The salvaged context-window screen (#52392) skips fallback candidates that
are too small, and the rate-limit/403 fixes skip candidates that are at
capacity. A third hard failure remained uncovered: a fallback that builds a
client fine but returns a 400 because it structurally cannot run the model.
The canonical case is a configured openai-codex / ChatGPT-account fallback
asked to compress a glm-5.2 conversation:
400 - {'detail': "The 'glm-5.2' model is not supported when using
Codex with a ChatGPT account."}
This is a request-validation error, so should_fallback was False and the
explicit-provider gate blocked it — the auxiliary task (compression) aborted
every turn, dropping middle turns without a summary and churning the session,
which is exactly what destroys the prompt cache.
Adds _is_model_incompatible_error() (400 + capability phrasing, excluding
not-found and billing 400s which the sibling predicates own) and treats it as
a fallback-worthy capacity error in both sync and async call_llm, so the chain
skips the incapable route and continues to the next viable candidate.
The runtime auxiliary fallback chain (_try_configured_fallback_chain and
_try_main_fallback_chain) returned the first reachable candidate without
checking whether the candidate's context window was large enough for the
task. For task='compression' this meant a reachable but undersized
fallback (e.g. 32K) could be selected and then fail, even when a later
larger-context fallback was available.
This adds two small helpers:
_task_minimum_context_length(task)
Returns MINIMUM_CONTEXT_LENGTH (64K) for compression, None for
other tasks (vision, web_extract, etc.).
_candidate_context_window(provider, model, ...)
Thin wrapper around get_model_context_length that returns None on
probe failure so unknown/custom endpoints pass through unchanged
(preserves the existing fallback surface).
Both fallback loops now skip reachable candidates whose resolved context
is below the task minimum and continue iterating. The success path
(first viable candidate wins) is unchanged. Return shape and ordering
for healthy candidates are preserved.
Six regression tests cover:
L2 configured chain skips too-small candidate
L2 chain continues after skipping, returns last viable
L3 main chain skips too-small candidate
L4 unknown-context candidate passes through
L5 non-compression task is not filtered
L6 minimum constant matches MINIMUM_CONTEXT_LENGTH (64K)
3/6 fail on upstream/main without the production change (verified); all
6 pass with the fix. Full test_auxiliary_client.py suite (231 tests)
and related compression tests (130 tests) remain green.
When an explicit aux provider cannot build a client before any request is
sent (missing raw env key, exhausted/unavailable OAuth or credential-pool
auth, resolver returning (None, None)), call_llm raised a misleading
"no API key was found" error and bypassed the configured fallback_chain
entirely. A provider authenticated through Hermes auth / the credential
pool (e.g. ollama-cloud) whose pool entry is exhausted hit this path, so
compression failed instead of routing to the configured fallback.
Adds _try_configured_fallback_for_unavailable_client() and wires it into
both sync and async call_llm before the raise, and into the startup
compression feasibility check.
Salvaged from #51835 by @herbalizer404.
Rate-limit (429) errors on explicit-provider auxiliary tasks were
silently failing instead of triggering the fallback chain. The
is_capacity_error gate only checked payment and connection errors,
excluding rate limits — so when a configured provider like
openai-codex hit its rate limit, auxiliary tasks (kanban_decomposer,
vision, web_extract, approval, etc.) had zero resilience.
Add _is_rate_limit_error() to is_capacity_error at both call sites
(sync and async paths) so rate limits trigger fallback regardless
of whether the provider was auto-detected or explicitly configured.
Fixes#52228
Ollama Cloud (and similar) return 403 with bodies like "this model requires
a subscription, upgrade for access" or "you have reached your session usage
limit, upgrade for higher limits". These are capacity/billing conditions
semantically identical to credit exhaustion, but _is_payment_error() did not
recognize them (403 missing from the status set; keywords missing), so the
configured fallback_chain was never tried and compression failed outright.
Adds 403 to the status set and the subscription/session-usage keywords.
Salvaged from #49076 by @herbalizer404.
Third-party OpenAI-compatible endpoints (self-hosted gateways, OpenRouter,
Azure proxies) fronting gpt-4o / gpt-4.1 / gpt-5+ / o1-o4 models silently
received max_tokens and 400'd with unsupported_parameter, because the three
kwarg-selection sites only checked base_url_hostname(...) == "api.openai.com"
and fell through to max_tokens on every other host. The constraint is
enforced server-side by the model family, not by the URL, so name-based
detection is required as a fallback.
Changes:
- utils.py: new shared helper model_forces_max_completion_tokens(model) that
prefix-matches gpt-4o, gpt-4.1, gpt-5, o1, o3, o4 families on normalized
(lowercased, vendor-prefix-stripped) names.
- run_agent.py: _max_tokens_param ORs the helper into the URL check.
- agent/auxiliary_client.py:
- auxiliary_max_tokens_param gains an optional keyword-only model arg.
- _build_call_kwargs inline branch applies the same check for both
provider == "custom" and non-custom paths.
Tests:
- tests/test_model_forces_max_completion_tokens.py: 31 new cases covering
positive families, negatives (classic gpt-4, claude, llama, mistral, qwen,
deepseek), vendor prefixes, case-insensitivity, whitespace, None/empty,
and substring-not-prefix guards.
- tests/run_agent/test_run_agent.py::TestMaxTokensParam: 5 new model-based
cases (custom + gpt-5.4, openrouter + gpt-4o-mini, custom + o1-preview,
classic gpt-4-turbo keeps max_tokens, llama3 keeps max_tokens).
- tests/agent/test_auxiliary_client.py::TestAuxiliaryMaxTokensParam: new
class, 7 tests covering the URL x model matrix.
A one-off transient transport failure (streaming-close / incomplete
chunked read / 5xx / 408) on an auxiliary LLM call escalated straight to
provider/model fallback (or, for context compression, dropped the summary
and entered cooldown), even when an immediate retry on the same provider
would have succeeded.
Add a single same-target retry at the top of call_llm() and
async_call_llm() — before the existing except-chain — gated on a new
_is_transient_transport_error() that reuses the canonical
_is_connection_error() detector plus a 5xx/408 status check. A second
failure (or any non-transient error: auth, other 4xx, malformed payload)
falls through to first_err and the existing fallback handling unchanged.
This lives in call_llm so every auxiliary task (compression, memory flush,
title generation, session search, vision) shares one transient-retry
surface, rather than each caller re-implementing it. The context
compressor needs no change — it calls call_llm and inherits the retry; its
existing fallback-to-main path (#18458) now composes naturally (retry the
aux model once, then fall back to main only if the retry also fails).
Co-authored-by: ARegalado1 <alberto.regalado@ymail.com>
The auxiliary Codex adapter maintained its own chat->Responses conversion
loop that forwarded every non-system message's role verbatim into
Responses input[]. When flush_memories()/compression replayed session
history containing assistant tool_calls + role=tool results, those tool
messages leaked into the request and the Responses API rejected them with
HTTP 400: Invalid value: 'tool'.
Route _CodexCompletionsAdapter.create() through the same shared converter
the main agent transport uses (_chat_messages_to_responses_input), so tool
calls become function_call items and tool results become function_call_output
items with a valid call_id. Single conversion path means no future drift.
Also remove the now-dead _convert_content_for_responses() helper — its only
caller was the private conversion loop this change deletes.
Co-authored-by: ProgramCaiCai <techxacm@gmail.com>
The salvaged conversion emitted type:"input_video", which MiniMax M3 rejects
just like the original video_url block. Per MiniMax's Anthropic-compat docs,
the video content block is type:"video" with an image-style source (base64 or
url). Fixes the block type, converts URL-based videos too, and adds 4 video
conversion tests (none shipped with the original PR).
A long-lived process (gateway, watcher) caches the Nous Portal's
recommended-models payload and can pin a model for its whole lifetime.
When that model is later dropped from the Nous -> OpenRouter catalog,
every auxiliary call 404s with 'model does not exist in our
configuration or OpenRouter catalog' until the process restarts.
Now such a 404 force-refreshes the Portal recommendation and retries
once with the current pick (or the gemini-3-flash-preview default).
Scoped to Nous-routed calls only.
- _is_model_not_found_error(): 404/400 'not found / does not exist /
not a valid model' predicate, excludes billing keywords so it never
overlaps _is_payment_error.
- _refresh_nous_recommended_model(): force-refresh fetch, returns a
model distinct from the one that failed, else the known-good default.
- Wired into both call_llm and async_call_llm error chains.
* fix(auxiliary): stop capping output with max_tokens by default
Auxiliary LLM calls (compression, titles, vision, etc.) no longer send
max_tokens on the OpenAI-compatible chat-completions path. Most providers
treat an omitted max_tokens as "use the model max", which is what we want;
an explicit cap only risks truncation or a wire-format 400.
This was surfaced by GitHub Copilot / GPT-5 (#34530): those models reject
max_tokens and require max_completion_tokens, so compression 400'd and fell
back to a static context marker. Omitting the param sidesteps that quirk
(and ZAI vision's error 1210) entirely.
The Anthropic Messages wire (MiniMax + /anthropic endpoints) keeps
max_tokens because it is a mandatory field there.
* test(auxiliary): update temperature-retry assertions for omitted max_tokens
The temperature-retry tests asserted retry_kwargs["max_tokens"] == 500 on an
api.openai.com endpoint. Now that auxiliary calls omit max_tokens on
OpenAI-compatible endpoints (#34530), that key is absent. Assert it's absent
in both first and retry kwargs and use model as the survives-the-retry witness.
Remove unused imports (F401) and duplicate/shadowed import
redefinitions (F811) across the codebase using ruff's safe
autofixes. No behavioral changes -- imports only.
- ~1400 safe autofixes applied across 644 files (net -1072 lines)
- __init__.py re-exports preserved (excluded from F401 removal so
public re-export surfaces stay intact)
- Re-exports that are imported or monkeypatched by tests but look
unused in their defining module are kept with explicit # noqa:
F401 (gateway/run.py load_dotenv; run_agent re-exports from
agent.message_sanitization, agent.context_compressor,
agent.retry_utils, agent.prompt_builder, agent.process_bootstrap,
agent.codex_responses_adapter)
- Unsafe F841 (unused-variable) fixes deliberately skipped -- those
can change behavior when the RHS has side effects
- ruff lints remain disabled in pyproject.toml (only PLW1514 is
selected); this is a one-time cleanup, not a config change
Verification:
- python -m compileall: clean
- pytest --collect-only: all 27161 tests collect (zero import errors)
- core entry points import clean (run_agent, model_tools, cli,
toolsets, hermes_state, batch_runner, gateway)
- static scan: every name any test imports directly from an edited
module still resolves
Closes#33368.
`_CodexCompletionsAdapter.create()` iterates `final.output` from the
Codex Responses stream. The event-driven consumer (introduced in #33042)
always sets `final.output` to a list, so this shape can't come from our
own code path. But:
- Mocked clients in tests can return a typed Response with `output=None`
- Third-party shims / compatibility layers that bypass the consumer can
do the same
- A future code path that wraps a different consumer could regress
The old code `getattr(final, "output", [])` returns `None` (not the
default `[]`) when the attribute EXISTS but is `None`. Iterating
`None` then raises `TypeError: 'NoneType' object is not iterable` —
the exact error logged by title-generation when this fires.
Fix: `getattr(final, "output", None) or []` — single-line defensive
coerce. Cheap; zero risk.
Regression test asserts the auxiliary path handles a final whose
`.output` is `None` (via monkey-patched consumer) without raising and
returns the expected chat.completions-shaped response.
Reporter: @pavegrid-1 (issue #33368).
* refactor(codex): drop SDK responses.stream() helper; consume events directly
The OpenAI Python SDK's high-level `client.responses.stream(...)` helper
does post-hoc typed reconstruction from the terminal
`response.completed.response.output` field. The chatgpt.com Codex
backend has been observed (today, gpt-5.5) to ship `response.output =
null` on terminal frames, which crashes the SDK with `TypeError:
'NoneType' object is not iterable` mid-iteration.
Carlton's #32963 patched the symptom by wrapping the helper in
try/except and recovering from the same per-event accumulator the SDK
was supposed to populate. This PR removes the helper from the call
path entirely: we now use `client.responses.create(stream=True)` (raw
AsyncIterable of SSE events) and assemble the final response object
ourselves from `response.output_item.done` events as they arrive. The
terminal event's `output` field is never read for content. Same
strategy OpenClaw uses for the same backend.
This makes Hermes structurally immune to the bug class, not patched.
The next time OpenAI ships a shape change to chatgpt.com's terminal
frame, our consumer keeps working because it doesn't read that frame
for content — only for usage/status/id.
Changes
- `agent/codex_runtime.py`: new `_consume_codex_event_stream()` shared
consumer; `run_codex_stream()` uses `responses.create(stream=True)`;
`run_codex_create_stream_fallback()` collapses into a thin alias
since the primary path now does what the fallback used to do.
- `agent/auxiliary_client.py`: `_CodexCompletionsAdapter` uses the
same consumer; old null-output recovery helpers deleted as
unreferenced.
- Tests migrated: fixtures that mocked `responses.stream` now mock
`responses.create` returning a raw iterable. New regression test
asserts the auxiliary path returns streamed items even when the
terminal event's `output` is literally `null`.
Validation
- Live: tested against fresh OAuth on `chatgpt.com/backend-api/codex`
with `gpt-5.5` — response built correctly with `response.output=null`
on the terminal frame, all events consumed, usage/reasoning tokens
propagated.
- `tests/run_agent/test_run_agent_codex_responses.py` +
`tests/agent/test_auxiliary_client.py`: 242 passed.
* test+fix(codex): migrate streaming tests, raise on truncated streams
CI surfaced 10 test failures across tests/run_agent/test_streaming.py
and tests/run_agent/test_codex_xai_oauth_recovery.py — both files had
their own `responses.stream(...)` mocks I missed in the first sweep.
agent/codex_runtime.py: _consume_codex_event_stream() now raises
"Codex Responses stream did not emit a terminal response" when the
stream ends without any terminal frame AND no usable content. This
preserves the signal callers used to get from the SDK's high-level
helper, which they distinguished from "completed with empty body"
in error handling.
Tests migrated:
- test_streaming.py: text-delta callback, activity-touch, and
remote-protocol-error tests all switch from mocking responses.stream
to responses.create returning an iterable of events.
- test_codex_xai_oauth_recovery.py: prelude-error tests are recast as
wire-error-event tests (the new path raises _StreamErrorEvent
directly when the wire emits type=error, which is strictly better
than the old two-phase "SDK RuntimeError → retry → fallback"). The
retry-on-transport-error test moves from responses.stream side-effect
to responses.create side-effect.
Verified live against chatgpt.com Codex with gpt-5.5 — AIAgent.chat()
through the full codex_responses path returns correctly, 319/319
targeted tests passing.
Aux callers (title generation, vision, session search, etc.) can reach
resolve_provider_client() without an explicit model when the user
picked their main provider via 'hermes model' and didn't bother
configuring a per-task auxiliary.<task>.model override. The
expectation in that case is universal: 'use my main model for side
tasks too.'
Before, the OAuth providers (xai-oauth, openai-codex) silently
returned (None, None) on an empty model — both lack a catalog default
because their accepted-model lists drift on the backend. That caused
_resolve_auto to drop to its Step-2 fallback chain (OpenRouter /
Nous / etc.), so aux tasks billed against the wrong subscription
without warning.
The fix is at the top of resolve_provider_client() — a single
3-step universal fallback that runs before any provider branch, so
no provider-specific empty-model guards are needed (now or for any
future provider we add):
1. caller-passed model (caller knew what they wanted)
2. provider's catalog default (cheap aux model, if registered)
3. user's main model from config.yaml
Behaviour by provider class:
- OAuth providers (xai-oauth, openai-codex) — no catalog default, so
step 3 applies. Title gen runs on grok-4.3 / gpt-5.4 against the
user's actual subscription instead of leaking to OpenRouter.
- API-key providers (anthropic, gemini, kimi-coding, etc.) — catalog
default wins at step 2, preserving the original 'cheap aux model'
behaviour. Anthropic users still get claude-haiku-4-5 for titles,
not opus.
- Explicit-model callers (auxiliary.<task>.model config, programmatic
callers) — caller wins at step 1, no surprise switching.
Salvaged from @wysie's PR #31845 which fixed the xai-oauth branch
specifically. The universal shape supersedes the per-branch fix
and covers openai-codex (same bug class) plus any future OAuth
providers.
4 new tests in TestResolveProviderClientUniversalModelFallback:
- empty_model_for_oauth_provider_falls_back_to_main_model
- empty_model_for_codex_also_uses_main_model
- empty_model_for_catalog_provider_uses_catalog_default
- explicit_model_takes_precedence_over_fallbacks
365/365 across tests/agent/test_auxiliary_*, tests/run_agent/test_codex_xai_oauth_recovery.py, tests/hermes_cli/test_auth_xai_oauth_provider.py, and tests/hermes_cli/test_plugin_auxiliary_tasks.py.
Co-authored-by: wysie <wysie@users.noreply.github.com>
* ci(tests): install ripgrep from prebuilt tarball instead of apt
apt-get update + install of ripgrep takes ~4 min on the GHA Ubuntu
runners (the apt-get update against archive.ubuntu.com is the slow
part; ripgrep itself is small). Switching to the upstream musl
binary tarball cuts the step to a few seconds.
- Pinned to ripgrep 15.1.0 with sha256 verification (same hash as
published in the releases sha256 sidecar file).
- Drops the `rg` binary into /usr/local/bin so it is on PATH for
every subsequent step without GITHUB_PATH manipulation.
- Applied to both the test and e2e jobs in tests.yml.
* fix(cli): compile syntax check to tempdir, not source __pycache__
`_validate_critical_files_syntax` runs `py_compile.compile()` on each
critical bootstrap file after a successful `git pull`. The default
`py_compile` writes the resulting `.pyc` next to the source under
`__pycache__/`, which causes two real problems:
1. Parallel test workers walking the same source tree (e.g. running
the suite under per-file process isolation) can race against each
other on the `__pycache__` write — manifests as flaky 'directory
not empty' errors during teardown.
2. In production, the post-pull syntax check leaves a `.pyc` behind
that the next interpreter run might pick up — fine when the
interpreter version matches, sketchy if it doesn't.
Fix: write the compiled output to a `tempfile.TemporaryDirectory()`
that's discarded on function exit. We only care about the compile-or-not
signal, not the artifact.
* test(runner): per-file process isolation, drop manual state reset + xdist
Replace fragile manual _reset_module_state test fixtures with robust
per-file subprocess isolation. Each test file runs in a fresh
`python -m pytest <file>` subprocess via ThreadPoolExecutor. No xdist,
no custom pytest plugin, no shared worker state.
Key changes:
* scripts/run_tests_parallel.py — new runner: discovers test files,
runs N in parallel via ThreadPoolExecutor, captures stdout per file,
treats exit code 5 (no tests collected) as pass, kills all children
on exit. Change from cpu_count to cpu_count*2. The runner is
I/O-bound (waiting on subprocess.communicate() from pytest children)
The parent process does almost no CPU work, so 2x oversubscription
keeps more pipes full. When a file fails, immediately show the last
30 lines of pytest output (stack traces + FAILED summary) plus a
ready-to-copy repro command:
python -m pytest tests/agent/test_auxiliary_client.py
* scripts/run_tests.sh — delegates to run_tests_parallel.py
* .github/workflows/tests.yml — test step: python
scripts/run_tests_parallel.py
* pyproject.toml — drop pytest-xdist, pytest-split; simplify addopts
* tests/conftest.py — remove ~200 lines of manual state-reset fixtures
* AGENTS.md — update Testing section for per-file design
* test(runner): speed gateway test antipattern scan up
* fix(test): web search provider plugin test missing xai
* fix(tests): make 14 test files pass under per-file subprocess isolation
Tests that relied on cross-file state pollution from xdist workers
fail when run in isolation (per-file subprocess model). Root causes
and fixes:
Tool registry not populated:
- test_video_generation_tool_surface_matrix: add discover_builtin_tools()
- test_web_providers_brave_free/ddgs/searxng/general: autouse fixtures
registering all 8 bundled web providers, reset after each test
- test_website_policy: same provider registration pattern
- test_web_tools_tavily: same pattern across 3 dispatch test classes
- Also add is_safe_url/check_website_access mocks where SSRF check
blocks example.com (DNS resolution fails in isolated envs)
Stale check_fn cache:
- test_kanban_tools: invalidate_check_fn_cache() + _clear_tool_defs_cache()
in both kanban guidance tests (prior test cached False for kanban_show)
- test_discord_tool: cache invalidation in setup/teardown
- test_homeassistant_tool: invalidate_check_fn_cache() before registry queries
Module-level state pollution:
- test_auxiliary_client: autouse fixture clearing _aux_unhealthy_until cache
- test_skill_commands: set_session_vars() instead of patch.dict(os.environ)
(ContextVar takes precedence over os.environ)
- test_dm_topics: overwrite sys.modules + separate telegram.constants mock
+ force-reimport of gateway.platforms.telegram
- test_terminal_tool_requirements: removed duplicate class declaration,
autouse _clear_caches fixture
* change(tests): run_tests.sh explicitly includes env vars
instead of manually dropping some vars, now we just only include some
* fix(tests): 5 more isolation/NixOS fixes
- test_approval_plugin_hooks: isolate HERMES_HOME so real user's
command_allowlist doesn't short-circuit the approval path
- test_google_chat: skipif when Platform.GOOGLE_CHAT not in enum
(feature not merged on this branch)
- test_write_deny: test systemd prefix against tmp_path instead of
/etc/systemd which resolves to /nix/store on NixOS
- test_pty_bridge: use shutil.which('cat') instead of /bin/cat
(doesn't exist on NixOS)
- profiles.py: rmtree onexc handler chmod's parent dirs too, fixing
profile deletion when copytree preserved read-only modes from
nix store
* fix(tests): clear unhealthy cache in autouse fixture for auxiliary_client
* fix(tests): skip send_message when telegram not installed; handle missing worker_id in browser_supervisor
* fix: py3.11 rmtree onexc compat + belt-and-suspenders unhealthy cache clear for expired codex test
* fix: address PR #29016 review feedback
- Remove tracked .pytest-cache/ artifact and add to .gitignore
- Fix stale 'xdist worker' comment in conftest.py
- Deduplicate web provider registration into tests/tools/conftest.py
shared helper (register_all_web_providers), replacing 8 copy-pasted
blocks across 6 test files
- Update PR description: remove stale recovered-test-files claim,
fix worker count to match code (cpu_count*2)
* fix: eliminate race in stale-cache achievements test
The background scan thread could complete and overwrite _SNAPSHOT_CACHE
before evaluate_all() returned the stale data — only 10 fake sessions
made the scan finish instantly. Added scan_delay param to _FakeSessionDB
and set it to 2s in the stale-cache test so the background thread can't
win the race.
7 new tests:
TestAuxiliaryFallbackLayering (3):
- configured_chain succeeds → main agent fallback NOT consulted
- chain returns nothing → main agent fallback runs and succeeds
- both exhausted → user-visible 'all fallbacks exhausted' warning
fires before the original error is re-raised
TestTryMainAgentModelFallback (4):
- returns (None, None, "") when main provider is 'auto'
- returns (None, None, "") when failed provider == main provider
(no point retrying the same backend)
- resolves the main provider's client when configured correctly
- skips when main provider is marked unhealthy
The two TestAuxiliaryClientPoisonedCacheEviction tests were written
when explicit-provider users got no fallback at all on connection
errors — they asserted ConnectionError propagated after eviction
because the fallback gate blocked the auto chain.
After the #26803 fix in the previous commit, capacity errors
(payment/quota/connection) now DO trigger fallback even on explicit
providers. The tests still verify cache eviction (their actual
contract) but now stub _try_payment_fallback so the fallback
machinery does not attempt a real network call.
Closes#26803
Root causes:
1. _is_payment_error() checked for billing keywords (credits, insufficient
funds, billing, payment required) but missed daily token quota exhaustion
phrases used by Bedrock, Vertex AI, and LiteLLM proxies — e.g.
'Too many tokens per day', 'quota exceeded', 'resource exhausted',
'daily limit'. These are functionally identical to credit exhaustion
(provider cannot serve the request) but don't trigger fallback.
2. The call_llm() fallback chain was gated on resolved_provider == 'auto'.
When a task resolves to a specific provider (e.g. 'custom' for a LiteLLM
proxy, or 'openrouter'), capacity failures (payment/quota/connection)
silently raise instead of trying alternatives. This is overly conservative:
capacity errors mean the provider *cannot* serve the request regardless of
user intent, so alternatives should always be tried.
Fixes:
- Add quota-related keywords to _is_payment_error(): quota_exceeded,
too many tokens per day, daily limit, tokens per day, daily quota,
resource exhausted (Vertex AI gRPC code).
- Allow fallback for capacity errors (payment + connection) even when
resolved_provider is not 'auto'. Rate-limit fallback stays gated on
is_auto to honour explicit provider constraints for transient limits.
- Apply both fixes to sync call_llm() and async acall_llm() paths.
- Add 6 targeted tests for the new quota-error detection cases.
#23482 fixed cache poisoning in the sync path: when a Codex auxiliary
timeout closes the underlying OpenAI client, _evict_cached_client_instance
walks CodexAuxiliaryClient wrappers via their _real_client attribute and
drops the cache entry so the next aux call rebuilds.
The cache key includes async_mode (see _client_cache_key), so the sync and
async clients for the same provider live in two distinct entries pointing
at the same underlying transport. The fix walked the sync wrapper's
_real_client correctly but the async wrappers
(AsyncCodexAuxiliaryClient, AsyncAnthropicAuxiliaryClient,
AsyncGeminiNativeClient) never exposed _real_client at all, so the async
entry survived eviction and kept handing out the poisoned client.
Effect on async aux callers: one timeout now poisons every subsequent
async aux call (compression, vision, session_search, title_generation)
with 'Connection error' until gateway restart -- even while the sync
route recovered as designed in #23482.
Mirror the sync wrapper's _real_client onto each async wrapper so the
existing eviction helper finds them. Three changes, one per wrapper:
- AsyncCodexAuxiliaryClient: self._real_client = sync_wrapper._real_client
(the underlying OpenAI client)
- AsyncAnthropicAuxiliaryClient: same shape
- AsyncGeminiNativeClient: self._real_client = sync_client (Gemini's
native facade is itself the leaf; no OpenAI client beneath it)
Update _evict_cached_client_instance docstring to reflect that it now
covers both sync and async wrappers via the same attribute walk.
Test: TestAuxiliaryClientPoisonedCacheEviction.test_evict_cached_client_instance_walks_async_wrapper
seeds both sync and async cache entries pointing at the same leaf and
asserts both are dropped on a single eviction call. Verified the test
fails without the wrapper changes ("async cache entry survived
eviction -- wrapper is missing _real_client") and passes with them.
Refs #23482, #23432
When an auxiliary provider returns HTTP 402 (credit / payment), every
subsequent compression / title-gen / session-search / vision call still
re-tried it as the FIRST entry in the chain — burning ~1 RTT to hit 402
again, then falling back. On a long Discord/LCM session that meant dozens
of doomed 402s per minute (issue #23570).
Add a per-process unhealthy-provider cache with a 10 min TTL. When any
caller observes a payment error against a provider, the label is marked
unhealthy and skipped by:
* _resolve_auto Step-1 (main provider use-as-aux path)
* _resolve_auto Step-2 (aggregator/fallback chain)
* _try_payment_fallback (used by call_llm/acall_llm on first 402)
Skip-logs are throttled to once per minute per label so a bursty session
doesn't spam agent.log. Entries auto-expire so a topped-up account
recovers without manual intervention. The cache is in-process only by
design — multi-profile users with different keys per profile must each
hit the 402 once.
Refs #23570
A Codex auxiliary timeout closes the underlying OpenAI client (so the
streaming hang doesn't sit until the user kills the session), but the
cached wrapper kept pointing at the now-dead transport. Subsequent
auxiliary calls (compression retry, memory flush, background review,
title generation routed via provider: main) reused that closed client
and failed fast with 'Connection error' until the gateway restarted —
even though the main agent route was healthy the whole time.
Sync `_get_cached_client` had no liveness check (async did, via loop
identity), and the connection-error fallback in `call_llm` only fired
on the auto provider path, so an explicit provider — including the
common `auxiliary.compression.provider: main` shape — never evicted.
Three fixes:
* New `_evict_cached_client_instance(target)` helper that drops the
cache entry whose stored client is target (or wraps it via
`_real_client`, for `CodexAuxiliaryClient`).
* `_CodexCompletionsAdapter._close_client_on_timeout` evicts the
wrapper after closing the inner OpenAI client.
* `call_llm` and `async_call_llm` evict on `_is_connection_error`
before re-raising, regardless of whether the provider is auto.
Net effect: one timeout costs one summary attempt + the existing 30s
compressor cooldown; the next compaction rebuilds the client and
works. Non-connection errors (4xx/5xx) do not evict, so cache hits
stay stable.
Closes#23432
## Summary
- Forwards chat-completions `timeout` into the Codex Responses stream call.
- Adds total elapsed-time enforcement while the Responses stream is still yielding events.
- Closes the underlying client on timeout to unblock stalled streams, then raises `TimeoutError`.
- Adds focused tests for timeout forwarding and total timeout enforcement.
## Why
The Codex auxiliary adapter can be used by non-interactive auxiliary work such as context compression. If the stream keeps yielding progress-like events but never completes, SDK socket/read timeouts do not necessarily protect the full operation. This makes the CLI look stuck until the user force-interrupts the whole session.
This is a refreshed upstream-ready version of the earlier fork fix around `d3f08e9a0` / PR #3.
## Verification
- `python -m py_compile agent/auxiliary_client.py tests/agent/test_auxiliary_client.py`
- `python -m pytest -o addopts='' tests/agent/test_auxiliary_client.py::TestCodexAuxiliaryAdapterTimeout -q`
- `git diff --check`
When a provider returns a 429 rate-limit error (not billing-related),
the auxiliary client's call_llm/async_call_llm previously did NOT trigger
the fallback chain. This caused auxiliary tasks like session_search to
exhaust all 3 retries against the same rate-limited endpoint, losing
session metadata that depended on the summarization completing.
Root cause: `_is_payment_error()` only matched 429s containing billing
keywords ("credits", "insufficient funds", etc.). Provider-specific
rate-limit messages like Nous's "Hold up for a bit, you've exceeded the
rate limit on your API key" didn't match, so `_is_payment_error` returned
False, `_is_connection_error` returned False, and `should_fallback` was
False — all retries hit the same rate-limited provider.
Fix:
- New `_is_rate_limit_error()` function that detects 429 + rate-limit
keywords, generic 429 without billing keywords, and OpenAI SDK
`RateLimitError` class instances (which may omit .status_code).
- Updated `should_fallback` in both `call_llm` and `async_call_llm` to
include `_is_rate_limit_error`.
- Updated the max_tokens retry path to also check for rate-limit errors.
- Updated the reason string to include "rate limit".
This complements the Nous rate guard (PR #10568) which prevents new calls
to Nous when already rate-limited — this fix handles the case where a
request is already in flight when the 429 arrives.
Related: #8023, #12554, #11034
Co-authored-by: Zeejay <zjtan1@gmail.com>
Copilot review on PR #17012 noted the docstring/comment lists `0`
among the falsy effort values that fall back to `medium`, but the
existing regression tests only cover `None` and `""`. Add the third
case to lock in the full contract.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
auxiliary.<task>.extra_body.reasoning, but the new translation path in
_CodexCompletionsAdapter.create() reads the effort with
``reasoning_cfg.get("effort", "medium")``. That returns the configured
value verbatim when the key is present, so ``effort: null`` /
``effort: ""`` (both common YAML shapes) flow through as
``{"effort": null, "summary": "auto"}`` and Codex rejects the request
with "Invalid value for parameter ``reasoning.effort``".
agent/transports/codex.py::build_kwargs() — which the new adapter is
documented to mirror — uses a truthy check (``elif
reasoning_config.get("effort"):``) so the same falsy values keep the
"medium" default. Switch the auxiliary adapter to the same
``or "medium"`` truthy form so identical config produces identical
requests on both paths.
- [x] Two new regression tests cover ``effort: None`` and
``effort: ""`` and assert the request goes out as
``{"effort": "medium", "summary": "auto"}``.
- [x] Old behaviour fails the new tests (``{'effort': None} !=
{'effort': 'medium'}``); fixed behaviour passes all 11 tests in the
``TestCodexAdapterReasoningTranslation`` class.
- [x] Adjacent suites green: ``tests/agent/test_auxiliary_client.py``
(108 passed) and ``tests/agent/transports/test_codex_transport.py +
test_chat_completions.py`` (73 passed).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
_try_anthropic() lacked the explicit_api_key parameter added to
_try_openrouter() in #18768. When resolve_provider_client() is called
with provider="anthropic" and an explicit key (e.g. from a fallback_model
entry with api_key set), the key was silently ignored — _try_anthropic()
always fell back to resolve_anthropic_token(), so the fallback returned
None,None for users without a default Anthropic credential configured.
Fix: add explicit_api_key: str = None to _try_anthropic() and use
explicit_api_key or <pool/env fallback> in both the pool-present and
no-pool paths. Pass explicit_api_key=explicit_api_key at the call site
in resolve_provider_client(). Symmetric with the _try_openrouter() fix.
No behavior change when explicit_api_key is None.
When resolve_provider_client() passes explicit_api_key for OpenRouter auxiliary
tasks, _try_openrouter() now accepts and honors this parameter instead of
silently ignoring it and falling back to OPENROUTER_API_KEY env var.
Root cause: _try_openrouter() had no explicit_api_key parameter, so even
when callers wanted to pass a runtime credential pool key, it could not be used.
Fix:
- Add explicit_api_key: str = None parameter to _try_openrouter()
- Prioritize explicit_api_key over pool key and env var
- Update resolve_provider_client() call site to pass explicit_api_key
Regression coverage:
- Test that explicit_api_key is passed to OpenAI client when provided
- Test that fallback to OPENROUTER_API_KEY still works when explicit_api_key is None
Closes#18338
Providers like Google Vertex, Azure, and Amazon Bedrock reject API
requests with duplicate tool names (HTTP 400: 'Tool names must be
unique'). The upstream injection paths in run_agent.py already dedup
after PR #17335, but two API-boundary functions pass tools through
without checking:
- agent/auxiliary_client.py: _build_call_kwargs() (all non-Anthropic
providers in chat_completions mode)
- agent/anthropic_adapter.py: convert_tools_to_anthropic() (Anthropic
Messages API path)
Add defensive dedup guards at both sites. Duplicates are dropped with
a warning log, converting a hard 400 failure into a recoverable
condition. This is intentionally conservative — the root-cause dedup
in run_agent.py is the primary defense; these guards add resilience
against future injection-path regressions.
Includes 8 new tests covering unique passthrough, duplicate removal,
empty/None edge cases.
Closes#18478
The _CODEX_AUX_MODEL constant had already rotated twice in 6 weeks
(gpt-5.3-codex -> gpt-5.2-codex -> now broken again at gpt-5.2-codex)
because ChatGPT-account Codex gates which models it accepts via an
undocumented, shifting allow-list that OpenAI publishes no changelog
for. Any pinned default will keep going stale. Issue #17533 reports
the current breakage: every ChatGPT-account auxiliary fallback fails
with HTTP 400 "model is not supported" and the 60s pause loop degrades
long sessions.
Rather than reset the clock with another stale pin (PR #17544 proposes
gpt-5.2-codex -> gpt-5.4), remove the hardcoded second-order Codex
fallback entirely:
- Delete `_CODEX_AUX_MODEL`.
- Drop `_try_codex` from `_get_provider_chain()` (the auto chain now
ends at api-key providers; 4 rungs instead of 5).
- Rename `_try_codex() -> _build_codex_client(model)` and require an
explicit model from the caller. No more guessing.
- `resolve_provider_client("openai-codex", model=None)` now warns and
returns (None, None) instead of silently guessing a stale model ID.
- Remove `_try_codex` from the `provider="custom"` fallback ladder
(same stale-constant trap).
- `_resolve_strict_vision_backend("openai-codex")` routes through
`resolve_provider_client` so the caller's explicit model is honored.
Codex-main users are unaffected: Step 1 of `_resolve_auto` already
uses `main_provider` + `main_model` directly and passes the user's
configured Codex model through `resolve_provider_client`, which never
touched `_CODEX_AUX_MODEL`. Per-task overrides (`auxiliary.<task>.provider/model`)
continue to work and are the supported way to route specific aux tasks
through Codex.
Users whose main provider fails with a payment/connection error and
who have ONLY ChatGPT-account Codex auth will now see the 60s pause
without a stale-model-rejection noise line in between -- same outcome,
cleaner failure.
Closes#17533. Supersedes #17544 (which resets the clock on the
same stale-constant problem).
* docs(anthropic): correct OAuth scope to Max plan + extra usage credits only
The previous docs pass (#17399) overstated what Anthropic OAuth works
with. In practice Hermes can only route against a Claude Max plan that
has purchased extra usage credits — the base Max allowance is not
consumed, and Claude Pro is not supported at all. Without Max + extra
credits, users must fall back to an ANTHROPIC_API_KEY (pay-per-token).
Updates the four pages touched in #17399:
- integrations/providers.md
- user-guide/features/credential-pools.md
- reference/environment-variables.md
- getting-started/quickstart.md
* fix(aux): skip kimi-coding in vision auto-detect (closes#17076)
Kimi Coding Plan's /coding endpoint (Anthropic Messages wire) has no
image_in capability — Kimi's own docs confirm and suggest switching to
a vision-capable model. Vision lives on the separate Kimi Platform
(api.moonshot.ai, OpenAI-wire, pay-as-you-go). When the user has
kimi-coding as main provider and auxiliary.vision.provider=auto,
resolve_vision_provider_client was handing back an AnthropicAuxiliaryClient
wrapped around /coding which 404'd on every vision request.
Add a _PROVIDERS_WITHOUT_VISION frozenset ({kimi-coding, kimi-coding-cn})
and gate the main-provider vision branch on membership. On a skip the
auto-detect falls through to OpenRouter → Nous like any other
main-provider-unavailable case.
Explicit per-task overrides (auxiliary.vision.provider=kimi-coding) are
unaffected — the skip only applies when the caller is in auto mode.
Tests: 4 new targeted tests in TestVisionAutoSkipsKimiCoding covering
the skip path, CN variant, explicit-override passthrough, and a guard
against accidental skip-list widening.
Auxiliary callers that configure reasoning via
auxiliary.<task>.extra_body.reasoning were having that config silently
dropped by the Codex Responses adapter — it only forwarded
messages/model/tools through to responses.stream(), never translating
chat.completions-shaped reasoning hints into the Responses API's
top-level reasoning + include fields.
Mirror the main-agent translation from agent/transports/codex.py:
- extra_body.reasoning.effort → resp_kwargs.reasoning.{effort, summary:"auto"}
- 'minimal' → 'low' clamp (Codex backend rejects 'minimal')
- Always include ['reasoning.encrypted_content'] when reasoning is enabled
- {'enabled': False} → omit reasoning and include entirely
- Non-dict reasoning values are ignored defensively
Reported by @OP (Apr 26 feedback bundle).
## Changes
- agent/auxiliary_client.py: _CodexCompletionsAdapter.create() now reads
and translates extra_body.reasoning before calling responses.stream()
- tests/agent/test_auxiliary_client.py: 9 new tests covering all effort
levels, the minimal→low clamp, the disabled path, the no-op paths,
and defensive handling of wrong-shape inputs
Co-authored-by: teknium1 <teknium@users.noreply.github.com>
- config.py: remove dead ENV_VARS_BY_VERSION[17] entry (current _config_version
is 22, so all users are past version 17 and would never be prompted for
GMI_API_KEY on upgrade — consistent with how arcee was added)
- auxiliary_client.py: use google/gemini-3.1-flash-lite-preview as GMI aux
model instead of anthropic/claude-opus-4.6 (matches cheap fast-model pattern
used by all other providers: zai→glm-4.5-flash, kimi→kimi-k2-turbo-preview,
stepfun→step-3.5-flash, kilocode→google/gemini-3-flash-preview)
- test_gmi_provider.py: fix malformed write_text() call in doctor test
(was: write_text("GMI_API_KEY=*** encoding="utf-8") → missing closing quote,
wrote literal string 'GMI_API_KEY=*** encoding=' to .env file)
- test_gmi_provider.py + test_auxiliary_client.py: update aux model assertions
to match new cheaper default
- docs/integrations/providers.md: add 'gmi' to inline 'Supported providers'
fallback list (was only in the table, not the inline list at line ~1181)
- docs/reference/cli-commands.md: add 'gmi' to --provider choices list
Keep auxiliary provider resolution aligned with the switch and persisted main-provider paths when models.dev returns github-copilot slugs.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
