| name | Incident Response |
| description | Use when responding to a production incident, running a post-incident review, or designing an on-call runbook. Triggers on 'incident', 'outage', 'postmortem', 'runbook', 'page', 'sev1', 'sev2'. |
Incident Response
When to invoke
- A page just fired.
- "Draft a postmortem for…"
- "Write a runbook for…"
- "Classify this incident severity."
Severity matrix
| Sev | Impact | Response target |
|---|
| Sev1 | Full outage or data loss risk | Page immediately, 15-min ack |
| Sev2 | Major feature degraded | Page during business hours |
| Sev3 | Minor feature or single-tenant issue | Ticket, next business day |
Active incident workflow (Sev1/Sev2)
- Declare the incident in the incident channel. One-sentence summary.
- Assign roles: Incident Commander (IC), Communications Lead, Ops Lead. IC does not debug.
- Stabilize first, understand later: roll back, failover, scale out, feature-flag off.
- Timestamped timeline in the channel - every action logged.
- External comms every 30 minutes on status page, even if "still investigating."
- Resolve = customer impact ended. Not "we found the bug."
- Handoff to postmortem owner before IC logs off.
Postmortem (blameless)
Write within 5 business days. Required sections:
- Summary - one paragraph, customer-facing language
- Timeline - UTC, copy-paste from channel
- Impact - users affected, duration, revenue/SLO burn
- Root cause - "5 whys"; contributing factors
- What went well / What went poorly
- Action items - each with owner, due date, priority; no "improve monitoring" without specifics
Runbook template
Every alert must link to a runbook with: symptom, dashboard links, likely causes, verification steps, mitigations (ranked by reversibility), escalation path.
References