Incident Response

Use this skill when a production incident, major repeated error, rollback decision, or on-call triage is involved.

Workflow

1. Stabilize first: stop risky changes, preserve logs, and identify the affected service or workflow.
2. Classify severity in docs/harness/INCIDENT_RESPONSE.md.
3. Capture timeline, impact, suspected trigger, current mitigation, and owner.
4. Prefer read-only diagnostics before mutating systems.
5. If rollback is safer than forward fix, document the rollback command and validation.
6. Update model-visible major errors only with information the next agent must read.
7. After mitigation, create or update docs/harness/POSTMORTEM_TEMPLATE.md.

Severity Guide

• SEV0: data loss, security exposure, total outage, destructive automation.
• SEV1: customer-visible outage or broken critical workflow.
• SEV2: partial degradation, failed deployment, repeated operational failure.
• SEV3: non-urgent defect, flaky check, noisy alert.

Required Output

Return:

• severity
• impact
• immediate mitigation
• diagnostics run
• rollback or forward-fix decision
• follow-up action items
• postmortem required: yes/no