[aw-failures] jsweep: investigate 8→60 turn explosion causing agent job failure

[github-actions]

Problem Statement

Run §24872562447 failed after 60 turns and 25.1 minutes, consuming 4.64M tokens. The last successful baseline run §24704249133 (2026-04-21) completed in only 8 turns. The audit-diff flagged this as changed with turns_increase as the primary signal.

Evidence

Audit comparison details

Metric	Baseline (2026-04-21)	Failing run (2026-04-24)
Turns	8	60 (+650%)
Duration	~5 min	25.1 min
Token usage	~low	4.64M
Conclusion	success	failure
Tool types used	multiple	1 (safeoutputs.create_pull_request)
Errors	0	1

Only 1 tool type was called across all 60 turns (a single safeoutputs.create_pull_request), suggesting the agent spent 59 turns iterating without making meaningful progress before failing.

Probable Root Cause

Something changed between 2026-04-21 and 2026-04-24 causing excessive iteration:

• A change in the JavaScript files being processed (more/larger files, or malformed content)
• A change in the workflow prompt (added complexity or unclear success criteria)
• A regression in how the agent determines task completion

Note: logs contained redacted suspicious-looking domains (evil.com, github.com.evil.com, etc.) — these are likely from JavaScript file content being analyzed, not actual outbound requests.

Proposed Remediation

1. Inspect agent-stdio.log at /tmp/gh-aw/aw-mcp/logs/run-24872562447/ for the specific error recorded at turn 60
2. Diff the workflow prompt between current and the 2026-04-21 baseline to find any added complexity
3. Add clear success/exit criteria to the jsweep prompt to prevent runaway iteration
4. Add a max_turns limit in the workflow frontmatter if supported

Success Criteria

• jsweep scheduled runs complete in ≤15 turns and ≤10 minutes
• No more 4M+ token runs for this workflow
• Agent job succeeds without exhausting its turn budget

References:

• Failing run: §24872562447
• Baseline run: §24704249133 (2026-04-21, success, 8 turns)
  Related to [aw] jsweep - JavaScript Unbloater failed #28211

Generated by [aw] Failure Investigator (6h) · ● 361.1K · ◷

• expires on May 1, 2026, 7:31 AM UTC

[github-actions]

🍪 *Issue Monster selected this for Copilotecho BEGIN___COMMAND_OUTPUT_MARKER ; PS1= ; PS2= ; unset HISTFILE ; EC=0 ; echo ___BEGIN___COMMAND_DONE_MARKER___0 ; }

I've identified this issue as a good candidate for automated resolution and requested assignment to the Copilot coding agent.

If assignment succeeds, the Copilot coding agent will analyze the issue and create a pull request with the fix.

Om nom nom! 🍪

🍪 Om nom nom by Issue Monster · ● 21.7K · ◷

[copilot-swe-agent]

The agent encountered an error and was unable to start working on this issue: Please try again later, or contact support if the problem persists. (Request id: 0A09:2D2F72:3682F89:DC888C5:69EB2091)