk.dilekakbas@gmail.com>
tool_progress: log keeps the chat silent and appends timestamped tool-call
lines to ~/.hermes/logs/tool_calls.log via a dedicated queue drained by an
async writer (RotatingFileHandler 5MB x 3, RedactingFormatter so secrets
never land on disk). Gateway-only by design; thinking_progress relaying and
the webhook gate are unaffected. /verbose now cycles
off -> new -> all -> verbose -> log.
Salvage of the surviving hunk of #3296 by @Mibayy. The PR's gateway
_handle_provider_command hunk targets code removed on main (/provider was
absorbed into /model + /status, which already read model.base_url); the
hermes status mislabel was the remaining live symptom:
_effective_provider_label() only checked the legacy OPENAI_BASE_URL env var,
so a custom endpoint configured canonically in config.yaml still displayed
as OpenRouter.
Replace the plugin-local _IMAGE_MAGIC_MIME table + _sniff_image_mime
body with a delegation to agent.image_routing._sniff_mime_from_bytes,
the canonical magic-byte sniffer already used across the codebase, then
gate its result to the raster formats gpt-image-2's Responses
input_image actually accepts (png/jpeg/gif/webp).
The shared sniffer also recognizes SVG/TIFF/ICO; without the allowlist
those would pass local validation and be rejected server-side with an
opaque HTTP 400. Gating locally fails them cleanly as invalid_image_input.
Adds a regression test for SVG rejection.
Follow-up on top of @CrazyBoyM's #55828.
Follow-up to #56874, which added the Camofox private-page SSRF guard
(_camofox_current_page_private_url) but wired it only into the Camofox
eval path (_camofox_eval). The other Camofox content-read tools —
camofox_snapshot, camofox_get_images, and camofox_vision — still read the
current page's accessibility tree / images / screenshot without the
guard, so on a non-local Camofox backend they can return the content of
an intranet or cloud-metadata page (e.g. 169.254.169.254) that the
terminal itself can't reach.
Apply the same guard, gated on _eval_ssrf_guard_active (non-local
backend, not a local sidecar, allow_private_urls unset) and fail-open on
probe failure, matching the eval-path guard and the main-browser
snapshot/vision guards. camofox_back is intentionally not changed: its
target is unknown until navigation completes, and the subsequent content
read is already guarded.
Adds regression tests covering the three read tools blocking on a private
page, the public-page pass-through, and the guard-inactive no-probe path.
Three CLI reliability fixes:
1. Interrupt reliability: chat() only re-queued the user's interrupt
message when the turn result carried interrupted=True. When the agent
thread raced past its last interrupt check (or finished) before the
interrupt landed, the message was silently dropped — and the stale
_interrupt_requested flag left on the agent instantly aborted the
NEXT turn. Un-acknowledged interrupt messages are now re-queued as
the next turn and the stale flag is cleared (only when the agent
thread actually exited). The clarify-race path also parks the message
in _pending_input instead of dropping it.
2. Slow exit (5+ min): stdlib ThreadPoolExecutor workers are non-daemon
and joined unconditionally by concurrent.futures' atexit hook — even
after shutdown(wait=False). One wedged tool worker (abandoned after
interrupt/timeout) held the process open forever. Promoted
async_delegation's daemon executor to a shared tools/daemon_pool
module and adopted it in tool_executor (concurrent tool batches),
memory_manager (background sync), delegate_tool (child timeout wrapper
+ batch fan-out), and skills_hub (source fan-out). Added a 30s exit
watchdog (HERMES_EXIT_WATCHDOG_S) armed at _run_cleanup start as a
backstop for wedged cleanup steps.
3. Exit jank: after prompt_toolkit tears down the input/status bars the
terminal sat silent for the whole cleanup window, looking hung. Print
'Shutting down… (finalizing session)' immediately at exit start.
E2E: live PTY interrupt of a foreground 'sleep 120' terminal tool now
aborts in ~1s and the typed message runs as the next turn; wedged-worker
+ wedged-cleanup subprocess exits in 5.8s (watchdog) instead of hanging.
Salvage of the surviving piece of #2696 by @tarunravi. The PR's other two
changes (tool progress streaming, SSE None-sentinel fix) were independently
superseded on main by the structured hermes.tool.progress SSE events and the
rewritten queue-drain loop.
Remote OpenAI-compatible frontends can't read server-local file paths, so
MEDIA:<path> tags (browser screenshots, generated images) were dead text.
_resolve_media_to_data_urls() now inlines small (<=5MB) local images as
markdown data URLs across all four response surfaces: chat completions
(non-streaming), session chat, session chat stream final event, and the
Responses API. Non-image, missing, or oversized paths pass through
untouched.
Salvage of #2794 by @CharmingGroot, ported to the relocated
plugins/platforms/email/adapter.py:
- Guard raw_email = msg_data[0][1] against IndexError/TypeError and
non-bytes payloads. UIDs are added to _seen_uids before fetch, so an
exception mid-batch permanently skipped every remaining message in
the batch — now the bad message is logged and skipped instead.
- Message-ID domain generation falls back to 'localhost' when
EMAIL_ADDRESS lacks '@' (now via a shared _message_id_domain() helper
covering all 3 send paths; the PR fixed 2 of 3).
- ChannelOverride + channel_overrides on PlatformConfig
- Resolve model/runtime: session /model, then channel_overrides, then global
- Thread/parent channel lookup; bridge discord.channel_overrides from YAML
- Drop unrelated test and delegate_tool changes from PR scope
delegation.max_concurrent_children is now the single cap for both a
batch's parallelism and concurrent background delegation units.
- _get_max_async_children() delegates to _get_max_concurrent_children();
a leftover max_async_children key logs a one-time deprecation warning
- config v32→33 migration removes the stale key, folding a raised
max_async_children into max_concurrent_children (max wins, no lost
headroom)
- capacity error messages now point at max_concurrent_children
- pool-at-capacity sync fallback now attaches an explanatory note so
the model/user know why the call blocked instead of dispatching async
Previously users who raised max_concurrent_children (e.g. to 15) still
hit the invisible default-3 async cap: the 4th background delegate_task
silently ran inline, blocking the turn with no signal.
CLAUDE_CODE_OAUTH_TOKEN is set and owned by the user's Claude Code
install (subscription OAuth), not a Hermes-managed inference
credential — Claude subscription auth is not a working Hermes provider
path. Blocklisting it broke agent-spawned claude CLIs: with no token in
the child env, claude fell through to the shared macOS Keychain /
~/.claude/.credentials.json store and, on auth failure, cleared it —
logging the user out of their interactive Claude sessions and the
desktop app.
Exempt it from _HERMES_PROVIDER_ENV_BLOCKLIST (it arrives via the
anthropic registry entry, so discard explicitly with rationale).
ANTHROPIC_API_KEY / ANTHROPIC_TOKEN and every other provider credential
remain stripped, and the GHSA-rhgp-j443-p4rf fail-closed passthrough
guard is unchanged for everything still on the blocklist.
Fixes#55878
Follow-up on the salvaged #56392 guard. The cherry-picked change matched
custom:<name> pool entries against the primary by raw base_url string
equality, which (a) can't disambiguate two named custom providers sharing
one gateway base_url and (b) left a latent bare-"custom" entry bypass.
Route the match through get_custom_provider_pool_key(rt[base_url]) compared
against the entry's custom:<name> key, mirroring the sibling guard in
recover_with_credential_pool. Use CUSTOM_POOL_PREFIX instead of the literal.
Add regression tests for the custom same-endpoint (swap) and cross-endpoint
(skip) branches, plus the plain-provider fallback-pool case from #56885.
Two related hardening fixes for auxiliary calls (which include MoA reference
advisors — a pinned-model path where provider fallback is not a meaningful
recovery):
1. Transient-transport retries: the same-provider retry on a connection reset /
timeout / 5xx / 408 was a single attempt, then fallback. For a pinned aux
call a second blip silently loses the call (root of the run2 double-advisor
'Connection error' collapse — a genuine upstream blip). Now retries N times
with exponential backoff, N = auxiliary.transient_retries (default 2 -> 3
total attempts, clamped [0,6]). Compression-on-timeout fast-fail carve-out
preserved.
2. Per-model client-cache isolation: _client_cache_key excluded the model, so
two concurrent auxiliary calls to the same provider/base_url/key but
different models (e.g. an opus + gpt-5.5 MoA fan-out) shared one cache entry
and could race each other's client lifecycle. Model now participates in the
key -> distinct clients, no cross-call races. Same-model reuse unchanged.
- agent/auxiliary_client.py: _transient_retry_count() + backoff loop; model in
_client_cache_key and both call sites.
- hermes_cli/config.py: auxiliary.transient_retries default (2).
- tests: new retry/isolation tests; updated 2 stale-expectation tests to the
corrected behavior (per-model resolve; N-retry escalation).
Backoff base is overridable (_TRANSIENT_RETRY_BACKOFF_BASE) so tests don't sleep.
_get_platform_tools reverse-maps a platform composite to configurable
toolsets with an all-tools subset test. Because get_toolset() merges
registry-registered tools into a toolset, a tool added to a toolset
(delegate_cli -> delegation; desktop-only read_terminal -> terminal) that the
static composite never listed made the subset test fail, silently dropping the
entire toolset on api_server and other inference-based platforms. Compare the
toolset's static membership at all three reverse-map sites.
Fixes#49622.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Adds include_registry=True kwarg to resolve_toolset/get_toolset. When False,
returns only the static TOOLSETS view with no registry-merged tools — the
composite-authored membership platform reverse-mapping must compare against.
Default True preserves all existing behavior; this is the enabling half of
the api_server toolset-drop fix (#49622).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Extends the browser private-network eval guard to the Camofox backend.
On main, _browser_eval() returned early in Camofox mode before running the
shared private-URL literal pre-scan and before re-checking the page URL
after eval, leaving Camofox as a sibling backend that could execute
browser_console(expression=...) against private/internal targets.
- move the eval private-URL literal pre-scan before the Camofox early return
- add a Camofox current-page private-URL probe via the evaluate endpoint
- withhold Camofox eval results when the page is now private/internal
Follow-up to browser private-network hardening in #56173, #56526, #56664.
Salvage of #56764 by @rayjun (rayoo), cherry-picked to preserve authorship.
The fake _SessionStore tracked peer_records but no test read it, leaving
#55300's peer-record behavior unasserted. Add a positive assertion on the
persist path and negative (== []) assertions on the two stale/moved-binding
skip paths, so the peer-record side effect is bound.
Mutation-verified: removing the production _record_gateway_session_peer call
makes the positive assertion fail.
Co-authored-by: João Vitor Cunha <jvsantos.cunha@gmail.com>
The #55300 peer-recording call now fires on the failed-turn compression
split path; the fake _SessionStore in test_compression_failure_session_sync
(carried in with #55721's test changes) lacked that method. Add a
call-tracking no-op so the combined salvage's tests pass.
Co-authored-by: João Vitor Cunha <jvsantos.cunha@gmail.com>
Manual /compress built a temporary AIAgent without the originating
platform / stable gateway session key, so an external context engine
ingested the retained transcript tail as source=cli during /compress
and again as the real platform on resume (duplicate cli,telegram rows).
Pass platform=_platform_config_key(source.platform) + the in-scope
gateway_session_key, mirroring the normal gateway turn. Assigned into
runtime_kwargs (single-valued, authoritative) so they neither collide
into a duplicate-kwarg TypeError nor lose to a stale resolver value.
Fixes#50422.
MoA per-turn latency is dominated by advisor GENERATION: turn wall time
correlates ~0.88 with output tokens and ~-0.03 with input tokens (measured over
52 turns). Each turn waits for the slowest advisor to finish writing, and
advisors were uncapped — writing multi-thousand-token essays the aggregator
only needs the gist of.
Add an opt-in per-preset reference_max_tokens knob (mirrors reference_temperature)
that caps ADVISOR output only; the acting aggregator is never capped. Default
None = uncapped, so existing presets are byte-for-byte unchanged (no regression).
Wired through both MoA execution paths (MoAChatCompletions.create and
aggregate_moa_context).
E2E: same task, closed preset uncapped vs reference_max_tokens=600 -> 59s to 33s
(~44% faster), final answer identical/correct.
- hermes_cli/moa_config.py: _coerce_int_or_none helper + reference_max_tokens
in _normalize_preset/_default_preset/flattened view
- agent/moa_loop.py: read preset.reference_max_tokens, pass to reference fan-out
- agent/conversation_loop.py: pass reference_max_tokens on the per-turn path
- tests + docs
Ben caught that the initial approach (widening _NOUS_PORTAL_ALLOWED_HOSTS to
include the staging host) was the wrong fix -- env vars are supposed to
override the allowlist, mirroring how NOUS_INFERENCE_BASE_URL already
bypasses _ALLOWED_NOUS_INFERENCE_HOSTS via _nous_inference_env_override().
The actual bug: both resolve_nous_access_token and
resolve_nous_runtime_credentials read
`_optional_base_url(state.get("portal_base_url")) or os.getenv(...) or ...`
-- a plain `or` chain where the STORED state value wins first (short-circuits
before the env vars are even read), and then whichever value won gets run
through the same _NOUS_PORTAL_ALLOWED_HOSTS gate regardless of its source.
So a hosted agent stamped with HERMES_PORTAL_BASE_URL=<staging> in its env
AND a staging portal_base_url already persisted to auth.json would still
get silently rewritten to prod on every refresh, because the env var never
even got a chance to be consulted.
Revert the previous _NOUS_PORTAL_ALLOWED_HOSTS widening entirely --
staying prod-only preserves the allowlist's actual job (rejecting an
untrusted network-provided portal_base_url persisted to auth.json by a
compromised Portal response).
Add _nous_portal_env_override() (mirrors _nous_inference_env_override())
and restructure both call sites so the env override is checked FIRST and,
when set, wins outright and skips the allowlist gate entirely -- the
allowlist only ever runs against the fallback (stored-state-or-default)
path now.
Rewrote tests/hermes_cli/test_nous_portal_staging_allowlist.py to test the
actual fix: the helper function, and an end-to-end
resolve_nous_access_token proof that the env override wins even when state
ALSO has the staging host stored (the exact incident shape), that it wins
over a stored PROD host too, and that the allowlist's heal-to-prod
behaviour for an untrusted stored value is preserved when no override is
set.
* fix(streaming): handle completed responses with empty/None choices
The streaming fallback guard added in #55932 recognized a completed
response object only when its `choices` was a non-empty list. But an
adapter can return a completed response whose `choices` is `None` or an
empty list (an error / content-filter / terminal frame) — still a whole,
non-iterable response, not a token stream. Those shapes fell through to
`for chunk in stream` and crashed with
'types.SimpleNamespace' object is not iterable
which is exactly issue #55933 (MoA `openai-codex` aggregator on
TUI/Desktop, where a stream consumer forces the streaming path).
Broaden the guard to discriminate on the PRESENCE of a `choices`
attribute (a genuine provider Stream object exposes none), disable
streaming for the session, and return the completed object so the outer
loop's normal invalid-response validation handles empty/None choices via
its retry path instead of iterating.
Based on the diagnosis in #56525 by @spiky02plateau (that PR normalized
the MoA aggregator return with a one-shot chunk iterator; the common
text/tool-call crash was already fixed at this seam by #55932, so this
extends the existing guard to cover only the remaining empty/None-choices
gap).
Fixes#55933
* refactor(streaming): simplify empty-choices guard body and parametrize tests
Post-review cleanup (no behavior change):
- Inline the single-use `response_choices` local and drop the redundant
`if first_choice is not None else None` guard (getattr(None, ...) already
returns the default safely).
- Collapse the two near-identical empty/None-choices regression tests into
one `@pytest.mark.parametrize` case.
Mutation-verified: reverting the guard to the old non-empty-list condition
still makes both parametrized cases fail with the historical
'types.SimpleNamespace' object is not iterable.
---------
Co-authored-by: spiky02plateau <155588579+spiky02plateau@users.noreply.github.com>
Follow-up to the salvaged fix: the regression test asserted a frozen
max_tokens == 128_000 literal, coupling it to the Opus-4-8 model table.
Assert against _get_anthropic_max_output("claude-opus-4-8") plus > 2000
instead, so the test survives model-table churn while still catching a
regression to the old `or 2000` fallback.
agent/vertex_adapter.py resolved VERTEX_CREDENTIALS_PATH,
GOOGLE_APPLICATION_CREDENTIALS, VERTEX_PROJECT_ID, and VERTEX_REGION via raw
os.environ.get() instead of the profile-scoped get_secret() every other
credential lookup in hermes_cli/runtime_provider.py uses. In a multiplex
gateway serving several profiles from one process, os.environ still holds
whichever profile's .env python-dotenv loaded at boot — so a raw read here
let one profile's turn silently mint a Vertex OAuth2 token from, and get
billed against, a different profile's GCP service account. No error, no
fail-closed guard: the multiplex UnscopedSecretError protection was bypassed
entirely because these reads never went through get_secret().
- _resolve_credentials_path/_resolve_project_override/_resolve_region now
call agent.secret_scope.get_secret(), matching the _getenv() pattern
already used for every other provider's credentials.
- get_vertex_credentials()'s ADC fallback (google.auth.default()) reads
GOOGLE_APPLICATION_CREDENTIALS from os.environ internally, bypassing
get_secret() entirely — closed with a narrow guard: when multiplexing is
active and this profile's scope has no Vertex credentials of its own, but
os.environ still carries a value (left by a different profile's boot-time
dotenv load), refuse ADC rather than silently authenticate as a stranger.
- Zero behavior change for single-profile installs: get_secret() falls
through to os.environ transparently whenever multiplexing is off.
Same bug class as the already-fixed _HERMES_OAUTH_FILE/_AUTH_JSON_PATH/
HOOKS_DIR cross-profile leaks, now closed for Vertex's OAuth2 credential
path.
The salvaged fix wired per-provider ssl_ca_cert / ssl_verify (and
HERMES_CA_BUNDLE) into the MAIN OpenAI client. This follow-up:
- Auxiliary client parity: process_bootstrap.build_keepalive_http_client
accepts and forwards verify; auxiliary_client._resolve_aux_verify mirrors
the main-client TLS resolution (via load_config_readonly, the read-only
fast path) so compression/vision/web_extract/title-gen/session_search
honor the same per-provider CA. Without this, chat worked against a
private-CA endpoint but every auxiliary call still failed APIConnectionError.
- switch_model now reads custom_providers from live config (load_config_readonly)
instead of the init-time agent._custom_providers snapshot, so ssl_ca_cert /
ssl_verify edits are honored on mid-session model switch — matching the
context-length reload (#15779).
- Drop the dead client-level verify= where a custom httpx transport is used
(httpx ignores it there); verify lives on the transport. Fix docstrings.
Applies to both run_agent._build_keepalive_http_client and process_bootstrap.
- resolve_httpx_verify: add CURL_CA_BUNDLE to the env chain (consistency with
agent/ssl_guard._CA_BUNDLE_ENV_VARS) and emit a loud logger.warning naming
the endpoint whenever ssl_verify:false disables verification.
- get_custom_provider_tls_settings: case-insensitive base_url match (config
dedup already lowercases; scheme/host are case-insensitive) so a mixed-case
entry doesn't silently drop its CA. Exact match preserved — no prefix bypass.
- Demote best-effort except Exception: pass in agent_init/switch_model to
logger.debug(exc_info=True).
- Tests for aux verify forwarding, _resolve_aux_verify, case-insensitive
match, and prefix-bypass rejection.
Wire ssl_ca_cert and ssl_verify through custom_providers config and env
vars into the keepalive httpx client, fixing APIConnectionError against
mkcert/self-signed Ollama proxies behind HTTPS.
Desktop chat bubbles render plain text, but a worker-routed command that
builds its own Rich Console (e.g. /journey) picks up truecolor from the
gateway's inherited COLORTERM and leaks raw escapes into the bubble. Strip
ANSI at the single worker-return choke point so every command renders cleanly.
The TUI opens /journey as an overlay, so it never travels this path.
In the interactive CLI, /journey dispatched straight to `args.func(args)`,
letting Rich write ANSI to stdout — which patch_stdout's StdoutProxy passes
through as literal `?[38;2;…m` garbage. Route the read-only views (default +
`list`) through a captured, force-color Console and re-emit via `_cprint`
(prompt_toolkit's ANSI parser), matching the `ChatConsole` idiom.
`delete`/`edit` stay on real stdio since they prompt / open `$EDITOR`.
parse_frontmatter's malformed-YAML fallback stores every value as a string,
so a skill's `metadata` can be a str. `_category`/`_related` chained
`.get("metadata", {}).get("hermes", {})` and blew up with `'str' object has
no attribute 'get'`, taking down `build_learning_graph()` (and thus /journey
and `hermes journey`) whenever any installed skill had bad frontmatter.
Extract a `_hermes_meta()` helper that returns the nested dict only when it
really is one. Fixes the whole class, not just the two call sites.
Two Hermes bots sharing a channel could volley replies at each other
indefinitely. Root cause: Discord reply-pings (allowed_mentions
replied_user=true) add the replied-to bot to message.mentions without a
literal <@bot> token in the body, so the existing bot-admission gate
treated a reply chip as an explicit @mention and re-triggered the peer.
Adds opt-in discord.bots_require_inline_mention (default false; env
DISCORD_BOTS_REQUIRE_INLINE_MENTION). When enabled, bot-authored
messages must carry a raw inline <@id>/<@!id> mention in the content;
reply-ping-only mentions no longer admit the message. Human messages and
all existing defaults are unchanged.
The new _self_is_raw_mentioned helper deliberately ignores the resolved
message.mentions list (which reply-ping populates) and checks only the
raw content token via the shared _raw_mentioned_user_ids primitive.
Vertex AI authenticates via OAuth2 (service-account JSON path / ADC), not
PROVIDER_REGISTRY, and VERTEX_CREDENTIALS_PATH is declared with
password=False (it's a path, not a bare key) under category="provider" —
a category the registry-derived blocklist loop never checks. Both it and
GOOGLE_APPLICATION_CREDENTIALS (the ADC fallback the adapter also reads)
fell through every existing blocklist source and leaked the on-disk
location of a GCP service-account key into every spawned subprocess
(terminal, codex/copilot app-server, browser workers) — the same leak
class already closed for every other provider's credentials in #53503.
Follow-up review fixes on the salvage of #54872 (原作者 张满良/@zmlgit):
1. [HIGH] Adapter selection now goes through the shared
_authorization_adapter chokepoint (gateway/authz_mixin.py) instead of a
local inline lookup that fell back to the DEFAULT profile's same-platform
adapter when the owning profile had a registry entry but no adapter for
that platform. That fallback re-introduced the exact cross-profile
mis-delivery ([230002] Bot can NOT be out of the chat) this change exists
to fix. Adds a mutation-verified guard test
(test_notifier_owning_profile_adapter_no_default_fallback).
2. [HIGH→documented] The creator-wake SessionSource cannot faithfully
reconstruct a DM/thread creator's session key because chat_type is neither
persisted on the subscription nor carried on the session-context bridge.
Documented the limitation inline; behavior degrades to a fresh group
session (never an exception). The end-to-end fix (stamp + persist
chat_type) is a scoped follow-up, not bundled into this salvage.
3. [MED] Documented that archived/unblocked are intentionally claimed (cursor
hygiene) but silent, and excluded from wake kinds.
4. [MED] Wake-injection failure now logs at WARNING with exc_info=True (the
cursor has already advanced, so a broken wake must not be a silent no-op).
The /codex-runtime slash command short-circuits with "openai_runtime
already set" when invoked with the same value as the current config,
and crucially skips the entire migration block below. The check
conflates two things: (a) "the config value is correct" and (b) "the
world state (managed block in ~/.codex/config.toml, hermes-tools MCP
callback, plugin discovery) is converged".
Common footgun this exposes: a user who pre-sets
`model.openai_runtime: codex_app_server` directly in config.yaml
(reasonable thing to do) and then runs /codex-runtime codex_app_server
to trigger migration sees "already set" and silently gets no migration.
~/.codex/config.toml never receives the managed block, the hermes-tools
MCP callback never registers, and codex falls through to its default
runtime instead of the app-server one — visibly successful but
functionally partial setup.
The migration is idempotent by design (it replaces its own managed
block in place between MIGRATION_MARKER and MIGRATION_END_MARKER), so
re-running it is safe and cheap. Fix the short-circuit to fall through
to migration when re-applying codex_app_server while skipping the
config persist (no value-level change needed). The disable case
(re-applying "auto") still short-circuits because disabling doesn't
touch ~/.codex/config.toml at all.
The user-visible message changes to "openai_runtime already set to
codex_app_server — re-applying migration" so re-runs surface what
happened.
Regression test (test_reapply_codex_app_server_runs_migration) asserts:
- migrate() was called when re-applying
- persist_callback was NOT called (no config write on no-op transitions)
- migration output (MCP servers, sandbox default) surfaces in the
user-visible message
- requires_new_session is True so callers know to /reset
Verified RED→GREEN: the test fails on origin/main with
"migration must run on reapply, not just first enable" and passes with
this fix. Full test_codex_runtime_switch.py suite: 31 passed.
Self-review follow-up on the salvaged approval-routing fix.
The initial adaptation re-read os.getenv("HERMES_YOLO_MODE") at session-build
time. That diverges from the repo's security invariant: HERMES_YOLO_MODE is
frozen into tools.approval._YOLO_MODE_FROZEN at import time precisely so a skill
running mid-process cannot set the env var and instantly flip the approval
bypass (a prompt-injection escalation path). A live re-read re-opened that hole
for the codex routing path.
- Add tools.approval.is_approval_bypass_active() — the canonical three-source
bypass check (frozen --yolo/HERMES_YOLO_MODE + session /yolo + approvals.mode
off) in one place. This is the 4th inline copy of that OR-chain (the three
sites in approval.py and tui_gateway/server.py:3121 all use the same idiom);
the helper is the shared chokepoint they can collapse onto.
- codex_runtime.py now calls is_approval_bypass_active() instead of the
hand-rolled mode-or-session check plus a runtime env re-read.
- Update the env-yolo test to patch _YOLO_MODE_FROZEN (the canonical test
pattern, e.g. tests/tools/test_yolo_mode.py) rather than setenv, which is
dead-on-arrival against the frozen constant.
Fail-closed default preserved on every branch; 28 integration + 77 session/yolo
tests pass; E2E confirms the real exec decision flips decline->accept only when
bypass is active.
On gateway/cron/non-CLI contexts the codex app-server runtime has no UI to
surface codex's exec/apply_patch approval requests, so they fail closed
(silently decline) — the bot appears responsive but cannot write files, with
no approval prompt anywhere ("patch rejected by user").
When the user has explicitly opted out of Hermes approvals (approvals.mode: off,
the /yolo session toggle, or HERMES_YOLO_MODE=1), collapse to codex's own
sandbox permission profile (~/.codex/config.toml) as the policy gate by passing
_ServerRequestRouting(auto_approve_exec=True, auto_approve_apply_patch=True) to
the session. Defaults (manual/smart/unset) preserve the current fail-closed
behavior — a no-op for users who have not opted out.
Reads the mode via the canonical tools.approval._get_approval_mode() (which
already normalizes the YAML-1.1 bare-'off'->False case) at session-build time,
so a mid-session /yolo toggle is honored too.
5 integration tests: each opt-out mechanism (config off, YAML False, env var,
session yolo) plus the default fail-closed regression guard.
Closes#26530
Co-authored-by: snav <jake@nousresearch.com>
Every other content-returning browser tool entry point
(browser_snapshot/vision/console/eval, and click/type/press via
_blocked_private_page_action) re-checks window.location.href against the
private/internal/cloud-metadata floor after the page could have changed --
because a redirect chain or client-side navigation can land on an address
the initial browser_navigate preflight never saw. browser_back was the one
navigation-triggering entry point missing this: it called
_run_browser_command(..., "back", []) and returned the resulting URL
straight to the model with no re-check.
On a cloud/CDP (non-local) backend, if browser history contains a
private/internal address (e.g. a prior redirect touched an internal host),
browser_back would navigate the live browser there and hand the URL back
to the model with no guard -- the exact class of gap the private-page
guard exists to close, just on the one entry point it hadn't reached yet.
Re-check happens after the navigation succeeds (not before, unlike
click/type/press) since it's the resulting page -- not the one being left
-- whose safety matters. A failed back navigation (no history) skips the
check entirely since nothing changed. Verified live: the new regression
test fails (returns the private URL instead of a blocked payload) on the
pre-fix code and passes after.
With the default busy_input_mode=interrupt, a burst of rapid gateway
messages arriving while context compression is in flight could interrupt
the current turn and start a fresh turn against the pre-rotation parent
session. Because compression is interrupt-immune (#23975), the still-
running compression later rotates the id out from under that new turn,
and if the new turn also grew past the compression threshold it started
its own uncancellable compression on the same stale parent — forking
multiple orphaned one-shot sibling continuations (#56391).
While a state.db compression lock is held for the session, demote
'interrupt' busy-input mode to 'queue' semantics (mirroring the subagent
protection in #30170), so the follow-up message waits for the in-flight
compression + its id rotation to land instead of racing a new turn
against the stale parent. Ack copy explains the compression demotion.
Fixes#56391.
Two independent MoA auxiliary-call fixes:
#53866 — auxiliary.moa_reference.timeout and auxiliary.moa_aggregator.timeout
were 600s while moa_agent was 120s. Raise both to 900s so a genuinely long
reference/aggregator turn (mixed providers, deep reasoning, long tool chains)
has headroom instead of being cut mid-generation.
#53735 — _CodexCompletionsAdapter (the Codex/Responses auxiliary path used by
the MoA acting-aggregator, compression, web_extract, session_search, etc.)
never set prompt_cache_key, so it stayed cache-cold while the MAIN Responses
transport (agent/transports/codex.py) was warm. Derive the same
content-addressed key via the shared _content_cache_key(instructions, tools)
helper and set it on the aux Responses request, with the same host guards the
main transport uses (xAI carries the key in extra_body; GitHub/Copilot opts out
of cache-key routing).
Tests: 5 new prompt_cache_key cases (set+prefixed, stable across identical
prefix, differs on different instructions, skipped for xai/github hosts).
tests/agent/test_auxiliary_client.py 279 pass; tests/hermes_cli/test_config.py
130 pass.
On the MoA path agent.model/provider are the virtual preset name (e.g.
"closed") and "moa", which have no pricing entry. estimate_usage_cost()
returned None for the aggregator turn, so the `if amount_usd is not None`
guard skipped it and the session's estimated_cost_usd reflected only the
advisor fan-out — a ~50% undercount when the aggregator does the full acting
loop (verified: $0.91 advisor-only vs $1.96 true, aggregator = 54%).
MoAChatCompletions.create() now stashes the resolved aggregator slot as
last_aggregator_slot (exposed via MoAClient); conversation_loop reads it to
price the aggregator turn at its real model/provider. cost_source flips from
'none' to 'provider_models_api'.
Follow-up to the END-MARKER reorder: moving the summary prefix after the
[PRIOR CONTEXT] wrapper meant _is_context_summary_content (prefix-at-start)
no longer recognized a merged-tail summary. That silently broke three
consumers — the last-real-user anchor (would pick the merged summary as a
real user turn, causing active-task loss), the carry-forward summary find,
and the auto-focus skip. _strip_summary_prefix would also carry the wrapper
+ stale tail content forward as the next summary body.
Extract the two delimiter strings into _MERGED_PRIOR_CONTEXT_HEADER /
_MERGED_SUMMARY_DELIMITER constants (writer + detector stay in sync), teach
_is_context_summary_content and _strip_summary_prefix to look past the
delimiter, and add a regression test. Standalone summaries unchanged.
When the compression summary is merged into the first tail message
(the alternation corner case where a standalone summary role would
collide with both head and tail), the old format was
SUMMARY + END_MARKER + OLD_TAIL_CONTENT — so the preserved tail content
appeared AFTER the end marker and the model could read it as a fresh
message to respond to.
Reorder so the END MARKER is always last: old tail content is wrapped in
[PRIOR CONTEXT ...][END OF PRIOR CONTEXT — COMPACTION SUMMARY BELOW]
delimiters, then the summary, then the END MARKER. _append_text_to_content
handles both string and multimodal-list content.
Salvaged from #56372 by @Gromykoss. Only the END-MARKER reorder half is
carried over. The PR's second change (a post-compaction pass that strips
user-role messages before the first summary marker on compression_count>=2)
was dropped: on 2nd+ compactions the protected head decays to system-only
(_effective_protect_first_n -> 0, #11996) so the targeted 'ghost head user'
does not occur, and where the strip does fire it deletes legitimate recent
tail user turns (data loss) and can leave consecutive assistant messages
(role-alternation violation).
Teknium
kshitijk4poor
isheng
williamumu
HexLab98
knoal
Ben Barclay
spiky02plateau
xxxigm
Fan Yang
helix4u
AIalliAI
Sanidhya Singh
sasquatch9818
Tranquil-Flow
Tanmay Choudhary
edgemass
brooklyn!
hellno
Taylor H. Perkins
Waseem Shahwan
Ahmett101
herbalizer404
giggling-ginger
izumi0uu
mbac
pierrenode
liuhao1024
allenliang2022
Frowtek
Flownium
davidgut1982
Marin Pesa
Guillaume Nodet
konsisumer
Ray
dodo-reach
yoma
Sam Beran
Michael Musser
Stephen Schoettler
luyifan
binhnt92
Alix-007
Jash Lee
Brett Bonner
Lohinth
falkoro
srojk34
heathley
Teknium
devatnull
Adrian Lastra
Sami Rusani
dsad
David Metcalfe
MorAlekss
Lord_dubious
Ahmett101
lEWFkRAD
Tuna Dev
Jigoooo
Gutslabs
Jonathan Almanzar
emozilla
Danilo Falcão
峯岸 亮
infinitycrew39
terry197913
Shashwat Gokhe
Michael Steuer
Marxb85
eliteworkstation94-ai
Lavya Tandel
SHL0MS
SHL0MS
Shannon Sands
Sabin Iacob
Jacky Zeng
Jiahui-Gu
LeonSGP43
Maxim Esipov
Que0x
Eugeniusz Gilewski
tt-a1i
nankingjing
Victor Kyriazakos
Hermes Trismegistus
kshitijk4poor
Gumclaw
Yingliang Zhang
Jaaneek
David Zhang
Jneeee
Mibayy
Morgan K
CrazyBoyM
Evo
Tarun Ravikumar
CharmingGroot
crazywriter1
VolodymyrBg
openhands
Nick Mason
Jake Present
João Vitor Cunha
snav
Gromykoss