[aw-failures] jsweep — Copilot CLI step times out at 20m while staging the cleanup PR #31652

Problem statement

Today's scheduled jsweep - JavaScript Unbloater run failed with ##[error]The action 'Execute GitHub Copilot CLI' has timed out after 20 minutes.. Examination of the agent stdio log shows the agent had completed the substantive work — lint, typecheck, and vitest run (21 tests pass) for action_conclusion_otlp.cjs / .test.cjs — and was already on the final tail of its plan: saving cache state, git checkout -b jsweep/action-conclusion-otlp, staging and committing the cleaned files, when the step-level timeout-minutes: 20 killed it before it could call create_pull_request.

No PR was created and no safe-output was emitted, so the work is lost despite the analysis succeeding.

Affected workflow and run

Workflow source: .github/workflows/jsweep.md. Step configuration: timeout-minutes: 20 on the Execute GitHub Copilot CLI step (line 38 of jsweep.md).

Evidence

2026-05-12T05:09:52.4010618Z ##[error]The action 'Execute GitHub Copilot CLI' has timed out after 20 minutes.

● Run tests for our file (shell)
  │ cd .../actions/setup/js && npx vitest run action_conclusion_otlp --no-file-parallelism 2>&1 | tail -30
  └ 26 lines...
All 21 tests pass. Now save cache state and create the PR.

● Save cache state (shell)
  │ STATE_FILE="/tmp/gh-aw/cache-memory/jsweep-state.json" ...
  └ 37 lines...

● Create branch and commit (shell)
  │ git checkout -b jsweep/action-conclusion-otlp && git add
  │   actions/setup/js/action_conclusion_otlp.cjs
  │   actions/setup/js/action_conclusion_otlp.test.cjs && git commit ...
  └ 14 lines...

● Confirm branch name (shell)
  │ git branch --show-current
  └ 2 lines...

Probable root cause

• jsweep.md sets the Copilot CLI step to timeout-minutes: 20, but the workflow's plan reliably consumes ~18–20 minutes on the analysis phase (npm install + lint/prettier + tsc + vitest + commit). There is no safety margin for the final git commit / create_pull_request to land.
• The Copilot engine does not appear to ratchet its own retry/timeouts based on the step's wall-clock budget, so when work is back-loaded (commit + PR last) any slowdown in the front half evicts the actual output.
• effective_tokens: 15.8M and 29 turns indicate the agent itself was healthy — no error_max_turns, no MCP failure, no firewall block.

Proposed remediation

Preferred (small, isolated):

• Raise timeout-minutes on the Execute GitHub Copilot CLI step in .github/workflows/jsweep.md from 20 → 30 to match its 30-min sibling steps and leave headroom for commit/PR creation.

If raising the timeout is undesirable, alternatives:

• Reorder the plan so the agent stages and commits early (after lint/typecheck pass) and runs the full test suite only on the already-staged branch; PR creation then has no large tail step gating it.
• Move PR creation to a post-agent step that consumes the cache-memory state file (/tmp/gh-aw/cache-memory/jsweep-state.json) and calls create_pull_request deterministically — the agent's job becomes "emit cache state file", not "branch + commit + PR".

Success criteria

• Two consecutive scheduled runs of jsweep - JavaScript Unbloater finish ≤ 18 minutes with at least one create_pull_request safe-output (or noop if nothing to clean).
• No ##[error]The action 'Execute GitHub Copilot CLI' has timed out for jsweep on a 30-day window after the fix.

Confidence and unknowns

High confidence on the proximate cause (step-level timeout vs. backloaded PR step). Unknown whether earlier jsweep runs were already brushing the 20-minute boundary or whether today's run was anomalously slow (e.g., npm install cache miss) — easy to check by comparing the prior 5 jsweep runs' Execute GitHub Copilot CLI step durations.

References

• Parent failure report: [aw-failures] [aw] Failure Report 2026-05-11 (6h): aw-portfolio-yield workflow defect blocks compile + own runtime + recurrences [Content truncated due to length] #31455
• Run: §25714027006
• Workflow source: .github/workflows/jsweep.md (line 38 — timeout-minutes: 20)
  Related to [aw-failures] [aw] Failure Report 2026-05-11 (6h): aw-portfolio-yield workflow defect blocks compile + own runtime + recurrences [Content truncated due to length] #31455

Generated by [aw] Failure Investigator (6h) · ● 15.3M · ◷

• expires on May 19, 2026, 7:51 AM UTC