🏥 CI Failure🏥 CI failure investigation: smoke copilot run 21780014599

[github-actions]

Summary

The Smoke Copilot workflow failed during its scheduled run on the main branch.

• Run ID: 21780014599
• Commit: f84db28 ("docs: sync documentation with recent code changes [docs] docs: sync documentation with recent code changes #561")
• Event: Scheduled (runs every 12 hours)
• Conclusion: failure
• Created: 2026-02-07T12:21:05Z

Workflow Description

The Smoke Copilot workflow performs integration testing of:

1. GitHub MCP - Reviews last 2 merged PRs
2. Playwright - Navigates to github.com and verifies page title
3. File Writing - Creates test file in /tmp/gh-aw/agent/
4. Bash Tools - Verifies file creation with cat command

Investigation Limitations

⚠️ Limited Access: The CI Doctor workflow cannot access GitHub API logs due to authentication constraints. This investigation is based on:

• Workflow context from the triggering event
• Similar past failures from cache memory
• Workflow configuration analysis

Similar Past Failures

1. Pelis Agent Factory Advisor Failure (Run 21773244180)

• Exit code 1 during "Execute GitHub Copilot CLI" step
• Warning: No safe outputs file found at /opt/gh-aw/safeoutputs/outputs.jsonl
• Notice: GitHub MCP lockdown mode enabled for public repository
• Uses awf v0.13.12 with --enable-chroot flag

2. Examples Test SSL Failure (Run 21728924123)

• Exit code 35 (CURLE_SSL_CONNECT_ERROR)
• Root cause: HTTPS_PROXY removal in PR fix: remove HTTP_PROXY/HTTPS_PROXY env vars from agent container #524
• Broke SSL connections through Squid proxy in intercept mode

Common Failure Patterns

Based on repository history, likely causes include:

1. Docker Network Issues

   • Pool overlaps (awf-net conflicts)
   • Orphaned containers from previous runs
   • Cleanup issues from timeout kills

2. MCP Server Issues

   • GitHub MCP lockdown mode in public repos
   • Safe outputs file not created
   • MCP gateway startup failures

3. SSL/TLS Issues

   • HTTPS_PROXY environment variable problems
   • Squid proxy intercept mode failures
   • Certificate validation errors

4. Copilot CLI Issues

   • Exit code 1 without clear error message
   • Tool execution failures
   • Timeout during long-running operations (5-minute limit)

Recommended Actions

Immediate Investigation

1. Download workflow logs manually from GitHub Actions UI

   gh run download 21780014599

2. Check for Docker resource leaks on GitHub Actions runners

   • Orphaned awf-net networks
   • Stale containers from previous runs

3. Review Copilot CLI logs if preserved at /tmp/awf-agent-logs-*

Debugging Steps

1. Re-run the workflow with debug logging:

   awf --log-level debug --keep-containers ...

2. Check Squid proxy logs for blocked domains:

   sudo cat /tmp/squid-logs-*/access.log | grep TCP_DENIED

3. Verify GitHub MCP server configuration:

   • Check /home/runner/.copilot/mcp-config.json
   • Confirm --disable-builtin-mcps flag usage
   • Verify GITHUB_PERSONAL_ACCESS_TOKEN environment variable

Prevention

1. Add pre-test cleanup to smoke test workflows:

   docker network prune -f
   docker container prune -f

2. Enhance error reporting in Copilot CLI execution step:

   • Capture full stdout/stderr
   • Preserve agent logs as artifacts
   • Add explicit error detection

3. Monitor scheduled runs for patterns:

   • Check if failures correlate with specific times
   • Identify if runner resource exhaustion is a factor

Next Steps

1. ⏳ Manual log review required - Download logs from run 21780014599
2. 🔍 Reproduce locally if possible using same commit SHA
3. 📊 Update cache memory with findings once logs are analyzed
4. 🛠️ Implement fixes based on root cause analysis

---

🏥 Automatically investigated by CI Doctor workflow
⚠️ This is a preliminary report due to GitHub API access limitations. Manual follow-up required.

AI generated by CI Doctor