postmortem

name	postmortem
description	Write incident postmortems using the DERP model (Detection, Escalation, Recovery, Prevention). Integrates with incident.io.
argument-hint	["find|list|from <path>|<incident-id-or-slug>"]
model	sonnet

Postmortem Skill

Write structured incident postmortems using the DERP model.

Arguments

Parse the argument to determine the mode:

Argument	Mode	Action
(none)	Interactive	Ask for incident details, then create/continue
find or list	List	Run postmortem-list.sh and display results
find <slug>	Find	Run postmortem-find.sh <slug>
from <path>	Generate from notes	Read notes file and generate postmortem
<incident-id-or-slug>	Create/Continue	Run postmortem-find-or-create.sh

Scripts

All scripts are in: ~/.dotfiles/ai/skills/postmortem/scripts/

• postmortem-find-or-create.sh <id-or-slug> - Find or return path for new postmortem
• postmortem-find.sh [slug] - Find specific or list all
• postmortem-list.sh - List all postmortems with metadata

Output format is tab-separated for easy parsing.

Storage

Postmortems are stored at: ~/dev/ai/postmortems/PostHog/{slug}.md

Template

Use the template at: ~/.dotfiles/ai/skills/postmortem/templates/derp-postmortem.md

Workflow: Create/Continue Mode

1. Run the find-or-create script with the incident ID or slug
2. If status is "found":
   • Read the existing postmortem
   • Show current status and what sections need work
   • Ask what the user wants to focus on
3. If status is "new":
   • Ask for basic incident details:
     • Title (one-line description)
     • Severity (SEV1-4 or Critical/High/Medium/Low)
     • Approximate duration
     • incident.io URL (if available)
   • Create the postmortem from the template
   • Begin the DERP interview

Workflow: Generate from Notes

When the user provides a path to existing notes (/postmortem from <path>):

1. Read the notes file at the provided path

2. Extract incident identifier:

   • Look for incident.io IDs (INC-123 pattern)
   • Look for explicit incident references
   • If not found, derive a slug from the notes title or ask the user

3. Map notes content to DERP sections:

   Notes Content	Maps To
   Timeline entries with times	Detection timeline, Recovery timeline
   "Root cause", "Why did this happen"	Prevention → Root Cause
   Actions taken, "We did X"	Recovery → Actions
   Team mentions, "@person"	Escalation → Who responded
   Alert mentions, monitoring	Detection → How detected
   "Action items", "TODO", "Follow-up"	Prevention → Action Items
   Impact statements, user counts	Summary → Impact

4. Generate draft postmortem:

   • Run postmortem-find-or-create.sh with the identifier
   • Create the postmortem file using the template
   • Fill in all sections that have data from the notes
   • Leave placeholders for missing information

5. Show the user what was extracted and what gaps remain

6. Offer to fill gaps through the interactive DERP interview

DERP Interview Process

Guide the user through each section with targeted questions. Only ask about sections that are incomplete. Ask one section at a time — don't present all questions at once.

Detection Questions

• How was the incident first detected? (alert, customer report, internal testing)
• What time was it detected? What time did it actually start?
• Were there any earlier warning signs we missed?
• What monitoring or alerting improvements would help?

Escalation Questions

• Who was paged or joined the response?
• How did communication flow? (Slack channel, video call, etc.)
• Was the initial severity assessment accurate?
• What could improve how we escalate incidents?

Recovery Questions

• Walk me through the steps taken to restore service
• What was the sequence and timing of recovery actions?
• Were there any false starts or rollbacks?
• What manual steps could be automated?

Prevention Questions

• What was the root cause?
• What contributing factors allowed this to happen?
• What specific action items will prevent recurrence?
• What lessons should the team take away?

Guiding Principles

• Capture exact quotes - Use the user's words; don't paraphrase
• Link to evidence - Ask for URLs to Slack threads, dashboards, and PRs
• Track action items - Each item needs an owner, priority, and due date
• No blame - Focus on systems and processes, not individuals

Status Progression

Update the status frontmatter field as the postmortem progresses:

• draft - Gathering information
• in-review - All sections complete, awaiting team review
• complete - Reviewed and finalized

Output Format

When displaying postmortem status, show title, status/severity/duration, file path, and per-section completion. For example:

Postmortem: {title}
Status: {status} | Severity: {severity} | Duration: {duration}
Path: {path}

Sections:
[x] Summary
[ ] Detection - needs timeline details
[x] Escalation
[ ] Recovery - missing automation opportunities
[x] Prevention (3 action items)