incidents

Load this skill when responding to a production incident, writing a post-mortem, executing a hotfix, or classifying incident severity. Auto-load whenever a production issue is reported, a P1/P2/P3 cla

SKILL.md

GitHub

Incident Response

Severity Levels

Level Definition Response Time
P1 Production is down or data is corrupted Immediate — all hands
P2 Major feature broken, no workaround for users < 1 hour
P3 Feature degraded, workaround exists < 1 business day

Response Steps

  1. Detect — identify from monitoring alerts or user report
  2. Triage — classify severity, identify scope
  3. Mitigate — roll back if available; disable the feature if possible
  4. Fix — hotfix process (minimal scope, all CI must pass, staging still required)
  5. Post-mortem — written within 48 hours for P1/P2

Hotfix Process

  1. Branch: fix/critical-<description> from main
  2. Scope: Minimal fix only — no unrelated changes
  3. Review: 1 approval minimum, fast-tracked
  4. CI: All checks must pass — do not skip CI under pressure. A broken hotfix is worse than a delayed one.
  5. Merge: Merge commit to main (never squash — preserve the commits)
  6. Tag: Bump PATCH version, apply tag
  7. Deploy: Staging → production (abbreviated but both still required)

Post-Mortem Format (Required for P1 and P2)

## Incident Post-Mortem: <Short Title>

**Date:** YYYY-MM-DD
**Severity:** P1 | P2
**Duration:** <how long production was impacted>

### What Happened

Brief timeline of events.

### Root Cause

The single underlying cause (not symptoms).

### Impact

What was broken. How many users/requests affected.

### Detection

How was the incident discovered? How long between impact and detection?

### Resolution

What fixed it?

### Prevention

What changes prevent recurrence?

- [ ] Action item (owner, due date)