Context
- Impact and failing endpoints
- Recent deploys and config changes
- Relevant logs, traces, and metrics
Agent instructions
- Reproduce or simulate failure path first.
- State expected vs actual behavior and likely scope.
- Generate hypotheses and test them one at a time.
- Apply the smallest fix and add a regression test.
Verification
- Root cause is supported by evidence.
- Fix is validated with targeted checks.
- Regression test prevents recurrence.
Expected result: A verified root cause and low-risk fix.
Compatible tools: GitHub Copilot, Claude Code, Cursor, Windsurf, Cline