Generate Post-Mortem

Path Resolution

1. Read workflow.json in the project root
2. If it exists and docsRepo is ".": this IS the docs repo — use local paths
3. If it exists and docsRepo is a repo name: resolve via pwsh .claude/skills/tool-worktree/scripts/resolve-repo.ps1 <docsRepo> to get the docs root path. Templates at <resolved>/templates/, output to <resolved>/postmortems/
4. If no workflow.json: templates at templates/, output to docs/postmortems/

Instructions

1. Resolve paths (see Path Resolution above)
2. Read the template at <templates>/postmortem.md
3. Gather incident details:
   • What happened? When?
   • What was the impact? (users, revenue, SLA)
   • What was the timeline? (detection → response → resolution)
   • What was the root cause?
4. Generate the post-mortem following blameless principles:
   • Focus on systems, not people — "the deploy pipeline didn't catch X" not "Bob forgot to check X"
   • Be specific about impact — numbers, not adjectives
   • Timeline must be precise — include timestamps
   • Action items must be actionable — assigned owner, due date, clear definition of done
5. Save to <output>/[date]-[slug].md

Blameless Principles

• Use passive voice or system-focused language for failures
• "What went well" section is mandatory — acknowledge good response
• "Where we got lucky" exposes hidden risks
• Action items should prevent recurrence, improve detection, AND improve response

Quality Checklist

• Summary is 2-3 sentences (scannable)
• Impact has concrete numbers
• Timeline has timestamps (not "then... later...")
• Root cause is technical and specific
• Contributing factors are listed (root cause is never the only factor)
• Detection section identifies the gap
• Action items have owners and due dates
• At least one action each for: prevent, detect, respond
• "Where we got lucky" section is filled in
• Tone is blameless throughout

Tips

• Write the timeline first — it structures everything else
• The "where we got lucky" section is often the most valuable
• Every post-mortem should update at least one runbook
• Link to monitoring dashboards and log queries for evidence