fix(agent): honor configured auxiliary.title_generation.timeout (#32729)

[Tranquil-Flow]

Bug fix: auxiliary.title_generation.timeout config was being silently ignored.

Reported in #32729: user updated auxiliary.title_generation.timeout in
config.yaml to a high value (e.g. 1800) for a slow local LLM, but the title
generation worker continued to time out at 30 s and surface
Auxiliary title generation failed: Request timed out. after every session.

Root cause

agent/title_generator.py::generate_title() (the worker entrypoint) had
timeout: float = 30.0 hardcoded in its signature. Its only caller
(auto_title_session) never passed an explicit timeout, so the hardcoded
default was always what reached call_llm() — which then refused to read
auxiliary.title_generation.timeout from config (the config lookup only
fires when call_llm is called with timeout=None).

In other words: the user's auxiliary.title_generation.timeout setting was
correct, complete, and respected by call_llm. The contract break was a
single hardcoded signature default.

Fix (single-line, two production files)

1. agent/title_generator.py:54 — timeout: float = 30.0 →
   timeout: Optional[float] = None. The Optional type matches call_llm's
   contract and the existing helper resolves the configured timeout when None
   is forwarded.
2. agent/auxiliary_client.py:5726 — timeout: float = None →
   timeout: Optional[float] = None. Pure annotation accuracy: the runtime
   default was already None; only the type was lying. Brings the signature
   in line with the runtime branching on None to consult the configured
   timeout.

Tests

tests/agent/test_title_generator_timeout_32729.py (new file, 5 tests):

• test_default_signature_is_optional_float_none — signature default is
  None, type is Optional[float].
• test_no_explicit_timeout_forwards_none_to_call_llm — without an explicit
  timeout, call_llm receives timeout=None (RED on upstream/main).
• test_explicit_timeout_from_caller_is_forwarded_unchanged — explicit
  caller-supplied timeout is passed through verbatim.
• test_auto_title_session_does_not_pass_timeout_to_generate_title —
  auto_title_session does not inject a timeout at the worker boundary
  (preserves the contract).
• test_auto_title_session_full_call_chain_preserves_none_timeout —
  end-to-end chain auto_title_session → generate_title → call_llm surfaces
  timeout=None to call_llm (RED on upstream/main).

3 of 5 fail on upstream/main; all 5 pass with the fix.

Verification (post-rebase onto current upstream/main)

tests/agent/test_title_generator.py .......... 23 passed
tests/agent/test_title_generator_timeout_32729.py ..... 5 passed
tests/agent/test_auxiliary_client.py ........ 274 passed
                                              (1 pre-existing timing flake
                                              deselected: fails on
                                              upstream/main in isolation)

Rebase history

Built against upstream/main = 44ddc552f5 at 2026-07-01T10:00:00Z.
Rebased onto current upstream/main = 1c350728ec before publication.
Rebase applied cleanly with no conflicts. Final HEAD:
36476b1b698717aae7463266294afac6117800fb. Branch is exactly one focused
commit ahead of upstream/main.

Cross-reference

PR #39035 by @rodboev (opened 2026-06-04, 28 days ago, still open) takes
the same approach for the title_generator fix but bundles an unrelated
.github/workflows/docker.yml change splitting PR validation from cached
publish builds (+22/-8 lines of CI infrastructure). This PR retains only
the surgical bug fix and adds the Optional[float] annotation accuracy
fix for call_llm's signature. Maintainer can choose to merge one, the
other, or both.

Notes

• One pre-existing test_enforces_total_timeout_while_stream_keeps_emitting_events
  timing flake in tests/agent/test_auxiliary_client.py is deselected.
  Verified to fail in isolation on upstream/main without the fix applied —
  pre-existing, unrelated.
• No new dependencies, no AI attribution, no config-file changes, no skill
  changes.

Auto-published by Moonsong via Path B automated pipeline.
Reviewed against CONTRIBUTING.md and AGENTS.md contribution rubric.

[alt-glitch]

This was generated by AI during triage.

Duplicate of #39035 — the same one-line fix (agent/title_generator.py timeout default 30.0 → Optional[None] so call_llm resolves auxiliary.title_generation.timeout from config) for #32729. #39035 is the earliest open PR in this cluster (canonical). Related competing open fixes: #41844, #41841. Earliest overall was closed #18129.

[kshitijk4poor]

Thanks — closing as superseded by #56322 (salvage of #41844), which lands the same core one-liner (generate_title timeout default 30.0 -> Optional[None] so call_llm resolves auxiliary.title_generation.timeout). We picked #41844 as the minimal creep-free base; your annotation tidy on call_llm (float=None -> Optional[float]=None) is a legitimate correctness improvement but not required for the fix, and #41844 is the smaller surface. Your dedicated test file was the most thorough in the cluster — credit for that. Feel free to push back or resubmit the annotation tidy standalone.