Context
- Failing job logs and commit SHA
- Workflow matrix and environment versions
- Recent merged changes and flaky test history
Agent instructions
- Identify first failing step, not just final error.
- Reproduce failure locally with equivalent commands.
- Classify failure type: test, build, lint, infra, flaky.
- Propose minimal fix and confidence level.
Verification
- Fix reproduces locally and in CI rerun.
- No unrelated pipeline behavior changed.
- Follow-up actions are documented.
Expected result: A concrete CI fix or clear escalation path with evidence.
Compatible tools: GitHub Copilot, Claude Code, Cursor, Windsurf