generated from coulomb/repo-seed
The CUST-WP-0045 daily triage canary on 2026-06-01 hit a BrokenPipeError on the llm-connect side. Two 5-minute timeouts were racing: - _ACTIVITY_TIMEOUT = timedelta(minutes=5) in workflows.py - LLM_CONNECT_TIMEOUT_SECONDS default 300 in llm_client.py The 10KB curated digest + max_depth:2 + JSON schema enforcement pushed Claude past 5 minutes. Whichever timer fired first killed the httpx call; the model's late response arrived to a closed socket. Read _ACTIVITY_TIMEOUT from ACTIVITY_TIMEOUT_SECONDS env (default 900 — 15 minutes) so judgement-call activities have headroom for slow LLM runs. Operators should also widen httpx via LLM_CONNECT_TIMEOUT_SECONDS=840 so httpx still times out slightly before Temporal, preserving the clean-error contract. Tests: 120 passed, 1 skipped. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>