Slash commands run on the event-loop thread, outside the per-turn
contextvars.copy_context() that pins the session cwd for the agent call.
/compress reaches agent._build_system_prompt(), whose "Current working
directory" line comes from resolve_agent_cwd() — so an unpinned handler
rebuilt the prompt against the Hermes install tree and PERSISTED it as
the session's cached prompt, re-poisoning every later turn even though
the turn itself is now pinned.
Pin inside a fresh context copy so the write cannot leak into other
concurrent ACP sessions on the shared loop and needs no teardown.
Some OpenAI-compatible endpoints — notably Tencent Copilot
(copilot.tencent.com) — only accept streaming chat requests; any
non-streaming call returns HTTP 400 (code 11101, 'Non-stream chat
request is currently not supported'). The main conversation loop already
streams, so interactive chat works, but every auxiliary task (title
generation, compression, web extraction) used the non-streaming path and
failed on each call.
_provider_requires_stream() detects stream-only endpoints
(copilot.tencent.com built in, plus user-configurable
auxiliary.stream_only_base_urls substring markers in config.yaml).
Matching sync auxiliary calls route through _create_with_progress
(force_stream=True) and async calls through the new
_acreate_with_stream, aggregating the chunk stream — including tool-call
deltas and reasoning deltas — into a complete response via the shared
_ChatStreamAccumulator.
Salvaged from PR #60686 by @kudi88 onto the progress-aware streaming
machinery from #71508, addressing both sweeper-review gaps: the async
path now consumes the stream with 'async for' (awaiting create() and
iterating synchronously raised on AsyncOpenAI streams), and tool-call
deltas are reassembled instead of dropped (MCP passes tools= through
call_llm). Under force_stream there is no silent non-streaming retry —
a stream-only provider rejects those by definition, so the original
error surfaces to the normal recovery chains.
The adapters wrapped every field in an arrow function, including the
optional ones. That makes an absent handler unconditionally truthy, and
several children gate on a handler's PRESENCE rather than just calling
it:
- onDismissError -> assistant-message.tsx renders the dismiss button
only when defined
- onRestoreToMessage -> thread/index.tsx gates the restore-confirm flow
- onTranscribeAudio -> use-voice-recorder / use-voice-conversation gate
recording on it
- onLoadMoreMessaging / onLoadMoreProfileSessions -> sidebar paging
So the adapter would paint a dead dismiss button and let voice recording
proceed into a no-op transcription path even when the controller had
deliberately left those handlers off.
Wrap an optional field only when it is currently present, and re-read the
latest value inside the wrapper so the stale-closure fix still applies.
Presence is stable for a given actions object (the controller mutates
fields in place rather than toggling a handler between defined and
undefined), while the closure is what churns — which is exactly what the
indirection re-reads.
Adds two regression tests: absent optional handlers stay undefined, and a
present optional handler still late-binds to the latest closure. The
first was verified to fail against the unconditional-wrapper form.
A background queue drain (fromQueue: true) whose runtime binding was
reaped by the gateway fires with sessionId=null. The expression
options?.sessionId ?? activeSessionIdRef.current
falls back to whichever runtime id the foreground happens to hold,
landing the queued prompt in the session the user is currently viewing
instead of the session that owns the queue entry — a cross-session
message leak.
Guard the fallback: only inherit the foreground runtime when the drain
targets the current view (no storedSessionId, or it matches the
foreground). A background drain (storedSessionId differs) is left with
sessionId=null so the existing session.resume path rebinds the correct
runtime before prompt.submit fires.
Includes a regression test: "a fromQueue drain with null runtime id
does NOT land in the foreground session (cross-session leak guard)".
All 53 existing tests pass.
Salvaged from PR #62805 (@floatingrain), whose diagnosis of the deterministic
frontend self-abort (createBackendSessionForSend mutating the selected ref +
route, then the caller's drift guard reading its own re-home as a user switch)
was correct — and correct about the sweeper misread that closed it: the
sweeper's implemented_on_main verdict cited submit.ts:245, which was inside
the session.resume block, while the raw post-create guard at the
createBackendSessionForSend site was still aborting every new chat on the
then-current main (4281151ae8). The mechanism itself has since landed via
8c288760d0 + 1bdd478efa, so what remains distinct is this regression: after
the pipeline adopts the created chat as its pinned target, a GENUINE user
switch (selection and route both moving to another chat during the
attachment-sync await) must still abort instead of being masked by the
re-baseline.
Part of the #63078 fix branch.
Salvaged from PR #62990 (@Roger--Han), a competing leg-1 fix for #63078. Its
mechanism (a pinned-route-token set accepting both the pre-commit and the
deterministic created-session token) was superseded by main's target-aware
drift predicate (1bdd478efa), but the PR pinned a timing case nothing on main
covered: React Router exposing the STALE new-chat route to the submit
continuation after create returns, committing the created session's URL only
before the next await settles. Both snapshots are the pipeline's own
transition — the first prompt must still reach prompt.submit.
Includes the contributors/emails mapping (added directly:
scripts/add_contributor.py's login regex rejects the consecutive hyphen in
the real GitHub login Roger--Han).
Part of the #63078 fix branch.
Follow-up hardening for the salvaged #69240 readiness gate (#67498):
- _start_polling_once now returns its (generation, progress_event) pair
so the strict cold-start gate binds to exactly the generation it
started, instead of re-reading self._polling_progress_event which a
concurrent recovery task may have replaced with a newer generation's
event (the G1/G2 race flagged in the #69240 review).
- Strict cold start no longer schedules background polling recovery: a
polling error during the readiness wait is captured by a strict
callback and fails the connect attempt immediately with a loud
OSError, so GatewayRunner disposes the partial adapter and retries
with a fresh one — no more waiting out the full readiness deadline on
a generation that already errored, and no G2-on-partial-app healing.
- After readiness is proven the strict callback delegates every later
polling error to the real background-recovery callback, preserving
the existing degraded/reconnect semantics for the polling lifetime.
- The readiness-timeout error message now states the deadline and that
the gateway will retry with a fresh adapter (loud failure, not a
silent wait).
- Regression tests: current-generation progress connects; a polling
error during strict cold start fails fast without scheduling
background recovery (the #67498 idle-threads shape); stale-generation
progress is rejected.
Progresses #67498
With the eager pre-warm removed (PR #66783), slash.exec is the only spawn
path — and it runs on the RPC thread pool, so two concurrent worker-routed
commands on a fresh session could both see slash_worker=None and each fork
a full stdio-MCP-fleet worker (the _attach_worker race loser leaking
unclosed). Add a per-session spawn lock with a double-check, plus a
regression test racing two slash.exec calls through handle_request.
Also maps Ne0teric's contributor email.
Three parity fixes between the API server and the native gateway's
agent-runtime resolution, integrated with the provider-aware request
routing that landed in #70853:
- Session-persisted model is honored: POST /api/sessions {"model": ...}
stores a model that the chat handlers previously fetched and threw
away. A stored value that matches a model_routes alias goes through
the route path (route provider/credentials apply); a raw model string
threads through as session_model, pinning the session's turns ahead
of per-request body values but below an explicit session /model
override.
- Empty-model recovery: provider-catalog default when config has no
model.default but a provider resolved, plus last-known-good model
recovery (#35314) keyed on gateway_session_key only (never ephemeral
session_id — no unbounded growth from one-off requests).
- Provider auth failures surface as controlled responses: RuntimeError
from _resolve_runtime_agent_kwargs() is re-raised as a dedicated
_ProviderAuthResolutionError at the call site, caught narrowly in
_run_agent() and the /v1/runs executor to return run.py's response
shape instead of an undifferentiated 500 (session-chat endpoints
previously returned a raw aiohttp 500 with no JSON body).
Salvaged from PR #57947 by @FvanW; session-model route-alias resolution
from PR #59941 by @kaishi00.
Co-authored-by: kaishi00 <kaishi00@users.noreply.github.com>
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.
Carry model, provider, and model_options through the API server's
execution surfaces (session chat, Chat Completions, Responses, /v1/runs)
without mutating global configuration. Precedence: session /model
override -> model_routes alias -> direct request selection -> global
defaults. Conflicting route/provider mixes fail closed with 400.
model_options stays request-scoped regardless of which selection wins.
Salvaged from PR #54426 by @abundantbeing.
Salvaged from PR #38820 by @bedirhancode, re-applied at current skill
locations (obliteratus and s6 moved to optional-skills/ since the PR):
- research-paper-writing: drop ml-paper-writing (never existed)
- touchdesigner-mcp: drop native-mcp (consolidated) + hermes-video (never existed)
- obliteratus: vllm -> serving-llms-vllm, gguf -> llama-cpp
- s6-container-supervision: drop hermes-agent-dev (not a repo skill)
4 of the original 8 hunks were dropped: heartmula already fixed in
#70453; native-mcp SKILL.md deleted from main; architecture-diagram and
comfyui hunks removed refs to concept-diagrams and
stable-diffusion-image-generation, which are valid optional skills.
Verification follow-up for the #51226 salvage: the host call site guarded
select_context with hasattr(), but the ABC defines a default on every
engine, so the built-in ContextCompressor (and any non-implementing
engine) still paid per-request shallow copies of the conversation
history plus a hook call on every provider request. Identity-check the
bound method against ContextEngine.select_context and return the
request untouched — mirroring the existing base-method short-circuit in
_notify_context_engine_turn_complete — so the default path does zero
work, not just produces an identical result.
Adds two pins: the base no-op is never invoked (patched-to-raise base
stays silent), and ContextCompressor.__dict__ contains neither new verb.
Also registers the contributor email mapping for @chaos-xxl.
Follow-up for salvaged PR #59753 rebased over the per-slot
reasoning_effort feature: _clean_slot now round-trips reasoning_effort
AND enabled together; add a normalize→normalize regression test, update
the validate/normalize agreement contract for the canonical enabled
default, restore the desktop per-slot toggle test on the current
autosave editor, and map oppenheimor's contributor email.
Salvaged from PR #47971 (LSP subset). On Windows, .cmd-wrapped language
servers (e.g. pyright-langserver.CMD launched via cmd.exe /c) and the
npm/go/pip LSP auto-installers spawn without CREATE_NO_WINDOW, so a
console window flashes whenever the spawn happens under a console-less
parent — e.g. a VS Code/Zed extension host running the ACP adapter.
- agent/lsp/client.py::_spawn: pass creationflags=windows_hide_flags()
to the language-server asyncio subprocess (inert 0 on POSIX;
start_new_session is kept — it is POSIX-only and ignored on Windows).
- agent/lsp/install.py: same flags on the npm and go installer
subprocess.run calls. The pip path goes through
hermes_cli.tools_config._pip_install, which already hides its windows.
Adapted from the PR's hand-rolled _NO_WINDOW constant to the repo's
hermes_cli._subprocess_compat.windows_hide_flags() convention.
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>
Adds test_moa_gemini_aggregator_sanitize_uses_real_model: drives a full MoA
tool-call turn (virtual-provider mode) with a Gemini aggregator and asserts
the strict-API sanitize pass is invoked with the resolved aggregator model
(gemini-3-pro-preview), never the virtual preset name once a slot is
resolved — the exact path that stripped extra_content/thought_signature and
made Gemini aggregators 400 (#65092).
Writing the test surfaced a gap in the salvaged #66212 fix: in virtual-
provider MoA mode (provider=moa, no moa_config threaded through
run_conversation) the conversation-loop branch never fired because it only
consulted moa_config. Extend it to fall back to the facade's
last_aggregator_slot — the same source the handle_max_iterations fix uses —
so both MoA entry modes resolve the real aggregator model.
Also adds the contributors/emails mapping for the #15676 credit base.
Follow-up fixes on top of the salvaged #22566 mechanism:
- N-collector now counts only REAL actionable user turns via
_is_actionable_user_turn + _is_synthetic_compression_user_turn —
the same filter pair _find_last_user_message_idx uses post-#69291.
The contributor's bare role=='user' + _is_context_summary_content
check let blank platform echoes and continuation/todo rows consume
N slots, silently degrading the guarantee.
- Default flipped 3 -> 1 (behavior-preserving): a default of 3 was
measured to change the tail cut on transcripts whose budget covers
only the last turn. min_tail_user_messages=1 delegates to the
existing single-user anchor; N>1 is opt-in, and the call site is
gated so the default path is byte-identical to main.
- Hardened config parse in agent_init (bool rejected, fractional
floats rejected, floor 1) matching the max_attempts parser shape.
- Wired the recurring external-PR config gaps: hermes_cli/config.py
DEFAULT_CONFIG + cli-config.yaml.example (PR only had cli.py).
- Regression tests: blank echoes / synthetic rows don't count toward
N; tool-call/result pairs never split by the N-boundary (no-orphan
both directions); N-guarantee wins over tail_token_budget and the
_MAX_TAIL_MESSAGE_FLOOR (floor is a minimum, not a cap); default
parity pin; DEFAULT_CONFIG pin.
21 tests pinning the salvaged #44166 behavior:
- marker emit + extractor round trip (patterns adapted from PR #32375
by @LeonSGP43, with credit)
- no-duplicate re-injection when the canonical marker survived (the
original PR's presence-check defect)
- Phase-1 protection for just-loaded / user-referenced skills, and the
Pass-4 pressure override that keeps #61932 fixed
- deterministic marker survival through a REAL compress() with a mocked
aux LLM: drop → re-injected, keep → not duplicated, static-fallback
path, iterative re-compression via rehydrated handoff
- markers never classify as handoff content (classify_summary_content /
_strip_context_summary_handoff_message untouched)
- SKILLS_GUIDANCE Skill Safety Rule renders with real newlines
Follow-ups on top of the cherry-picked #27748 mechanism:
- move the cap constant to module level with full rationale comment
(class attribute aliases it so subclasses/tests can override)
- bound the iterative-update path too: the PREVIOUS SUMMARY block is
passed through _bound_summary_input so a pathological rehydrated
handoff cannot blow up the prompt (previous summary + new turns each
capped)
- extra regression tests: byte-identical small-input passthrough
(identity), direct bound+marker unit check, bound-after-per-message-
truncation shape (hundreds of under-_CONTENT_MAX turns), iterative
path bounded, marker vs classify_summary_content non-collision
- contributor email mapping for @robgfl45
MoA reference_max_tokens is preset-level — one cap for all reference
models. When mixing a verbose model with a terse one, a single cap is
either too tight for the terse model or too loose for the verbose one.
Now each reference slot can optionally carry its own max_tokens:
reference_models:
- provider: openrouter
model: deepseek/deepseek-v4-pro
max_tokens: *** # per-slot cap, overrides preset-level
- provider: openai-codex
model: gpt-5.5
# no max_tokens → falls back to preset-level reference_max_tokens
_clean_slot (moa_config.py) preserves an optional max_tokens field on
the slot dict, coerced via _coerce_int_or_none. _run_reference
(moa_loop.py) reads slot-level max_tokens first, falling back to the
preset-level cap passed by the caller. Slots without the field are
unaffected — backward compatible.
Type hints on slot-handling functions updated from dict[str, str] to
dict[str, Any] to reflect the now-heterogeneous slot shape.
_build_call_kwargs omitted max_tokens for every provider except
anthropic-compat endpoints and NVIDIA NIM. Gemini's native
generateContent maps max_tokens -> maxOutputTokens and, when it is
omitted, applies a fixed 65,535-token ceiling (not "the model's full
budget"), so dropping the value made MoA's reference_max_tokens a
silent no-op for gemini advisors — they ran effectively uncapped
(observed ~2900 output tokens against a configured cap of 600),
inflating per-turn MoA latency.
Forward max_tokens for the gemini-native path (provider name or native
base_url). Gemini supports maxOutputTokens, so the cap is safe here;
providers that reject max_tokens (Copilot, GPT-5 max_completion_tokens,
ZAI vision) are unaffected — they still omit it as before.
PR #56756 added reference_max_tokens to cap MoA advisor output and cut
turn latency. The value is correctly threaded through five layers of MoA
code (moa_config → conversation_loop → aggregate_moa_context →
_run_references_parallel → _run_reference → call_llm(task='moa_reference',
max_tokens=800, ...)).
However, _build_call_kwargs() in auxiliary_client.py silently drops
max_tokens for all OpenAI-compatible providers (PR #34845, which fixed
endpoints and NVIDIA NIM keep it. This means reference_max_tokens never
reached the API for the vast majority of providers.
The bug affects every OpenAI-compatible MoA reference/aggregator slot:
Z.AI (coding plan), OpenRouter, OpenAI, GitHub Copilot, and local
providers. Only Anthropic-compat endpoints (MiniMax, /anthropic URLs)
worked — by coincidence, not MoA-aware design.
Fix: thread the 'task' parameter through all six _build_call_kwargs()
call sites. When task starts with 'moa_', max_tokens is always included
in the request kwargs regardless of provider. Non-MoA auxiliary tasks
(compression, titles, vision, etc.) keep PR #34845 behavior unchanged.
Verified end-to-end:
- Z.AI GLM-5.2 with max_tokens=50 → returned exactly 50 tokens
- Z.AI GLM-5.2 with max_tokens=20 → returned exactly 20 tokens
- Z.AI GLM-5.2 uncapped → returned 315 tokens
- 7 new regression tests covering 4 providers, Anthropic wire, non-MoA
tasks, and prefix-matching boundary
- 288 auxiliary_client tests pass (was 281, +7 new), 84 MoA tests pass
- Zero regressions
Note: #51899's commit was authored as a placeholder identity
(hermes-agent@example.invalid); it was re-stamped to @byshubham's
numeric noreply on this branch, so no mapping file is needed. Other
salvaged commits use GitHub noreply addresses.
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).
When _api_key_passes_startup_guard() rejects the key (missing,
placeholder/too short, or fail-closed unverifiable strength), connect()
returned a bare False with no fatal-error info. gateway.run's reconnect
watcher treats that as transient and re-queues with backoff forever —
each retry re-instantiating the adapter and its ResponseStore sqlite
connection. Observed in production (#37011): ~501 leaked connections
(1002 fds) over ~2.5 days until EMFILE made the whole gateway
unresponsive.
Set a non-retryable fatal error (api_server_key_invalid) in connect()
when the guard rejects, covering all three rejection branches, so the
platform drops from the reconnect queue; recover with
`/platform resume api_server` after fixing the key. Same treatment as
the port-conflict guard (api_server_port_in_use, #65665 / bda8bd76a8).
Tests mirror the port-conflict precedent: each rejection path asserts
connect() is False, has_fatal_error True, fatal_error_retryable False,
and fatal_error_code api_server_key_invalid, plus a strong-key control.
Re-implementation of #38803 by @cifangyiquan against current main —
their patch targeted the old inline guard in connect() which was since
extracted to _api_key_passes_startup_guard() (and gained the
fail-closed branch in 683059feb5), so the original diff no longer
applies. Their production diagnosis and non-retryable direction
preserved.
Refs: #38803, #37011
Salvaged from PR #36180 (commits 68dfeb4b16 and 86f437509b by arimu1),
re-applied onto current main with the incidental black-reformat churn
stripped out (~1,700 lines -> the semantic change + tests).
Previously gateway/config.py enrolled the api_server platform on
`api_server_enabled or api_server_key`, so API_SERVER_ENABLED=true with
no key (or a weak/placeholder key) still loaded the platform: the
adapter is instantiated (ResponseStore/SQLite opened in __init__), the
reconnect watcher spins, and the startup guard refuses at connect() —
logging errors forever. Now the platform is enrolled only when
API_SERVER_KEY passes the same strength bar as the adapter's startup
guard (has_usable_secret, min_length=16), via a shared
_has_usable_api_server_key() helper.
The no-op `lambda cfg: True` connected-checker for API_SERVER is also
replaced with the same key check, so get_connected_platforms() only
reports the platform "up" when it could actually start.
Known limitation (intentionally out of scope): a YAML config with
`platforms.api_server.enabled: true` and no key still loads the
platform; this gate covers the env-override path only.
Dropped from the original PR: EMAIL/SMS checker additions (scope creep
beyond the PR title; absent on current main) and the wholesale black
reformat of gateway/config.py and tests.
Fixes#36111
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
