incident-response
Installation
SKILL.md
Incident Response
Purpose
Respond to live production incidents: stabilize first, capture a trustworthy timeline, keep communication factual, and convert every incident into durable follow-ups once the fire is out.
Workflow
- MUST declare the incident: state severity (SEV1 / SEV2 / SEV3), current impact, and who is affected; if severity cannot be determined, declare the higher severity and downgrade later.
- MUST mitigate before diagnosing: stop the bleeding (rollback, feature flag, failover, rate limit) before hunting root cause. If no safe mitigation exists, say so and escalate rather than experimenting in production.
- MUST capture the timeline as events happen: UTC timestamp, actor, action, observation. Reconstructed timelines are marked as estimates.
- MUST keep comms factual: what is impacted, what is being done, next update time. No speculation, no blame.
- MUST identify root cause only after mitigation is verified; distinguish trigger, root cause, and contributing factors.
- MUST end with follow-ups that have owners; an incident without follow-ups is not closed.