asyncio.ensure_future(result) creates a task with only a weak ref in
the event loop's task table. After the carrier raises CancelledError,
the local 'task' variable goes out of scope and the loop can GC the
handler before it finishes — the exact 'handler killed mid-flight'
class we are fixing, just via GC instead of cancellation.
Add _detached_fatal_tasks set on BasePlatformAdapter (matching the
gateway-level pattern in _handle_adapter_fatal_error). Uses getattr
fallback for test stubs built via object.__new__().
When an adapter escalates a retryable fatal error from inside one of its
own tasks (e.g. Telegram's _polling_error_task after exhausting polling
network retries), the gateway's _handle_adapter_fatal_error tears the
adapter down via disconnect() — which cancels that very task. The
propagating CancelledError killed the handler between popping the
adapter from the adapter map and queueing the platform in
_failed_platforms, leaving a zombie gateway: process alive, zero
connected platforms, zero pending retries, until a manual restart.
Run the handler as a detached task under asyncio.shield so carrier
cancellation no longer aborts teardown/queueing mid-flight. The carrier
still observes CancelledError (teardown semantics unchanged); only the
handler is protected. A done-callback consumes the detached task's
exception to avoid 'Task exception was never retrieved' noise.
Fixes#81335
e81d18dfb collapsed six per-surface copies of reasoning resolution onto
resolve_reasoning_config() and, in its own words, "fixes the gateway
resolving reasoning against config model.default instead of the session's
effective model". It did not touch gateway/platforms/api_server.py, which
kept that defect.
_create_agent() called GatewayRunner._load_reasoning_config() with no
model on its first line — before the model precedence chain (browser lock
-> session /model -> session row -> route -> per-request -> defaults) has
run. Per-model agent.reasoning_overrides therefore keyed off model.default
on the one surface where every request names its own model: a request for
a model with an override silently got the global effort instead.
Resolve after the chain settles, so the override follows the model the
request actually runs. An explicit per-request reasoning parameter still
takes precedence over config.
The existing test stub for _load_reasoning_config took no arguments (it
mirrored the old call); it now matches the real signature, as the sibling
stub in the same file already did.
Follow-ups on the salvaged commit (#37207 by @charzhou):
- Persistent /root home mount translates too: an agent writing
/root/out.png produced a real host file under
<sandbox>/docker/default/home the gateway could not find.
- /root/.hermes cache mounts translate to the HOST cache (longest-prefix
beats the home mount), so MEDIA:<agent_visible_image> paths deliver.
- /root/.hermes/* OUTSIDE a cache mount never translates through the home
mount: those are the sandbox's credential copies (.env, auth.json) that
sit outside the host-side denylist prefixes — fail closed.
- Run the idempotent terminal-config->env bridge before mount parsing so
in-process gateways (Desktop backend, hermes serve) see the active
backend and docker_volumes (covers #42299's /output case there too).
Translate MEDIA paths under configured Docker volume mounts (and the
default persistent /workspace) to host paths before media delivery
validation, using longest container-prefix match so host:/workspace and
/output export mounts work.
Salvage of PR #17973 by @TKCen (Sebastian Hänisch), re-implemented on
current main to preserve speed/instructions/provider params,
prepare_spoken_text normalization, OPUS_VOICE_PLATFORMS, is_write_denied
path security, microsecond timestamps, and the streaming-TTS gate.
- Split long TTS text into provider-safe chunks instead of truncating
- Pack generated audio against platform upload limits (Discord 10MB,
Telegram 50MB, configurable via tts.delivery_profiles)
- Combine chunks with ffmpeg (OGG/Opus re-encoded, MP3 stream-copied)
- Multi-file delivery when combination fails or would exceed limits
- Remove hard [:4000] truncation from all callers (cli.py, voice.py,
gateway/run.py, gateway/platforms/base.py)
- Gemini TTS raises ValueError instead of silently truncating when
composed prompt exceeds the provider limit
Simplify-code fixes: removed dead all_touched_paths set, added
try/finally for scratch file cleanup on exception, clean error response
on chunk failure instead of leaking stale file_path.
The non-streaming /v1/responses path built function_call and
function_call_output output items with no status field (and no item id),
while the SSE streaming path correctly emits status in_progress ->
completed. Spec-strict OpenAI clients reading the non-streaming output
array could interpret the status-less function_call items as pending
calls the CLIENT must execute — but these tools were already executed
server-side by the Hermes agent and are replayed for structured tool UI
only. Reported by a community user whose GPT-5.6 client concluded 'a
server should not tell an OpenAI client to execute a tool the server
already executed itself'.
- _extract_output_items now stamps status: completed and spec-shaped
item ids (fc_/fco_) on replayed items, matching the streaming path
- test updated to pin status + id shape
- docs example updated + explicit note that output tool calls are
replayed, never pending
Thread.start() failure is RuntimeError; catching BaseException here
swallowed KeyboardInterrupt/SystemExit without re-raise (unlike _worker,
which forwards them into the future).
The Matrix adapter built MessageEvent from inbound room events but dropped
the sender's MXID and display name on the event itself -- only 'source'
carried them. Other adapters (signal/slack/telegram/discord/mattermost/irc)
have the same gap; this PR fixes matrix and adds the supporting
top-level MessageEvent fields so the rest can follow.
Downstream effects for matrix specifically:
- gateway prompt assembly can now read event.user_name (or source)
without having to dig into source per platform
- reply context (reply_to_text / reply_to_author_id /
reply_to_author_name) is parsed from the inline > <@user:server> ...
Matrix fallback format before stripping, instead of discarded
- the gateway's existing [Replying to: "..."] renderer can now show
who the user was replying to (was always anonymous for matrix)
MessageEvent gains two optional top-level fields (user_id, user_name,
both default None) so non-IM producers (cron/webhook/autonomous) remain
unaffected. Source still carries the same values for callers that
already read from there.
Tests cover:
- non-reply message carries sender user_id/user_name on MessageEvent
- different senders (alice, bob) both propagate
- reply message carries reply_to_message_id + reply_to_text +
reply_to_author_id + reply_to_author_name, parsed from the
> <@carol:example.org> original question\n\nactual reply shape
- non-reply message does NOT spuriously set reply_to_* fields
Sibling matrix tests (148 across test_matrix*.py) remain green.
Authored by WintleChoung <cwt@users.noreply.github.com>
Salvaged from PR #80293.
The shutdown drain ACCOUNTS for API-server work but never INTERRUPTS it.
`_drain_active_agents()` folds `_active_api_run_count()` into both its wait
loop and its `timed_out` verdict, while `_interrupt_running_agents()` iterates
`self._running_agents` only -- a dict no API turn ever enters, because the
API server owns its own agent lifecycle. `gateway/run.py` states the gap
against itself: "API-server / desk sessions have the same structural gap
(#63529)."
The user-visible result is that every gateway restart with a live API or
desktop turn burns the full drain timeout and then runs
`_kill_tool_subprocesses("post-interrupt")`, which amputates the turn's tool
subprocesses with no cooperative interrupt and no resume marker.
There are seven API agent-entry points. Six funnel through `_run_agent()`
(both session-chat routes, and `/v1/chat/completions` + `/v1/responses` in
streaming and non-streaming form) and are counted by `_inflight_agent_runs`;
the seventh, `/v1/runs`, runs its own lifecycle and is counted through
`_active_run_tasks`. None of the six has a run_id, so the run_id-keyed
`_active_run_agents` cannot reach them, and only two pass `agent_ref` -- which
lands in a caller-local list, not a registry.
So register once at the single unconditional creation site inside
`_run_agent`, beside the existing `_publish_turn_process_ownership()` call,
and unregister in the same `finally` that already clears it. That one
symmetric pair covers all six callers. The registry is adapter-owned and
keyed by object identity, kept separate from `_active_run_agents` because
that dict is run_id-keyed and scoped to the public `/v1/runs` stop API.
`interrupt_active_runs()` then walks both registries, deduped by identity, so
the interrupt set matches the set the drain waits on. The settle window after
the interrupt now polls API work as well: the interrupt is cooperative, and
without this the window closes the instant `_running_agents` is empty -- which
it always is for API turns -- and the tool kill lands on a turn that was asked
to stop microseconds earlier.
Sibling-site fix for #80216: yuanbao recall redaction also calls
rewrite_transcript() and was subject to the same archived-history
data loss when active_only defaulted to False. Pass active_only=True
at both yuanbao call sites — load_transcript only returns active
rows, so the redacted content is in the active set and the archived
pre-compaction history should survive the rewrite.
Also drops the stale 'callers that mean to purge (e.g. yuanbao
recall redaction) keep the default' note from the rewrite_transcript
docstring — no caller intentionally purges archived rows.
Pinned was capped at half the viewport by its own nested scroller, so past
roughly a dozen pins the rest were reachable only by scrolling inside a
scroller — a pin you have to go hunting for isn't doing its job.
Drop the cap and let the section grow into the sidebar's existing scroll,
and stop virtualizing Pinned: virtualization needs a bounded viewport to
measure against, which is exactly what's being removed. No count badge, no
"show more" — pin as many as you want and they all render.
Also back-fill pins on the API-server list route, which was the one list
path still windowing purely on recency.
PATCH /api/sessions/{id} only accepted title and end_reason, so the
`pinned` flag the desktop sends was rejected as an unsupported field —
and the client swallows that error. Pins lived in one app's localStorage
and never reached state.db, which also meant the server-side auto-archive
sweep was free to hide the chats a pin exists to keep.
Accept pinned and archived as booleans, route them to the SessionDB
setters that already existed, and include both in the serialized session
so clients can reconcile against server truth.
The session event stream (api_server.py:~2236) was the one genuinely
unicode-distinct SSE writer — json.dumps(payload, ensure_ascii=False) +
.encode('utf-8'). Every other writer uses plain json.dumps. Route it
through _sse_frame(..., ensure_ascii=False) so _sse_frame is now the single
source of truth for ALL SSE frame serialization in the module (chat-
completion, responses._write_event, /v1/runs, and the session stream).
Byte-identical for non-ASCII payloads: verified against the historical
inline encoder (raw bytes preserved). The ensure_ascii=False path is now
exercised by test_sse_frame_ensure_ascii_false_reproduces_session_event_stream.
_extend _sse_frame with an explicit ensure_ascii param (default True,
byte-identical to a bare json.dumps) and route the two sibling writers
through it: _write_sse_responses._write_event and the /v1/runs event
stream. This completes the dedup PR #65009 — previously only the five
_write_sse_chat_completion sites used the helper, leaving the other two
writers on inline json.dumps with no shared shape.
No behavior change: every writer's emitted bytes are unchanged (verified
byte-for-byte, including non-ASCII payloads where the default
ensure_ascii=True matches the original inline encoders). The ensure_ascii
option is exposed so a future writer can opt into raw non-ASCII bytes
without fractalizing the format again.
Adds tests/gateway/test_sse_frame.py asserting the byte-contract
invariant between _sse_frame and the historical inline encoders.
_write_sse_chat_completion had five near-identical
f"data: {json.dumps(...)}\n\n".encode() (and one event-tagged variant)
scattered across its role/content/finish/error chunk writes. Pure
extract-method, no behavior change: encoding is byte-identical for every
call site touched.
Left the pre-serialized-string writers elsewhere (_write_event's
json.dumps(..., ensure_ascii=False) path, the /v1/runs SSE writer) alone
— routing them through this helper's plain json.dumps(data) would
silently change their unicode-escaping behavior, which is out of scope
for a pure dedup.
_write_sse_chat_completion and _write_sse_responses bridged their
stream_delta_callback queue into the event loop via
`await loop.run_in_executor(None, lambda: stream_q.get(timeout=0.5))`
in a while-True poll — a thread-pool round trip on every 0.5s tick even
when idle, plus up to 500ms of tail latency between a delta landing in
the queue and it reaching the SSE response.
Add ThreadSafeAsyncQueue (asyncio.Queue + a put_threadsafe() that wraps
call_soon_threadsafe), used by both streaming producer closures
(_on_delta, tool start/complete callbacks — all invoked from the worker
thread running run_conversation via loop.run_in_executor). Consumers
now do a plain `await asyncio.wait_for(stream_q.get(), timeout=0.5)` —
woken immediately when a delta arrives, no executor hop, no poll
interval.
Updated tests/gateway/test_sse_agent_cancel.py's 7 call sites to
construct ThreadSafeAsyncQueue inside the running loop (required, since
it captures asyncio.get_running_loop() at construction) instead of a
bare queue.Queue() at test-method scope.
Follow-up on the salvaged pair: the original guard's `not msg_id` arm let an
id-less internal/synthetic event erase a tracking entry a concurrently-queued
id-bearing message's drain task still needs for recall matching (id-less
events never write entries in _dispatch_inbound_event, so they must never
pop). Tests cover: normal cleanup, id-less non-erasure, overwritten-entry
ownership handoff, TTL eviction + fresh-entry survival.
_build_msg_body_with_mentions() checks the TTL of each _member_cache
entry and returns an empty member list when the entry is stale, but
never removes the entry from the dict. Over time every group_code the
bot has ever queried accumulates a permanent entry, retaining the full
member list (potentially thousands of records per group) until
disconnect().
Fix: delete the stale entry at the point it is detected as expired.
The next call to get_group_member_list_raw() for the same group will
repopulate the cache with fresh data as before.
Symmetric with the existing TTL pattern in MessageDeduplicator, which
evicts on access.
_dispatch_inbound_event() writes session_key → msg_id/raw_text into
_processing_msg_ids and _processing_msg_texts so RecallGuardMiddleware
can find and interrupt the currently-processing message. These entries
were never removed after a message finished processing, causing both
dicts to grow unboundedly — one persistent entry per unique session key
for the lifetime of the bot.
Fix: clear both entries in the _process_message_background() finally
block, after super() returns. The guard compares the stored msg_id
against event.message_id before popping: a concurrent pending message
may have already overwritten the entry in _dispatch_inbound_event while
we were running, in which case the drain task owns it and we must not
clear it. When msg_id is absent (nothing was written at dispatch time)
the pop is a safe no-op.
Note: _msg_content_cache already bounds itself to 200 entries at the
same write site; _processing_msg_ids and _processing_msg_texts had no
such bound.
CI slices failed the offload tests with 0.5s witness timeouts: on a
loaded shared runner the event loop thread can take >0.5s to get
scheduled even when NOT blocked, making the probe report a false
positive. A genuinely blocked loop can never set the progress event at
any timeout (the witness coroutine can't run at all), so 5s only
absorbs scheduler flake without weakening the invariant. Mutation
re-verified: reverting the offload still fails all 4 tests.
The sweep-path test parametrizes over _runner/_adapter, which live on
TestGatewayRedeliverySweep; main later added
TestUnconnectedPlatformKeepsItsBudget at the cherry-pick anchor point and
the test landed in that class, where the helpers don't exist
(AttributeError x2). Placement-only move.
Replace the process-global HERMES_CRON_SESSION env var with a per-session
ContextVar so a cron tick in the gateway process cannot leak into unrelated
live gateway/API/TUI turns. The cron scheduler now sets the ContextVar
inside the job's try/finally scope and resets it on cleanup. Gateway, API
server, ACP adapter, and TUI gateway all pass cron_session='' to explicitly
mark their sessions as non-cron, masking any stale process env.
Co-authored-by: hinablue <hinablue@gmail.com>
Closes#37968
_handle_cron_fire verified the NAS-minted fire JWT by calling the
fire-verifier inline on the event loop. That verifier resolves the NAS
signing key from a JWKS URL — a synchronous HTTP GET on a cache miss (a
cold PyJWKClient, or a rotated kid the cached client doesn't know) — so a
slow or rate-limited portal stalls the whole event loop and starves every
other adapter sharing it. #64641 already documented this exact symptom
(relay 504s on high-job-count instances) and cut the fetch frequency by
caching the client per URL, but the residual cache-miss fetch still ran
inline on the loop.
Dispatch the verifier the same way the platform HTTP event verifier was
hardened: await a coroutine verifier directly, run a sync one via
asyncio.to_thread so its blocking I/O stays off the loop, and fail closed
(reject with 401, never admit the fire) if the verifier raises — this is
the only inbound that can trigger remote job execution. The verifier's
JWK-client cache is already thread-safe (threading.Lock), so moving the
call to a worker thread is safe.
Adds regression tests: a sync verifier runs on a worker thread rather
than the loop thread, a crashing verifier yields 401 with no fire, and a
coroutine verifier is awaited.
The QQ adapter read QQ_APP_ID, QQ_CLIENT_SECRET, the QQ_STT_* backend
config and the QQ_ALLOW_ALL_USERS policy flag through raw os.getenv,
bypassing the active profile secret scope. In multiplex mode a secondary
profile whose secret lives in its own .env (installed as an isolated
scope, not into os.environ) would silently fall back to the
default/primary profile's value — the same cross-profile collision fixed
for the WeChat/weixin adapter in #59662.
Route these reads through a scope-aware resolver that reads the profile
scope when one is installed (secondary profiles and per-turn inbound) and
falls back to os.environ otherwise. The fallback is deliberate: the
primary/active profile is constructed without a scope and owns
os.environ, so a bare get_secret would raise UnscopedSecretError and
break its startup. Mirrors gateway.config._getenv.
Adds regression tests including active-profile-no-scope construction (the
fail-closed case), plus scope-wins-over-environ, two-profile isolation,
single-profile fallback, explicit-config precedence and STT key scoping.
api_server's __init__ API_SERVER_KEY read now matches the scoped
_expected_api_key path. tests/gateway/test_adapter_startup_secret_scope.py
asserts, for every migrated module: helper exists, scoped read wins, scoped
miss returns default (no environ borrow), unscoped-under-multiplex falls back
to environ without raising, and legacy single-profile reads still work.
The adapter's __init__ and send_weixin_direct read WEIXIN_ACCOUNT_ID/
TOKEN/BASE_URL/CDN_BASE_URL via bare get_secret, which raises
UnscopedSecretError when the DEFAULT profile's adapter constructs or
sends unscoped under multiplexing (corrects the direction of #66073 /
#68854, which tried to solve this by borrowing os.environ on every
read — a cross-profile leak).
Add a module-level _wx_secret helper following the established Slack
SLACK_APP_TOKEN pattern (#59739) and WhatsApp's _get_wsecret: a SCOPED
miss returns the default (the scope is authoritative — no environ
borrow), while an UNSCOPED read under multiplex falls back to
os.environ, which is the default profile's own value.
Regression tests cover both directions: scoped construction reads the
scope's value and a scoped miss yields empty (no borrow); unscoped
construction falls back to os.environ instead of raising.
Makes interrupt-protected context compression cancellable by an explicit
user or lifecycle stop, without weakening protection against ordinary
incoming messages, voice interjections, or active-turn redirects.
Separates explicit hard cancellation from ordinary interrupt/redirect
state with a dedicated threading.Event; introduces
AuxiliaryExplicitCancellation as an attempt-local frozen-cause signal;
isolates the synchronous provider callback in a bounded daemon worker
during protected compression; atomically linearizes Codex timeout
cleanup against explicit cancellation; propagates hard cancellation
through child agents and explicit stop surfaces; serializes hard-cancel
admission against compression commit admission with
CompressionCommitFence; aborts before session rotation or late DB commit,
restores in-place transcript mutations and compressor state, and releases
the heartbeat and compression lease.
Based on #74449 by @suparious. Resolved merge conflicts in
agent/context_compressor.py (feasibility check + try/except) and
tui_gateway/methods_session.py.
The API server intentionally lets concurrent runs share a client-provided
session_id (= process task_id), so the SSE-disconnect reap could kill a
process a still-live concurrent run spawned after the disconnecting run's
baseline — the same stale-reaper bug class the gateway path gates via
run_generation.
- Per-task-id run epochs (monotonic counter): each run claims the epoch at
publish; a reaper holding a superseded epoch declines to kill. A missing
entry (the run's own clear pruned it) still reaps, so the leak fix isn't
silently disabled.
- _publish_turn_process_ownership / _clear_turn_process_ownership helpers
replace the copy-pasted marker set/clear blocks, so attribute names and
epoch bookkeeping can't drift between surfaces.
- /v1/runs — the third own-lifecycle surface — now records ownership and
reaps on POST /v1/runs/{id}/stop and on server-side SSE cancellation,
closing the remaining sibling paths of #76115.
Addresses the hermes-sweeper review on #76188:
1. task_id is session-scoped (task_id == session_id), not turn-scoped,
and the reap runs on a detached thread. A replacement turn could
claim the same session and spawn a legitimate process before the
previous turn's reaper thread actually enumerates its targets,
killing that new process by mistake.
Fixed by gating the reap on the existing run_generation mechanism
(_is_session_run_current) instead of inventing a new ownership
token: the timeout path captures its own run_generation at turn
start, the interrupt path captures the generation immediately after
invalidating it. If a newer turn has since claimed the session, the
reap is skipped — that newer turn owns its own baseline, so nothing
is left permanently unreaped.
2. gateway/platforms/api_server.py's SSE handlers for chat-completions
and the /api/sessions responses endpoint run their own agent
lifecycle via _run_agent() and never passed through TurnRunner, so
client-disconnect abandonment there had no baseline and no reap —
contradicting the PR's stated disconnect coverage. Both disconnect
handlers now snapshot/reap through the same
tools.process_registry primitives, via a small
_reap_disconnected_agent_processes() helper shared by both call
sites.
Follow-ups on the #75382 salvage (review findings):
- _wenv/_get_wsecret now catch UnscopedSecretError and fall back to
os.getenv for the DEFAULT profile's adapter, which constructs and sends
outside any _profile_runtime_scope under multiplexing — a bare
get_secret would crash its WhatsApp path (fixing one profile by
breaking another). Same pattern as Slack SLACK_APP_TOKEN (#59739) and
the Matrix recovery key. Scoped misses still return the default — no
cross-profile borrow.
- bridge_env overlay extended to the full WHATSAPP_* set bridge.js
consumes (DEBUG, FORWARD_OWNER_MESSAGES, REPLY_PREFIX,
MAX_MESSAGE_LENGTH, CHUNK_DELAY_MS, SEND_TIMEOUT_MS).
- Removed the always-true conditional on WHATSAPP_MODE injection.
Fix#75349
Root cause:
Under multiplex_profiles, secondary profiles run inside
_profile_runtime_scope which installs a per-profile secret scope via
set_secret_scope. The WhatsApp adapter (and the shared
WhatsAppBehaviorMixin + Cloud API adapter) read WHATSAPP_MODE,
WHATSAPP_DM_POLICY, etc. via raw os.getenv(), bypassing the secret
scope. Since os.environ doesn't contain secondary profile .env values,
the bridge silently falls back to 'self-chat' and rejects all inbound
messages with self_chat_mode_rejects_non_self.
Fix:
- Add _wenv() helper in adapter.py that reads WHATSAPP_* vars through
get_secret() (agent.secret_scope), which honors the active scope.
- Replace all os.getenv('WHATSAPP_*') calls in adapter.py,
whatsapp_common.py, and whatsapp_cloud.py with get_secret()-based
equivalents.
- Inject resolved WHATSAPP_* values into the bridge subprocess
environment so the Node.js bridge (which reads process.env) sees the
profile's own configuration.
Changes:
- plugins/platforms/whatsapp/adapter.py: 37 lines (+ helper, bridge_env
injection, 2 os.getenv→_wenv)
- gateway/platforms/whatsapp_common.py: 13 lines (6 os.getenv→_get_wsecret)
- gateway/platforms/whatsapp_cloud.py: 21 lines (9 os.getenv→_get_wsecret)
- New regression test: 6 test cases covering scope isolation, fallback,
and cross-profile non-leakage.
Extend phone/JID alias matching to whatsapp_cloud and treat a removed
allowlist env key as empty so sole-entry revoke cannot revive a stale
adapter snapshot.
Track which source seeded the DM allowlist so live intake does not let a
stale env carrier override explicit config, while env-seeded adapters still
reread pairing mutations.
ForwardedRecordsParseMiddleware.handle() called the coroutine function
_send_loading_heartbeat() without awaiting it, so the coroutine was built
and dropped. The RUNNING heartbeat never reached the client and Python
raised "coroutine was never awaited".
Forwarded-record parsing is the slow inbound path, which is where the
loading bubble matters most: the user sees nothing while the deep parse
runs. Awaiting is safe, since the helper already swallows every exception
and the call sits inside the middleware's own try block.
kshitij
gregorustar-maker
Drexuxux
Teknium
CharZhou
kinsolee
HenryG
Gille
WintleChoung
briandevans
dsad
kshitij
Brooklyn Nicholson
zabih-sudo
EloquentBrush
szzhoujiarui
ibaldr89
Coffee☕️
Frowtek
Da7-Tech
Shaun Prince
evgyur
joaomarcos
x7peeps
fangliquanflq
MaxFreedomPollard