SKILL: debugging

Purpose

Diagnose errors with a deterministic workflow: reproduce → isolate → fix → verify.

When to Use

• Tests are failing.
• Runtime errors occur (stack traces, crashes, misbehavior).
• Performance regressions or unexpected output is reported.

Inputs

• symptom (required, string): error message, failing test, or observed behavior.
• repro_steps (optional, string): how to reproduce.
• logs (optional, string): relevant logs/traces.
• environment (optional, object|string): OS, versions, env vars (no secrets).

Steps

1. Reproduce deterministically:
   • capture exact command(s)
   • capture versions
2. Localize failure:
   • narrow to smallest failing unit/test
   • add targeted logging/diagnostics if needed
3. Form hypotheses and test them one by one.
4. Implement the smallest fix that resolves the root cause.
5. Add/adjust tests to prevent regression.
6. Verify:
   • original repro no longer fails
   • related tests still pass

Validation

• Root cause is stated explicitly (not only the symptom).
• Fix is verified by test/repro.
• No unrelated behavior changes are introduced.

Output

root_cause: "<what broke and why>"
fix: "<what changed>"
verification: ["<commands/tests>"]
follow_ups: ["<optional hardening>"]

Safety Rules

• Do not “fix” by disabling tests or checks unless explicitly approved and justified.
• Prefer reversible debugging changes (guarded logs, small diffs).

Example

Symptom: “TypeError: cannot read property 'x' of undefined in production.” Output: root cause (missing null check), fix, and regression test.