Salvaged from #31671 (@Shizoqua). The config-cache half of that PR was
superseded on main (9b8b054c2d gave _load_config_safe a readonly path),
but this cron-scanner half is still live: _strip_cron_safe_constructs
used re.search + a single str.replace, which only scrubbed occurrences
IDENTICAL to the first match. A cron job loading several GitHub skills
carries heterogeneous auth-header curl forms (-H vs --header, quoting,
token var names) — every non-identical block tripped the
exfil_curl_auth_header detector on every tick, blocking legitimate
GitHub cron jobs.
Now re.sub scrubs every occurrence; the trailing [^\n]* consumes the
URL path so no dangling fragment remains. Sabotage-verified: the old
implementation false-blocks the heterogeneous two-skill prompt the new
regression test pins; exfil to a non-GitHub host is still blocked.
79/79 cron tool tests green.
Follow-up to the #68246 salvage. The backend permission-mode resolution
only checked the DB session_id the tool path passes, but gateway /yolo
keys approval bypass off the gateway session_key (contextvar). Consult
both namespaces so /yolo works on messaging platforms, not just CLI/TUI.
Adds a regression test driving the real approval contextvar + yolo
toggle path E2E.
- Remove tests/-shadowing sys.path.insert(dirname/'..') from 11 test files:
it prepended the tests/ dir itself to sys.path, so 'import agent' /
'import hermes_cli' resolved to the test packages and collection died
with ModuleNotFoundError depending on import order (2 files failed in
every full-suite run; 9 more were latent).
- Patch call_llm in 5 context-compressor tests that called compress()
unmocked: each burned ~50s attempting live LLM traffic through the
relay before falling back (572s file — the slowest in the suite, and
flaky under the 300s per-file timeout). File now runs in ~5s.
- agent/redact.py: fix two catastrophically-backtracking regexes hit by
the compressor's redaction pass on large payloads —
_STRICT_URL_USERINFO_RE anchors on the mandatory '//' (optional-scheme
prefix backtracked O(n^2): ~55s on a 320KB payload, now sub-ms;
output-equivalence fuzz-verified on 20k random strings), and the
_CFG_DOTTED_RE/_CFG_ANCHORED_RE subs gain an exact linear keyword
pre-gate so secret-free text skips the quadratic pattern entirely.
- tests/gateway/test_feishu.py: version-guard the extra_ua_tags SDK
signature check; the repo pins lark-oapi==1.6.8 but stale local
installs (1.5.3) fail the assertion — skip below the pin.
- tests/tools/test_managed_browserbase_and_modal.py: stub
agent.redact + agent.credential_persistence in the fake agent package
(empty __path__ blocks all real agent.* imports added since the fake
was written).
- tests/gateway/test_startup_restart_race.py: raise wait_for timeouts
2s -> 30s; 2s wall-clock on a loaded 40-worker box flaked in the
baseline run (passes instantly when the box is quiet).
TUI (tui_gateway/server.py _load_approval_mode): now delegates to
tools.approval._get_approval_mode instead of re-reading config raw via
_load_cfg + _deep_merge(DEFAULT_CONFIG, ...) and normalizing locally.
Behavior fix, not pure refactor: the canonical load_config path applies
managed-scope config overlays and ${VAR} env expansion, plus a legacy
max_turns lift, which the TUI's raw YAML read bypassed — under a managed
config that sets approvals.mode, the TUI previously reported/toggled a
different mode than the approval gate actually enforced. Both surfaces
now agree by construction. Name/signature and the mode-vocabulary clamp
are preserved.
Codex (agent/transports/codex_app_server_session.py): read confirmed the
_decide_exec_approval/_decide_apply_patch_approval paths carry NO
Hermes-side mode/timeout reads — the Hermes resolution already flows in
from agent/codex_runtime.py via tools.approval.is_approval_bypass_active()
(auto_approve_* routing) and via the shared approval-gate callback. So no
code extraction was needed; added docstrings pinning that invariant and a
cross-reference on the protocol-semantic choice mapping
(_approval_choice_to_codex_decision), which intentionally stays local.
Adds tests/tools/test_approval_mode_parity.py: cross-surface invariant
test asserting the core resolver, the TUI path, and the codex bypass
derivation agree for synthetic configs (unset defaults, global mode set,
YAML-bool off, malformed values, whitespace/case), plus a delegation-seam
test proving the TUI has no independent config read left.
Note: gateway/run.py has sibling raw reads but is intentionally untouched
(multiple in-flight PRs); flagged as follow-up.
`_canon_key_combo` (the `_BLOCKED_KEY_COMBOS` gate in
`handle_computer_use`) split key strings on `+` only, but the cua-driver
backend's `_parse_key_combo` splits on both `+` and `-`. So a model could
issue `{"action":"key","keys":"ctrl-alt-delete"}` (or `alt-f4`,
`cmd-shift-q`): the gate saw a single unknown token and let it through
while the backend executed the real destructive shortcut.
Split the gate on both `+` and `-` so it canonicalizes combos the same
way the backend does. Non-destructive hyphen combos (`cmd-c`) and the
literal `-` zoom key (`cmd+-`) are unaffected.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
cua-driver 0.7.x can return list_windows/list_apps payloads under
structuredContent.windows, data.windows, data._legacy_windows, or
top-level windows/_legacy_windows (direct CLI responses). The wrapper
only read structuredContent.windows, so discovery came back empty
(capture 0x0, list_apps []) while raw cua-driver calls worked.
- add _windows_from_tool_result(): walks the known envelope shapes in
priority order, skipping empty higher-priority envelopes
- route _load_windows() (MCP + CLI re-fetch paths) through the helper,
covering capture() and focus_app()
- harden _ingest_windows(): skip non-dict members, normalize untrusted
app_name/title/z_index fields
- list_apps(): prefer structuredContent.apps, fall through populated
data/top-level envelopes, derive unique apps from window-shaped
payloads via _apps_from_windows(), keep the text-line fallback last
- tests for every envelope shape, precedence, malformed records, and
app derivation
Salvaged from #63037 (Reaper-Legion), which itself preserved the
original implementation from #57961 (kohoj); #73007 (umi008)
independently proposed the same normalization later.
Fixes#57905
Co-authored-by: Reaper <248977840+Reaper-Forge@users.noreply.github.com>
Co-authored-by: Ulises Millan Guerrero <ulises.millanguerrero@gmail.com>
Replaces the half-duplex per-playback barge monitors with ONE listener
that runs for the entire agent turn in continuous voice mode: armed at
utterance-submit, disarmed when the turn is fully done (response + TTS
finished). Fixes Teknium's live report that voice interruption never
works: (a) not while the LLM is generating, (b) not while TTS plays.
Root causes:
- HALF-DUPLEX GAP: the barge monitor only spawned when TTS playback
STARTED (cli.py streaming/whole-file paths, gateway _tts_stream_begin).
During LLM generation there was NO microphone listener at all.
- PLAYBACK DEAFNESS: the monitor calibrated its VAD noise floor WHILE the
speaker was blasting TTS (speaker bleed baked into the floor), then
multiplied it by 8x with a 1s strictly-consecutive block requirement —
normal speech could rarely reach the trigger, and the 2s grace
swallowed early interjections.
New model — tools/voice_mode.full_duplex_listen():
- Pre-playback calibration: quiet-room noise floor established at turn
start and HELD through playback (never recalibrated against bleed).
- Phase-aware trigger: generation = floor x voice.barge_in_threshold_multiplier
(new config, default 3.0, justified by synthetic-frame tests);
playback = additionally clamped to a 1500-RMS minimum so bleed alone
can't trip; 4000-RMS ceiling keeps speech always reachable.
- Windowed-majority detection (>=80% of a 300ms window) instead of the
strictly-consecutive counter that reset on intra-word energy dips.
- Grace on playback ONSET only (voice.barge_in_grace_seconds, default
down 2.0 -> 0.5) — suppresses the onset transient, not the mic.
- Debug diagnostics at every decision point (calibrated floor, per-window
RMS above 50% of trigger, trip/no-trip, grace suppressions) — always
logger.debug, mirrored to stderr under HERMES_VOICE_DEBUG=1.
Phase behavior (CLI cli.py + tui_gateway/server.py, same model):
- generation: speech interrupts the in-flight turn via the SAME seam the
typed/Ctrl+C interrupt uses (agent.interrupt()), cuts any pending TTS
pipeline so the stale reply never plays, and submits the captured
interjection (pre-roll capture, first syllable kept) as the next turn.
- playback: cuts TTS (streaming pipeline stop + fallback speak stop
events + file player) and submits the capture.
- stop phrase honored in BOTH phases: mid-generation 'stop' interrupts
the turn AND ends the voice chat (stop everything).
- one listener instance spans generation -> playback (no re-arm race);
double-arm refused (CLI _voice_fd_active / gateway _fd_listener_active).
Gateway specifics: _arm_full_duplex_listener() at _run_prompt_submit turn
start and inside _tts_stream_begin; _speak_text_with_barge registers its
(stop, done) pair in _fd_speak_pipelines so fallback speaks are cut and
tracked; _tts_stream_barge_in_monitor kept as a shim that arms the new
listener. Desktop renderer owns its own mic path (voice-barge-in.ts) and
is unaffected; if desktop backend-mic mode is used it inherits via the
gateway.
Tests: full_duplex_listen synthetic-RMS suite (speech-over-bleed trips,
bleed alone doesn't, quiet floor held through playback, grace window,
multiplier math 3x vs 8x, windowed-majority dips), CLI listener phase
tests (generation interrupt seam, playback cut, lifecycle spans phases,
double-arm, config forwarding, stop-phrase-mid-generation), gateway
generation-phase interrupt + stop-phrase tests. Generation-interrupt
test sabotage-verified.
tools/computer_use/tool.py keeps the CLI approval flow in module-globals:
_approval_callback plus the per-session unlock stores _always_allow /
_session_auto_approve. Any test that installs a callback (or drives CLI
init far enough that the real one is registered) and does not reset it
poisons every later computer-use test in the same process:
* a leaked callback that raises — dead UI infra or a stale two-argument
signature (the contract is (action, args, summary)) — becomes
verdict='deny' in _request_approval, so dispatch tests fail with an
empty backend call list;
* a leaked callback that blocks (the real CLI one waits on an answer
queue) hangs a single-process run forever.
Both are order-dependent: tests/tools/test_computer_use.py passes 220/220
in isolation but shows dispatch failures in single-process full-suite runs
(and, with a blocking leak, a permanent hang observed via py-spy inside
_request_approval -> callback -> queue.get with no timeout).
Fix: an autouse teardown-only fixture resets callback + unlock stores
after every test; tests that install their own callback keep it for their
own duration. Regression pair included: a 'forgetful' test leaves a stale
two-arg callback behind, the next test asserts dispatch still routes to
the backend — red without the fixture (1 failed), green with it (225
passed together with the whole computer-use file).
_start_lifecycle_locked reassigns a fresh threading.Event() at line
786, so patching the pre-made instance's wait() is lost and the real
30s wait races pytest-timeout. Patch threading.Event itself with a
FakeEvent whose wait() returns False immediately (#69372).
Premise check on live main: barge-in machinery EXISTS for the per-turn
STREAMING pipeline only — cli.py chat() arms _voice_barge_in_monitor and
tui_gateway _tts_stream_begin arms _tts_stream_barge_in_monitor. What was
actually broken for spoken interruptions:
1. CLI whole-file fallback (_voice_speak_response_async — used whenever
streaming TTS cannot start: sounddevice missing, requirement probe
fails): NO monitor was ever armed, so talking over the reply did
nothing. Now arms _voice_barge_in_monitor in continuous voice mode.
2. Gateway fallback speak (tts_queue None → speak_text thread) and the
voice.tts RPC (desktop-triggered speech): speak_text ran bare, and
its internal streaming dispatch created a PRIVATE stop event nothing
could reach — uninterruptible even by stop_playback(). New
_speak_text_with_barge() runs the same barge monitor beside the speak
thread; hermes_cli.voice.speak_text/_speak_text_streaming accept an
external stop_event so a barge cuts the streaming pipeline too.
Stop-phrase handling and voice.transcript submission are inherited
from the shared monitor (merged #73933 behavior preserved).
3. False barge during TTS (the reason interruption "worked" then
self-cancelled or fired randomly): salvaged PR #71083 by @beardedeagle
(previous commit, kept authorship) — rolling-window VAD floor, 8x
multiplier, 4000-RMS trigger ceiling, barge_in_grace_seconds (2s)
before the mic opens, min-floor clamp. barge_in_grace_seconds is now
documented in DEFAULT_CONFIG.
Desktop spoken barge (renderer mic via voice-barge-in.ts) already covers
both its live-stream and fallback speech paths — verified, no change.
Long thinking/tool stretches in a voice conversation are dead air — the
user cannot tell whether the agent is alive. New: quiet, repeating soft
bubble blips while the agent works and no speech audio is flowing.
- tools/voice_mode.py: numpy-synthesized blips (no binary assets) — two
alternating low pitches (G4/E4) with pitch glide + smooth attack/decay
envelopes, ~0.8-1.2s randomized spacing, volume = voice.beep_volume * 0.5.
start_thinking_sound(should_play=...) / stop_thinking_sound() daemon-loop
lifecycle; macOS-TCC-safe (sounddevice output gated there → silent skip,
no per-second afplay churn). New mark_audio_output_active()/
is_audio_output_active() ref-count wraps play_audio_file and the
streaming OutputStream sentence writes so "audio is flowing" is accurate.
- Config: voice.thinking_sound (default true) off-switch.
- cli.py: starts when a voice-mode turn begins, per-blip gate skips while
TTS speaks / mic records / barge capture owns the mic; stopped in the
chat() finally on every exit path.
- tui_gateway/server.py: same lifecycle around _run_prompt_submit turns
(voice mode on), gated on is_audio_output_active + continuous capture.
- Desktop: renderer owns voice-conversation audio, so a matching WebAudio
implementation (src/lib/thinking-sound.ts, same envelope/pitches) runs
while conversation status === "thinking"; honors voice.thinking_sound
(via config store) and the shared sound-mute toggle; stops instantly on
speaking/listening/end.
One owner for the wording: voice_stop_hint() in tools/voice_mode.py —
sources the phrase from voice.stop_phrases (first entry) so a custom
phrase renders correctly, and returns "" when the feature is disabled
(stop_phrases: []) so no surface shows a hint.
- CLI: printed in /voice on output (style-matched dim notice).
- TUI: voice.toggle action=on now carries stop_hint; the Ink client
renders it in the "Voice mode enabled" block (older gateways omit
the field — no hint, no crash).
- Desktop: the renderer voice loop never touches tools/voice_mode.py,
so the phrase is read from config (voice.stop_phrases → $voiceStopPhrase
store, seeded in use-hermes-config) and shown as an info toast when a
voice conversation starts. i18n: en/ja/zh/zh-hant/ar.
Replace one-shot VAD calibration with a rolling deque window that
continuously recalibrates the noise floor throughout TTS playback,
preventing false barge-in triggers from stale calibration. Add a
grace period before VAD activates so TTS playback establishes first.
Suppress duplicate text rendering when token streaming is enabled.
Mirror the barge-in and TTS stream stop logic to the TUI gateway path
so both CLI and gateway use the same VAD semantics.
Rolling-window VAD:
- 90th percentile of rolling window (~3s) for noise floor
- 8x multiplier (was 5x) for TTS volume variation headroom
- 4000 RMS trigger ceiling so genuine speech can still trip
- min_floor clamped to SILENCE_RMS_THRESHOLD * 2
- sustained_ms=1000, calibration_ms=800
Barge-in grace period (barge_in_grace_seconds, default 2.0s):
Delays VAD activation so TTS playback establishes before the mic opens.
Duplicate render suppression:
When streaming_enabled, pass display_callback=None to
stream_tts_to_speaker so the token stream is the sole display path.
TUI gateway mirror:
Mirror _tts_stream_stop and _tts_stream_barge_in_monitor changes to
tui_gateway/server.py so both code paths use the same VAD parameters,
grace period, and TTS CUT diagnostic logging.
Profile-scoped session DB and MoA progress events in tui_gateway/server.py
were necessitated by the TTS pipeline changes affecting session state
and event routing.
Normal-exit flag and TTS CUT diagnostic logging at all cut paths.
Regression tests:
- test_quiet_then_loud_playback_does_not_trip
- test_8x_multiplier_absorbs_tts_volume_spikes
- test_trigger_ceiling_lets_genuine_speech_trip
- test_silence_calibration_does_not_false_trip_on_tts
- test_tts_stream_stop_latches_interruption_for_next_turn
- test_tts_stream_stop_after_natural_finish_does_not_latch
- Profile-scoped session DB tests (10 tests)
Skill Sync had no default base URL, so a user with no `sync.base_url` in
config.yaml and no HERMES_SYNC_BASE_URL got:
sync inert: no sync base URL configured (config.yaml sync.base_url
or HERMES_SYNC_BASE_URL).
Every sync command was unusable out of the box. The URL was left unset
because the plane did not exist yet when the client was written; it does now.
- Adds DEFAULT_SYNC_BASE_URL = "https://gateway-gateway.nousresearch.com" and
returns it as the last step of resolve_sync_base_url().
- Resolution order is unchanged otherwise: HERMES_SYNC_BASE_URL ->
config.yaml sync.base_url -> production default. The env var and config key
now exist to point a dev/staging build at another plane rather than to make
the feature work at all.
- Follows the existing precedent for production endpoints in this codebase
(DEFAULT_NOUS_PORTAL_URL in hermes_cli/auth.py, HERMES_DIAGNOSTICS_BASE_URL
in diagnostics_upload.py): a module constant with env/config override.
The "no sync base URL configured" guards are kept — they are now unreachable
in practice but remain correct if the default is ever blanked.
Tests: 3 new — the default is returned when nothing is configured, config
still overrides it, and the constant is a bare https origin (no trailing
slash, no path) since the client appends /v1/sync/. 2349 passed / 0 failed
across 56 suites via scripts/run_tests.sh.
Verified against a temp HERMES_HOME with no config: resolves to the
production plane; HERMES_SYNC_BASE_URL and sync.base_url both still win, and
trailing slashes are stripped.
The exact kwargs snapshot broke when VAD hardening added keys — the
test's real contract is 'null stt.local: must not crash or force
language/prompt'. Baseline kwargs are pinned by the dedicated suite.
Local faster-whisper called model.transcribe with bare {'beam_size': 5}:
no VAD, cross-window conditioning on, no confidence filtering. Pure
silence produced hallucinated tokens (E2E: 5s anullsrc WAV -> 'You',
no_speech_prob=0.705) and noisy clips could produce runs of junk, often
in other languages.
Three-layer class fix, one shared owner for every local-whisper call
site (build_local_transcribe_kwargs):
1. Silero VAD filter (bundled with faster-whisper) on by default —
silence never reaches the model. stt.local.vad: false restores the
raw behavior for music/ambient transcription.
stt.local.vad_min_silence_ms tunes chunk splitting (default 500).
2. condition_on_previous_text=False — one hallucinated token can no
longer seed a self-reinforcing run; negligible cost for
voice-note-length audio.
3. Segment confidence gate (_join_confident_segments): drop a segment
only when no_speech_prob > 0.6 AND avg_logprob < -1.0 (openai-whisper's
own heuristic shape; both must hit so quiet-but-real speech survives).
Config: stt.local.no_speech_prob_threshold / logprob_threshold.
The WHISPER_HALLUCINATIONS blocklist in voice_mode.py stays as
last-resort defense but should now almost never fire.
E2E (real faster-whisper 'base', CPU int8):
silence.wav before 'You' -> after ''
noise.wav before '' -> after ''
speech.wav before/after 'Hello World, this is a test of the
transcription system.' (unchanged)
Docs (EN + zh-Hans), DEFAULT_CONFIG, cli-config.yaml.example updated;
19 unit tests (kwargs contract, off-switch, confidence gate incl.
quiet-speech survival, _transcribe_local wiring), sabotage-verified.
Saying OR typing a configured stop phrase (voice.stop_phrases, default
"stop") now ends the voice chat everywhere, not just classic CLI PTT:
- hermes_cli/voice.py: new explicit on_stop_phrase callback through
start_continuous/stop_continuous. The force-transcribe path previously
DISCARDED the stop phrase silently — with auto_restart=False the client
re-arms the next capture, so the conversation never ended. Both halt
paths now fire on_stop_phrase (fallback: on_silent_limit for legacy
callers) as user intent, distinct from the no-speech timeout.
- tui_gateway/server.py: voice.record wires on_stop_phrase and emits
voice.transcript {stop_phrase: true} after flipping HERMES_VOICE(_TTS)
off and stopping streaming TTS — same teardown as /voice off. The TTS
barge-in monitor stop-checks its transcript too. prompt.submit consumes
a TYPED bare stop phrase at the server-side choke point when voice mode
is on (returns {voice_stopped: true}, no turn starts).
- ui-tui: voice.transcript {stop_phrase} ends voice mode with a clear
'voice chat ended' notice (distinct from the no-speech-limit message);
submitPrompt releases the busy latch on a consumed voice_stopped reply.
- cli.py: _typed_voice_stop in process_loop — typing a bare stop phrase
while voice mode/continuous is active ends voice mode instead of
sending 'stop' to the agent; typed 'stop' outside voice mode is
unchanged. Voice transcripts skip the check (already stop-checked).
- desktop: interceptsTypedVoiceStop — the composer's onSubmit ends the
live voice conversation (same path as clicking end on the pill) when a
bare stop command is typed with no attachments; renderer-owned loop, so
handled client-side like the existing spoken isVoiceStopCommand.
- tools/voice_mode.py: transcribe_recording never lets the Whisper
hallucination filter swallow a configured stop phrase (e.g. 'bye'
configured as a stop phrase is both a hallucination-blocklist entry and
a stop phrase — stop-phrase check now wins).
Tests: continuous-loop signal (sabotage-verified), force-transcribe stop
signal + legacy fallback, hallucination-filter ordering, typed-stop CLI
unit tests (voice on/off/longer text), prompt.submit typed-stop gateway
tests, TUI vitest for stop_phrase event handling, desktop vitest for the
typed-stop interceptor.
The Photon iMessage sidecar needs node_modules under
plugins/platforms/photon/sidecar/, but hosted/managed images keep the
whole install tree under an immutable /opt/hermes — every install and
self-heal path (setup CLI, stale-deps reinstall, cold install) died on
EROFS, and hosted users have no shell to work around it.
Three-layer fix, mirroring the WhatsApp bridge resolver pattern:
1. Bake the deps into the image. The Dockerfile now runs npm ci for the
sidecar in the layer-cached dependency stage (deterministic installs
from the committed lockfile; the postinstall spectrum-ts patch runs
at build time). Hosted happy path needs no runtime install at all.
2. New sidecar_paths.resolve_sidecar_dir() decides where the sidecar
runs from: PHOTON_SIDECAR_DIR override > writable source dir (dev
installs, unchanged) > read-only dir with baked fresh deps (managed
image) > mirror to $HERMES_HOME/photon/sidecar (writable data
volume) when deps are missing or stale in a read-only tree. The
mirror refreshes changed source files on image updates while
keeping node_modules, so the existing lockfile-staleness self-heal
works there.
3. connect() can now cold-install: _start_sidecar() runs the bounded
npm ci bootstrap when node_modules is missing instead of raising
immediately, and check_requirements() reports available when a
self-install is possible (npm present + writable resolved dir) so
the gateway actually creates the adapter on hosted instances. A
failed bootstrap still raises the actionable error, which connect()
surfaces as the retryable SIDECAR_FAILED fatal state on the
dashboard.
Tests: resolver decision table (env override, in-place, mirror,
refresh, fail-open), cold-install lifecycle paths, and a Dockerfile
contract test guarding the baked-deps + no-chown invariants.
Fixes NS-606.
Adds gpt-transcribe (OpenAI's new file-transcription model, $0.0045/min)
to the OpenAI STT provider:
- OPENAI_MODELS set: gpt-transcribe is recognized so provider
auto-correction keeps it on OpenAI and rejects it on Groq
- Language hint wiring: gpt-transcribe replaces the singular
'language' field with a 'languages' list; the API rejects the legacy
field, so the hint is sent via extra_body {languages: [..]}
- Config comment (DEFAULT_CONFIG), cli-config.yaml.example, desktop
settings enum, and docs (en + zh-Hans) updated
- Tests: model pass-through, languages-list hint shape, legacy singular
hint preserved for gpt-4o-transcribe, Groq auto-correction
gpt-live-transcribe (realtime WebSocket, $0.017/min) is NOT wired here:
the file-based STT pipeline has no realtime session path; it belongs in
a future realtime/voice-mode integration.
Installing a memory provider (Honcho, mem0, hindsight, ...) from the
dashboard Plugins page failed on hosted deployments with a permission
error: the setup endpoint shelled out to
`uv pip install --python sys.executable`, which targets the sealed
read-only venv under /opt/hermes (immutable hosted image, NS-579/#49113).
The correct mechanism already exists: tools/lazy_deps.py redirects
installs to the writable durable target on the data volume
(HERMES_LAZY_INSTALL_TARGET=/opt/data/lazy-packages) when the venv is
sealed (HERMES_DISABLE_LAZY_INSTALLS=1), appends the target to the END
of sys.path (core venv always wins collisions), and constrains shared
deps to core-venv versions. The dashboard installer simply never used
it.
Fix:
- tools/lazy_deps.py: new public install_specs() — installs arbitrary
manifest-declared pip specs through the same environment routing as
ensure(): venv-scoped by default, durable-target on sealed images,
refused with an actionable reason when gated off (config kill switch
or sealed venv without a target — never surfaces raw EROFS/EACCES).
Specs are validated with _spec_is_safe(); post-install it invalidates
import/metadata caches so availability rechecks in the same process
see the new packages without a restart. Never raises.
- hermes_cli/web_server.py: _install_memory_provider_pip_dependencies
now calls install_specs() instead of building its own uv/pip
subprocess. Blocked installs surface the gate reason in the setup
results; the response's status block reflects post-install
availability (stale 'missing deps' state clears immediately).
- hermes_cli/memory_setup.py, plugins/memory/honcho/cli.py,
plugins/memory/mem0/_setup.py: CLI setup wizards routed through
install_specs() too — same sealed-venv failure mode, same fix.
No hosted setup path writes to /opt/hermes anymore; provider discovery
and installation now use the same environment (sys.path activation is
shared with the lazy-install bootstrap in hermes_bootstrap).
Tests:
- tests/tools/test_lazy_deps.py: TestInstallSpecs — gating matrix
(sealed+no-target blocked with immutable-deployment reason, config
kill switch, sealed+target proceeds), spec-safety rejection before
any subprocess, venv-scoped vs --target command display, failure
stderr passthrough, never-raises contract.
- tests/hermes_cli/test_web_server.py: setup endpoint routes pip
through lazy_deps (regression guard asserts no direct 'pip install'
subprocess), blocked-reason surfacing, same-response availability
recheck clears stale missing state.
Fixes NS-605 (Plain T-1111).
Follow-up integration for the #47588 salvage, aligning the new streamers
with the post-campaign invariants:
- All streaming key lookups go through _resolve_key -> tts_tool.
_resolve_provider_key -> resolve_provider_secret (config > env/.env >
credential pool, profile-scoped) — never bare get_env_value. xAI
resolves via resolve_xai_http_credentials so OAuth users stream too.
- _capped(): every provider's chunk iterator is bounded at 16 MiB per
sentence, mirroring _read_tts_response_bytes' bounded-upstream-body
invariant on the sync paths.
- Tests updated for the resolver contract + new coverage for credential
routing and the cap.
Salvaged from PR #47588 and rebased onto the post-campaign streaming core:
the StreamingTTSProvider ABC/registry and the ElevenLabs/OpenAI streamers
already live on main (tools/tts_streaming.py), so this ports the pieces
main lacked:
- GeminiStreamer: streamGenerateContent?alt=sse -> base64 PCM chunks
(24 kHz mono int16), reusing main's DEFAULT_GEMINI_TTS_* constants.
- XAIStreamer: WebSocket wss://api.x.ai/v1/tts -> binary PCM frames,
async->sync bridged via the _collect_async test seam.
- tts.streaming.provider config knob: pin one streamer, or 'auto' to
walk the priority list elevenlabs -> gemini -> openai -> xai. Unset
keeps the never-swap-the-user's-voice default.
- docs/streaming-tts.md: architecture, capability matrix, how to add
a provider.
- Unit tests for the knob, SSE parsing, and the WS bridge; key-gated
E2E tests (skipped without credentials).
Refs: #47588
Two follow-ups from the voice PR (#70509).
1. macOS ARM64 onnx migration. Existing users who pinned
openwakeword.inference_framework=onnx before the tflite fix landed kept a
wake word that arms but never fires (ONNX's embedding model is broken on
Apple Silicon, upstream #336). New resolve_inference_framework() honors an
explicit framework everywhere ONNX actually works, but coerces the one
provably-dead combination (explicit onnx + macOS ARM64) to tflite with a
one-time warning. No config mutation; empty still falls back to the platform
default. Both read sites (engine init + requirements check) route through
the shared resolver.
2. Voice turn-timeout leak. Each listen cycle reassigned turnTimeoutRef
without clearing the prior 60s timer, so a stale timer from an earlier
cycle could fire handleTurn() mid-way through a later listen — after enough
idle re-listens this wedged the loop into a non-re-arming state (the
'voice chat deactivates after ~a minute' report). Clear before re-arm.
Tests: 64 wake tests (added onnx-coercion / intel-kept / tflite-kept /
empty-default cases; updated the stale 'explicit onnx kept on ARM64' test
that encoded the old broken behavior), 39 desktop voice/wake vitest, tsc +
eslint clean.
'photon:any;-;+1555...' targets matched no parser pattern, so
_handle_send bounced them off the channel directory and failed
resolution even though the adapter accepts the GUID verbatim (the
react handler already passed them through). Recognize the DM chat
GUID shape (mirrors the adapter's _DM_CHAT_GUID_RE) in
_parse_target_ref for photon only.
Command providers legitimately reference their own API keys in shell
templates (curl one-liners). The #70342 scrub removes ALL provider keys,
which would break such setups. Add a per-provider env_passthrough list
(TTS + STT) that copies named variables back from the parent env, plus
docs and tests. Scrub stays the default; passthrough is explicit opt-in.
Port the progress-based idle-timeout pattern from _run_command_tts
(PR #50087, @CleanDev-Fix) to _run_command_stt: the timeout resets on
any stdout/stderr output, so a slow-but-alive STT provider survives
while a silently stalled one is killed. Stuck detection stays
progress-based, never wall-clock.
`transcribe_audio` reads a local file and hands it to the configured STT
provider — for the hosted providers (Groq, OpenAI, Mistral, xAI, ElevenLabs)
that ships the file's bytes to a third-party API. The same local-input read
guard was added to image-gen (587be5b5b) and xAI video-gen (104232979) to keep
the agent from feeding credential/secret stores to a provider, but STT was
missed.
Call `get_read_block_error(file_path)` at the top of `transcribe_audio`, before
validation/dispatch, so a `.env`, `auth.json`, `.anthropic_oauth.json`,
`mcp-tokens/`, etc. is refused up front instead of being transcribed (and, for
hosted providers, exfiltrated). This is defense-in-depth, not a security
boundary — the guard's own message says so — but it restores parity with the
image/video-gen tools.
Regression test: a `.env` file is refused with the shared read-guard message
before any provider dispatch (mutation-verified).
HERMES_LOCAL_STT_COMMAND rendered quoted placeholders into a
user-configured template and passed the result to shell=True. Shell
metacharacters in the template therefore remained executable syntax even
though the placeholder values themselves were quoted.
Tokenize the rendered template and invoke it as an argv list while
preserving the existing timeout, closed stdin, and Windows creation flags.
Lock the invocation contract with metacharacter regression coverage and
document explicit shell wrapping for trusted templates that need it.
Salvages #32694
Co-authored-by: Ernest Hysa <takis312@hotmail.com>
Command-type TTS providers validated output_format against a hardcoded
{mp3,wav,ogg,flac} set; any other value was silently coerced back to mp3,
which then mismatched the output path the post-run check expects. This
blocked common ffmpeg-producible containers/codecs — notably m4a (AAC),
the portable choice for WeChat/iOS/mobile voice files — with no
config-only path (only a local source patch, lost on every update).
Widen COMMAND_TTS_OUTPUT_FORMATS to add m4a, aac, amr, opus. This only
permits a command provider to declare these; the user's command still
produces the file (e.g. via ffmpeg). No built-in provider behavior
changes and no new required config.
Update the two tests that pinned the old set, and add a positive case
covering the new formats. Document the supported output_format values.
## Summary
- Spawn system audio players (`ffplay` / `afplay` / `aplay`) with `hermes_subprocess_env(inherit_credentials=False)`.
- Prevent gateway tokens and provider API keys from leaking into OS media helpers.
- Add a regression test asserting scrubbed env on `Popen`.
## Salvage / credit
Sibling of #70342 / incomplete #56332 (TTS/STT command scrub) on the voice-mode playback path.
Salvage incomplete #56332: route command TTS/STT through hermes_subprocess_env
while preserving delegated-child lineage, and close the sibling local-whisper
subprocess.run path that still inherited the full process environment.
Co-authored-by: Cursor <cursoragent@cursor.com>
The client called its gate "the DEV-PHASE gate (tool_gateway_admin)", which
reads as though Skill Sync is gated on an unrelated service's admin right.
It isn't. NAS populates that claim from Permissions.ADMIN_ACCESS — the global
portal admin permission that guards /admin/* — so the gate is "is this user a
Nous admin?". The claim is simply named for its first consumer, the tool
gateway.
Renamed on this side to say what it means, while keeping the wire string
(other services read it):
- DEV_GATE_CLAIM -> NOUS_ADMIN_CLAIM (value unchanged: "tool_gateway_admin",
with a comment recording why the wire name differs).
- identity/status key dev_gate_ok -> nous_admin, across the client, the CLI
consumers, and the tests.
- The module docstring now states where the claim comes from, that the wire
name is misleading, and that this gate is pre-launch containment rather
than the shipping entitlement — admin status conflates "may administer
Nous" with "has Skill Sync enabled" and has no middle setting for a beta
cohort. Choosing the real entitlement is left as a separate decision.
Naming only — no behaviour change, and no change to which accounts can sync.
The user-facing messages stay deliberately vague ("not enabled for your
account yet") rather than telling users they need portal admin.
Verified: 2346 passed / 0 failed across 56 suites via scripts/run_tests.sh;
`hermes sync status` against a live token reports "nous_admin": true. Zero
stale dev_gate_ok / DEV_GATE_CLAIM references remain.
Encapsulates the feature behind one command for launch, and adopts the
official product name.
One command:
- `propose` moves from `hermes skills propose` to `hermes sync propose`, so
the whole feature is one command to learn and one to document. Its handler
moves from cmd_skills to cmd_sync accordingly.
- The `hermes sync` parser now documents both halves plainly: personal sync
across your devices, and sharing with your organisation. Added an examples
epilog; rewrote the verb help in user language ("Include a skill in your
sync" rather than "Opt a skill into sync").
- Every user-facing string that pointed at `hermes skills propose` now points
at `hermes sync propose` (8 sites, including the agent-visible guidance
returned by skill_manage and the org provenance header).
This also clears the way for #39343, which adds its own top-level `sync` for
git-repo profile backup — that feature nests under `skills`, this one owns
`sync`.
Naming:
- HSP / "Hermes Sync Protocol" is gone from prose, docstrings, and comments.
The feature is "Skill Sync".
- Public identifiers renamed: HSPClient -> SyncClient, HSPError -> SyncError,
HSPConflict -> SyncConflict, hsp_address -> wire_address, HSP_VERSION ->
WIRE_VERSION.
- The WIRE names are deliberately NOT renamed: the `hsp_version` capability
field and the `x-hsp-object-type` response header are set by the deployed
gateway-gateway sync plane (verified in src/sync/syncRouter.ts), so
renaming them client-side would break sync against a live server. A comment
at the version constant records why they differ from the product name.
- The version-mismatch error is now actionable ("this server speaks sync
version X, but this Hermes speaks Y — update Hermes to sync with it")
instead of leaking the protocol acronym.
Also fixes a wiring gap found on the way: the gateway housekeeping tick
pulled personal skills but never org skills — the same defect already fixed
for the CLI. Org pull now runs there too, gated on real org membership.
Tests: the jargon guard now also fails on a bare "HSP". The two tests that
asserted the old cross-command structure are replaced by three asserting the
new one (propose IS under sync, propose is NOT under skills, sync usage
lists it). 2294 passed / 0 failed across all 51 suites that import the
changed modules, via scripts/run_tests.sh.
Verified by running the real CLI: `hermes sync --help` lists all eight verbs,
`hermes skills --help` no longer mentions propose, `hermes sync propose
--help` parses, and `hermes sync status` still reports live org state.
CI (Linux) failed with ModuleNotFoundError: portalocker — it's a
win32-only dependency (concurrent-log-handler chain). Tests now lock
handles via a platform helper (fcntl on POSIX, portalocker on Windows),
mirroring production _try_acquire_mcp_discovery_lock.
Review follow-up: test_micro_pause_tolerance_during_speech was left on real
sleeps, so it kept the wall-clock dependency the rest of this PR removes.
Its three sleeps (50 ms, 50 ms, 60 ms) feed the same monotonic gate at
tools/voice_mode.py:561 that the other two tests were converted for.
It was measured rather than assumed before being left out: 0 failures in 40
runs, because on Windows sleep() rounds up to the timer tick, so sleep(0.05)
really costs ~62.5 ms and the three sleeps clear the 150 ms gate by ~37.5 ms
-- wider than the 15.625 ms GetTickCount64() quantum that made the other two
flake. That margin is an accident of sleep granularity rather than anything
the test guarantees, so it is not worth keeping the bet.
Driving it from fake_clock makes the timeline exact (0.05 + 0.05 + 0.06 =
0.16 against the 0.15 gate, dip 0.05 under the 0.1 tolerance) and drops the
real sleeping from the suite. Mutation-checked: widening the gate at :561
fails the test.
test_silence_callback_fires_after_speech_then_silence and
test_custom_threshold_and_duration space their audio frames with
time.sleep(0.06) and check the result against 50 ms thresholds. That
leaves a 10 ms margin, which only survives if the clock is finer-grained
than the margin — and on Windows it isn't. time.monotonic() there is
GetTickCount64() with 15.625 ms resolution until CPython 3.13 moved it to
QueryPerformanceCounter(), so a sleep that really lasts 62.5 ms measures
as 46.9 ms often enough to matter, landing under the threshold. Depending
on which sleep got clipped, either the speech-confirm gate misses or the
silence timer never matures — which is why the same flake shows up on two
different asserts.
Both tests now advance a hand-driven clock instead of sleeping, so the
arithmetic is exact on every platform and neither test waits on real time.
The fixture patches the `time` name inside tools.voice_mode rather than
time.monotonic itself: voice_mode.time IS the stdlib module, so patching
through it would swap the clock out from under every other importer for
the duration of the test.
Measured on Windows 11 / Python 3.11.9, 20 runs of each test: 6/20 and
3/20 failed before, 0/30 after. The same machine on Python 3.13, where
monotonic() is QueryPerformanceCounter(), never failed either test — that
comparison is what pinned the clock resolution as the cause rather than
the detection logic. Linux CI never sees this: clock_gettime is
nanosecond-resolution there.
Running the suite could take over the machine it ran on. Two real
effects, both now closed:
- **The suite spoke through the speakers.** Once any test drove the
`voice.toggle` RPC with `action="tts"`, the handler set
`HERMES_VOICE_TTS=1` in the *live process environment*, and the flag
outlived that test. Every later test that drove a turn to completion
then fed its final response text to `hermes_cli.voice.speak_text` on a
background thread - real synthesis, real playback, no API key needed
(the default `edge` provider is keyless). A developer heard the fixture
string "partial answer complete" out loud. Because the flag is set from
inside the process, `scripts/run_tests.sh`'s `env -i` never protected
against this.
`tests/conftest.py` now blanks `HERMES_VOICE`/`HERMES_VOICE_TTS` per
test, and a new autouse `_audio_playback_guard` stubs `speak_text` and
its playback binding outright, so the speakers stay shut even inside the
test that sets the flag itself. `@pytest.mark.real_audio_playback` opts
out.
- **The suite launched Chrome.** `tests/tools/test_browser_supervisor.py`
spawned a real browser on any machine with Chrome on `PATH`. Its
docstring promised a `HERMES_E2E_BROWSER=1` gate that existed nowhere in
the code. That gate is now real, and the file is marked `integration`
so the default marker filter excludes it. `scripts/run_tests.sh`
forwards `HERMES_E2E_BROWSER` so the documented manual run still works.
Miscellanea
- `tests/test_audio_playback_guard.py`: regression cover for both defences,
driving the real `voice.toggle` handler rather than a stand-in.
- Move U+FFFC placeholder detection before _record_last_inbound() so the
placeholder message id is not recorded as the reaction target
- Cancel pending U+FFFC tasks in disconnect() to prevent task leaks
- Check mimeType audio/x-caf in addition to filename for CAF→VOICE promotion,
fixing unnamed attachments that default to '(unnamed)'
- Add 7 Photon adapter tests: CAF named/unnamed promotion, U+FFFC no-dispatch,
U+FFFC not recorded as last inbound, U+FFFC+attachment cancel, timeout fire,
disconnect cleanup
- Add 6 transcription tests: ffmpeg conversion, afconvert fallback, all-fail,
CAF→WAV before Groq, conversion failure error, local provider skip
_generate_xai_tts passes stream=True to requests.post since the TTS
speaker pipeline moved to the streaming core. Four fake_post mocks in
test_tts_xai_speech_tags.py still had the pre-streaming signature and
raised TypeError: unexpected keyword argument 'stream'. The other 20+
mocks in the same file already accept stream=False; this aligns the
stragglers.
Fixes the 4 failures in Python tests slice 5/8 on main.
Two in-house micro-fixes from issue triage:
- #49883: voice.beep_enabled gates in cli.py and hermes_cli/voice.py used
bool() on the config value, so a quoted YAML string like "false" or
"off" kept beeps on. Route through utils.is_truthy_value.
- #18432: AudioRecorder.start() collapsed OSError from _import_audio into
the 'pip install sounddevice numpy' hint — but OSError means the
PortAudio SHARED LIBRARY is missing, which pip cannot fix. Mirror
detect_audio_environment's system-package hint (libportaudio2 /
brew portaudio / Termux pkg install portaudio) on that path.
Fixes#49883Fixes#18432
Addresses review on #62601. Applies a single rule — no sounddevice for
audio OUTPUT on macOS (PortAudio/CoreAudio init triggers a
kTCCServiceMediaLibrary prompt) — consistently at all three output sites:
- play_audio_file: WAV playback (already routed to afplay) now uses the
shared _sounddevice_output_allowed() helper.
- play_beep: synthesize the tone with numpy only, then on macOS play it
via a temp WAV through afplay instead of sounddevice.
- stream_tts_to_speaker (tts_tool): on macOS, skip the sounddevice
OutputStream so playback falls through to the existing tempfile/afplay path.
Audio INPUT (recording) is untouched — it legitimately needs mic permission.
Tests: TestMacOSAudioOutputPolicy (voice_mode) proves WAV + beep routing
does not import sounddevice on Darwin and still uses it off Darwin;
test_tts_macos_output proves streaming TTS skips the OutputStream on Darwin.
Existing test_play_wav_via_sounddevice pinned to non-Darwin for determinism.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Two test classes lock in the new detection contract:
1. TestTermuxApiAppInstalledProbeLadder — drives _termux_api_app_installed
with a fake subprocess.run dispatcher and walks each rung of the
ladder:
- non-Termux env returns False (no probes run),
- pm-confirms returns True (back-compat),
- pm-clean-miss → cmd-confirms returns True,
- pm-FileNotFoundError → cmd-confirms returns True,
- pm-TimeoutExpired → cmd-confirms returns True,
- pm-nonzero-exit → cmd-confirms returns True,
- both probes inconclusive + binary on PATH returns True (the
core #31015 case),
- both probes inconclusive + no binary returns False,
- both probes clean-miss returns False (the genuine "CLI without
app" case keeps the existing warning),
- case-insensitive package match for ROMs that capitalise differently.
2. TestDetectAudioEnvironmentTermuxFallback — end-to-end through
detect_audio_environment, asserting the misleading "Termux:API
Android app is not installed" warning no longer fires when probes
are inconclusive but the binary is on PATH (the user-reported #31015
symptom), AND that the warning still fires when probes can
conclusively report the app is missing.
Refs: NousResearch/hermes-agent#31015
Ports #63768 forward onto current main per teknium1's review.
On WSL2 without a PulseAudio bridge, ffplay and aplay have no audio
device and TTS playback silently fails (issue #17608). When
powershell.exe and ffmpeg are available, convert the audio to a
uniquely-named WAV in the Windows %TEMP% directory and play it via
Media.SoundPlayer.
Per review, this fixes two gaps in the original port:
1. Exit-status masking: the cleanup subshell was
'( ffmpeg && powershell ); rm -f wav' -- the shell's exit status is
the LAST command's (rm -f, which is always 0), so a real
ffmpeg/PowerShell failure could never be detected by the rc-checking
fallback logic added to the player loop. Now captures the real
status before cleanup and re-exits with it:
'( ffmpeg && powershell ); rc=0; rm -f wav; exit '.
2. The no-Pulse WSL gate in detect_audio_environment() still hard-blocked
voice mode entirely (input AND output) even when the PowerShell
fallback made TTS output viable. Added _wsl_powershell_tts_available()
and use it to downgrade the WSL-without-Pulse case from a hard
'warnings' block to a non-blocking 'notices' entry when the fallback
is available -- the same PulseAudio-bridge recording guidance is still
surfaced (mic capture genuinely still needs it), it just no longer
blocks /voice on for TTS-only usage. cli.py's existing
env_check['available'] gate needed no changes since it already
respects this flag.
Also fixed the flaky uniqueness test (the original asserted
len(filenames) >= 2, which passed trivially on zero captured
filenames) and added a real fallback-triggering regression test for
the exit-status fix.
10 new/fixed tests pass in TestWSL2PowerShellFallback and the new
TestWSLAudioEnvironmentGate; 80/80 in the full tests/tools/test_voice_mode.py file.
AudioRecorder hard-codes SAMPLE_RATE (16 kHz) when opening the input
stream, but some capture devices (e.g. USB microphones exposed through
ALSA hw) reject 16 kHz outright — sd.InputStream fails with
PaErrorCode -9997 (Invalid sample rate) and voice recording is broken.
Query the default input device for its native default_samplerate at
recording start and open the stream / write the WAV at that rate,
falling back to the Whisper-friendly 16 kHz constant when the backend
does not expose a usable rate. STT providers accept standard WAV rates,
so downstream transcription is unaffected.
Closes#55908. The CLI voice-mode beep amplitude is hardcoded at 0.3 inside
tools.voice_mode:play_beep(), which makes the record start/stop cues too
quiet on low-volume systems and headphones. Users couldn't adjust it
without editing source.
Move the literal into a configurable voice.beep_volume setting (clamped to
0.0-1.0, default 0.3 to preserve prior behaviour). The new
_get_beep_volume() helper reads via the same load_config() pattern used by
cli.py's _voice_beeps_enabled() and hermes_cli/voice.py's _beeps_enabled(),
keeps bools / out-of-range / non-numeric / NaN values safely on the default,
and falls back silently if config can't load so the audio cue never breaks
the voice loop on a degenerate config.yaml.
Covered by tests/tools/test_voice_mode.py:
- TestGetBeepVolume (12 cases: missing key, custom value, boundary 0.0/1.0,
out-of-range clamp, type coercion, bool guard, NaN guard, exception
guard, dict-typed voice section)
- TestPlayBeepVolumeWiring (guards against re-introducing a hardcoded 0.3
literal in play_beep)
Docs: website/docs/user-guide/configuration.md mentions the new key.
Other locale translations (zh-Hans etc.) intentionally untouched —
handled by the regular i18n sync pipeline as a separate change.
No change in default behaviour: existing users hear exactly the same beep.
Review follow-up: the sweeper is right that the first commit only cured
half the dead config — the TUI gateway builds its recorder params
explicitly, so the cap never reached recordings started from the TUI.
- start_continuous() grows a max_recording_seconds param (default 0.0 =
disabled, so existing callers keep today's behaviour) and applies it to
the shared recorder next to the silence params
- tui_gateway voice.record start forwards the validated cap; corruption
semantics now mirror the silence params everywhere: non-numeric/bool
falls back to the documented 120 default (a hand-edited
`max_recording_seconds: true` must not become a 1-second cap), while
an explicit numeric <= 0 disables the cap
- cli.py wiring updated to the same corruption semantics
New coverage, per review:
- TestMaxRecordingCap drives the mocked InputStream callback past the
cap during continuous loud speech (the silence branch physically can't
fire there) and asserts the one-shot callback fires exactly once; plus
the disabled-cap negative
- TestMaxRecordingSecondsConfigReal pins the CLI config assignment for
the valid / zero / bool / garbage cases via _voice_start_recording
- test_voice_record_start_forwards_max_recording_seconds pins the TUI
forwarding for the same matrix
Salvaged from PR #68063 (@malaiwah). MatrixAdapter.send_voice now transcodes
any non-Ogg audio to Ogg/Opus at the adapter boundary (best-effort — the
original file is sent unchanged when ffmpeg is unavailable), so MSC3245
voice bubbles render even when a caller hands the adapter MP3/WAV audio.
_matrix_voice_metadata_for_file probing now runs via asyncio.to_thread so
ffprobe/ffmpeg subprocess timeouts can't stall the adapter event loop.
The PR's tools/tts_tool.py want_opus hunk was dropped: main's
OPUS_VOICE_PLATFORMS set (PR #73072) already includes matrix; the Matrix
opus-routing test is kept.
Refs #14841
Consolidates all TTS text-preparation paths onto
tools/tts_text_normalize.prepare_spoken_text:
- strip_nonspoken_blocks: removes <think> reasoning blocks (#34213,
incl. unterminated streaming blocks) and the end-of-turn
file-mutation verifier footer emitted by run_agent.py (#40772).
- flatten_newlines_for_payload: collapses newlines into sentence
breaks so newline-sensitive OpenAI-compatible providers (Kokoro)
speak the whole script instead of truncating at the first newline
(#9004).
- tools/tts_tool._strip_markdown_for_tts (voice-mode streaming + web
dashboard path) now delegates to the shared cleaner, with the legacy
regex pipeline kept as a best-effort fallback.
- hermes_cli/voice.py speak_text and cli.py _voice_speak_response now
use the shared cleaner instead of their own duplicated regex
pipelines.
- gateway auto-TTS fallback also strips think blocks.
Tests: tests/tools/test_tts_prepare_spoken.py covers think blocks,
verifier footer, emoji, newline flattening, and the shared-cleaner
wiring on the tool/streaming/gateway paths. Updated the header
expectation in test_voice_cli_integration.py for the heading-fold
behavior of the shared cleaner.
Closes#34213, #9004, #40772
Address review feedback on #31693:
- Regression tests assert extra_body == {"lang_code": ...} is forwarded
when tts.openai.language is configured, and omitted when unset/empty
- Document tts.openai.language as intended for OpenAI-compatible
endpoints that support lang_code (e.g. Kokoro-FastAPI)
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
xAI TTS supports a text_normalization boolean that normalizes
written-form text (numbers, abbreviations, symbols) into spoken-form
before synthesis. This parameter was not being sent, leaving users
with literal number/symbol pronunciation.
- Add DEFAULT_XAI_TEXT_NORMALIZATION_DEFAULT = False
- Read tts.xai.text_normalization from config (via _xai_bool_config)
- Attach text_normalization: true to the POST payload when enabled
- Omit the field entirely when unset or explicitly false (API default)
- Add 3 tests: default omission, enabled passthrough, explicit false
The `text_to_speech` tool schema accepted only `text` and `output_path`,
so style direction (tone, emotion, pacing, whispering) could never reach
the OpenAI backend — even though `gpt-4o-mini-tts` (Hermes's OpenAI
provider default) treats `instructions` as its primary voice-design
control.
This plumbs an optional `instructions` argument through the tool schema,
the handler lambda, and `text_to_speech_tool()` into
`_generate_openai_tts`, where it is forwarded to
`client.audio.speech.create()` only when truthy. Empty/None values still
omit the key entirely, preserving behavior on `tts-1`/`tts-1-hd` and
strict OpenAI-compatible servers.
The same passthrough unblocks self-hosted OpenAI-compatible voice-design
servers (Qwen3-TTS-VoiceDesign on oMLX, etc.) that are already wired in
via `tts.openai.base_url` — the established convention per #9004 and the
TTS config docs — without inventing a new provider backend.
Tests: `tests/tools/test_tts_instructions.py` covers backend passthrough,
tool-level threading, schema declaration, and the empty-string/absent
omission cases. `tests/tools/test_tts_max_text_length.py` fake_openai
signature widened to accept the new kwarg.
Refs NousResearch/hermes-agent#14196
Add optional 'speed' parameter (0.25-4.0) to the text_to_speech tool
schema and handler. When provided by the model, it overrides the
config-level tts.speed setting, enabling per-request speed control
without config changes.
Use cases:
- Language learning: slow playback (0.5x) for pronunciation practice
- Accessibility: adjustable speed for hearing preferences
- Content review: accelerated playback for long text
The speed value is clamped to [0.25, 4.0] and injected into tts_config
before dispatching to any provider (Edge, OpenAI, MiniMax, etc.), so
all existing provider-level speed handling works transparently.
Includes 3 new tests for tool-level speed injection, clamping, and
config preservation when speed is not specified.
Auto-TTS previously fed raw chat Markdown and compact symbols straight to the
speech provider, so units were read as stray letters and headings or bullets ran
together. This routes spoken text through a new normalizer,
tools/tts_text_normalize.prepare_spoken_text, that expands units (for example a
temperature written with the degree symbol becomes "degrees Celsius") and
flattens Markdown into a transcript-like script with sentence pauses.
The normalizer is best-effort: if it ever fails the code falls back to the
previous markdown-strip behavior, so auto-TTS keeps working. The Telegram voice
caption uses the same normalized text. Includes a unit test.
Salvaged from PR #62977 (@Vissirexa) — TTS model-cache half only (the
hindsight turn-buffer half is a different subsystem and was dropped).
_piper_voice_cache and _kittentts_model_cache were keyed by voice/model
with no eviction, and each entry is a whole loaded model (tens of MB).
A surface that sweeps voices pinned one model per voice for the process
lifetime. New _tts_cache_get_or_load() get-or-loads through a small LRU
(_TTS_MODEL_CACHE_MAX=3), refreshing recency on a hit and evicting the
least-recently-used model on a cold miss.
Class-level sweep following the ElevenLabs salvage (#66311): every cloud
TTS provider section now honors tts.<provider>.base_url. xAI, MiniMax,
Gemini, OpenAI and DeepInfra already did; Mistral (SDK server_url) was
the remaining gap. Adds per-provider config tests locking in the
ElevenLabs environment plumbing and the Mistral server_url passthrough.
Follow-up to salvaged PR #70307 (@aml1973): OpenAIStreamer now checks
tts.openai.api_key (config.yaml) ahead of the env resolver in both
available() and stream(), completing config parity between the sync
and streaming OpenAI TTS paths.
Use the shared OpenAI audio key resolver and prefer tts.openai.base_url over the global environment fallback in the Desktop streaming path. Add focused regression coverage for credential and endpoint propagation.
Salvaged from PR #26233 (@LeonSGP43), rebased onto the current 3-tuple
_resolve_openai_audio_client_config (is_managed flag, post-#73072 layout).
Same fix independently submitted earlier in PR #26209 (@zccyman) — credit
to both.
Resolution order now mirrors the STT resolver: tts.openai.api_key/base_url
from config.yaml -> VOICE_TOOLS_OPENAI_KEY/OPENAI_API_KEY env (still
honoring config base_url) -> managed gateway. _has_openai_audio_backend
also counts a config api_key as an available backend.
Fixes#26175
- parametrize the OAuth retry regression over HTTP 401 and 403 so both
documented rejection statuses are exercised
- reword the retry-failure warning: the except block also covers the
retried request, not just the credential refresh
- drop docs/proof/ from the PR diff; the live-proof screenshot now lives
on the fork's proof-assets-xai-stt-oauth-retry branch and stays linked
from the PR description
Follow-ups for the salvaged wave: the auto-detect legacy-error test now
stubs the split validators, the unknown-command-provider test expects
the new provider_not_registered error, and _transcribe_local tolerates
a null stt.local config section again.
Two small fresh fixes on top of the salvage wave:
- Wrap the check-then-load of the module-global faster-whisper model in a
double-checked threading.Lock so concurrent voice messages can't both
download/load the model (#24767).
- Treat an empty stt.openai.api_key as no-auth when stt.openai.base_url
points at a loopback/RFC-1918/.local host, so local OpenAI-compatible
STT servers (faster-whisper-server, speaches, vLLM whisper) work
without a sham api_key value. Reimplements the idea from PR #25193 —
credit @nnnet.
Co-authored-by: nnnet <nnnet@users.noreply.github.com>
Decode WeChat/QQ SILK v3 voice notes to WAV inside transcribe_audio so
any platform that caches a .silk file gets STT for free (same central-
normalization philosophy as the outbound container repair). pilk is
lazy-installed on first use (stt.silk in tools/lazy_deps.py) instead of
being added to the voice extra.
Fixes the inbound half of #32196.
(cherry picked from commit e5db79369d; reworked to compose with the
provider-scoped upload size cap and to lazy-dep pilk)
Normalize the structured <asr_text> marker after extracting text from string, SDK object, and dictionary transcription responses. Preserve the current provider-aware STT configuration architecture.
Refreshes #8773 on current main.
Co-authored-by: angelos <angelos@oikos.lan.home.malaiwah.com>
Assisted-by: Codex:gpt-5.6
Add isolated test where ensure('stt.mistral') raises FeatureUnavailable
but the raw mistralai.client.Mistral import succeeds, verifying the
transcription_tools.py fallthrough path introduced in the same PR.
Replace
aise ImportError(str(e)) with pass in the except Exception
handler of _import_edge_tts(), _import_elevenlabs(), and
_import_mistral_client() so packages installed via PYTHONPATH or Docker
layered filesystems still work when lazy_deps.ensure() raises.
Also fix the Mistral STT path in transcription_tools.py which only
caught ImportError, not FeatureUnavailable.
Adds 6 regression tests using sys.modules fixtures (no
builtins.__import__ patching).
Force CPU (int8) for faster-whisper on Apple Silicon / Rosetta, where
ctranslate2's device=auto path can hard-abort in native code. Salvaged
from PR #28624 without the numpy pin change (main already moved on).
(cherry picked from commit 7edf2d5196, pyproject.toml hunk dropped)
- Add Blackwell-specific cuBLAS error marker to _CUDA_LIB_ERROR_MARKERS
- Allows CPU fallback on RTX 5090 (sm_120) when faster-whisper
reports CUBLAS_STATUS_NOT_SUPPORTED instead of loading successfully
- Add regression test for CUBLAS_STATUS_NOT_SUPPORTED path
Closes#17526
The local STT transcription function hardcoded device="auto" and
compute_type="auto" when instantiating WhisperModel, ignoring the
user's stt.local.device and stt.local.compute_type config values.
Closes#8319
Rework of #68509 per triage: hoist the duplicated per-tool
_resolve_provider_key helpers into one owner,
tools.tool_backend_helpers.resolve_provider_secret(), and migrate every
STT/TTS key lookup site to it.
Resolution order: explicit config.yaml value > profile secret scope /
env / ~/.hermes/.env > credential pool (checks both '<provider>' and
'custom:<provider>' pool keys, so keys added via 'hermes auth add
mistral' or declared under providers.<name> both resolve). Under an
active multiplex turn the profile scope stays authoritative — no pool
or .env fallback that could borrow another profile's key (composes with
the #69469 scope fix).
Coverage now includes GROQ_API_KEY, MISTRAL_API_KEY, ELEVENLABS_API_KEY,
DEEPINFRA_API_KEY, MINIMAX_API_KEY, GEMINI_API_KEY/GOOGLE_API_KEY, the
XAI_API_KEY fallback in resolve_xai_http_credentials, and the OpenAI
audio key (resolve_openai_audio_api_key now pool-aware for
OPENAI_API_KEY via 'hermes auth add openai-api').
Unit tests: fake pool entry proves each provider resolves from the pool
when env is empty; env still wins when set; config wins over both; a
multiplex scope miss never borrows the pool; pool read failures never
raise; tool-level wiring for STT, TTS, xAI, and OpenAI audio.
Fixes#68003
`resolve_openai_audio_api_key()` reads the key that authenticates the audio
client straight from the process environment:
return (
os.getenv("VOICE_TOOLS_OPENAI_KEY", "")
or os.getenv("OPENAI_API_KEY", "")
).strip()
That value is not advisory. It flows through
`_resolve_openai_audio_client_config()` into `OpenAIClient(api_key=...)` for
TTS, and through `transcription_tools` for voice-note STT — both on the
per-turn tool path, inside the profile secret scope the gateway installs.
`agent/vertex_adapter` states the contract this breaks:
in a multiplex gateway serving several profiles from one process,
os.environ reflects whichever profile's .env happened to be loaded at
boot, not the profile the current turn belongs to. Reading it directly
here would let one profile mint tokens from — and get billed against —
a different profile's service-account file.
Reproduced with the real resolver, multiplexing on and profile A's scope
installed:
scope-aware get_secret -> sk-PROFILE-A-key
voice/STT resolver -> sk-PROFILE-B-key
So profile A's spoken reply and its users' voice notes are sent to OpenAI on
profile B's account, and billed there.
Route both reads through `agent.secret_scope.get_secret`, the same fix already
merged for the WeChat send path (#59662) and pending for QQ (#60420) — neither
covers the audio credential family. Under multiplexing the scope stays
authoritative, so a scope miss now yields no key instead of borrowing another
profile's; with multiplexing off `get_secret` falls through to `os.environ`
exactly as before, so single-profile deployments are untouched. The
VOICE_TOOLS_OPENAI_KEY > OPENAI_API_KEY precedence is unchanged.
Deliberately narrow: `fal_key_is_configured()` and
`has_direct_modal_credentials()` in this file are presence checks, not
authentication, and the former is already being reworked in open PR #20929.
tests/tools/test_tool_backend_helpers.py: the scope wins over another
profile's `os.environ`; a scope miss does not borrow another profile's key;
voice-key precedence holds inside a scope; and a control proves the
single-profile path still reads `os.environ`. The three isolation tests fail
on main; the control passes there. 320 passed across the helper, secret-scope,
and consumer suites (the fluctuating voice_mode/voice_cli failures are
pre-existing PulseAudio/ordering artifacts — the differing test passes 3/3 in
isolation on both main and this branch).
Per-job cron inference pins are now user-owned: the agent-facing cronjob
tool schema no longer exposes model/provider/base_url, and the registered
handler ignores them even if a model hallucinates the old parameters.
Users set pins via the dashboard, hermes cron create/edit --model/--provider,
or jobs.json directly — and once set, a pin sticks until the user changes it.
Existing agent-era pins are grandfathered untouched.
New cron.model / cron.model_provider config keys give the cron fleet its
own default model, independent of the chat model. Fire-time resolution:
per-job pin > cron.model > HERMES_MODEL > model.default. An axis covered
by the explicit cron-fleet default is deliberate routing, not drift, so
the #44585 fail-closed guard skips it — switching your chat model with
/model or hermes model no longer breaks unpinned cron fleets.
- tools/cronjob_tools.py: drop model param from agent schema + handler;
remove now-dead _resolve_model_override
- cron/scheduler.py: cron.model/model_provider resolution + per-axis
drift-guard skip
- cron/jobs.py: snapshot resolution mirrors the new precedence
- hermes_cli/subcommands/cron.py + hermes_cli/cron.py: --model/--provider
on hermes cron create/edit
- hermes_cli/config.py: cron.model / cron.model_provider defaults
- docs: cron.md model-resolution tip rewritten
One sniffer owns magic-byte container detection (Teknium's one-concept-
one-owner rule): the new tools/audio_container.py is used by
- gateway/platforms/base.py _sniff_audio_ext (inbound cache — PR #36166's
central sniffer, now covering AAC/ADTS, MP4-brand disambiguation, webm)
- gateway/platforms/signal.py _guess_extension (audio/AV branches
delegated; RIFF/WAVE fix from PR #50690 and M4A-brand fix from
PR #72490 now live centrally)
- tools/tts_tool.py _sniff_audio_container (outbound repair, PR #73072)
cache_audio_from_url inherits the sniff via cache_audio_from_bytes.
Adds tests/tools/test_audio_container.py covering every magic-byte type,
wrong-extension repair on the inbound cache, unknown passthrough, the
URL path, and Signal's delegation.
Telegram sends voice notes as .oga (OGG/Opus). SUPPORTED_FORMATS listed
.ogg but not .oga, so transcribe_audio rejected every Telegram voice note
with "Unsupported format: .oga" before reaching any STT backend. Add .oga
and .opus to the allowlist, with a regression test.
'hey hor' triggered the wake word: the default sensitivity was 0.5, which
for openWakeWord IS the raw per-frame score threshold — openWakeWord's own
permissive baseline that near-misses clear. Raised the default to 0.6 so
phonetic near-misses fall short while real 'hey hermes' (typically 0.9+)
still fires easily.
Also fixed a real cross-engine inconsistency found while checking: the
sensitivity knob is documented 'higher = stricter' and behaves that way
for openWakeWord (threshold = sensitivity) and sherpa (0.05 + 0.4*s), but
Porcupine's own 'sensitivities' param runs the opposite way (higher = MORE
false alarms, per Picovoice). Turning sensitivity up made Porcupine looser
— backwards. Now inverted (1 - sensitivity) so 'higher = stricter' holds
for every engine.
- tools/wake_word.py: default 0.6; _sensitivity fallback uses _DEFAULTS;
Porcupine sensitivity inverted with rationale
- hermes_cli/config.py + docs: default + consistent-direction note
- tests: Porcupine inversion, default>=0.6 regression, fallback-to-default
openWakeWord scores one ~80ms frame at a time and the detector fired the
instant a SINGLE frame crossed threshold — so a stray phoneme in background
conversation could trigger the wake word unintentionally (reported in
testing). A real utterance of the phrase holds a high score across several
consecutive frames; an ambient blip spikes just one.
_OpenWakeWordEngine now requires N consecutive over-threshold frames
(wake_word.confirmation_frames, default 3) before firing. The streak resets
on any sub-threshold frame and on engine reset() (pause/resume), so a
pre-pause frame can't count toward a post-resume fire. confirmation_frames=1
restores the old single-frame behavior; clamped 1..10.
Only openWakeWord is affected — sherpa (streaming transducer) and porcupine
decode the whole phrase internally and already reject single-frame spikes.
- tools/wake_word.py: _confirmation_frames() accessor, streak logic in
process()/reset(), config default
- hermes_cli/config.py: wake_word.confirmation_frames documented default
- tests: 5 new (spike rejected, sustained fires once, =1 legacy behavior,
reset clears streak, config clamp) — 58 wake tests green
- docs: 'Reducing false triggers on ambient speech' section
Follow-up to benbarclay's macOS ARM64 tflite fix:
- test_openwakeword_bundled_model_matches_framework stubs
ensure_tflite_runtime()=True so it exercises artifact selection, not
runtime availability — off-Darwin the bridge legitimately returns
False and the engine falls back to onnx (that path is covered by its
own tests). Was failing on Linux CI otherwise.
- check_wake_word_requirements now ANDs both this branch's STT/TTS gate
and Ben's tflite_ok into 'available' (cherry-pick conflict resolution).
openWakeWord's ONNX backend returns near-zero scores on Apple Silicon
(dscripka/openWakeWord#336), so "Hey Hermes" never crossed the 0.5
threshold: the listener armed, the microphone worked, and nothing fired.
Bisecting the pipeline puts the fault in exactly one stage — feeding the
same audio through both backends, the melspectrogram front-end is
bit-identical (maxdiff 0.00000) and the wake classifier agrees on
identical features, while the shared embedding model diverges by 45.44.
Cross-feeding confirms it: tflite features scored through the *onnx*
classifier give 0.9948 vs 0.000009 for onnx features. A telling
secondary symptom is that scores fall as input gets louder (0.5x ->
0.00031, 8x -> 0.000066), which is garbage inference rather than a weak
detection.
Selecting tflite in config alone does not fix it. openWakeWord hardcodes
`import tflite_runtime.interpreter` but declares tflite-runtime for
`platform_system == "Linux"` only; on macOS the equivalent wheel is
ai-edge-litert, so that import always fails and model.py silently
downgrades back to onnx. The result is a detector that reports itself
listening and can never fire.
- default the backend per platform (tflite on macOS ARM64, onnx
elsewhere) instead of hardcoding onnx, and pick the matching bundled
model artifact
- bridge tflite_runtime -> ai_edge_litert through sys.modules, in-process,
with no writes to site-packages
- refuse the silent onnx downgrade on macOS ARM64 and report the missing
runtime through check_wake_word_requirements() so the GUI surfaces an
actionable hint rather than arming a dead ear
- lazy-install ai-edge-litert via its own feature key, because lazy-dep
specs cannot carry PEP 508 markers (_spec_is_safe rejects ";")
An explicit `inference_framework` in config still wins, so anyone pinning
a backend keeps it.
Verified on macOS 26.5.2 / M-series: "hey hermes" scores 0.0005 on onnx
and 0.9423 on tflite from the same clip, with cross-phrase controls at
0.0003. Live over-the-air through the real microphone fires 4/4
utterances (peak 0.9532).
Two fixes from live testing (Teknium):
1. Desktop installs now ship the wake/voice stacks up front.
install.sh + install.ps1 desktop stages run 'uv pip install
-e .[wake,voice]' (best-effort, lazy-install remains the fallback)
before building the app, so the first ear-click arms instantly
instead of sitting through a multi-minute onnxruntime download.
CLI-only installs keep the lazy path — [all] curation unchanged.
2. The vanished ear: the STT/TTS gate made wake.status call
check_tts_requirements(), whose edge path runs _import_edge_tts →
lazy_deps.ensure — a synchronous PIP INSTALL inside a status poll.
On a venv without edge-tts that blew the desktop's 30s RPC timeout,
armWakeWord caught the error, the atom never learned enabled=true,
and the ear unmounted. _tts_ready is now a pure probe: deps missing
+ lazy installs allowed counts as ready (installs at first speak)
WITHOUT touching pip; check_tts_requirements only runs once deps
are present. Regression test asserts the probe never calls it while
deps are missing.
The wake loop is wake → record → STT → agent → TTS. Arming without
either end configured gives a mic that hears you and then does nothing
perceivable — a useless experience. check_wake_word_requirements now
probes both (same probes /voice uses: stt.enabled + provider != none;
check_tts_requirements) and refuses with a pointer to `hermes tools`
naming exactly which half is missing. The desktop ear hides (available:
false already hides the button), /wake on prints the hint on CLI/TUI.
wake.start also validates requirements BEFORE persisting
wake_word.enabled, so a refused gesture can't leave config claiming on
while nothing can ever arm.
Tests: per-half and both-missing hint assertions; existing requirements
tests pinned via _voice_loop_ready so they don't depend on the test
venv's installed voice stack. E2E: stt.enabled=false in a real config
-> unavailable with the speech-to-text hint.
Internal testing (macOS): clicking the ear froze the button for ~30s,
then it went blue but 'hey hermes' never fired even though STT worked.
Two distinct bugs:
1. Frozen-then-timeout button: first-use wake.start lazy-installs the
detection engine (onnxruntime is a large wheel), which blows the
desktop's default 30s WS request timeout — the RPC 'failed' client-
side while the backend kept installing and armed later on its own.
wake.start now gets a 180s budget and the pending state says
'arming — first use may take a minute while the engine installs'
instead of a silent disabled button.
2. Armed but deaf: macOS grants mic permission per PROCESS. The
renderer having mic access (STT working) does not grant the Python
backend anything — CoreAudio hands an unentitled process a 'working'
stream that delivers zeros forever, so the listener looks healthy
and can never hear the phrase. The detector now tracks frame peaks:
10s of consecutive near-zero frames sets audio_silent, logged with
the exact macOS Settings path, cleared automatically when audio
appears. Surfaced everywhere: wake.status (audio_silent + hint),
desktop ear tooltip (kept visible while listening), /wake status on
TUI (⚠ line) and classic CLI, plus a docs troubleshooting section.
Tests: detector silent-flag set/recover cycle (fake silent/loud
streams), desktop tooltip keeps the dead-mic hint while listening.
44 wake Python tests, 22 desktop + 14 TUI vitest green.
check_wake_word_requirements() gated 'available' on the audio probe,
but the probe imports sounddevice + numpy — two of the packages the
lazy installer would install. On a fresh machine deps_ok was False, so
audio_ok was always False and /wake on printed the manual pip hint and
bailed before the engine constructors' lazy_deps.ensure() could run.
Now the audio probe only runs once deps are installed; with deps
missing and lazy installs allowed (the default), /wake on proceeds and
ensure() installs the pinned engine deps in-process — no restart. The
manual pip hint remains for security.allow_lazy_installs=false, and a
mic hint still blocks when deps are present but no audio device works.
The CLI announces the one-time engine install so the pause is explained.
numpy is a lazy voice-extra dep absent from CI's hermetic env; the two
engine process() tests imported it for real. Stub asarray/float32 in
the fixture (verified against a blocked-numpy import, matching CI).
One sherpa listener now enrolls every wake-enabled profile's phrase
(defaulting to "hey <profile name>") and reports WHICH phrase fired.
wake.detected gains a profile field; the desktop live-switches to the
matching profile (same path as the profile rail), opens a fresh session
there, and starts hands-free voice. The single-profile CLI/TUI print the
hermes -p switch command for foreign-profile phrases instead of
answering as the wrong profile. Opt out per listener with
wake_word.profile_routing: false.
New "sherpa" wake_word provider: the configured phrase is BPE-tokenized
at runtime against a small streaming zipformer KWS model (~13 MB English,
one-time download cached under HERMES_HOME), so ANY typed phrase works
with zero training — including per-profile phrases like "hey coder".
wake.sherpa lazy-dep group + [wake] extra grow sherpa-onnx/sentencepiece;
requirements probe routes per provider; sensitivity maps onto sherpa
keywords_threshold. E2E-verified on real audio (target phrase fires,
foreign phrase stays silent, reset drops buffered state).
From #53378: ships hey_hermes.onnx/.tflite (openWakeWord pipeline,
Apache-2.0) under tools/wakewords/, resolves the default (and hey_hermes
aliases) to the bundled file, ensures openWakeWord base feature models
are fetched for custom paths too, and updates config defaults + docs
from hey_jarvis to hey hermes.
Two bugs surfaced by the desktop wake conversation:
1. Runaway loop: wake -> voice -> resume -> wake fired again within
~200ms. openWakeWord keeps its rolling feature buffer across
pause/resume, so on resume it immediately re-scored the "hey jarvis"
captured before the pause and re-fired, reopening a session and
restarting voice in a tight cycle. Reset the engine buffer on every
detector (re)start so resume begins from clean audio.
2. Empty-transcript toast: a silent re-listen returns
success:false / "… STT returned empty transcript", which the desktop
transcribe endpoint turned into a 400 -> thrown error -> "Voice
transcription failed" notification on every silent gap. Treat an empty
transcript as no-speech: return {ok, transcript: ""} so the voice loop
quietly re-listens. Real failures still 4xx/5xx.
Makes the wake word a tri-surface feature with one configurable owner.
- wake_word.surface ("auto" | "cli" | "tui" | "gui") + shared
wake_surface_enabled() gate consulted by every surface, so exactly one
place owns the listener and the new session it opens.
- tui_gateway: wake.start/stop/pause/resume/status RPCs + a wake.detected
event, sharing one server-side detector for both TUI and desktop. The
detector yields the mic to voice.record (pause on capture start, resume
on terminal) and to the desktop's browser mic (wake.pause/resume).
- TUI (Ink): arm wake.start on gateway.ready; on wake.detected open a
fresh session and start voice capture.
- Desktop (Electron): arm wake.start on connect; on wake.detected open a
fresh session.
- CLI now gates on wake_surface_enabled("cli"); /wake status shows surface.
- Tests for the surface gate; docs cover the surface knob + cross-surface.
Adds an opt-in, on-device hotword listener for the CLI. With
wake_word.enabled (or /wake on), Hermes listens in the background for a
wake phrase; on detection it starts a fresh session, captures one
utterance through the existing voice pipeline, and answers — the
"Hey Siri" pattern.
- tools/wake_word.py: provider-pluggable detector (openWakeWord, free
local default; Porcupine, premium) over the shared 16 kHz sounddevice
capture path. Background daemon thread with pause/resume so it yields
the mic during a voice turn.
- CLI wiring: startup listener (off-thread), on-wake flow, an idle
watchdog that resumes the detector after each turn, cleanup hook, and
a /wake [on|off|status] command.
- config.yaml wake_word section; PORCUPINE_ACCESS_KEY as an optional
secret. Engines lazy-install via the [wake] extra.
- Hands a transcript to the input queue exactly like voice mode, so no
system-prompt/cache mutation. No new core model tool.
- Tests (mocked, no live audio/network) + feature docs.
Skip hashing active copies when the bundled origin is unchanged, build rename and optional migration indexes lazily, and reuse one optional-skill directory index per sync.
Add regression coverage for unchanged copies, deferred modification detection, lazy rename recovery, and optional provenance scans.
Figma's mcp.figma.com register endpoint is a client_name allowlist
(Claude Code / Codex succeed; Hermes Agent 403s) and returns a client
secret while advertising auth_method=none, then requires the secret on
token exchange. Auto-set client_name + client_secret_post for Figma
hosts, pass oauth cfg through login/add paths, force interactive OAuth
for hermes mcp login from non-TTY desktop shells, and ship a catalog
entry. Proven: hermes mcp login figma → 26 tools.
Saying EXACTLY a configured stop phrase (default: 'stop') and nothing
else now ends the voice conversation instead of being sent to the agent
as a prompt. Match is deliberately strict — whole utterance,
case-insensitive, surrounding punctuation stripped — so 'stop doing
that and try X' still reaches the agent.
- tools/voice_mode.py: is_voice_stop_phrase() + voice.stop_phrases
config loader (default ['stop'], [] disables, malformed config falls
back safely).
- hermes_cli/voice.py: shared continuous loop (TUI + desktop) halts on
a stop phrase exactly like the silent-cycle limit (fires
on_silent_limit so every UI turns voice off); stop_continuous
force-transcribe path swallows the phrase without counting a silent
cycle.
- cli.py: classic CLI push-to-talk/continuous path and barge-in
utterance path disable voice mode on a stop phrase.
- Config default voice.stop_phrases: ['stop']; docs updated.
Tests: tests/tools/test_voice_stop_phrase.py (27 tests,
sabotage-verified: loop test fails when detection is disabled);
voice suites green (110 + 44 passed).
Whisper auto-detection frequently misidentifies short/accented clips,
which users experience as voice notes transcribed in the wrong language
(Teknium + CTO both hit this). The unified resolver from #73067 made a
global hint possible; this makes it the DEFAULT so stock installs stop
guessing. Non-English users set stt.language once; '' restores
auto-detect for multilingual use.
Deep-merge gives existing configs the new default automatically (no
_config_version bump needed); any explicit per-provider or global
language setting still wins.
Root-cause fix for the 'TTS voice bubble broken' issue family (#57048,
#54589, #57213, #58845, #14841, #45557, #57049). Two class-level defects:
1. Several backends silently write MP3/WAV bytes into a .ogg output path
(Edge only emits MP3, Piper writes WAV, xAI writes MP3, some
OpenAI-compatible servers ignore response_format=opus). Platforms that
need real Ogg/Opus render 0-second/broken voice bubbles. Instead of
per-provider patches, text_to_speech_tool now sniffs magic bytes once
after synthesis (_sniff_audio_container) and repairs the container
centrally (_repair_ogg_container): ffmpeg transcode in place, or rename
to the honest extension when ffmpeg is unavailable. Covers every
current and future provider, including command providers and plugins.
2. want_opus only recognized Telegram, so Matrix/Feishu/WhatsApp/Signal
auto-TTS voice replies were synthesized as MP3 and delivered as broken
attachments. New OPUS_VOICE_PLATFORMS set covers all voice-bubble
platforms.
_convert_to_opus refactored onto a shared _ffmpeg_transcode_to_opus that
supports safe in-place transcodes (-f ogg forced muxer, temp file +
os.replace).
Tests: tests/tools/test_tts_container_repair.py (13 tests incl. a live
ffmpeg round-trip); full TTS suite green (153 + 187 passed).
Class-level fix for the 'STT transcribes the wrong language' issue family
(#55551, #50181 and siblings). Previously language handling was per-provider
chaos: local honoured stt.local.language, Groq/OpenAI/Mistral/DeepInfra sent
no language hint at all, xAI silently forced 'en', ElevenLabs used its own
language_code key, and there was no global setting.
- New _resolve_stt_language() helper: stt.<provider>.language >
stt.language (new global key) > HERMES_LOCAL_STT_LANGUAGE > auto-detect.
- Threaded through ALL providers: local, local_command, groq, openai,
mistral, xai, elevenlabs, deepinfra (shared OpenAI handler), command
providers, and plugin dispatch.
- xAI no longer forces English when nothing is configured (auto-detect).
- Mistral Voxtral now receives a language hint when configured.
- stt.groq.model is now honoured from config (previously env-only).
- DEFAULT_CONFIG gains stt.language, stt.groq, stt.xai, stt.mistral.language.
- Tests: tests/tools/test_stt_language_resolution.py (11 tests, sabotage-
verified) + full transcription suite green (236 passed).
Builds on cherry-picked contributor work from #19786 (@zombopanda),
#23161 (@materemias), #50684 (@BlackishGreen33).
`stt.groq: null` in config.yaml yields `groq_cfg = None`, so the
subsequent `.get("language")` raised AttributeError. Use `or {}`
(matching main's widened xai/local provider guards) and add a
`{"groq": None}` regression test confirming auto-detect stays intact.
Addresses hermes-sweeper review on #23161.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_019Zc7Tgei4kav4n6WSsBq5F
- Normalize stt.groq.language: cast to str, strip, treat
empty/whitespace as unset (parity with xAI's str().strip()).
- Clarify "blank = auto-detect" inline comments in 4 docs/configs to
reflect the env-var fallback (HERMES_LOCAL_STT_LANGUAGE).
- Document that HERMES_LOCAL_STT_LANGUAGE also drives the local
faster-whisper provider, not just the CLI fallback.
- Add unit tests covering: omitted language when unset, config-supplied
language, env fallback, config-over-env precedence, whitespace
normalized to unset.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Add optional stt.openai.language config for OpenAI transcription. Forward non-empty hints to the API while preserving auto-detection when unset. Document the config-only setting, add its default, and cover configured and unset request arguments.
Tool-schema assembly at CLI/Desktop startup runs the browser-family
check_fns (browser, browser_cdp, browser_dialog, browser_vision). Each
of those gates called _get_cdp_override(), which resolves the configured
endpoint over HTTP (GET /json/version, timeout=10) — so a *stale*
browser.cdp_url pointing at a dead debug browser cost ~7 serial blocking
socket connects before the banner rendered. Measured on a real Windows
install with a dead http://[::1]:9222 config: 15.1s of an 18s launch,
with no warning or error — just mystery slowness. The value is easy to
leave behind: /browser connect writes a session-scoped env override, but
'hermes config set browser.cdp_url' persists forever while the debug
Chrome it pointed at dies on the next browser restart.
Split the helper:
- _get_cdp_override_raw() — returns the configured value (env var or
config.yaml) with zero network I/O. Used by every is-it-configured
gate: check_browser_requirements, _browser_cdp_check, _is_local_mode,
_is_local_backend, _navigation_session_key, _should_inject_engine
(via _is_local_mode), and the hermes doctor chromium-skip check.
- _get_cdp_override() — unchanged contract (raw + /json/version
resolution), now only called on paths that are about to connect:
session creation and the dialog-supervisor attach.
This follows the existing rule in check_browser_requirements ('do not
execute agent-browser --version here') and the browser.manage status
path, which already banned _get_cdp_override for exactly this reason
(test_browser_manage_status_does_not_call_get_cdp_override): schema
assembly must not perform blocking I/O.
A/B on the same machine, same dead endpoint: get_tool_definitions()
15.08s unpatched -> 1.89s patched, with browser_cdp/browser_dialog
still advertised (gate now keys off configuration, not reachability —
matching the documented lazy-supervisor contract in
_browser_dialog_check).
Adds a regression test asserting the browser_cdp check_fn never touches
the network.
PR #72425 added getattr(agent, '_incremental_persistence_failed', False)
checks at the top of execute_tool_calls_{sequential,concurrent,segmented}.
A bare MagicMock auto-creates a truthy value for any attribute access,
so the interrupt-skip test's MagicMock agent short-circuited before
appending cancelled-tool messages — assert len(messages)==3 got 0.
Production is unaffected: run_conversation resets the flag to False
explicitly at turn start (conversation_loop.py:~1028).
The change-detector test asserted 'hasattr(signal, "SIGPIPE")' in source.
PR #72677 replaced inline hasattr+signal.signal with _install_signal()
which does the same guard via getattr internally. Update the test to
accept either form.
Port from nearai/ironclaw#5149 (the describe-first live-hardening fix in
their progressive tool disclosure work): when a model invokes a deferred
tool through the tool_call bridge without the schema-required arguments,
return the tool's parameter schema instead of dispatching blind.
Pre-fix, a blind call produced an opaque downstream failure
("[TOOL_ERROR] Tool execution failed: KeyError: 'document_id'") that
teaches the model nothing about what the tool expects — IronClaw observed
cheap models looping ~30 identical invalid calls until the iteration
budget died. Post-fix, the model repairs the call in one round-trip.
- tools/tool_search.py: new validate_deferred_call_args() — key-absence
check of schema 'required' fields only; no type checking (coerce_tool_args
already repairs types downstream); fails open on any validator error so
it can never block a legitimate dispatch.
- model_tools.py: probe after the scope gate in the bridge dispatch.
- agent/tool_executor.py: probe in both unwrap sites (concurrent +
sequential) before the underlying tool replaces the bridge; sequential
path flattens the payload to match its {"error": str} wrapping.
- tests: TestDeferredCallSchemaProbe — blind call returns schema (not
KeyError), valid/optional calls dispatch, unvalidatable tools fail open,
out-of-scope rejection unchanged.
Port from openai/codex#33464: GNU rm permutes options, so
`rm build/ -rf`, `rm build/ -r -f`, and `rm build/ --recursive
--force` are equivalent to the flags-first spellings — but every
existing rm pattern required the flag group BEFORE the path, so these
spellings ran with no approval prompt at all (proven live on main).
The hardline floor was NOT affected: protected paths (/, system dirs,
$HOME) match regardless of flag position because the hardline path
matcher does not require flags. The gap was the approval-prompt layer
for arbitrary paths.
New DANGEROUS_PATTERNS entry with a tempered operand run: cannot cross
command separators (; | & newline), quotes, or a bare -- end-of-options
separator (after --, -rf is a literal filename). Flag token must be
whitespace-anchored so the r inside long options like --registry does
not count.
9 positive + 7 negative shapes in tests; approval cluster (868 tests)
green.
Port from anomalyco/opencode#38133/#38134 (patch Unicode matching corpus):
extend UNICODE_MAP with the Zs space-separator family (en/em quad, en/em/
three-per-em/four-per-em/six-per-em/figure/punctuation/thin/hair spaces,
narrow NBSP, medium mathematical space, CJK ideographic space) and the
Unicode minus sign U+2212.
Before: a file containing typographic spacing (French narrow NBSP, CJK
ideographic spaces, math minus) never matched a model's ASCII old_string
via the precise strategies — the edit only succeeded through the
similarity-based context_aware fallback, which (a) can pick the wrong
region (#54572 family) and (b) silently flattens the file's Unicode to
ASCII on replacement. After: these match at unicode_normalized (strategy
7), whose _preserve_unicode_in_replacement keeps the file's typographic
characters in unchanged spans.
All additions are 1:1 mappings, so the existing position-mapping and
preservation logic apply unchanged. Proven live before/after with a
multi-line probe; 59 fuzzy-match + 188 file-tools/patch/skill-manager
tests pass.
Inspired by Claude Code 2.1.214, which added permission prompts for
container-CLI commands (including the Podman shim) carrying
daemon-redirect flags (--url, --connection, --identity, remote mode)
that previously ran without one.
A daemon redirect makes a local-looking command operate on a different
(often remote) daemon, silently acting on production infrastructure.
Any container-CLI invocation carrying a redirect now requires approval
regardless of subcommand:
- -H/--host and --context global flags (value required, global-flag
position only — bare -h help and run-level -h <hostname> stay allowed)
- context use (persistently switches the default daemon)
- podman --url/--connection/--identity and -r/--remote
- DOCKER_HOST=/DOCKER_CONTEXT=/CONTAINER_HOST=/CONTAINER_CONNECTION=
environment prefixes
Sibling-site widening: the existing container lifecycle rules matched
only the verb directly adjacent to the binary name, so a global flag or
a compose -f file flag slipped past the guard, and the legacy hyphenated
compose binary was never covered. They now tolerate global flags — the
same treatment the 'hermes ... gateway' rule already has — and match the
hyphenated compose binary.
Validation: 33 new tests; 339 pass in test_approval.py + new file;
442 pass across the adjacent guard suites; E2E battery of 12 dangerous
+ 17 safe commands via real imports, hot path ~330us/call.
Regression for #71296: dump+source must not execute continuation lines from
HERMES_SESSION_CHAT_NAME/USER_NAME, and the export snippet must unset by
name instead of grepping declare lines.
Widen the cherry-picked /diff base (#4839 by @SHL0MS) into one
cross-surface implementation, folding in the review feedback and the
best ideas from the two sibling PRs (#22703, #53527):
- tools/working_diff.py: shared git collection layer — unstaged
(default), staged, and all (vs HEAD) modes; untracked files folded in
via `git diff --no-index` so new files appear as additions (Codex
/diff parity); shlex-split arguments preserve quoted paths.
- CLI: handler moved to hermes_cli/cli_commands_mixin.py per the
current god-file decomposition (dispatch stays in cli.py), renders
through the rich console with a 400-line terminal-flood guard.
- Gateway: _handle_diff_command in gateway/slash_commands.py + dispatch
in gateway/run.py; fenced ```diff output truncated to 60 lines /
3000 chars before the platform senders apply their own per-platform
message clamps (tool-progress-style layered truncation). Localized
strings in all 17 locale catalogs.
- /diff session (from #53527): cumulative checkpoint-baseline diff of
everything Hermes changed, via new CheckpointManager.session_diff();
docstring records the retained-baseline approximation caveat from
review. Works on both surfaces; degrades with an actionable message
when checkpoints are off.
- Slack: /diff routed via /hermes diff (50-slash cap; keeps
telegram-parity test green and /version native).
- Registry: cross-surface CommandDef with staged|all|session
subcommands; docs: slash-commands reference (CLI + gateway tables +
both-surfaces list) and hermes-agent skill reference.
- Tests: tests/tools/test_working_diff.py (real git repos),
tests/hermes_cli/test_diff_command.py (real git + stubbed checkpoint
manager), tests/gateway/test_diff_command.py (end-to-end handler,
real checkpoint store), TestSessionDiff in
tests/tools/test_checkpoint_manager.py.
Salvaged from the /diff PR cluster #4839 + #22703 + #53527.
Co-authored-by: Ninso112 <ninso112@proton.me>
Co-authored-by: Harshkamdar67 <harshkamdar67@gmail.com>
After N consecutive guardian denials in a session the deny message escalates to a hard-stop instruction. Inspired by ChatGPT Work auto-review circuit breaker.
Cross-surface coverage #23768 missed (per review feedback):
- tools/clarify_gateway.py: _ClarifyEntry carries a multi_select flag
(register() accepts it; signature() exposes it to adapters).
_coerce_text_response now parses multi-select replies — comma- or
space-separated numbers ('1,3' / '1 3'), exact labels, dedup — into a
JSON array string that _parse_multi_select_response decodes into a
list. Out-of-range/unknown tokens reject the reply (native button UI)
or fall back to custom text (awaiting_text/'Other' mode).
- gateway/run.py: _clarify_callback_sync accepts multi_select and
registers it on the pending entry.
- gateway/platforms/base.py: default numbered-list text fallback tells
the user multiple selections are allowed and how to reply.
- tui_gateway/server.py: clarify_callback passes multi_select through
the clarify.request payload as a hint; renderers without checkbox
support ignore the field and remain single-select-compatible.
- tests: 13 new gateway tests (flag storage, comma/space/single-number
parsing, label matching, out-of-range rejection, dedup, end-to-end
resolve, single-select regressions).
Follow-ups to the salvaged #23768 commit, which targeted a pre-79559214
codebase:
- agent/tool_executor.py + agent/agent_runtime_helpers.py: pass
multi_select at both current clarify dispatch points (the PR's
run_agent.py edits landed on dead code paths).
- tools/clarify_tool.py: replace the broad TypeError-retry in
_invoke_callback with inspect.signature detection, so a compatible
callback that raises TypeError internally is not invoked twice
(addresses hermes-sweeper review feedback on #23768).
- tests: cover single-invocation on internal TypeError, legacy 2-arg
callbacks, **kwargs callbacks, and registry handler multi_select
pass-through (schema arg → handler → callback).
hermes update's lazy-refresh pass re-asserts LAZY_DEPS pins whenever the
package is present (active_features() is presence-based). The
tool.trace_upload pin huggingface-hub==1.2.3 sat below transformers'
>=1.5.0,<2 requirement, so every update force-downgraded the shared
package and broke Hindsight local embeddings on daemon startup (#60783).
Keep the exact-pin security posture — no ranges — but move the pin to
1.24.0 (current) and bump uv.lock in lockstep (uv lock --upgrade-package
huggingface-hub: hub 1.4.1->1.24.0, hf-xet 1.3.1->1.5.2, click
8.3.1->8.4.2, drops typer-slim), so the entire tree converges on ONE hub
version. The refresh pass now reports 'current' with zero churn.
Invariant tests (not snapshots): the lazy pin must equal the uv.lock
resolved version, and must sit inside transformers' accepted window.
HfApi surface used by trace upload (whoami/create_repo/upload_file)
verified present with identical kwargs on 1.24.0 in a live venv.
Completes #51690 on top of the salvaged #60378 timeout metadata:
- async_delegation: terminal 'stalled' events now carry structured
stall context (stalled_after_quiet_seconds, stall_threshold_seconds,
stall_phase idle|in_tool, stall_grace_seconds) on both single and
batch paths, persisted in the durable row so restart-restored events
keep it. Mirrors the sync path's timeout_seconds/timed_out_after_
seconds/timeout_phase from #60378.
- list_async_delegations(): exposes seconds_since_progress and live
children_activity (per-child api_calls, current_tool,
seconds_since_activity) sampled from the dispatch's progress_fn
outside the records lock; private monitor bookkeeping and callables
never leak.
- /agents (CLI + gateway): background delegations render per-child
activity rows, quiet-time hints, and the stalling state; gateway
section is new (previously async delegations were invisible there).
New locale key gateway.agents.background_delegations in all 17
catalogs.
Tests: stall-metadata event shape, live-listing projection, gateway
/agents rendering (real registry dispatch, sabotage-verified), sync
timeout metadata fields, non-timeout None contract.
Include last_activity_ts in the progress token sampled from each child.
_touch_activity ticks on every streamed chunk ('receiving stream
response'), every tool transition, and API-call start/completion — so a
child mid-stream on a long response is alive even though api_call_count
only advances when the call completes. Same liveness signal as the
compaction inactivity budget (PR #71508): if tokens are flowing it never
dies; staleness is measured from the last streamed token / tool activity
/ API call.
Replace the wall-clock timeout watchdog (from #60234) with progress-based
staleness detection, on by default with zero config:
- The async registry now accepts a progress_fn per dispatch; delegate_task
wires a sampler over the batch's child agents (api_call_count +
current_tool from get_activity_summary()).
- A single monitor thread sweeps running delegations: a child whose
progress token keeps advancing is never touched, no matter how long it
runs. A frozen token past the stale threshold (450s idle / 1200s
in-tool, mirroring the sync-path heartbeat monitor) marks the record
'stalling' and interrupts the child.
- A stalling child that unwinds within the grace window (120s) finalizes
through the NORMAL path, preserving its partial results. One that never
returns is force-finalized with a terminal 'stalled' completion event so
the owning session hears an outcome and the async slot frees.
- Late runner returns after force-finalization are deduped by the
begin/push/finish finalization split (kept from #60234).
Why not a timeout: delegation.child_timeout_seconds defaults to 0 by
deliberate design (DEFAULT_CHILD_TIMEOUT rationale) — a timeout-based
watchdog never arms for default configs, leaving the reported silent-
profile symptom (#60203) unfixed, and when armed it kills legitimately
slow heavy subagents mid-task. Progress detection distinguishes 'wedged
at first API call' from 'grinding through a 2h review'.
Builds on izumi0uu's finalization-atomicity work from #60234.
Async background delegation can leave gateway sessions holding only a dispatched handle when the detached runner wedges before it can return and enqueue a completion. Enforce the configured child timeout in the async registry so the parent observes a terminal timeout event and the async slot is released.
Constraint: Issue #60203 reports long-lived gateway processes with background child delegates that never produce completion events despite child_timeout_seconds being configured.
Rejected: Relying only on _run_single_child timeout handling | it cannot finalize the async registry when the outer runner thread itself never reaches normal completion.
Confidence: high
Scope-risk: narrow
Directive: Keep background delegation completion owned by the async registry whenever detached workers can outlive the caller's immediate control.
Tested: .venv/bin/python -m pytest tests/tools/test_async_delegation.py tests/tools/test_delegate_subagent_timeout_diagnostic.py tests/tools/test_delegate.py -q
Tested: .venv/bin/python -m ruff check tools/async_delegation.py tools/delegate_tool.py tests/tools/test_async_delegation.py
Tested: git diff --check
Not-tested: Multi-day real gateway degradation; covered with deterministic stuck-runner registry tests.
CI slices 2 and 7 caught three tests broken by always-defer:
- /tools (CLI show_tools + TUI gateway tools.show) now passes
skip_tool_search_assembly=True — it's a discovery/inspection surface,
so users verifying an MCP installed must see deferred tools, not a
collapsed bridge row. This also fixes
test_slash_worker_mcp_discovery (profile MCP tool visible in /tools).
- test_plugins.py::test_plugin_tools_in_definitions: 'visible' becomes
'reachable' — direct schema OR listed in the bridge description;
scope-exclusion assertions unchanged (not direct AND not listed).
- test_discord_tool.py dynamic-schema-rebuild test reads the
pre-assembly list (the rebuilt schema is what tool_describe serves).
Banner/status tool counts intentionally keep the post-assembly view —
they reflect what the model actually sees.
The include/exclude filter matched exact names only — glob-style entries
('*_radar_*') silently matched nothing, so a Cloudflare flat-mode config
meant to trim 3,320 tools to ~1,900 actually registered 3,319. Unmatched
patterns produced no warning.
- matches_name_filter(): exact membership first (O(1) for literal lists),
then fnmatch.fnmatchcase for entries containing * ? [ — same pattern
semantics as approvals.deny. Entries without metacharacters stay
strictly literal ('docs' never matches 'docs_search').
- _should_register() uses it for both include and exclude (symmetric)
- hermes mcp tools picker (mcp_config.py) pre-selection uses the same
matcher so the UI agrees with runtime registration
E2E against the live Cloudflare capture with the real exclude list:
3,320 -> 1,905 surviving (1,415 excluded); radar/DLP gone,
purge_cache/dns_records kept. 220/220 mcp_tool tests (4 new).
Teknium review changes on the tiered policy:
1. threshold_pct default 10 -> 5 (listing budget = min(5% of context,
listing_max_tokens)); unknown-context fallback 20K -> 10K.
2. Tier 2 no longer leaves the model blind: when even names-only doesn't
fit, the bridge description now carries a one-line-per-server summary
('cloudflare (3320 tools)') plus an instruction to search FIRST rather
than substitute a generic tool or claim the capability is missing —
the measured tier-2 failure mode (core-tool substitution) at zero
meaningful token cost (~50 tokens/server).
3. Listing degradation is now PER SERVER, largest first: one oversized
server (Cloudflare) collapses to its summary line while small
co-attached servers (Linear) keep their full per-tool listings
('mixed' form). Previously global: attaching Cloudflare next to
Linear silently cost Linear its listing. Greedy fit is deterministic
(size then label) so the rendered block stays byte-stable per catalog
— prompt-prefix cache safe.
E2E on real captures (defaults, 200K ctx): linear alone -> tier 1 full;
unreal alone -> tier 2 groups (5% budget) / tier 1 names at 1M;
cloudflare alone -> tier 2 groups; linear+cloudflare -> tier 1 MIXED
(linear fully listed, cloudflare summarized). 48/48 tests.
Cloudflare's flat API MCP ships 61 property keys that violate Anthropic's
^[a-zA-Z0-9_.-]{1,64}$ pattern (query-filter params like 'issue_class~neq'
and 'meta.<field>[<operator>]'). One bad key anywhere in the tools array
400s the ENTIRE request — measured live: Anthropic, Bedrock, Google Vertex,
and Azure all rejected an eager 3,320-tool Cloudflare request at validation,
before token limits even applied.
- schema_sanitizer: rename non-conforming property keys deterministically
(bad chars -> '_', 64-char truncation, collision dedup with numeric
suffixes), nested schemas included; required[] remapped alongside
- unrename_tool_args(): reverse map applied in coerce_tool_args at dispatch,
so the MCP server receives the original wire names; recurses into object
values and array items
- deterministic on both sides: the rename map is recomputed from the
registry's original schema at dispatch time, no state carried
E2E on the real capture: 61 -> 0 violations across 3,320 tools; round-trip
verified on get_accounts_intel_attacksurfacereport_issues (5 '~neq' keys).
Tier 0: no MCP/plugin tools -> everything eager (pass-through).
Tier 1: deferred tools whose catalog listing fits min(threshold_pct%
of context, listing_max_tokens) -> bridge + skills-style
listing, degrading to names-only over budget.
Tier 2: listing over budget even names-only (Cloudflare's flat API
surface: 3,320 tools, names alone ~32K tokens) -> bare bridge,
discovery through tool_search only.
The old activation threshold (defer only when schemas > threshold_pct
of context) let mid-size catalogs ride eager and pay full schema cost;
with servers like Cloudflare (~597K tokens of schema, would not even
fit a 200K window) the binary gate is the wrong shape. Activation is
now driven purely by deferrable-tool presence; threshold_pct is
repurposed as the listing budget's context-relative leg.
- AssemblyResult gains tier + listing_form for observability
- listing_max_tokens default 4000 -> 20000 (cap 60000) so an 830-tool
catalog keeps a names-only listing while Cloudflare-scale drops to
bare bridge
- E2E verified against real captures: Linear 24 tools -> tier 1 full,
Epic UE 5.8 830 tools -> tier 1 names-only, Cloudflare 3,320 tools
-> tier 2 (both 200K and 1M context)
Deferred MCP/plugin tools become invisible once the tool_search bridge
activates — live benchmarking (48 runs, Claude Haiku 4.5) showed models
substituting visible core tools (terminal/web_search/browser) for deferred
capabilities or declaring them nonexistent instead of searching: 16/24
task success vs 24/24 with eager loading.
Skills never had this failure mode because every skill keeps a ~21-token
name+description listing line in the system prompt. This ports that exact
pattern to the tool bridge: when tool_search activates, a grouped manifest
of every deferred tool (name + first sentence of description, clipped to
60 chars, grouped per MCP server / toolset) is embedded in the tool_search
bridge description.
- tools/tool_search.py: build_catalog_listing() with deterministic
ordering (byte-stable across assemblies -> prompt prefix stays
cacheable); token-budget fallbacks full -> names-only -> legacy bare
count; bridge_tool_schemas(listing=...) embeds it and instructs the
model to skip tool_search when the exact name is visible (one fewer
round-trip per use)
- config: tools.tool_search.listing auto|on|off (default auto),
listing_max_tokens (default 4000, clamped 200..20000); legacy bool
shapes keep working
- tests: 8 new tests (config parsing/clamps, short-desc clipping,
deterministic rendering, budget fallbacks, bridge embedding, assembly
on/off paths); full file green (47 passed)
- docs: tool-search.md config table + rationale
- scripts/tool_search_livetest2.py: benchmark harness v2 with real
per-call token accounting (normalize_usage spy) and a third 'listing'
mode for A/B/C comparison
- require positive agent ownership proof before any background-review mutation
- fail closed when usage record is missing, malformed, or unreadable
- preserve foreground user-directed mutations and valid agent-owned curator flows
- cover patch, edit, delete, write_file, and remove_file end to end
Closes#67073
The background write guard decided ownership from `isinstance(usage_rec, dict)`,
so a local skill with NO usage record passed. That successful write called
bump_patch(), which created a `created_by: null` record — and the identical
write was refused from then on. "Allowed exactly once, then never" is a race
with our own bookkeeping, not a policy. Reproduced on main: patch #1 succeeds,
patch #2 with the same arguments is refused.
Option B from the issue. Option A (split `session_review` from
`scheduled_curator` and let the session fork patch user-owned skills it
consulted) would widen autonomous write permission onto skills the user owns
with no user present to consent — wrong direction for a no-user-present actor.
- skill_manager_tool: missing and explicit-null records now resolve
IDENTICALLY, both fail closed. The refusal names the reason and points at
`hermes curator adopt <name>`.
- background_review: both review prompts told the reviewer to patch any skill
consulted in the session and claimed pinned skills could be improved, while
enforcement refused both. Prompts now list pinned, external, and user-owned
skills as protected, and tell the reviewer to RECOMMEND adoption instead of
attempting a write that will be refused.
- skill_usage: document that `created_by` is a curator-management policy flag,
not a provenance claim, and add `is_curator_managed()` so call sites read as
the question they ask. Field name retained — it is on disk in every
`.usage.json` and renaming would strand those records.
- curator CLI: `hermes curator list-unmanaged` itemizes unmanaged skills with
the reason each is unmanaged (completes the #67139 spec).
Foreground writes are untouched: a user-directed edit to a user-owned skill
still works, including on pinned skills.
Sibling tests: 9 failures in test_skill_manager_tool.py were fixtures that
created record-less skills to exercise OTHER guards (consolidation-delete,
read-before-write) and relied on ownership falling through. Fixed at the
fixture, since the real curator only ever operates on managed sediment. One
test asserted the old "manually authored" wording; rewritten to assert the
behavior contract instead of the string.
Validation: 274 targeted tests + all 7 background-review files (60 tests) pass.
E2E on a temp HERMES_HOME (30 checks) covers the flip, foreground writes,
adoption unblocking, pin semantics, prompt/enforcement parity, and the new verb.
Each new test sabotage-verified: revert the fix, confirm it goes red.
Fixes#67140
`hermes curator status` reported only the skills it manages, staying silent
about curation-eligible skills it can never touch. On a 237-skill library that
meant 112 skills were invisible to every automatic transition with no signal
anywhere — the curator looked broken when it was working as designed.
A skill becomes curator-managed only when `created_by: agent` lands on its
usage record, and only the background review fork writes that marker. Two
populations therefore never qualify: records written before the marker existed
(no key at all, authorship unknowable) and every foreground
`skill_manage(create)` (unset by design — those skills belong to the user).
- skill_usage: `list_unmanaged_skill_names()` / `unmanaged_report()` enumerate
the blind spot, tagging each row with `has_provenance_key` so the two causes
are distinguishable. `adopt_skill()` writes the marker on user declaration
and refuses bundled, hub-installed, external, and protected built-ins.
Adoption never resets the inactivity clock.
- curator CLI: status prints an `unmanaged (no provenance marker)` block on
BOTH the managed and no-managed-skills paths; new `adopt` verb takes names or
`--all-unmanaged`, with `--dry-run` and a confirmation prompt on bulk.
- skills_sync: `_backfill_optional_provenance()` matched candidates by
repo-derived path only, so a skill installed at `mlops/chroma` that upstream
later moved to `mlops/vector-databases/chroma` was skipped forever and
`hermes skills repair-optional` could not fix it. Falls back to an
unambiguous name match, still gated on identical content, and records the
ACTUAL install path.
Provenance stays a declaration, never an inference: a high patch count proves
the agent MAINTAINS a skill, not that it authored one, since Hermes edits
user-written skills on the user's behalf routinely. An "looks agent-made"
heuristic would eventually archive hand-written work.
Validation: 347 targeted tests pass. Each new test verified via sabotage run
(revert the fix, confirm the test goes red) — one initially passed for the
wrong reason because the fixture pinned `prune_builtins` off, masking the guard
under test; fixed to force the shipped default on.
`sync_skills()` keys the bundled manifest by frontmatter name but computes
the destination from the bundled path. When upstream renames or
recategorizes a skill, the manifest key still matches while the new dest
does not exist yet, so the loop fell into its "in manifest but not on
disk" branch and misread the skill as user-deleted: the user's copy was
stranded at the old path forever and never received another update.
Three skills hit this in the July 2026 reorg (computer-use,
evaluating-llms-harness, serving-llms-vllm) — silently frozen at their
pre-rename content on every machine that ran `hermes update`.
Recovery only moves a stale copy when it is byte-identical to the origin
hash recorded the last time sync wrote it, which proves the directory is
ours rather than the user's work. User-modified copies are kept in place
with a warning, hub-installed paths are never touched, and a genuine
deletion (no copy anywhere on disk) is still respected.
- tools/skills_sync.py: add _recover_renamed_skill() plus the
_index_active_skills() / _read_hub_install_paths() indexes; call it
before classification and report moves via a new `relocated` key.
- hermes_cli/main.py: surface relocations in both `hermes update` skill
sync reporting sites.
- tests: 4 cases covering relocate, user-modified preservation,
hub-installed exemption, and genuine-deletion respect.
Follow-up for salvaged PR #69380. The snippet
'export -p | grep -vE ... > $tmp.$BASHPID || true' attaches the redirect to
the grep pipeline segment, so $BASHPID expands inside grep's pipeline
subshell — a DIFFERENT pid than the parent shell that expands the follow-up
'mv $tmp' operand. The dump landed in an orphaned temp file, mv failed
silently (2>/dev/null), and the shared snapshot never updated again:
exported env / venv activation stopped persisting between commands
(tests/tools/test_local_shell_init.py TestSnapshotEndToEnd caught it).
Wrap the pipeline in a brace group and attach the redirect to the group so
the expansion happens in the current shell, matching mv's expansion. Also
rename the shell-init probe var HERMES_SESSION_ENV_PROBE ->
HERMES_STICKY_ENV_PROBE: it matched the HERMES_SESSION_ prefix the salvaged
fix now intentionally strips from snapshots, and the snippet-shape test is
updated to pin the brace-group contract.
A single long-lived backend serves many sessions through one
_active_environments['default'] LocalEnvironment (the messaging gateway, TUI,
and desktop/web dashboard all collapse the terminal to 'default'). That
environment persists a bash session snapshot file and sources it before every
command. 'export -p' dumped the FIRST session's HERMES_SESSION_ID into the
snapshot, so every LATER session sourced that stale value and its
'echo $HERMES_SESSION_ID' reported a FOREIGN session's id — overriding the
correct per-command Popen env injected by _inject_session_context_env.
Confirmed on staging-fp: session A read its own id, session B (same backend)
read A's id. Reproduced locally with two threads sharing one LocalEnvironment;
the snapshot held 'declare -x HERMES_SESSION_ID=...' from the first session.
Fix: strip the per-session bridged vars (HERMES_SESSION_* / HERMES_UI_SESSION_ID
/ HERMES_CRON_AUTO_DELIVER_*, i.e. gateway.session_context._VAR_MAP prefixes)
from the snapshot at both dump sites in base.py. They are re-injected fresh on
every command, so a snapshot should carry only the user's own shell state, not
Hermes' per-turn session identity.
Complements the _set_session_context session_id fix: that ensures the ContextVar
carries the right id; this ensures the shared snapshot can't override it with a
neighbour's. Adds tests/tools/test_snapshot_session_id_leak.py (regex unit +
real two-session LocalEnvironment integration) and updates the export-shape
assertions in test_base_environment.py for the new 'export -p | grep' dump.
Asked to link to a session, the agent had no way to know the @session
reference syntax exists — every mention in the tool schema described
consuming a link the user dropped, never writing one — so it answered
with the title and timestamp as prose and the desktop had nothing to
render.
Every result now carries a ready-to-copy `link`, and the schema says to
write it inline instead of restating the title around it. The profile
segment is omitted when the active profile can't be named confidently;
a bare id still resolves.
Also skip linkifying a ref a model already wrapped in a markdown link,
which would otherwise rewrite into a nested link.
A delegate_task fan-out batch occupies ONE async-pool slot by design, so
active_count() (unit/slot count) reports a 3-task batch as 1 — which
undercounts real concurrent subagent load on the background_work metric.
Add active_task_count() that expands a running batch to its child count
(N-task batch -> N, single -> 1) and switch the background_work reader to
it. active_count() is unchanged (capacity semantics preserved). Adds a
contract test for the unit-vs-task distinction and documents both counts.
Follow-up to the cherry-picked #69063: egilewski's review found that the
_dir_has_any_entry(parent) guard treats ANY entry in the mount point's
parent as proof the volume is attached — but unmounting exposes the
UNDERLAY directory's own files (e.g. a .keep placeholder), so a populated
underlying mount-point dir still classified the project as an orphan and
deleted its ref/index/metadata. Reproduced on both main and the PR head.
Attachment evidence is now positive instead of circumstantial:
* _volume_evidence() records the parent directory's (st_dev, st_ino)
identity in the project's metadata while the workdir is observably
live (at _register_project/_touch_project time). A mount point
resolves to the mounted filesystem's root while attached and to the
underlay directory after detach — same path, different directory,
different identity.
* _workdir_is_observably_gone() now requires the parent visible at
prune time to match that recorded identity before the populated-parent
check can classify an orphan. A mismatch means a different directory
(the underlay) is showing through — a detached volume, not an
observed deletion.
* Metadata without a recorded identity (written by older versions) is
never orphan-classified — unsure never deletes; the retention/stale
rule still reclaims genuinely abandoned projects off last_touch.
* The frozen pre-v2 layout has no metadata channel for the identity, so
it keeps the structural checks only (require_parent_identity=False).
* A failed evidence probe on re-registration preserves the previously
recorded identity — stale evidence can only make pruning MORE
conservative.
Windows: st_dev/st_ino of 0 (filesystems without file IDs, some network
shares) is treated as "no evidence recorded", which falls into the
conservative never-orphan path. os.path.ismount and Path.stat are
cross-platform; no POSIX-only calls added.
tests/tools/test_checkpoint_manager.py: adds egilewski's exact
regression (checkpoint history for mnt/volume/project, detach exposes
mnt/volume/.keep, prune with orphan deletion enabled → NOT deleted;
fails on the bare cherry-pick, passes with this fix), plus
no-recorded-identity conservatism and probe-failure identity
preservation. His absent-parent/empty-parent/retention/genuine-deletion/
live-project controls all still pass.
Reported-by: egilewski (review on #69063)
Addresses @egilewski's review: the parent-directory check still deleted
checkpoint history for the most common unmount layout.
Detaching storage removes the parent outright in some layouts
(`/Volumes/Ext/proj` on macOS, `/media/<user>/<label>/proj`), which the first
commit handles. But in the classic static layout — `/mnt/volume/proj`, an
fstab entry, a container bind-mount — unmounting removes the contents and
leaves the mount point behind as an empty directory. `parent.is_dir()` is then
true, the project is absent, and the startup sweep deletes its ref, index and
metadata: exactly the case this PR set out to protect.
Reproduced against the real predicate before this commit:
mount root vanished (macOS) -> False ok
empty surviving mount point -> True <-- history deleted
really deleted (siblings) -> True ok
An empty parent carries no information: it looks identical whether the volume
was detached or the project was deleted. So require the parent to actually say
something — it holds some other entry (we observed a populated directory that
does not contain the project), or it is itself a live mount point (the volume
is attached right now and demonstrably does not hold the project).
The cost is that a project deleted out of an otherwise-empty parent is no
longer reclaimed by the orphan rule. It is not leaked: the retention rule
reads `last_touch` rather than probing the filesystem and still collects it,
so reclamation is deferred, not lost. That is the right direction for a
predicate whose false positive destroys a user's restore points unattended.
`_dir_has_any_entry` stops at the first entry via `os.scandir` instead of
materializing a listing, since a project root can hold a large tree.
tests/tools/test_checkpoint_manager.py: `test_surviving_empty_mountpoint_
keeps_its_checkpoints` pins the reviewed case, and `test_empty_parent_project_
is_still_reclaimed_by_retention` pins the deferral above so the safety valve
cannot silently regress into a leak. Both fail on the previous commit. The
real-orphan control now seeds a sibling so it exercises a populated parent
rather than the ambiguous empty one. 80 passed in the checkpoint suite; the 2
remaining failures (`TestGitEnvIsolation`, `TestClearFunctions`) fail
identically on clean main.
Orphan pruning decides a project is gone from a single probe:
if delete_orphans and (not workdir or not Path(workdir).exists()):
reason = "orphan"
then deletes its ref, index, and metadata — the project's entire checkpoint
history. `Path.exists()` is False for a deleted directory, but it is equally
False for one whose storage is not attached right now: an unplugged external
drive, a share behind a downed VPN, a bind-mount absent from this container,
an offline Windows mapped drive. The project is fine; only our view of it is.
This is not an opt-in maintenance command. `maybe_auto_prune_checkpoints`
runs unattended at startup from both `cli.py` and `gateway/run.py`, with
`delete_orphans=True` by default. So starting Hermes once while the drive is
unplugged silently destroys the restore points for every project on it — the
one thing checkpoints exist to provide, and there is nothing to restore from
afterwards.
Reproduced against the real store: a project registered under an unmounted
path and one on local disk, then a startup prune —
prune: {'scanned': 2, 'deleted_orphan': 1}
unreachable project index still on disk: False
The legacy pre-v2 branch has the same flaw plus a second one: a
`HERMES_WORKDIR` marker that exists but cannot be read leaves `workdir = None`,
which the same condition treats as an orphan. Failing to read a file is not
evidence that a project was deleted.
Require corroboration before deleting: the workdir's parent must be present,
so its absence is something we actually observed. A missing parent means the
volume is not there and we know nothing, so the entry is left alone — and an
unreadable marker never deletes at all. Genuinely abandoned projects are still
reclaimed, both by the unchanged orphan path (parent present, project gone)
and by the retention/stale rule, which runs off `last_touch` rather than a
filesystem probe.
tests/tools/test_checkpoint_manager.py: a project whose whole mount disappears
keeps its history; controls prove a genuinely deleted project is still pruned
and a live project is untouched. The data-loss test fails on main; both
controls pass there. 81 passed across the checkpoint suites (2 failures in
test_checkpoint_manager.py are pre-existing and fail identically on clean
main).
_read_hub_installed_names() reads ~/.hermes/skills/.hub/lock.json with a
strict utf-8 decode. Hub skill descriptions can carry Windows-1252
typographic bytes (em-dash 0x97, smart quotes, bullets) as single high
bytes; read_text(encoding="utf-8") then raises UnicodeDecodeError, which
is a ValueError sibling not caught by the function's
except (OSError, json.JSONDecodeError). It escapes and 500s the whole
/api/skills endpoint, blanking the desktop Skills panel.
Decode with errors="replace" so the offending byte degrades to U+FFFD
and the structurally valid JSON — and every other skill — stays readable.
Fixes#68053
Address P1 from PR review: cmd_prune()'s y/N preview reads
store_status() but the confirmed deletion re-scans both the v2 and
pre-v2 layouts from scratch. A workdir that goes missing while the
human is answering the prompt gets swept in as if it had been shown
and approved.
prune_checkpoints() now accepts orphan_allowlist — a set of v2 project
hashes and/or pre-v2 shadow repo paths. When set, only orphans whose
identity is in the set are deleted; anything newly orphaned since the
scan survives the run. cmd_prune() builds this set from the exact
projects it just displayed and passed confirmation for. --force still
passes None (no preview shown, so nothing to bind to).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
When a background terminal() command backgrounds its own long-lived
child (`node server.js &`, `sleep 300 &`), the grandchild inherits the
write end of the reader thread's stdout pipe. The direct bash child
exits promptly, but the pipe never reaches EOF while the grandchild
lives — so `_reader_loop`'s blocking `read1()` parked the thread
forever, `session.exited` never flipped on its own, and
`notify_on_complete` was silently lost. `_reconcile_local_exit`
(#17327) only runs lazily from poll()/wait(), so nothing autonomous
ever surfaced the exit; each occurrence also leaked a reader thread
and pipe fd for the grandchild's lifetime.
Fix: on POSIX, drain via select() with a short poll interval and stop
shortly after the direct child exits even if the pipe hasn't EOF'd —
the same pattern the foreground path uses in
tools/environments/base.py::_wait_for_process (#8340). Windows pipes
don't support select(), so the blocking path is kept there with the
existing lazy reconcile as the safety net; mocked/iterator stdout
streams (no usable fileno) also keep the historical path.
Fixes#68915
Follow-up hardening on top of the salvaged #69745 guard, addressing both
review findings:
- Drift detection no longer re-reads the file. _reload_target performs ONE
checked read and derives both the drift check and the entry parse from that
same raw snapshot (_detect_external_drift now takes the raw text). The old
second read swallowed OSError as 'no drift', so a read failure between the
two reads let replace/remove/apply_batch rewrite the file from a stale view,
discarding externally added entries.
- Invalid UTF-8 now counts as unreadable: the checked read catches
UnicodeDecodeError and mutations return the preservation refusal instead of
raising (or worse, rewriting bytes we can't round-trip).
- USER.md is covered by the same guard (shared _reload_target path) and now
pinned by an explicit test.
Tests: read-once structural invariant, invalid-UTF-8 refusal with
byte-identical file, user-store refusal.
`_read_file` degraded any read failure to `[]`, conflating "file exists but
couldn't be read" with "empty store". That is a silent, total data-loss bug on
the `add` path.
`add` re-reads the file under lock, appends the new entry, and rewrites the
WHOLE file from the parsed entries. It deliberately skips the drift guard
(#42874: "appending never clobbers existing content") — but that reasoning only
holds when the reload actually saw the file. When `read_text` raises
(an external editor momentarily holding the file on Windows, a permission
change, a filesystem/EINTR blip), `_read_file` returns `[]`, so `add` treats
the store as empty and rewrites the file down to just the new entry — every
prior memory gone — while returning `success: True`.
Reproduced with a transient read failure during `add`:
entries on disk before : 3 (dark-mode pref, deadline, deploy target)
add("A brand new fact") : success=True
entries on disk after : 1 ("A brand new fact") <-- the other 2 wiped
replace/remove/apply_batch were shielded only incidentally — an empty view
means `old_text` never matches, so they abort before writing — but they still
returned a misleading "no entry matched" instead of naming the real problem.
Fix: distinguish unreadable from empty. `_read_entries_checked` returns
`(entries, read_ok)`, with `read_ok=False` only when the file exists but can't
be read; absent/empty stays a clean `([], True)`. `_reload_target` returns a
`_READ_FAILED` sentinel in that case without touching in-memory state, and all
four mutation paths (add, replace, remove, apply_batch) refuse the write with a
clear "retry in a moment" error. This is the same posture as the drift guard
and the pairing/checkpoint fixes: never rewrite a file from a view that isn't
the real one. `_read_file` keeps its `[]`-on-error contract for the read-only
`load_from_disk` caller, which never persists.
tests/tools/test_memory_tool.py: new TestUnreadableFileDoesNotWipeMemory —
add/replace/remove/apply_batch all refuse and leave the file byte-identical on
a transient read failure, plus controls that an absent file is still a clean
empty store and the happy path is undisturbed. The four refusal tests fail on
main. Full suite: 90 passed, 1 pre-existing failure (`test_deduplication_on_load`,
a UnicodeDecodeError unrelated to this change, identical on clean main).
Three durable ledgers used `with _connect() as conn:` where the sqlite3
connection context manager commits/rolls back but never closes, leaking the
db/-wal/-shm file descriptors on every call. On a long-running gateway this
exhausts RLIMIT_NOFILE and fails unrelated components with
`[Errno 24] Too many open files`. Same bug class as the cron execution ledger
(#69567 / PR #69594), which the connection helpers here are modeled on.
Fix: route every ledger operation through a `_transaction()` context manager
that guarantees `conn.close()` on exit. `_connect()` keeps its
schema-on-connect contract (several tests call it directly) and now self-closes
if schema init fails.
Adds per-module regression tests asserting every opened connection is closed,
including the no-op-update and exception-mid-transaction paths.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
On Windows, pipe I/O can deliver non-UTF-8 bytes at chunk boundaries,
causing `UnicodeDecodeError` when the MCP SDK's `TextReceiveStream`
uses `errors="strict"`. Set `encoding_error_handler="replace"` on
`StdioServerParameters` so undecodable bytes become U+FFFD instead
of crashing.
- Strip the salvaged commit's inline encoding kwargs where main had since
gained its own (process_registry, local env, cua doctor, gateway,
commands, gateway_windows — the latter keeps its locale-aware
_schtasks_encoding() from #38186)
- Revert encoding kwargs mistakenly applied to non-subprocess APIs
(exa get_contents, tempfile.mkstemp in webhook.py)
- Guard the ddgs worker Popen (new on main since #55339)
- Update two kwarg-snapshot test assertions for the new kwargs
Hermes-sweeper review on #60741 flagged that _op_version (the paired
op probe used by the same setup/status CLI flow at lines 127 and 205)
still ran text=True without explicit encoding/errors.
Add encoding='utf-8', errors='replace' to match _op_whoami and the
production op read path at agent/secret_sources/onepassword.py:271-278.
Also extend the regression test to cover _op_version alongside
_op_whoami, and update the module docstring to reflect the widened
scope. Test sensitivity verified: reverting the source change makes
test_op_version_passes_utf8_encoding fail with encoding=None.
PR #55339 adds encoding='utf-8', errors='replace' to 26 subprocess.run(text=True)
call sites across the codebase. The triage review (thanks @alt-glitch) diffed
this PR against #55339 and found that 5 of the 6 originally-touched call sites
are already covered there byte-identically:
- hermes_cli/main.py::_probe_container
- hermes_cli/setup.py SSH probe
- tools/tts_tool.py::_generate_neutts
- tools/transcription_tools.py::_prepare_local_audio
- tools/transcription_tools.py::_transcribe_local_command (both branches)
The one genuinely net-new site — hermes_cli/onepassword_secrets_cli.py::_op_whoami
(the 1Password op CLI whoami probe) — is NOT in #55339 and is fixed here.
Without explicit encoding=, text=True decodes child output with
locale.getpreferredencoding(False) — cp936 on Chinese Windows — which crashes
_readerthread on non-GBK bytes, cascading into pipe buffer fills, event loop
stalls, and TUI freezes (issues #47939, #53428, #57238).
Scope narrowed per triage feedback: the other 5 sites should land via #55339.
Refs #53428 (together with #55339).
Follow-up to salvaged PR #70549:
- Replace fragile 'or' assertions with single precise checks that catch
partial-rewrite regressions (would have masked a missing closing brace)
- Add test_pty_path_uses_rewritten_command covering the PTY spawn path
that was modified but previously untested
Issue #68915: when the agent runs a compound command with trailing & (e.g.
`cd /app && node server.js &`), bash parses it as `(A && B) &` — a subshell
that holds the stdout pipe open forever when B is a long-running server.
The existing _rewrite_compound_background in terminal_tool.py correctly
rewrites this to `A && { B & }` to avoid the subshell fork, but it was only
applied in the foreground execute() path (tools/environments/base.py).
The background spawn_local() path bypasses base.py entirely and passed the
raw command directly to Popen/PTY, leaving the deadlock unmitigated.
Fix: apply _rewrite_compound_background in spawn_local() before the command
is passed to Popen or PTY spawn. Uses a lazy import to avoid circular
dependency (terminal_tool imports process_registry).
- PTY spawn path: now uses safe_command (rewritten)
- Popen spawn path: now uses safe_command (rewritten)
- Session.command still stores the original (unrewritten) command for display
- Simple `cmd &` is left unchanged (no subshell bug)
Tests: 4 regression tests verifying (1) compound is rewritten, (2) simple bg
is preserved, (3) multi-line compounds are rewritten, (4) session.command
stores original.
Follow-up on the salvaged #68469 commit:
- Literal-IP hostnames never take the proxy DNS-delegation path (a
getaddrinfo failure on a literal IP is not a proxy-environment
symptom, and IPs need no DNS) — keeps the private-IP/metadata floor
intact under proxy env vars.
- Adds TestProxyEnvironmentDnsDelegation: delegation fires only for
hostnames, metadata hostname/IP floor holds, DNS-success path
unchanged, empty proxy var ignored.
- Guards the three pre-existing DNS-failure tests against ambient
proxy env vars so they don't flake on developer machines.
test_docker_network_config.py landed on main after the #58489 revert and
stubbed docker ps with the 2-field ID\tState format. The re-landed
egress-aware reuse probe requests ID\tState\tEgressLabel when egress is
off, so the fake line failed to parse and the reuse path never fired.
Fixture-only change; production behavior is unchanged.
The blanket MAX_DESCRIPTION_LENGTH=1024->60 change is narrowed:
create-time validation now rejects new skills whose description
exceeds SKILL_PROMPT_DESC_LIMIT (60) with actionable guidance, while
edit/patch paths stay permissive (warning via system_prompt_preview)
so existing over-limit skills remain maintainable. Runtime display
truncation in skills_tool is left at 1024 (display behavior is a
separate concern from authoring validation).
Boundary tests: 60 accepted, 61 rejected at create; edit/patch on
over-budget skills still succeed.
When a skill is created or edited with a description longer than
SKILL_PROMPT_DESC_LIMIT (60 chars), the tool response now includes a
system_prompt_preview field showing exactly what the system prompt
skill index will display. This gives the agent immediate feedback to
self-correct truncated trigger phrases.
Also adds tool schema guidance about the 57-char window and fixes a
stale docstring in skill_commands.py that incorrectly claimed the
system prompt renders the full description.
Make the x_search / xurl boundary explicit in the skill, feature docs,
toolset metadata, setup note, and reference pages, while keeping the
model-facing x_search schema generic (no static xurl name).
Regression tests assert behavioral routing invariants rather than frozen
prose snapshots. Drop the stale CI-only plugin/hangup hunks already on
main so this rebases cleanly.
A direct run (cronjob action='run' / webhook-triggered manual fire) takes
the job's fire_claim and advances next_run_at. When it races the external
provider's scheduled fire for the same occurrence, Chronos loses the claim
and — by design — does not re-arm (the winner owns the re-arm). But the
direct-run winner never notified the provider, so the NAS one-shot for the
consumed occurrence was left stale forever and the recurring job silently
stopped firing.
Observed in production: a managed 1-minute review job stalled for 20 hours
because a GitHub-webhook direct run claimed the job 2s before the Chronos
fire arrived; every subsequent occurrence was orphaned while /api/status
stayed green.
Fix: after a *claimed* direct execution completes (success or failure —
next_run_at advances at claim time either way), call
_notify_provider_jobs_changed_safe() so the active provider re-arms the
post-run next_run_at. No-op for the built-in ticker; claim-lost direct
runs still never notify (the winning scheduler owns the re-arm).
skills/dogfood/SKILL.md sat at the root of the bundled skills tree,
making it one of the few uncategorized skills (Discord /skill
autocomplete listed it under 'uncategorized'; hermes_cli/commands.py
cited it as the example). Move it to
skills/software-development/dogfood/ alongside the other QA/testing
skills (test-driven-development, systematic-debugging,
requesting-code-review).
- git mv skills/dogfood -> skills/software-development/dogfood
(history preserved)
- fix test fixture path in tests/tools/test_browser_console.py
- update root-level-skill docstring examples (commands.py,
generate-skill-docs.py) to cite computer-use, which is still root-level
- drop 'dogfood' from the category list in hermes-agent-skill-authoring
SKILL.md (en + zh-Hans docs mirrors)
- docs: regenerate/move the bundled page to
software-development-dogfood (en + zh-Hans), update sidebars.ts,
skills-catalog.md, and the adversarial-ux-test related-skills link
E2E validated: fresh sync_skills() copies to the new nested path;
existing installs with the old flat copy keep it untouched (manifest is
name-keyed, hash unchanged -> skipped, no duplicate);
_get_category_from_path resolves 'software-development'; docusaurus
build green.
The Windows /restart watcher's outer Popen spawns the watcher with
windows_detach_popen_kwargs() (which carries CREATE_BREAKAWAY_FROM_JOB),
but a restrictive parent job object can reject that bit with OSError and
the current call has no retry. Preserve the current watcher
implementation and add a focused breakaway-denied fallback.
Preserved from current main: watcher_python / pythonw.exe selection, the
str(restart_after_s) deadline, the scrubbed watcher_env, the intentional
no-breakaway inline respawn, and the entire POSIX setsid/bash path.
- primary keeps **windows_detach_popen_kwargs()
- on OSError, retry the same argv/env with
creationflags=windows_detach_flags_without_breakaway()
- on dual failure, log a definitive, path-safe warning (interpreter
basename + numeric winerror/errno only) and return without crashing
Replace the superseded breakaway-first inline design and its AST tests
with focused behavioral coverage that drives the real coroutine with a
mocked subprocess.Popen (retry, argv/env/DEVNULL preservation, POSIX
single-session kwarg, no-breakaway inline respawn, secret-safe logging).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Two related Slack delivery fixes for send_message text sends:
- Route Slack text delivery through _send_via_adapter so the live
in-process gateway adapter (multi-workspace aware, channel→client
mapping, adapter-side gates) is preferred, with the plugin's
_standalone_send as the out-of-process fallback — matching how the
media path already behaves.
- _standalone_send: SLACK_BOT_TOKEN can be a comma-separated list in
multi-workspace installs and slack_tokens.json carries OAuth
per-workspace tokens; the standalone Web-API path used to send the
literal comma-joined string, which Slack rejects as invalid_auth.
Try each token individually, retrying on token-scoped errors
(invalid_auth / not_in_channel / channel_not_found …) and stopping on
terminal ones. User-DM resolution (U…/W… targets) also tries each
token.
Adapted from #47547 by @replygirl — the original patched the legacy
tools/send_message_tool.py::_send_slack helper, which moved to the
Slack plugin's _standalone_send in #41112.
Salvaged from #47547
origin_session_id (the api_server wake self-post target) lived only in
the in-memory record: durable dispatch persistence and abandoned-
delegation recovery omitted it, leaving completions recovered after a
process restart unroutable to api_server sessions. Persist it in the
async_delegations table (CREATE TABLE + ALTER TABLE migration for legacy
DBs), restore it on recovery, and expose it via get_durable_delegation.
Also adds the contributors mapping for ianks (PR #64998 author).
Follow-up to #64998 (sweeper review F3).
Wake-ups for kanban notifications and background delegation completions were
injected via handle_message() using a build_session_key()-derived key, which
can never match the raw X-Hermes-Session-Id key that api_server sessions run
under — so the wake landed in a session nobody was reading. On top of that,
ApiServerAdapter.send() reports failure without raising, and that was treated
as a successful delivery, so the notify cursor advanced past events that were
permanently lost; and background delegation was forced synchronous on
api_server since there was no way to wake the session afterward.
Fix: route wake-ups for non-push adapters through a self-post to
/v1/chat/completions with the original session id, treat non-raising send
failures as failures (rewind instead of advancing the cursor), and re-enable
background delegation whenever a session id is available to wake.
The origin session id is captured from the request-scoped api_server chat_id
binding rather than HERMES_SESSION_ID: constructing a child agent calls
set_current_session_id() with the subagent's internal id, clobbering that
variable right before dispatch would read it and misrouting the wake into
the subagent's own session.
Related: #56580, #64609, #53027, #63169, #56531, #50319, #64113
Several platform fetch paths called is_safe_url before constructing ordinary httpx clients, leaving a second DNS lookup at connection time. This preserved the rebinding window for Slack batch images, Feishu documents, Telegram URL-photo fallback, and WeCom remote media.
Route each path through create_ssrf_safe_async_client and the shared redirect guard so direct connections validate and dial vetted IPs while configured proxies remain an explicit trusted egress boundary. Add per-path regressions that change DNS from public at preflight to metadata at connect time.
The Skills Hub provenance fixture intentionally serves content over loopback. Opt that test-scoped server into private-address access so it keeps exercising the real HTTP transport without weakening production blocking.
Related #8033
Co-authored-by: teknium1 <127238744+teknium1@users.noreply.github.com>
Install connect-time DNS validation for Hermes-owned direct httpx clients so SSRF-sensitive fetch paths dial a vetted IP instead of re-resolving after preflight. This preserves Host/SNI semantics for direct HTTP(S) connections and keeps proxy routing as an explicit trusted egress boundary.
Wire the guarded clients into media cache downloads, vision downloads, Skills Hub direct/raw fetches, and platform attachment fetch paths that already perform SSRF preflight and redirect validation.
Fixes#8033
Co-authored-by: Tom Qiao <zqiao@microsoft.com>
Extends the desktop backend's root-cause fix (aa2ae36c3f) to all remaining
console-less parent launch paths. The Windows console-flash class
(#54220/#56747) is governed by the PARENT's console: a DETACHED_PROCESS or
pythonw.exe daemon has no console, so every console-subsystem descendant
(git, gh, cmd, node, wmic, powershell) allocates its own visible conhost —
one flash per spawn, unreachable by any per-call-site CREATE_NO_WINDOW
sweep. Worse, MSDN specifies CREATE_NO_WINDOW is IGNORED when combined
with DETACHED_PROCESS, so the hide bit in the old detach bundle was dead.
Changes:
- _subprocess_compat: drop DETACHED_PROCESS from windows_detach_flags()
and windows_detach_flags_without_breakaway(); the daemon now owns a
single hidden console (CREATE_NO_WINDOW) that all descendants inherit.
- gateway_windows: _resolve_detached_python() returns the venv console
python.exe (no pythonw/base-interpreter detour — the uv-shim flash
premise only held while DETACHED_PROCESS was masking the hide bit);
UAC handoff launches console python under SW_HIDE; cmd/vbs launchers
render console python (vbs runs it window-style 0).
- gateway/run.py: restart watcher keeps sys.executable instead of
swapping in GUI-subsystem pythonw.
- web_server: dashboard actions spawn sys.executable (already carries
windows_detach_flags()).
Tests updated to pin the new invariants, including an explicit
DETACHED_PROCESS-must-stay-out regression guard.
Default kanban_create children now keep fresh scratch paths, while explicit dir sharing remains supported and project context resolves to a per-task worktree. Surface resolved workspace fields in create responses/events and cover scratch mutation, nesting, explicit sharing, and project inheritance.
Fixes#67567
Dispatcher-spawned Kanban workers are finite one-shot processes, so detached delegation completions can outlive their only consumer. Mark that runtime as unable to deliver async completions and reuse the synchronous delegation fallback, returning required child results before the worker exits.\n\nAlso make unsupported-session notes runtime-generic and cover the delayed-child lifecycle regression.\n\nRefs #63169
Treat cua-driver's Linux `is_on_screen: null` as unknown instead of
off-screen, and skip GNOME Shell desktop/backdrop helper windows
(ding "Desktop Icons", @!x,y;BDHF) when selecting the default capture
target — they are targetable X11 windows but capture as empty.
Reconciled with the _NET_ACTIVE_WINDOW fallback from #58030: helper
windows are filtered out of the candidate pool first, then the tied
z-order active-window probe runs on the remaining real app windows.
Also falls back to the requested app name for _last_app when Linux
windows carry no app_name.
Salvaged from #54173 by @dnth.
The three subprocess tests spawn 'sys.executable -c' children that import
hermes_cli. From a worktree, the child resolved the MAIN checkout's editable
install instead of the tree under test, so the new DB/CLI guards appeared
missing and the tests failed with rc=0. Route the spawns through a helper
that pins the repo root under test on PYTHONPATH.
Client leg of M2 org-shared skills (hsp-1-contract.md §11), pairing with
gateway-gateway #162 and NAS #768 (both merged).
- resolve_org_identity(): org_id + org_role from the token claims. NO
org_role claim (personal org — NAS only stamps it for multi-member orgs)
=> SyncInertError => every org surface is inert; personal M1 sync
untouched (contract §11.1 REFINED). org_sync_available() for callers.
- pull_org_skills(): materialize the org canonical set (refs/org/<org_id>/
HEAD) into ~/.hermes/skills/_org/<org_id>/ — fast-forward only, no client
merge on the org path (design.md §2.6); read-only mirror by convention
(§7.1: a local edit is a personal fork until proposed). maybe_pull_org_
skills() best-effort hook (never raises; inert without the claim).
- propose_skill(): snapshot the LOCAL skill dir, splice it into the org
HEAD's skill-tree map (per-skill delta, never wholesale replace), upload
?scope=org, CAS the org HEAD. ADMIN => direct merge; MEMBER => server
converts to a proposal — cas_ref now surfaces 202 as
{proposal_pending: True, proposal_id, ref} (success-shaped, NEVER
presented as live). Non-interactive by design for the future automated
submitter (Ben's trajectory note).
- put_objects(org_scope=True) adds ?scope=org (contract §11.5).
- is_sync_eligible(): skills under _org/ are excluded from PERSONAL sync —
enterprise content never rides a personal push (§11.11).
- CLI: hermes skills propose <name> [-m msg] — prints 'pending admin
review' for 202, 'merged' for admin, and a plain 'org sync unavailable'
for personal orgs instead of a raw 403.
Tests: 56 in the sync client suite (+10 org: identity gate both ways, _org/
personal-sync exclusion, admin direct merge, member 202 w/ HEAD untouched +
proposal ref parked + never-merged, splice-not-replace root, pull mirror
materialization, no-head noop, org-feature gate, maybe_pull inert). Mock
server extended (org feature flag, member-CAS→202). 103 across skills
suites. Live CLI smoke: propose --help + personal-org inert path verified.
cua_driver_update_check now short-circuits to None when no driver
resolves; CI has none installed, so pin resolve_cua_driver_cmd in the
update-check and env-sanitization tests.
- Add autouse fixture to TestMcpInvocationResolution to disable
--no-overlay flag so existing tests assert baseline args
- Make test_config_load_failure_fails_safe and test_missing_section_enables
platform-aware (Linux auto-detect returns True, macOS/Windows False)
Post-rebase consolidation over the merged C7/C16/C10 work:
- _ensure_dm_conversation now records workspace ownership via
_remember_channel_team and bounds _dm_conversation_cache (cap 5000)
- module-level _slack_dm_cache (C7 standalone path) bounded oldest-first
- _user_is_bot_cache (C10) bounded with _trim_oldest_dict_entries
- caption-mode contract tests updated for the C7+C8 merged media path
* test(clarify-gateway): cover signature, timeout fallback, and notify paths for 100% coverage
Fixes#36531
(cherry picked from commit 5265dfe2f5)
* fix(clarify): one canonical timeout across CLI, TUI/desktop, and gateway
The clarify wait timeout was resolved three different (wrong) ways:
- CLI (`cli.py`, `hermes_cli/callbacks.py`) read a non-existent top-level
`clarify.timeout`, so it always fell through to a hardcoded 120s instead of
the canonical `agent.clarify_timeout` (default 3600) the gateway uses (#42969).
- The TUI/desktop bridge called `_block("clarify.request", …)` with no timeout,
so it used the hardcoded 300s `_block` default and ignored config (#51960).
- There was no way to disable the auto-skip: a user who wanted the agent to wait
indefinitely while they think couldn't get it.
Collapse all of this onto a single resolver:
- `tools.clarify_gateway.resolve_clarify_timeout(config)` is the one source of
truth. Order: explicit legacy `clarify.timeout` (back-compat) → canonical
`agent.clarify_timeout` → 3600. `<= 0` is preserved verbatim as "unlimited".
- CLI, callbacks, and the TUI bridge (`_clarify_timeout_seconds`) all route
through it, so the three surfaces can't drift.
- `<= 0` means unlimited everywhere: `wait_for_response` and `_block` drop the
deadline (heartbeat still fires), and the CLI hides its countdown.
Tests: resolver order / default / non-numeric / unlimited-sentinel; an
unlimited `wait_for_response` blocks until resolved rather than auto-skipping;
the TUI clarify bridge passes the configured timeout to `_block`.
Supersedes #42974 (CLI key), #51993 (TUI honors config), and #68986 (unlimited
wait); folds in #52031 (clarify_gateway coverage).
Co-authored-by: liuhao1024 <liuhao1024@users.noreply.github.com>
Co-authored-by: lkevincc0 <lkevincc0@users.noreply.github.com>
Co-authored-by: theone139344 <theone139344@users.noreply.github.com>
Co-authored-by: baauzi <baauzi@users.noreply.github.com>
---------
Co-authored-by: Christopher-Schulze <210261288+Christopher-Schulze@users.noreply.github.com>
Co-authored-by: liuhao1024 <liuhao1024@users.noreply.github.com>
Co-authored-by: lkevincc0 <lkevincc0@users.noreply.github.com>
Co-authored-by: theone139344 <theone139344@users.noreply.github.com>
Co-authored-by: baauzi <baauzi@users.noreply.github.com>
Commit author.device was an opaque uuid4 hex, so the sync console showed a hash
per device. Make it human-friendly:
- Default: seed new devices from the short hostname + a short random suffix
(e.g. bens-macbook-a1b2c3) instead of a bare uuid. Existing .sync_device_id
files are honored verbatim (a machine keeps its id) — backward-compatible.
- hermes sync device [--name N]: show or set an explicit label
(set_device_name(), written to ~/.hermes/skills/.sync_device_id). New commits
use it; past commits keep their old label (author.device is immutable).
- Hermes Cloud: HERMES_SYNC_DEVICE_NAME env seeds the first-use label so a
hosted instance shows a recognizable name with no CLI call. Precedence:
explicit .sync_device_id file > HERMES_SYNC_DEVICE_NAME env > hostname default.
Env seeds first-use only, then persists, so a later --name still wins locally.
Tests: 46 pass (+5: hostname default, file-wins precedence, env first-use seed +
persistence, set/trim, empty-name reject). CLI verb smoke-verified end-to-end.
Integration test for the two compaction layers landing this sweep:
#63144's discovery-scope fix (archived rows surface from the current
session) and #69334's bookend bounding (summary exclusion + content
caps). A single compacted session exercises both: the archived FTS hit
must surface while the compaction handoff at the session tail stays out
of bookend_end and long messages stay capped.
Addresses @teknium1's second review round on #63144:
1. _is_compacted_message now checks both active=0 AND compacted=1.
Previously checked active=0 alone, which also matched rewind/undo rows
(active=0, compacted=0) that must stay hidden.
2. Added _is_compression_ended() — checks only the session's own
end_reason, not the lineage-wide has_compression_hop flag. This prevents
delegation children living under a compression continuation from leaking
through the lineage filter.
3. _discover lineage skip now uses is_ended_session (session-level check)
instead of has_compression (lineage-level flag).
New tests:
- TestRewindExclusion: rewind rows stay hidden alongside compacted rows
- TestCompressionEndedHelper: session-level end_reason checks
- TestLegacyContinuationPlusDelegation: delegate child excluded while
compression ancestors surface
78 passed (71 + 7 new).
After context compaction, pre-compaction content was invisible to
session_search — a memory black hole. The _discover() skip logic
filtered both same-session and same-lineage hits unconditionally,
without distinguishing compression-summarised content (gone from live
context) from delegation children (still visible to the parent agent).
Reworked _resolve_to_parent to return (root_id, has_compression_hop),
checking end_reason='compression' on every hop during the same
db.get_session() traversal — zero extra queries.
_discover() now has three compression-aware paths:
- In-place compaction: FTS hits on active=0 (compacted=1) rows pass
through even when raw_sid == current_session_id
- Legacy rotation: lineage hits pass through when has_compression_hop
is true on either side of the chain
- Delegation children: still excluded (no compression edge)
18 new tests covering all three scenarios + unit tests for the helpers.
Addresses Teknium's review feedback on #6256.
Closes#13840, #13841.
mark_speech_interrupted() / take_speech_interrupted(): a one-shot,
TTL'd (120s) latch plus SPEECH_INTERRUPTED_NOTE. Barge-in paths mark
it when they cut live speech; the next turn's submit path pops it and
prepends the note to the model-bound message — API-call local, never
persisted, so history and prompt caching are untouched.
The streaming gate broadens from ElevenLabs-only to any provider that
passes check_tts_requirements(). In continuous voice mode a mic monitor
runs during playback: talking over the agent cuts TTS at detection
while the monitor keeps recording, then the captured interruption is
transcribed and queued as the next turn (process_loop's auto-restart
stands down while the capture owns the mic). New voice.barge_in config
key, default true.
listen_for_speech(): sustained-RMS speech detection with the noise
floor calibrated against audible playback (speaker bleed doesn't
self-trigger). With capture=True it keeps a rolling pre-roll buffer and
records through to a silence endpoint, so the interruption is
transcribed from its first syllable — detection alone loses the opening
words to the detector's sustain window plus the mic re-open.
transcribe_recording() maps no_speech provider results to a successful
empty transcript (silence, not an error).
An STT provider that hears no words is reporting silence, not failing.
ElevenLabs/xAI empty-transcript errors now carry no_speech so live
voice loops can re-listen quietly instead of surfacing an error on
every pause.
StreamingTTSProvider ABC + registry + resolve_streaming_provider() —
ElevenLabs (pcm_24000) and OpenAI (response_format=pcm) stream chunked
PCM; every other provider keeps its configured voice and falls back to
per-sentence sync synthesis. SentenceChunker is the one incremental
cutter every surface shares: sentence-boundary cuts on the delta
stream, <think> blocks stripped even when split across deltas, short
fragments merged forward so they never stall as tiny clips.
Two changes:
1) Rename the client-local head-bookkeeping file .sync_manifest -> .sync_state
(read_sync_state/write_sync_state) to remove the name collision with the
§2.8 plane 'sync-manifest' OBJECT. read_sync_state migrates an existing
.sync_manifest on first read so no device loses its head record.
2) Make the knobs a Hermes Cloud instance needs env-configurable, so an
instance can be set up to use sync BY DEFAULT with no config.yaml edit or
per-skill CLI call. Precedence: HERMES_SYNC_* env -> config.yaml sync.* ->
built-in default (mirrors the existing HERMES_SYNC_BASE_URL bridge).
- HERMES_SYNC_ENABLED -> sync.enabled (master on/off; def off)
gated in maybe_push/maybe_pull alongside the dev-gate + base_url.
- HERMES_SYNC_DEFAULT_OPT_IN -> sync.default_opt_in (M1-D policy; def off)
opt-out mode: every eligible skill syncs unless usage rec says sync:false;
'your skills follow you with no setup' — the Cloud default. opt-in mode
(default) unchanged: only sync:true skills sync.
sync_status() + 'hermes sync status' surface feature_enabled/default_opt_in.
Cross-repo byte-compat re-verified: client manifest bytes still parse cleanly
through gateway-gateway parseSyncManifest.
Tests: 41 pass (+7: env precedence, opt-out/opt-in policy, rename migration).
Replace the device-local opt-in model (`.usage.json` `sync` flag as the sole
source of truth) with the §2.8 content model: a root-level `sync-manifest` blob
in the tree at `refs/user/<owner>/HEAD` recording per-skill {name, enabled},
matching gateway-gateway src/sync/manifest.ts byte-for-byte.
- build/parse_sync_manifest: canonical {type,version:1,skills:[{name,enabled}]},
strict parse (malformed != empty).
- snapshot_profile embeds the manifest as a root-level blob alongside skill
subtrees; the skill walk skips it (blob, not a SKILL.md-bearing tree).
- pull reconciles local opt-in intent FROM the plane manifest, so a skill opted
in on one device becomes opted in on the others (opt-in is now cross-device,
not per-device). Never silently disables a locally-enabled skill on pull.
- .usage.json `sync` flag kept as the editable local intent; plane manifest is
authoritative.
Cross-repo byte-compat verified: the Python client's manifest bytes parse
cleanly through gateway-gateway's real parseSyncManifest (tsx harness).
Tests: 34 pass incl. 5 new (roundtrip, wire shape, strict-reject, root-blob
embed, pull adopts opt-in from manifest).
Extract the open_preview emitter into a shared tools/desktop_ui bridge
(one gateway-injected sink, routed by HERMES_UI_SESSION_ID) and add a
second desktop-gated tool on top of it:
- focus_pane(chat|files|terminal|review|sessions) -> pane.reveal event.
The desktop runs each pane's own reveal path (revealDesktopPane table)
and only acts on the active window -- a background turn never moves the
user's focus (desktop AGENTS.md: offer, don't hijack).
open_preview now emits through the same bridge. Both tools are check_fn
on HERMES_DESKTOP (zero footprint elsewhere), sitting beside
read_terminal/close_terminal in _HERMES_CORE_TOOLS.
Deliberately not adding run_slash: letting the agent fire slash commands
mid-turn (/model, /new, /clear) fights prompt-cache + conversation
invariants.
Add a desktop-gated open_preview tool so 'open cnn.com in the preview
pane' works. The tool (check_fn on HERMES_DESKTOP, zero footprint
elsewhere) emits a preview.open event through a gateway-injected emitter,
mirroring the close_terminal -> terminal.close bridge. The desktop
handles it in usePreviewRouting, normalizing the target and opening the
pane for the active session only -- a background turn never hijacks it.
Bare domains and localhost are coaxed into fetchable URLs (www.cnn.com ->
https://, localhost:3000 -> http://); file paths and schemes pass through
to the renderer's normalizer.
In Slack threads, ordinary follow-up messages sent while a native
multi-choice clarify was pending were consumed as clarify answers —
_coerce_text_response accepted arbitrary text for any pending entry, so
the gateway text-intercept swallowed unrelated thread messages and the
user's messages appeared to be ignored.
Tighten resolve_text_response_for_session for native interactive
multi-choice prompts (buttons rendered, awaiting_text=False):
- numeric selections ("2") still resolve to the canonical choice
- exact choice-label matches (case-insensitive) still resolve
- arbitrary prose is now REJECTED (returns False) so the message
continues as a normal turn instead of vanishing into the clarify
Behavior is preserved everywhere free text is legitimately the answer:
open-ended clarifies, explicit 'Other' text-capture mode, and the base
adapter's numbered-text fallback (which flips awaiting_text at send
time).
Salvaged from PR #62042 by @liuhao1024.
Fixes#62034
Context-compaction handoff summaries (prefixed with [CONTEXT COMPACTION])
were being returned as normal bookend_start/bookend_end messages in
session_search discovery mode. A single compaction handoff could be 57K+
chars, immediately bloating a fresh session prompt to 73K+ chars from one
search hit.
Changes:
- Add _COMPACTION_PREFIXES and _is_compaction_summary() helper
- Filter compaction summaries from bookend_start and bookend_end in _discover()
- Cap bookend content to 1200 chars and window content to 4000 chars
- Add content_truncated/original_content_chars metadata when truncation occurs
- Add 6 regression tests covering prefix detection, bookend filtering,
content capping, and legacy [CONTEXT SUMMARY] prefix
Fixes#43175
Children inherit the parent's env, repo, and toolsets but were denied
execute_code ('children should reason step-by-step, not write scripts').
That forces subagents doing mechanical multi-step work (batch file
reads, fetch-N-pages loops, filter-before-context reductions) to burn
reasoning iterations one tool call at a time.
- Remove execute_code from DELEGATE_BLOCKED_TOOLS
- Stop stripping the code_execution toolset from child bundles
- No recursion risk: the sandbox bridges only the 7 SANDBOX_ALLOWED_TOOLS
(web/file/terminal) — delegate_task and execute_code itself are not
reachable from inside a sandbox script
- Update schema text, AGENTS.md, and delegation docs
- Tests: blocked-constant, strip, and child-assembly tests updated;
new test pins execute_code as intentionally unblocked
Surgical reapply of PR #37523 onto current main (the original branch
predates the SecretSource registry refactor). _build_safe_env() now
forwards env vars tagged in env_loader._SECRET_SOURCES — widened from
Bitwarden-only to any registered secret source (Bitwarden, 1Password,
plugin backends), since the provenance map is source-agnostic.
Explicit server env: config still wins; untagged secrets stay filtered.
Fixes#37499.
test_session_expired_retry_waits_for_new_session hardcoded
_server_breaker_opened_at["hindsight"] = 123.0 to simulate a circuit breaker
whose cooldown has already elapsed. But the breaker in tools/mcp_tool.py
compares that stamp against time.monotonic() (age = monotonic() - opened_at,
elapsed when age >= _CIRCUIT_BREAKER_COOLDOWN_SEC). time.monotonic()'s origin
is arbitrary and small on a freshly-booted CI container, so age worked out to
only a few seconds there (< the 60s cooldown) — the breaker stayed open, the
half-open probe never fired, and the retry returned the "unreachable" error
instead of "bank ok". It passed on long-uptime dev boxes (large monotonic)
and failed under CI, with the reported "Auto-retry available in ~Ns" drifting
run to run as the container's monotonic clock varied.
Stamp opened_at relative to the same clock the code reads
(time.monotonic() - _CIRCUIT_BREAKER_COOLDOWN_SEC - 1.0) so the cooldown is
provably elapsed regardless of the monotonic origin, exercising the intended
half-open transition deterministically.
Builds on diffen77's #66547 (cherry-picked as the previous commits).
Extend _is_session_expired_error's iterative traversal to follow
__cause__/__context__ in addition to ExceptionGroup .exceptions — SDK
wrappers often raise a generic RuntimeError *from* the message-less
ClosedResourceError, leaving the transport signal reachable only via
the chain. The identity-visited set guards chain cycles (handlers
re-raising previously seen exceptions), and a bounded node budget
(_EXC_TRAVERSAL_MAX_NODES) caps pathological acyclic graphs.
Adds regression tests: cause/context chain detection, interruption
precedence through chains, cyclic cause/context termination, and
budget-bounded termination.
Builds on trevorgordon981's #50589 (cherry-picked as the previous commit).
The #50394 cooldown reset only ran inside the async _shutdown coroutine,
which is skipped on the empty-_servers fast path — the most common state
when a server failed to connect (failed servers are never recorded in
_servers). It was also skipped when the MCP loop wasn't running.
Clear _server_connect_retry_after/_server_connect_failures on the fast
path and in a final unconditional sweep so a full shutdown/restart always
re-attempts every configured server immediately. Adds regression tests
for both paths.
Log-storm hygiene for the reconnect machinery (#65673, #66092):
- State transitions carry exactly one WARNING each:
connected → degraded (keepalive failure), degraded → parked
(budget exhausted / rapid-drop park / permanent error),
parked → connected (revival proven healthy, via
_mark_session_proven).
- Per-attempt retry logs (initial-connect attempts, connection-lost
reconnect attempts, per-cycle rebuild notices, park-wake probes)
demoted to DEBUG.
- Backoff sleeps get +/-20% uniform jitter (_jittered) so a herd of
servers that lost the same backend doesn't retry — and log — in
lockstep.
- Add module-level _unwrap_exception_group (ported from
hermes_cli/mcp_config.py, adapted): handles nested groups, prefers
non-cancellation leaves over the CancelledErrors anyio sprays across
sibling tasks, and re-raises KeyboardInterrupt/SystemExit leaves.
- Add _classify_mcp_failure(exc) -> 'permanent'|'transient'.
Permanent: auth 401/403, NonMcpEndpointError, InvalidMcpUrlError,
FileNotFoundError/ENOENT on the stdio command. Transient: everything
else (network/EOF/ClosedResource/TaskGroup drops). Permanent failures
park immediately without burning the retry ladder — every retry
against a missing binary or a revoked credential hits the same wall.
- Every log site in run() and the keepalive failure log now logs the
unwrapped root cause as 'TypeName: message', so a dead stdio pipe
says 'BrokenPipeError' instead of the opaque
'unhandled errors in a TaskGroup (1 sub-exception)' (and empty
str(exc) dead-pipe errors are no longer blank).
Harden the immediate-reconnect path from #66271:
- _reconnect_or_reraise_group re-raises when the transport TaskGroup
carries a KeyboardInterrupt / SystemExit leaf — fatal signals must
propagate to the interpreter, never be converted into a reconnect.
- Rapid-drop budget: a completed handshake alone no longer clears
_reconnect_retries/backoff on clean transport return. A session is
UNPROVEN until it demonstrates real health — survived >=1 full
keepalive interval (keepalive success path) or served >=1 successful
tool call (_mark_session_proven). Unproven clean returns are charged
against _MAX_RECONNECT_RETRIES, so a flapping transport that
handshakes fine and drops moments later still reaches the park
instead of respawning forever (#62212: 6212 spawns in 63h).
Proven sessions keep the #57604 behaviour: budget clears, transient
blips over a long-lived session never accumulate toward parking.
Streamable-HTTP / SSE MCP transports run their stream pump inside an anyio
TaskGroup. A transient stream drop (idle timeout, brief backend blip,
server-side TCP close) surfaces as a BaseExceptionGroup escaping the
transport context manager. It reached run()'s error path, which applied
exponential backoff (1s..16s) and eventually parked the server for 300s and
deregistered its tools — turning a sub-second glitch into a multi-minute
tool outage even though the POST path was still healthy.
_run_http now wraps all three transport branches (SSE, new + deprecated
Streamable-HTTP) and routes a BaseExceptionGroup through a new
_reconnect_or_reraise_group() helper that returns 'reconnect' (immediate
rebuild, no backoff/park/deregister). It re-raises instead of masking when
shutdown is in progress, when the group carries a real CancelledError
(cancellation must propagate, cf. #9930), or when no live session was
established this attempt (a connect/handshake failure that should fall
through to run()'s backoff rather than hot-loop).
Fixes#66092
Co-authored-by: 雨哥 <hanyu1212@users.noreply.github.com>
Update the Matrix reaction-seeding contract to the four-reaction default
(once/session/always/deny), add tirith-tier (session without always) and
no-session-tier cases, and assert allow_session=True in the tirith
gateway payload.
Three related messaging-approval fixes:
1. approvals.timeout default 60 -> 300. PR #63501 collapsed the gateway
wait onto the canonical approvals.timeout (previously
gateway_timeout=300), silently shrinking messaging approval windows
to 60s. Push-notification approvals routinely arrive later than a
minute; taps landed after the wait had already failed closed.
2. Stale-tap honesty: adapters resolved the approval AFTER rendering
'<checkmark> Approved by <user>' (Telegram/Discord/Slack), or ignored a zero
resolve count (WhatsApp Cloud/Feishu). A tap on an expired prompt
claimed approval while the command had already been denied. All
button paths now resolve first and render 'Approval expired -
command was not run' when nothing was waiting.
3. Mixed-warning prompts (dangerous pattern + tirith finding) now offer
Always: the persistence layer already permanently allowlists the
pattern key and downgrades the tirith key to session scope, but the
UI hid Always whenever ANY tirith warning was present. Pure-tirith
prompts still withhold Always (content findings are session-max by
design), and Smart-DENY overrides remain once-only.
_resolve_active_context_length() called get_model_context_length() with the
model id alone, so provider-enforced windows (e.g. Codex OAuth's 272K for
gpt-5.x vs the direct API's 1.05M) never reached the tool-search activation
gate — it sized against generic metadata for the same slug.
Resolve the runtime provider for the configured model and pass provider,
base_url, and api_key through. If credential resolution fails (offline, no
keys), degrade to a provider+base_url-only lookup so the static
provider-aware fallbacks still apply; explicit model.context_length keeps
short-circuiting as before (#46620). Gap flagged during review of #16735.
Follow-up to #67640: move the agent.file_safety import to module top
(stdlib-only, no circular-import concern), replace the over-broad
except Exception + logger.warning with an import sentinel plus
logger.exception so a guard failure is debuggable instead of silently
swallowed. Adds fail-closed tests asserting the diagnostic is emitted.
The live transcripts added with delegate_task write each child's events to
<hermes_home>/cache/delegation/live/<delegation_id>/. Nothing on that path is
redacted, and the rendered events are precisely the secret-bearing surfaces:
tool args, tool results and streamed assistant text.
That location is not incidental — delegate_tool.py:1620 documents cache/
delegation as "mounted read-only into remote backends", so every line lands
in a file readable from inside the sandbox.
Observed on the writer today:
tool | -> terminal(curl -H "Authorization: Bearer sk-ant-api03-...")
result | terminal ok 0.4s: OPENAI_API_KEY=sk-proj-... AWS_SECRET_ACCESS_KEY=wJalr...
The same three values through the canonical redactor:
curl -H "Authorization: Bearer ***" https://api.internal
OPENAI_API_KEY=*** AWS_SECRET_ACCESS_KEY=***
Every other sink for this data already routes through that redactor — search
results via redact_sensitive_text(file_read=True), terminal output via
redact_terminal_output — so the transcript was the one place an operator's
keys reached disk in the clear.
Redact at three points, covering every artefact the dispatch writes into that
directory:
* event() — every typed helper (assistant_text, thinking, tool_start,
tool_result, marker, finalize, the stream flush) funnels through it, so
one call covers them all and a helper added later cannot bypass it.
* the .log header — written directly rather than through event(), and a
caller can paste a key into the task text.
* manifest.json — _write_manifest serialises the same goal, and the manifest
sits in the same mounted directory, so redacting only the header would
have left the credential exposed one file over.
force=True because this is a safety boundary and must redact even with the
global toggle off. On the (impossible-in-practice) import failure the line is
withheld instead of written raw: losing a debug line costs less than writing
a live credential into a sandbox-readable file.
Key names, tool names, statuses, durations and ordinary prose are untouched,
so tail -f stays as useful as before — pinned by its own test, alongside a
whole-directory sweep asserting no file under live/<id>/ carries the raw key.
register_credential_file() takes a skill-declared relative path from
required_credential_files frontmatter and bind-mounts it read-only into the
remote sandbox the skill's own code runs in. It validates that the resolved
path stays inside HERMES_HOME — the docstring names the threat directly:
so that a malicious skill cannot declare
required_credential_files: ['../../.ssh/id_rsa'] and exfiltrate
sensitive host files into a container sandbox
Containment is the wrong boundary on its own, because HERMES_HOME is exactly
where the master credential stores live. Traversal is blocked; asking for the
keys by name is not:
skill declares mounted? agent may read it?
.env YES DENIED
auth.json YES DENIED
.anthropic_oauth.json YES DENIED
cache/bws_cache.json YES DENIED
mcp-tokens/srv.json YES DENIED
google_token.json YES allowed
../../.ssh/id_rsa no n/a
Every row marked DENIED is refused by the canonical read guard
(agent.file_safety.get_read_block_error) — the agent cannot read_file them —
yet one line of hub-installed skill frontmatter gets them bind-mounted where
that skill can cat them. .env alone is every provider API key.
Reuse the canonical deny-list as the mount bar: what the agent is forbidden
to read is not mountable either, so the mount surface cannot hand a skill
what the read surface denies it. Fails CLOSED — if the guard can't be
consulted the mount is refused rather than risked.
The module keeps doing its job: a skill still mounts its own service token
(google_token.json, skills/*), and a refused entry is reported back through
register_credential_files' missing list instead of failing the batch.
The three prior PRs here (#3946, #3951, #4316) all hardened traversal; this
closes the half that traversal validation never covered.
An orphaned pip descendant holding the probe's inherited stdout/stderr
pipe handles wedged subprocess.run's post-timeout communicate() (which
joins the pipe reader threads with NO timeout on Windows). The warm
probe thread then hung holding the module-level _CACHE_LOCK, so every
new session's prompt build blocked indefinitely (#67964).
Two layers:
- _run(): replace subprocess.run with Popen + communicate(timeout); on
TimeoutExpired kill the process TREE (taskkill /T on Windows) and
reap the direct child bounded — never re-read the pipes, so an
orphaned descendant holding them open can't block us.
- get_environment_probe_line(): the probe now runs in a single
background worker publishing via a threading.Event; callers wait at
most _PROBE_WAIT_TIMEOUT (10s) then fail open with "". After one
full timeout, later callers only peek. If the stuck worker ever
finishes, its line resumes appearing in new prompts.
Regression tests: hung probe with 4 concurrent callers returns bounded;
late recovery publishes; repeat callers skip the wait; _run() returns
promptly despite a pipe-holding descendant (real subprocess E2E).
Fixes#67964
judge_goal() returns (verdict, reason, parse_failed, wait_directive) since
the goals.py wait-directive change, but the kanban goal-mode completion gate
at tools/kanban_tools.py still unpacked 3 values. Every judge call raised
ValueError, the defensive except swallowed it, and the pre-initialized
verdict='done' let every completion through — the acceptance gate was
silently disabled.
Now unpacks all 4 values; the test mock is updated to match the real
contract. The other two judge_goal consumers (hermes_cli/goals.py) already
use 4-value unpacks.
Reported and diagnosed by @bill3wits in PR #57276; reimplemented under
project authorship because the original commit was authored under a
non-existent local identity (bash@hermes.local) that cannot be carried
into history. Also fixes#58066 (duplicate report by @Gibcity).
Follow-up for salvaged PR #39946: file_sync.py already aliases time.sleep
as _sleep specifically to avoid tests mutating the shared stdlib module
object. Apply the same convention to the rate-limit clock (_monotonic)
and point the new regression test at it.
FileSyncManager.sync() is rate-limited to once per _sync_interval via
_last_sync_time, and its docstring promises that on failure "state rolls
back so the next cycle retries everything". But the except handler also
set _last_sync_time = time.monotonic() on failure, so the next non-forced
sync() within the interval hit the rate-limit guard and returned early —
suppressing the retry the rollback had just prepared.
Because the non-forced sync() runs before every command on the SSH, Modal
and Daytona backends, a single transient upload failure (network blip,
dropped channel) left the remote with stale files for the next command
(up to _sync_interval, default 5s). Forced syncs bypass the guard, which
is why it was intermittent.
Remove the failure-path timestamp bump so the clock only advances on a
successful or no-op cycle, matching the documented contract. Add a
regression test that fails before this change and passes after.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
grok-4.5 is xAI's newest release (their versioning is non-monotonic:
4.5 > 4.20) and is the model xAI's own docs use for the server-side
x_search tool. Users who explicitly pinned x_search.model keep their
choice; everyone else picks up the new default via the config
deep-merge — no _config_version bump needed.
- tools/x_search_tool.py: DEFAULT_X_SEARCH_MODEL
- hermes_cli/config.py: DEFAULT_CONFIG x_search.model + comment
- agent/reasoning_timeouts.py: 300s stale-timeout floor entry for
grok-4.5 (grok-4.20-reasoning entry kept for pinned users)
- docs: x-search.md en + zh-Hans (config sample + troubleshooting)
- tests: default-model assertion + timeout-floor positive case
* feat(delegation): live-viewable subagent transcripts for delegate_task
Each child now streams an append-only, human-readable log to
<hermes_home>/cache/delegation/live/<delegation_id>/task-<n>.log while it
runs, and the dispatch return includes the paths so the caller can tail
them immediately instead of waiting blind for the consolidated summary.
- New tools/delegation_live_log.py: LiveTranscriptWriter (per-event append
+ flush, one-line rendering with truncation, never raises into the agent
loop), wrap_progress_callback (tees the child's existing
tool_progress_callback events into the log, preserves the _flush
contract), dispatch-time creation with pre-headered files so tail -f
attaches immediately, manifest.json (goals/task count/per-task status),
and 7-day retention pruning on new dispatches.
- delegate_task: wraps each child's progress callback with the writer;
sync results and background dispatch responses gain live_transcripts
(+ hint field on dispatch); per-task result entries carry
live_transcript; transcripts finalized with exit-reason markers.
- async_delegation: dispatch_async_delegation_batch accepts an optional
delegation_id so the live/ dir name matches the returned handle; the
completion event carries live_transcripts.
- process_registry: consolidated batch-completion block references each
task's live transcript path.
- Tool schema description documents the live_transcripts return surface;
docs gain a 'Live Transcripts' section with a tail -f example.
Placement under cache/delegation means the logs are mounted read-only
into remote terminal backends for free. Side-channel only: zero changes
to message content, so prompt caching is unaffected. Transcript-OUT only
— no overlap with the subagent control surfaces of PR #66046.
* fix(delegation): label the kickoff transcript line as user — it is the child's one user message
The salvaged unit tests hand-construct completed futures to lock the
_run_on_mcp_loop contract. Add a live-loop test that reproduces the actual
field trigger: an inner asyncio.wait_for expiry stores a real TimeoutError
on a real future scheduled on the MCP loop. Asserts the fixed loop surfaces
it once, promptly -- not spinning to the outer deadline (which both leaked
memory and masked the real error behind the generic wrapper message).
Bug 1 of #55048: when the MCP connection dropped (driver crash / restart),
_lifecycle_coro exited but left _started=True, so the next list_apps/capture
passed _require_started() and then operated on a None session — hanging
forever instead of reconnecting.
- _lifecycle_coro's finally now resets _started=False on ANY exit, so a dead
session is re-enterable (idempotent no-op on the normal stop() path; atomic
bool write, safe from the bridge-loop thread without the lock stop() holds).
- call_tool() re-enters start() when the session isn't active, rebuilding it
before the call. The start_session/end_session handshake (driven by start()/
stop() themselves) is exempted so bootstrap doesn't recurse.
Tests: two cases in test_computer_use_delivery_ladder.py — finally resets
_started, and call_tool restarts a dead session exactly once. Full
computer_use suite green (233).
Refs #55048 (Bug 1). Bug 2 (expose foreground dispatch) is covered by the
delivery_mode work in #67123.
Hermes' computer_use wrapper dropped cua-driver's structured action verdicts,
exposed no delivery_mode, and injected background-only guidance — so the agent
reported unverified no-ops as success and concluded cua-driver 'cannot drive'
Electron/Chromium surfaces (observed live on tldraw offline). Fixes#67052.
Phase A — preserve the result contract:
- ActionResult carries verified/effect/escalation/path/degraded/code/delivery_mode
- CuaDriverBackend._action() reads structuredContent (was data-only); a helper
normalizes it, additive and None-safe on old drivers
- _text_response surfaces the fields additively (ok stays transport-only)
Phase B — bounded, model-reachable foreground:
- delivery_mode (background|foreground) + bring_to_front on the schema, dispatcher,
ABC, and all input methods
- foreground is capability-gated (input.delivery_mode); old drivers get a
structured foreground_unsupported refusal, never a silent background downgrade
- no automatic/hidden foreground retry — the model selects it from the signal
Phase C — guidance + isolation:
- system prompt (prompt_builder) and bundled skills/computer-use/SKILL.md go from
background-ONLY to background-FIRST, teaching the AX→PX→foreground ladder driven
by returned effect/escalation, not predicted from the app being Electron
- foreground approval scoped by (action, delivery_mode): a background approval
never silently authorizes foreground
- approval state keyed per session_id so concurrent gateway runs don't leak unlocks
Tests: tests/tools/test_computer_use_delivery_ladder.py (15) cover confirmed/
unverifiable/suspected_noop/degraded/old-driver verdicts, delivery_mode gating +
foreground_unsupported, and session-scoped foreground approval. Existing 265
computer_use tests still green.
Live E2E (real cua-driver 0.8.3 + tldraw offline on Linux/X11): a background click
returned effect='unverifiable'/path='ax' (no fabricated success), and a foreground
request returned code='foreground_unsupported' — correct on a driver that predates
the input.delivery_mode capability.
Adds two tests for the _background_review_write_guard manual-skill check:
- refuses delete on a skill with created_by=None (manually authored)
- allows delete on a skill with created_by='agent' (agent-created)
Use the direct POSIX parent relationship instead of process creation time and pid_exists checks. Remove the dead create-time argument chain while preserving process-group cleanup and signal forwarding.\n\nRefs #62505
A leaf subagent is meant to be denied delegate_task, execute_code, memory,
clarify, cronjob, and send_message. _strip_blocked_tools() only drops a
toolset when EVERY tool in it is blocked, so mixed platform bundles
(hermes-cli, hermes-telegram, and every other gateway bundle) survived
stripping and re-exposed the blocked tools after composite expansion. A
leaf child spawned from any gateway platform could recursively delegate,
run code, and write memory.
Pass exact one-tool deny toolsets into the child's disabled_toolsets so
model_tools subtracts the blocked names AFTER composite expansion, and the
restriction survives later registry/MCP refreshes. Orchestrators regain
only delegate_task.
Salvaged from #66036 by Mason Tanguay (@DictatorBacon); scoped to the
authority fix + its regressions (docs/interrupt changes dropped).
Co-authored-by: Teknium <127238744+teknium1@users.noreply.github.com>
* feat(attribution): conflict-free contributor mappings via contributors/emails/ directory
The AUTHOR_MAP dict in scripts/release.py was a merge-conflict magnet:
every concurrent salvage PR appended entries to the same lines of the
same file, so parallel PRs re-conflicted on every merge to main.
New system: one file per email under contributors/emails/ — filename is
the commit-author email, first non-comment line is the GitHub login.
File additions never conflict, so any number of PRs can add mappings
concurrently.
- scripts/release.py: AUTHOR_MAP is now LEGACY_AUTHOR_MAP (frozen)
merged with the directory at import time (directory wins). All
existing consumers (resolve_author, contributor_audit.py) unchanged.
- scripts/add_contributor.py: idempotent CLI to add a mapping; refuses
conflicting reassignments (incl. against the legacy map), validates
email/login shapes.
- contributor-check.yml: attribution gate now accepts a mapping file OR
a legacy entry; failure message prints the exact add_contributor
command. Also auto-resolves bare <login>@users.noreply.github.com
emails is intentionally NOT added (kept id+login form only, matching
previous behavior).
- contributor_audit.py: guidance now points at add_contributor.py.
- tests/scripts/test_contributor_map.py: 12 tests covering loader,
merge precedence, CLI idempotency/conflict/validation, subprocess E2E.
* feat(ci): one-shot per-file flake retry in the parallel test runner
A failing test FILE is re-run once in a fresh subprocess. Pass-on-retry
counts as green but is loudly reported in a '⚠ FLAKY' summary section
(with both attempts' output preserved) so the flake gets fixed instead
of eating a full-run rerun. Deterministic failures fail both attempts —
regressions cannot be laundered green.
- --file-retries N / HERMES_TEST_FILE_RETRIES (default 1, 0 disables)
- E2E verified: simulated first-run-fail flake goes green with banner;
deterministic failure still exits 1; retries=0 restores old behavior.
This converts the dominant CI failure mode (one timing-sensitive test
flaking a 4600-test shard, requiring a manual 10-minute rerun and an
agent triage loop) into a self-healing retry that costs one file's
runtime.
* test(approval): loosen wall-clock perf bounds 0.15s -> 2.0s
These guard against catastrophic regex backtracking (seconds-to-minutes
class), but 0.15s is within scheduler-stall noise on loaded shared CI
runners — test_max_accepted_separator_free_input_is_fast failed a CI
shard this week on runner load alone. 2.0s still catches the regression
class with zero flake surface.
* fix(ci): job timeouts everywhere + retries on all network installs
Reliability pass over every workflow:
- timeout-minutes on all 21 jobs that lacked one (a hung job previously
burned the 6-hour default runner budget)
- ./.github/actions/retry wrapped around every network-fetching install
that lacked it: pip installs (deploy-site, skills-index), npm ci
(deploy-site website, upload_to_pypi web + ui-tui), uv sync (docker
test deps). Deterministic build steps (npm run build) deliberately
NOT retried — split into separate steps so a real build failure fails
fast instead of retrying 3x.
* docs(agents): document the file-retry flake policy
* fix(ci): curl retries on deploy hook + skills-index probe
* fix(ci): kill the remaining transient-failure classes in workflows + Dockerfile
From the workflow reliability audit:
- tests.yml: duration-cache restore had NO restore-keys while saves use
run_id-suffixed keys — the cache never matched once, so LPT slicing
always ran blind and unbalanced slices pushed heavy files toward the
per-file timeout. One-line restore-keys fixes slice balancing.
- Label gates (lint ci-reviewed, supply-chain mcp-catalog-reviewed):
'gh pr view || true' turned an API blip into 'label absent' → false
BLOCKING failure. Now 3x retry, and API failure is reported as an API
failure instead of a missing label.
- detect-changes action: compare API retried before failing open (was
silently running all lanes on any blip).
- uv-lockfile-check: 'uv lock --check' resolves against PyPI — retried
so registry blips don't read as 'lockfile stale'.
- docker.yml merge job: imagetools create retried (Docker Hub eventual
consistency on just-pushed digests).
- Dockerfile: apt-get Acquire::Retries=3; s6-overlay ADDs converted to
curl --retry 3 (ADD cannot retry; checksums still enforced); npm
--fetch-retries=5; playwright chromium fetch retried 3x.
- Advisory artifact uploads (per-slice durations, ci-timings report)
get continue-on-error so an artifact-service blip can't fail a green
test slice.
* fix(tests): kill the two root-cause flakes — leaking pre-warm timer + env-dependent provider list
- test_tui_gateway_server.py: session.create / non-eager session.resume
arm a 50ms threading.Timer (_schedule_agent_build) that outlives its
test and fires into the NEXT test's _make_agent mock, racily
corrupting captured state (the recurring session_resume shard
failures). Replaced the per-test whack-a-mole stub with a module-wide
autouse fixture; the 3 worker-lifecycle tests that genuinely need the
deferred build opt back in via @pytest.mark.real_agent_prewarm (new
marker in pyproject).
- test_api_key_providers.py: PROVIDER_ENV_VARS is now derived from the
live PROVIDER_REGISTRY instead of a hand-list that had drifted
(missing HF_TOKEN / DEEPINFRA_API_KEY) — resolve_provider('auto')
tests failed on any machine with HF_TOKEN exported. E2E-verified with
HF_TOKEN/DEEPINFRA_API_KEY set: 42/42 pass.
* test: de-flake 30 timing-sensitive test files for loaded CI runners
Root-cause fixes from the flake audit (session-DB mining + repo sweep):
Event-based sync instead of sleep-sync:
- title_generator: mock sets threading.Event, wait(10) replaces
sleep(0.3) hoping the daemon thread got scheduled
- docker zombie_reaping / profile_gateway: poll-for-state helpers
replace fixed 1-3s sleeps (s6 transitions + SIGCHLD reaping are async)
- process_registry tree test: select()-bounded readline replaces an
unbounded blocking read (parent wedge now fails THIS test with a clear
message instead of an opaque rc=124 file kill); SIGTERM grace 1s->2s
(the 1s partition window mid-interpreter-startup is how a child PID
escaped the live-system guard in CI)
Timeout raises (loaded 8-way-sliced runners see ~5s scheduling floors;
all of these complete in ms-to-1s when healthy so the raises cost
nothing on green runs):
- subprocess/thread waits <= 2s raised to 10-15s across mcp_tool,
mcp_circuit_breaker, mcp_reconnect_retry_reset, mcp_parked_self_probe,
mcp_cancelled_error_propagation, registry, clarify_gateway, interrupt,
voice_cli_integration, docker_environment, session_store_lock_io,
planned_stop_watcher, cli_interrupt_subagent, thread_scoped_output
(joins now also assert not is_alive() so stragglers fail loudly)
- wall-clock discrimination ceilings loosened where the guarded hang is
10x larger: local_background_child_hang 4s->10s, interrupt_cleanup
setup 5s->20s + pgid-exit 30s->60s, mcp_stability grandchild spinup
5s->15s, protocol/gil-starvation fast-handler 0.5s->2s,
iso_certify_seam 1.5s->5s, wait_for_mcp_discovery 0.1s->1s
- narrow assertion windows widened: honcho first-turn wait 0.4..0.65 ->
0.25..2.0 (property is bounded-not-hung, not an exact wall-clock);
compression fork-lock TTL 1s->3s (12 refresh chances per lease);
compression-lock expiry margins symmetric (ttl 0.05->0.5, sleep 1.0)
- telegram hung-DNS bound 1.0->1.4 (fake hang is 1.5s — must stay under)
* fix(tests): repair indentation from de-flake batch edit
* fix(tests): harden env isolation and replace remaining sleep-sync races
The full 42k-test run and complete npm check surfaced three more classes:
- Environment isolation: local ~/.honcho defaultHost and SSH_* variables
leaked into Python/TUI tests. Pin the default Honcho host in the
hermetic fixture, isolate the one fallback test from ~/.honcho, and
blank SSH_* around terminalSetup tests. This flipped 20 false failures
back to deterministic behavior on developer machines.
- Background-thread sleep-sync: Honcho async writer tests patched
time.sleep globally, then busy-polled with that same mocked sleep. Under
full-suite load the poller could starve the writer. Each test now waits
on an Event emitted by the exact flush/retry transition; 30/30 passed
under 15-way contention.
- Desktop streaming: the test slept 80ms and assumed a 500ms timer could
not fire before its assertion. A loaded runner descheduled the test for
>500ms and both chunks arrived. Producer controls now gate second-chunk
and completion transitions explicitly.
Also make file-retry observability complete: a self-healed flaky file now
prints BOTH attempts' full output in the FLAKY summary. Two behavioral
runner tests prove pass-on-retry is green+loud+traceback-preserving, while
a deterministic failure remains red.
* refactor(ci): use gh bot pat, better retries
refactor(ci): use retry action for PR label fetch
the retry action now captures stdout as a step output, so it can serve
double duty: retry + output capture for commands like 'gh pr view' whose
result must be consumed by later steps.
Retry action gains:
- 'stdout' output (heredoc-delimited to preserve newlines)
- tee to temp file so stdout still streams to the job log
- step id 'retry' for output reference
Both lint.yml and supply-chain-audit.yml now use the retry action
directly with 'command: gh pr view ...' and read
steps.<id>.outputs.stdout.
ci: use AUTOFIX_BOT_PAT for all gh CLI / GitHub API auth
Replace secrets.GITHUB_TOKEN and github.token with
secrets.AUTOFIX_BOT_PAT across all workflows and composite actions
that use the gh CLI or GitHub API. The PAT has consistent permissions
across fork PRs (where GITHUB_TOKEN is read-only), avoids API rate
limit sharing with the default token, and is already used by
js-autofix.yml for the same reasons.
19 sites swapped across 9 files:
- lint.yml (3): label fetch, comment post/edit, comment update
- supply-chain-audit.yml (5): scan, critical comment, unbounded dep
comment, label fetch, mcp-catalog comment
- lockfile-diff.yml (1): PR comment post/update
- skills-index-freshness.yml (1): issue creation on degraded probe
- skills-index.yml (2): index build, trigger deploy workflow
- upload_to_pypi.yml (2): release view poll, release upload
- ci.yml (1): timings report
- deploy-site.yml (2): skills index crawl
- detect-changes/action.yml (1): compare API call
---------
Co-authored-by: ethernet <arilotter@gmail.com>
A root-launched CLI session can leak /root into the terminal cwd state a
non-root gateway/cron process later resolves (#65583). os.path.isdir('/root')
is True for a non-root user — stat only needs search permission on / — so
_resolve_safe_cwd returned it and subprocess.Popen(cwd='/root') died with
PermissionError: [Errno 13], failing EVERY cron job's terminal/file/search
tool on every command until restart.
_resolve_safe_cwd now requires X_OK (new _cwd_usable helper) and climbs to
the nearest enterable ancestor, logging a WARNING that names the leak class
when an existing-but-denied cwd is skipped. Missing-cwd recovery (#17558)
behavior unchanged.
E2E-verified: LocalEnvironment constructed with an unenterable cwd now runs
commands from the fallback directory instead of raising.
test_build_child_agent_ignores_acp_command_when_binary_missing patches
shutil.which globally and asserted the LAST call was which("copilot").
That is order-dependent: an unrelated which("uv") reached later in the same
process (which happens under some CI test-slice orderings) becomes the last
call, so assert_called_with("copilot") fails even though the copilot binary
was probed exactly as intended. Switch to assert_any_call("copilot"), which
verifies the actual intent and is robust to unrelated which() calls. The
behavioural assertions (provider, acp_command, acp_args) are unchanged.
Port from anomalyco/opencode#35439/#35500: preserve full MCP catalogs
across paginated tools/list responses.
The MCP spec allows servers to paginate tools/list, resources/list, and
prompts/list via an opaque nextCursor token. The Python SDK's
ClientSession.list_* methods fetch exactly one page per call, and hermes
never passed the cursor back — on a paginated server every tool,
resource, and prompt past page 1 was silently invisible to the agent.
Adds _paginate_full_list() (cursor-draining helper with a 50-page
runaway cap and spec-correct opaque-string cursor validation) and
applies it at all three discovery sites: _discover_tools(), the
tools/list_changed refresh handler, and the list_resources/list_prompts
utility handlers. The keepalive probe intentionally keeps its
single-page call (liveness only).
E2E: real stdio MCP server serving 3 pages (2/1/1 tools) — old code
discovered 2 tools, new code discovers all 4.
Port from openai/codex#31494: user-visible history replay must strip CSI
sequences and control characters. Stored conversation history can carry
raw terminal escapes (pasted content, gateway-origin text, model output
echoing injected tool results). Replaying it via /resume's recap panel or
build_recap (/status on CLI + gateway) wrote those bytes straight to the
terminal — an injected message could clear the screen, retitle the window,
move the cursor, or restyle the recap UI. Rich's Text() does not neutralize
raw escape bytes.
- tools/ansi_strip.py: add sanitize_display_text() — strip_ansi() plus
bare C0/C1 control removal, preserving \n and \t, normalizing \r to \n
(adapted to Python from Codex's sanitize_user_text; reuses the existing
ECMA-48 stripper instead of transcribing their char-walk)
- hermes_cli/cli_agent_setup_mixin.py: sanitize user + assistant text in
_display_resumed_history() before building the Rich recap panel
- hermes_cli/session_recap.py: sanitize preview lines in build_recap()
(_truncate choke point) so /status recaps are clean on every platform
- tests: 10 new sanitize_display_text cases (incl. the exact codex#31494
fixture), recap + resume-display leak assertions
A UTF-8 BOM saved into a SKILL.md (e.g. Notepad or PowerShell `>`) is kept by
read_text(encoding="utf-8"), so the string handed to parse_frontmatter starts
with the BOM and the startswith("---") fence check fails. The whole frontmatter
is then silently dropped: the skill loads with no name/description, `platforms`
gating falls open (a macOS-only skill becomes visible everywhere), and
required_environment_variables / metadata.hermes.config setup never fires.
Strip a single leading BOM at the top of parse_frontmatter, the shared
chokepoint for every local skill-loading path (_parse_skill_file,
discover_all_skill_config_vars, DESCRIPTION.md parsing, _inject_skill_config,
and the tools/skills_tool._parse_frontmatter re-export), so the whole class is
covered, not just the reported site. Only the leading marker is removed; a BOM
mid-content is left as data. Mirrors the existing file-tools BOM handling
(#35278) and CONTRIBUTING.md "File encoding".
Adds tests: BOM'd frontmatter parses identically to plain, the body is
BOM-free, platforms gating and config-var extraction survive a BOM, and an
end-to-end BOM-write / plain-read round trip (mirroring _parse_skill_file).
* feat(browser): store full snapshots on truncation; make eval denylist opt-in
Two harness fixes motivated by BU_Bench results where fixed-verb + lossy
observation cost Hermes heavily vs code-driven browser agents:
1. Snapshot truncation no longer loses content. When a snapshot exceeds
the 8000-char threshold, the complete accessibility tree is saved to
cache/web (same truncate-and-store pattern as web_extract) and the
truncated view / LLM summary includes the file path plus a ready-made
read_file call. Element refs beyond the cut are recoverable without
re-snapshotting. Stored copies are force-redacted and capped at 2MB;
content-hash filenames dedupe repeated snapshots of the same page.
2. The browser_console(expression=...) sensitive-primitive denylist is
now opt-in via browser.restrict_evaluate (default false). The
names-based denylist blocked legitimate DOM extraction — any selector
or expression containing 'fetch', 'cookie', 'input', etc. — which
crippled the agent's only programmatic page-inspection path. The
SSRF/private-URL egress guards in _browser_eval are independent of
this policy and remain always-on. browser.allow_unsafe_evaluate keeps
its meaning (bypass the denylist) for configs that already set it.
* test: update None-guard test for stored-snapshot pointer in _extract_relevant_content
test_normal_content_returned pinned the exact return value; the summary
now carries a pointer to the stored full snapshot. Assert the summary
passes through and the pointer is present instead.
* feat(browser): align snapshot threshold with web_extract's 15k char budget
SNAPSHOT_SUMMARIZE_THRESHOLD 8000 -> 15000, matching
web_tools.DEFAULT_EXTRACT_CHAR_LIMIT so the snapshot and web_extract
truncate-and-store paths give the model the same per-page budget.
_truncate_snapshot's default max_chars now follows the constant.
Invariant test added; docs (en+zh) and CLI tip updated.
Keep each xAI OAuth auth-add login as an independent manual device-code pool entry and recognize xAI personal-team spending-limit 403 responses as billing exhaustion. Preserve the structured top-level error message so the failed credential is quarantined and the next healthy account is selected without attempting a pointless token refresh.
Route direct xAI HTTP consumers through the credential pool as well. Proactive and 401-reactive refreshes update the exact issuing manual entry, preserve validated xAI base URL overrides, and serialize single-use refresh-token rotation across concurrent pool instances.
terminal_tool reads all settings from TERMINAL_* env vars, bridged from
config.yaml by the CLI, gateway, and TUI-PTY launchers. Processes that
skip every launcher bridge — hermes serve / the Desktop app backend's
in-process agents, the desktop cron ticker — saw an unset TERMINAL_ENV
and silently ran every command on the host even when config.yaml selects
terminal.backend: docker. A user who configured Docker isolation got
unsandboxed host execution with no warning.
Two layers:
- _ensure_terminal_env_bridged() in _get_env_config(): when TERMINAL_ENV
is unset, backfill TERMINAL_* from config.yaml via
apply_terminal_config_to_env(override=False). Explicit env always wins
(honor explicit choice; only fix the accidental fallback). One-shot,
fail-open to the historical local default.
- cmd_dashboard/serve: run the same bridge at startup so every consumer
in the backend process (in-process agents, desktop cron ticker,
tui_gateway cwd resolution) sees the bridged env directly.
Fixes#63141, #54449, #61115, #65696.
_download_url_with_cap called urlopen() after only a scheme check, so a
model-controlled URL could reach loopback services, RFC1918/CGNAT hosts,
or cloud metadata endpoints (169.254.169.254), and a public host could
302 to any of those unvalidated.
Route the fetch through the repo's canonical SSRF guard instead:
validate every hop with tools.url_safety.is_safe_url() and follow
redirects manually (httpx, follow_redirects=False, 5-hop limit) so each
Location target is re-checked before it is fetched — the same pattern
as tools/skills_hub._guarded_http_get. The streaming size cap is
unchanged. Local-fixture tests opt in via HERMES_ALLOW_PRIVATE_URLS
(the guard's documented escape hatch); new tests pin rejection of
loopback, cloud-metadata, and private-range URLs, a mocked
public→loopback redirect, and a mocked public happy path.
The salvaged attachment-toolset commit predated main centralizing the
25 MB cap as kanban_db.KANBAN_ATTACHMENT_MAX_BYTES and re-introduced a
private _MAX_ATTACHMENT_BYTES alias. Drop the duplicate: kanban_db's
store_attachment_bytes(), the dashboard upload endpoint, and the
kanban_attach_url tool all reference the one shared constant now, and
the tests monkeypatch that same name.
The kanban board has had full attachment storage and a dashboard HTTP
API (upload/list/download/delete) since #35338, but there was no agent
toolset tool and no `hermes kanban` CLI verb for attachments. Agents and
scripts that don't go through the dashboard server (or can't touch the DB
directly) had no way to create or read real attachments — only links in
comments.
Close that gap by mirroring the existing comment surface:
- `kanban_db.store_attachment_bytes()` — one shared write path (validate
name, enforce the 25 MB cap, write the blob under the per-task dir with
collision-free naming, insert the metadata row, clean up an orphan blob
if the insert fails). `_MAX_ATTACHMENT_BYTES`, `_safe_attachment_name`,
and a new `_collision_free_path` move here so the dashboard, the tool,
and the CLI all share one implementation and can't drift.
- Tools (`tools/kanban_tools.py`): `kanban_attach` (inline base64),
`kanban_attach_url` (server-side http/https fetch with the same cap),
`kanban_attachments` (list). Write tools respect worker task-ownership;
list is read-only. Registered in the `kanban` toolset.
- CLI (`hermes_cli/kanban.py`): `attach <id> <path>`, `attachments <id>`,
`attach-rm <attachment_id>`.
- Dashboard `upload_task_attachment` now imports the shared helpers and
uses `_collision_free_path` — behavior identical (still streams to disk
with the cap, still 413 on overflow).
- Docs (AGENTS.md, kanban-worker skill) and toolset membership updated.
Tests: tool round-trip + oversize + bad base64 + ownership; attach_url
against a local HTTP fixture incl. oversize-mid-stream and non-http
scheme rejection; CLI attach/attachments/attach-rm; shared-helper unit
tests; dashboard parity preserved.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
_wait_for_callback still read the legacy module-level _oauth_port, so
with two concurrent OAuth flows, flow A's callback wait bound flow B's
port while A's redirect URI pointed at A's port — the callback-side
half of the cross-flow collision that #65622 fixed on the redirect
side. _make_callback_waiter(port) closes over each flow's resolved
port; both provider construction sites (build_oauth_auth and
MCPOAuthManager._build_provider) now wire per-flow waiters. The legacy
_wait_for_callback delegates for backwards compatibility.
Direction credit to @LeonSGP43 (#34280) and the #34260 analysis.
_wait_for_callback catches OSError on bind with a comment claiming the port is
held by a server build_oauth_auth started, and promising to fall back to polling
it. build_oauth_auth never starts a callback server (this is the only listener),
so there is nothing to poll: the branch just raised a misleading "OAuth callback
timed out" when the real cause is a busy port (a concurrent login, a leftover
listener, or a fixed oauth.redirect_port that collided).
Fix the stale comment and raise an accurate, actionable message (names the port,
suggests freeing it or setting a free oauth.redirect_port), chained from the
original OSError. Behavior is otherwise unchanged. Adds a regression test.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- redirect_host tests: localhost swap, precedence of full redirect_uri,
empty-value fallback, client-metadata propagation
- The salvaged redirect-uri hint tests predate the #57836
OAuthNonInteractiveError guard; monkeypatch _is_interactive like the
sibling tests in TestRedirectHandlerSshHint
The PR added a configurable `redirect_uri` (proxy/Funnel callbacks) but
shipped without tests, and the loopback SSH-tunnel hint stayed hardcoded —
actively misleading the exact proxy user the feature targets.
- Extract `_resolve_redirect_uri(cfg, port)` so the client-metadata and
pre-registration paths derive an identical callback (a mismatch makes the
authorization server reject the redirect).
- Make `_redirect_handler` redirect_uri-aware: a configured proxy callback
reaches this machine on its own, so it no longer prints the `ssh -N -L`
loopback guidance. Wired via `functools.partial` — no new global state.
- Document `redirect_uri` in the config block.
- 14 new tests (red/green TDD): helper resolution + empty-string fallback,
metadata + pre-registration for configured/default, AnyUrl normalization,
no-client_id skip, client_secret combo, and both SSH-hint branches.
ruff clean · 81 passed (tests/tools/test_mcp_oauth.py) · ty baseline unchanged
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
_find_free_port() closed its probe socket before HTTPServer re-bound
the port minutes later, leaving a window where another process could
steal it (#22161 by @amathxbt). _reserve_callback_port() now keeps the
selected socket bound (bounded FIFO pool) until _wait_for_callback
adopts it via bind_and_activate=False. Also sets allow_reuse_address
BEFORE binding — the cherry-picked #44872 set it after the constructor
had already bound, where it is a no-op.
Also updates the three #57836 non-interactive-guard tests to the
closure-factory API from #44872.
Two related OAuth fixes:
1. Replace module-level _redirect_handler with _make_redirect_handler()
closure factory that closes over the resolved port. This prevents
cross-server state pollution when multiple MCP servers run OAuth
concurrently (#44588).
2. Set server.allow_reuse_address = True on the ephemeral callback
HTTPServer so the socket doesn't stay in TIME_WAIT after the flow
completes. This prevents 'Address already in use' errors on the
next OAuth flow for the same port (#44590).
Fixes#44588Fixes#44590
Follow-up for salvaged PRs #56055 + #56803, adapted to the per-session
cwd record store (PR #65213): the record (live cd state) is rung 1,
the registered session.cwd.set override rung 2, TERMINAL_CWD rung 3 —
the same ladder file tools and the terminal resolve against. Covers
record-over-override, record-only, and stale-record fall-through.
Follow-up for salvaged PR #63255: with LocalEnvironment._update_cwd
delegating to the stdout marker parser, the cwd temp file has zero
readers left. Drop the 'pwd -P > file' writes from the bootstrap and
_wrap_command so every command stops paying a pointless file write
(and stops littering temp dirs with hermes-cwd-*.txt).
Shizoqua
Alex Fournier
Teknium
Francesco Bonacci
Andrew Chen
Koho Zheng
ai-ag2026
Jasmine Naderi
victor-kyriazakos
beardedeagle
Ben Barclay
Shannon Sands
Carlos Diosdado
Frowtek
dsad
Eugeniusz Gilewski
Sherman
峯岸 亮
CleanDev-Fix
Atakan
3ASiC
Solitud1nem
obelisk-complex
Vivaan Dhawan
Florian Valade
kshitij
Dean Chen
simonmmafs
xxxigm
ygd58
Brice
Kewe63
YAMAGUCHI Seiji
Michel Belleau
Gabriel Anzziani
Alcibiades Athens
Hu
Alexander Russell
Tikkanaditya Siddartha Jyothi
luyifan
aml1973
LeonSGP43
Ben Sheridan-Edwards
Tobias Safaie geb. Schmidt-Philipp
Dennis
Zehua Wang
LauraGPT
Tushar
Charles Cha
AnthonyAssistantAi
liuhao1024
Tranquil-Flow
Drexuxux
Elie BRUNO
Brooklyn Nicholson
Omid Saadat
Booker
墨綠BG
Mate Remias
zombopanda
Jeffrey Quesnelle
rob-maron
embwl0x
Jeffrey Quesnelle
HexLab98
Vite Balloons
Ghislain LE MEUR
tymrtn
Gille
webtecnica
webtecnica
bo.fu
Victor Kyriazakos
PRATHAMESH75
joaomarcos
Dhruv Raajeev
liuhao1024
jinglun010-cpu
Hermes Agent
flyingdoubleg
Alan Harman-Box
Julien Talbot
Gabriel Steenhoek
replygirl
Ian Ker-Seymer
Willie Peacock
Mustafa
Dickson Neoh
trkim
David Metcalfe
Tianqing Yun
davesecops
kandotrun
MacroAnarchy
Andy
SeoYeonKim
Börje
trevorgordon981
yungchentang
hanyu1212
João Vitor Cunha
Taylor H. Perkins
Jupiter363
Bruce-anle
Mason Tanguay
Dusk1e
Que0x
cresslank
Erosika
Tim Roth
otsune
doxe0x
Florian Burka
Code-suphub
nima20002000
Gaurav Saxena