The banner update-check ran an unscoped 'git fetch origin', transferring
all ~1,400 remote heads (measured 3.0s dry-run vs 0.55s scoped, and up to
70s on a cold ref store) and frequently burning its full 10s timeout on
slow links. cmd_update already scopes its fetch for exactly this reason.
A scoped 'git fetch origin main' updates both the origin/main tracking
ref (full-clone count path) and FETCH_HEAD (shallow compare path), so
behind-count semantics are unchanged — verified empirically on a full
clone (rewound tracking ref restored to tip, count correct) and a
--depth 1 shallow clone (FETCH_HEAD updated, boundary preserved).
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).
Three cold-start cuts, measured with .venv python, PYTHONPATH=worktree,
median of 3-5 fresh subprocesses:
1. tools/mcp_oauth.py — availability now via importlib.util.find_spec("mcp");
SDK classes (OAuthClientProvider et al.) import lazily on first use via
_ensure_sdk_loaded(). Module-level names kept as None placeholders so the
test-patch surface (patch.object(mcp_oauth, "OAuthClientProvider", ...))
still works. import tools.mcp_oauth: 242ms -> 56ms (mcp SDK no longer
loaded at import time).
2. tools/registry.py — discover_builtin_tools AST scan memoized in an
mtime_ns+size-keyed disk cache at ~/.hermes/cache/tool_discovery_cache.json
(atomic write via utils.atomic_json_write, best-effort/never raises;
corrupt or missing cache -> full rescan + rewrite; per-file stat mismatch
-> rescan just that file). Scan of 100 files: 158ms cold -> 3ms warm.
3. tools/browser_tool.py — top-level `import requests` and
agent.auxiliary_client.call_llm moved to lazy first-use (PEP 562
__getattr__ preserves patch("tools.browser_tool.requests.get") and
patch("tools.browser_tool.call_llm") surfaces; internal call sites go
through _lazy_call_llm which reads module globals so patches are honored).
Entry-point imports (median ms, before -> after):
import model_tools 392 -> 245 (warm discovery cache)
import cli 152 -> 151 (unchanged; cli doesn't hit these paths)
import gateway.run 234 -> 230
Functional verification:
- get_tool_definitions(quiet_mode=True) under temp HERMES_HOME: identical
sorted 30-tool name set before vs after (empty diff).
- Discovery cache: delete cache -> 158ms, second run -> 3ms; corrupt cache
-> clean full rescan; touching one file -> single-file rescan (12ms).
- Tests green: tests/tools/test_registry.py, all test_mcp_oauth*,
test_mcp_dashboard_oauth, test_mcp_tool_401_handling, all
tests/tools/test_browser*.py, tests/hermes_cli/test_mcp_{config,startup,
dashboard_oauth}.py, test_skills_tool_discovery_cache.py.
Fixes#67278. Since config v12 (see #8776), custom_providers: list is
legacy — the migration converts it to the providers: dict and the
resolver reads the dict first (runtime_provider.py). Docs still taught
the legacy list as primary.
- integrations/providers.md: all 8 YAML examples converted to
providers: dict (field mapping code-verified: api, default_model,
transport), one consolidated legacy-format note
- configuring-models.md, configuration.md, credential-pools.md,
migrate-from-openclaw.md, provider-runtime.md, faq.md: examples and
prose flipped to dict-first; legacy list noted as still-read
- adding-providers.md untouched (sole mention is a literal test
filename)
Review follow-up on the salvaged #47772 work. Three defects made the
filter a no-op or actively harmful in production, plus both Copilot
review items.
1. The blank-line burst filter never fired. pty_bridge.py spawns via
ptyprocess.PtyProcess.spawn() and never calls setraw(), so the PTY
line discipline runs with ONLCR: every LF the child writes reaches
xterm as CRLF. The /\n{50,}/ pattern requires consecutive LF, so a
real 1000-row burst matched nothing (verified against a live PTY:
b"A"+b"\n"*5+b"B" is read back as b"A\r\n\r\n\r\n\r\n\r\nB").
Now matches /(?:\r?\n){50,}/.
2. Bursts split across WebSocket frames were not collapsed. bridge.read()
does os.read(fd, 65536) per drain tick and each read is forwarded as
its own frame, so a burst spans frames and each fragment fell under
the 50 threshold (3000 rows survived in a 40-byte-read simulation).
The sanitizer now holds back a trailing newline run — including a lone
trailing CR, since a frame can split a CRLF pair — and resolves it on
the next frame or on flush.
3. Erase-code stripping was permanent, not resume-scoped. resumeParam is
the durable session identity and is never cleared after connect, so
every spinner/progress/status redraw in a resumed session lost its
ESC[K and left stale glyphs. Suppression is now bounded to
PTY_RESUME_SANITIZE_WINDOW_MS (30s) after connect; burst collapsing
still applies for the life of the socket.
Copilot review items:
- flush() no longer writes a buffered partial CSI into xterm. #pending
only ever holds an incomplete sequence, and emitting one leaves the
parser in an in-escape state that swallows output after reconnect. A
buffered newline run is still emitted (collapsed).
- Test expectations updated accordingly.
Tests: 29 cases (was 18), now using CRLF fixtures that match real PTY
output, plus cross-frame burst reassembly, CRLF-pair frame splits, and
post-window erase preservation. Full web suite 135 passing.
Addresses Copilot review:
- Reuse a single TextDecoder instance instead of allocating per message
- Only filter erase codes during session resume (resumeParam != null)
- Extract filter chain into named helper sanitizeResumeOutput()
Refs #47313
Ink two-pass virtual scrolling during session resume floods the
PTY output with \x1b[K (erase-line), \x1b[NX (erase-char), and
thousand-line \n bursts. In the Dashboard INLINE mode, xterm
main scrollback buffer absorbs these as blank rows.
Filter all three in ws.onmessage before they reach xterm.
Refs #47313.
The upstream cua-driver installer scripts on trycua/cua@main carry a
baked default version that Release Please bumps in the release PR
*before* the release assets are published. During that window an
unpinned installer run 404s on the asset download and the
`hermes update` cua-driver refresh fails with:
error: download failed: The remote server returned an error: (404) Not Found.
⚠ cua-driver refreshing did not complete. Re-run manually: ...
Observed live 2026-07-29: baked version 0.14.0 vs latest published
release 0.13.1 — every `hermes update` run with an out-of-date driver
hit the warning until upstream publishes the assets.
We already know the correct version: `cua-driver check-update --json`
returns `latest_version` straight from the GitHub Releases API, whose
entries by definition have published assets. When the check positively
confirms an update, export that version as CUA_DRIVER_RS_VERSION into
the installer child env (both install.sh and install.ps1 honour it over
their baked default), so the refresh downloads the release that
actually exists instead of racing the upstream release pipeline.
Malformed / missing latest_version values fall back to the previous
unpinned behaviour. The explicit `hermes computer-use install
--upgrade` force path and fresh installs are unchanged.
The import-agent subcommand (24c3c27ba8) shipped with its parser builder
and handler logic but build_import_agent_parser() was never called from
main.py, making the documented and unit-tested command uninvocable.
Register it alongside the other subcommand builders.
Streamed responses no longer insert real newlines at terminal width —
logical lines are emitted whole and the terminal soft-wraps them, so
highlight-copy rejoins the full line (emulators only keep linebreaks
the app actually printed). This is the CLI equivalent of the TUI's
selection copy, which reads logical source lines from its screen
buffer. TTFT perception is preserved by mirroring the unfinished
line's tail into the spinner status text instead of chunk-printing.
/copy now prefers OSC 52 when running over SSH (SSH_CONNECTION /
SSH_TTY / SSH_CLIENT) — native tools there write the REMOTE clipboard,
which is never what the user wants. The CLI's OSC 52 writer also gains
tmux/screen DCS passthrough wrapping, mirroring the TUI's
wrapForMultiplexer. Fixes#31528 for the CLI surface.
Sabotage-verified: restoring the old chunk emitter fails 3 of the new
tests (hard-wrap detection, spinner mirror, unbreakable-run split).
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.
The continuous-voice no-speech counter (3 strikes -> voice off) counted
every silent capture cycle unconditionally. During a long agent turn
(thinking/tool-calling for minutes) or while TTS is speaking, the user
is CORRECTLY silent — those cycles ended the voice chat under them.
- hermes_cli/voice.py: new set_voice_busy_probe() seam + _voice_activity_held()
(TTS-playing via the existing _tts_playing Event, agent-busy via the
registered probe). Both the continuous-loop strike path and the
force-transcribe single-shot strike path skip counting while held.
Fail-open: a broken probe counts cycles as before.
- tui_gateway/server.py: registers _any_session_running() as the probe
on voice.record start (voice is process-global; any running session holds).
- cli.py: classic CLI strike path skips counting while _agent_running
or TTS playback is in flight.
Stop phrase and barge-in still work during the hold (own paths).
Includes a fixture fix for the #71083 cherry-pick: the fake tools.tts_tool
module needs _load_tts_config (main's tts_streaming imports it).
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)
Fixes#73973.
A finite one-shot whose dispatch was claimed (claim_dispatch increments
repeat.completed BEFORE execution) but whose run died before mark_job_run
was left permanently wedged: completed==times, last_run_at null, state
'scheduled'. The run-claim TTL blocked re-dispatch, and once it expired
the dispatch-limit guard silently removed the job with no output and no
error.
Two complementary fixes:
- cron/scheduler.py run_one_job: the outer handler now catches
BaseException, not just Exception. The inner run_job handler re-raises
CancelledError/KeyboardInterrupt/SystemExit after agent teardown, and
none of those are Exception subclasses, so the outer 'except Exception'
missed them and mark_job_run(False) was never called. Failures are now
recorded first (mark_job_run + finish_execution, each independently
guarded), then non-Exception BaseExceptions are re-raised to preserve
teardown semantics. Plain Exceptions keep the existing behavior
(recorded, return False, no re-raise). Empty str(e) (bare
CancelledError) falls back to the exception class name.
- cron/jobs.py: when either removal site (claim_dispatch or the
get_due_jobs dispatch-limit guard) drops a one-shot whose claimed run
never completed (last_run_at null), _write_wedged_oneshot_diagnostic
now writes an operator-visible .md into cron/output/<job_id>/ instead
of vanishing silently. Best-effort: diagnostics can never break the
removal. No diagnostic when last_run_at is set (normal completion
race).
Issue #65773: run_one_job installs a <home>/.env secret scope around every
job; before c758ded6d (#69057) an installed scope was authoritative even
with multiplexing off, so provider keys injected only via the process
environment (container env vars, systemd Environment=) resolved to empty
inside cron and every provider call went out with the no-key-required
placeholder -> HTTP 401, while interactive turns kept working.
The fix landed in agent/secret_scope.py (scope-miss fallthrough to
os.environ when multiplex is off) with unit tests at that layer only.
These two tests pin the end-to-end contract where the bug actually
surfaced - cron's run_one_job:
- env-injected key resolves during run_job with multiplex OFF (fails on
pre-fix code, verified by mutation against c758ded6d~1)
- .env value still wins when both sources define the key (precedence)
Implementation-agnostic: passes whether the fix is the get_secret
fallthrough (main today) or a multiplex guard at the installation site
(the approach in #65801/#65802/#73037).
The gateway's /reset cleanup path called shutdown_memory_provider without
first draining the memory manager's serialized background write worker.
shutdown_all only gives that worker a bounded (~5s) drain and abandons
whatever is still queued past it, so a /reset could silently drop writes
the session had already handed off -- the next session then loaded
stale MEMORY.md.
The CLI exit path already drains via MemoryManager.flush_pending before
shutdown; this PR pins the same contract on the gateway cleanup path.
Cleanup now calls agent._memory_manager.flush_pending(timeout=10) before
the existing shutdown_memory_provider step. The flush is best-effort:
a flush failure must never block teardown, so it is wrapped in
try/except and the existing shutdown path remains the fallback.
Closes#73297
- extract _commit_registry/_note_refresh_failure shared by the background
worker and foreground stage-4 (identical 4-step success + failure paths
were duplicated); worker now commits under _models_dev_fetch_lock so a
failing background refresh can never re-arm the backoff immediately
after a successful force_refresh committed (unsynchronized-write race)
- add should_clear_context_pin_async to hermes_cli/route_identity.py
(matching the get_model_context_length_async precedent) and use it at
the 4 async gateway sites instead of inline asyncio.to_thread wraps;
the sync _format_session_info site keeps the sync call (already
off-loop via its callers' to_thread)
- test the background-refresh success path (the PR's primary new
behavior): disk saved, mem cache swapped, backoff cleared, in_flight
reset — mutation-checked
- replace the race-prone spin-wait on _models_dev_refresh_in_flight with
a named-thread join in the backoff test
The branched call shape in get_provider_info/get_provider is deliberate:
~69 test sites across tests/hermes_cli and tests/gateway monkeypatch
fetch_models_dev (and get_provider_info) with zero/single-arg lambdas.
Passing allow_network= unconditionally broke 5 tests in CI slices 2/3/7.
Documented the constraint inline.
- _mark_stale_cache_grace only moves cache_time forward so a completed
background refresh is never rewound to a 5-minute grace window
- clear _models_dev_refresh_in_flight if Thread.start() raises so a
one-off thread-exhaustion failure doesn't disable refresh forever
- move empty-registry validation into _fetch_models_dev_from_network
(was duplicated in the background worker and the foreground fetch)
- pass allow_network through as a plain kwarg in get_provider_info and
hermes_cli.providers.get_provider instead of the branched call shape
- refresh the stale module docstring (no bundled snapshot exists; the
resolution order now describes stale-serve + background refresh)
