name: incident-triager
description: Evidence-driven debugging/incident workflow: reproduce, isolate, minimize, fix, verify, and prevent regressions.
Incident Triager — Debug Without Drift
Use this skill when:
- Something is broken, flaky, or non-deterministic
- You're doing "investigation work" that can easily drift
- You need to hand off a bug efficiently
Goal: get to a minimal reproducible case, a minimal fix, and a verifiable outcome.
Triage protocol (must follow in order)
-
Define the incident
- Symptom (what is observed)
- Expected behavior
- Severity (user-blocking / degraded / cosmetic)
-
Reproduce
- Exact steps
- Environment (OS/version/build, branch, config)
- Is it deterministic? (Y/N)
-
Isolate
- Smallest scope that still reproduces (file, module, step, config)
- Identify the first bad point (when it starts failing)
-
Hypothesize (bounded)
- List 1–3 plausible causes max
- For each, define a test that could disprove it
-
Fix (minimal)
- Prefer the smallest change that addresses the root cause
- Avoid "refactor while here" unless explicitly requested
-
Verify
- Rerun the reproduction steps
- Add/extend a cheap regression check if possible (Tier 1)
-
Document
- Record the outcome and the prevention check
- Summarize in a handoff note using session-summarizer if handing off
Output format (for handoff)
- Repro: steps + determinism
- Root cause: (if known)
- Fix: what changed (summary)
- Verification: what was run and result
- Next: if still open, what to try next (bounded)