Objective

Diagnose and fix the safe_outputs job failures in the Smoke Update Cross-Repo PR and Smoke Multi PR smoke test workflows.

Context

From the Daily Audit Report (discussion #22237):

Two smoke test workflows both failed at the safe_outputs job (not the agent job), suggesting a possible regression in safe-output processing:

Both ran close in time and both use the copilot engine. The simultaneous failure of two smoke tests at the same job stage (safe_outputs) suggests a systemic issue rather than an isolated incident.

Investigation Steps

1. Use gh aw audit 23392658504 and gh aw audit 23392726144 to inspect the failed runs
2. Check logs for both safe_outputs jobs to identify the error message/root cause
3. Check git log around the time of the failures for recent changes to safe-outputs logic
4. Look at actions/setup/js/ for any safe-output processing code that may have a bug
5. If a regression is found, fix it and add a test to prevent recurrence

Files to Investigate

• actions/setup/js/ — JavaScript safe-output processing scripts
• .github/workflows/smoke-update-cross-repo-pr.md (or similar)
• .github/workflows/smoke-multi-pr.md (or similar)
• Recent commits touching safe-output logic

Acceptance Criteria

• Root cause of safe_outputs failures identified
• Fix applied (or confirmed as a transient/external failure)
• Both smoke workflows pass on re-run (or regression prevented)
• make agent-finish passes

Generated by Plan Command for issue #discussion #22237 · ◷

• expires on Mar 24, 2026, 8:48 AM UTC