debugger skill

Debugging

Use the reported symptom, observed behavior, code, and tests to establish what failed. Derive the expected behavior from the user's request and reliable project evidence. State uncertainty where those sources conflict.

Investigation and repair

1. Capture the failing command or interaction, error, environment, revision, inputs, and reproduction frequency. Keep sensitive values in protected local storage and use sanitized evidence in reports.
2. Reproduce the smallest failure. Compare a known working path or revision when that helps isolate the cause.
3. Trace the responsible data and control flow. Inspect recent changes, assumptions, boundary conditions, dependency behavior, and error handling.
4. Test a concrete root-cause hypothesis. Distinguish the triggering condition from the code responsible for the failure.
5. Make the focused repair within the requested scope. Add regression coverage that fails for the original defect and exercises the intended behavior.
6. Re-run the reproduction and relevant adjacent checks. Review the diff and summarize the cause, fix, evidence, and any remaining uncertainty.

When the user requests diagnosis only, deliver the investigation and proposed repair as an evidence-based report. A repair request includes implementation and verification within its scope. Clarify a product decision when the evidence supports materially different expected outcomes; continue useful investigation while that decision is pending.

For intermittent failures, collect timing, concurrency, retry, and resource signals. For environment-only failures, compare versions and configuration without exposing secrets. When reproduction is unavailable, label hypotheses and explain the observation that would distinguish them.