[aw-failures] Daily Model Inventory Checker: Copilot CLI silent startup crash (exit code 1) #30043

Problem Statement

The Daily Model Inventory Checker workflow fails with a 100% failure rate over the last 6 hours. The Copilot CLI agent binary exits silently with code 1 — producing zero stdout and zero stderr — preventing the agent from executing any logic.

Parent report: #30042

Affected Runs

Both failures are on main branch. The pattern is consistent across different trigger types, indicating the issue is environment/config-level, not trigger-specific.

Symptoms

• All data-collection jobs succeed: collect_anthropic_models, collect_openai_models, collect_gemini_models, collect_copilot_models all exit cleanly
• models.json artifact: 54 bytes (effectively empty — no model data written)
• Agent job: 0 turns, 0 tool calls, 0 tokens consumed
• Copilot CLI binary starts (PID assigned) then immediately exits with code 1
• Harness message: "no output produced — not retrying (possible causes: binary not found, permission denied, auth failure, or silent startup crash)"
• Network: 2 requests to api.githubcopilot.com:443 (allowed), 0 blocked — no substantive API calls

Probable Root Cause

The harness identifies four possible causes for zero-output exit code 1:

1. Auth failure — Copilot CLI token expired or invalid (most likely given api.githubcopilot.com was reached but no calls were made)
2. Binary incompatibility — Copilot CLI v1.0.40 may have a runtime dependency issue in the current container image
3. Empty input — the 54-byte models.json (empty) may cause the agent's prompt-injection step to abort before generating output
4. Permission denied — less likely since PID was assigned

The most actionable hypothesis is Copilot CLI auth token expiry — the binary connects to the Copilot API endpoint but fails to authenticate before producing any output.

Proposed Remediation

1. Verify Copilot CLI credentials: Check that the Copilot auth token used in the scheduled workflow is valid and not expired. Rotate if necessary.
2. Add startup health check: Run copilot auth status or equivalent as a pre-agent step to surface auth failures with a clear error message before the agent job.
3. Diagnose empty models.json: Add validation to data-collection jobs to fail loudly if models.json is < 100 bytes, rather than silently passing an empty artifact to the agent.
4. Enable harness retry for startup crashes: Consider adding one retry with a forced re-auth step when exit code 1 + zero output is detected.

Success Criteria

• Daily Model Inventory Checker completes at least one full successful run on schedule
• models.json contains non-empty model inventory data
• Agent produces > 0 turns and creates an output artifact
• Root cause (auth vs binary vs empty input) confirmed from diagnostics

Generated by [aw] Failure Investigator (6h) · ● 594.5K · ◷

• expires on May 11, 2026, 1:33 AM UTC

Closing: workflow confirmed working

Fresh evidence (2026-05-11 within the [aw] Failure Investigator 6h window):

• Daily Model Inventory Checker run §25643617414 completed with conclusion: success at 2026-05-11T00:10:30Z.
  • Duration: 6.7m
  • Turns: 15 (vs. 0 in the failure pattern described above)
  • Token usage: 770k (vs. 0 in the failure pattern)
  • Engine: GitHub Copilot CLI, schedule trigger on main

The 100%-failure / zero-output / exit-code-1 signature documented in this issue is no longer reproducing. Closing as resolved. If the silent-exit pattern returns, the safe-output auto-tracker will recreate a fresh issue.

Action taken by [aw] Failure Investigator (6h).

Generated by [aw] Failure Investigator (6h) · ● 15.7M · ◷