Postmortem

Postmortem Tools

Overview

Core tools for reading postmortem documents during investigation, and writing them during the dedicated "Generate Postmortem" action. get_postmortem is available in all sessions. save_postmortem is only available during the "Generate Postmortem" action — not during RCA.

Tools

get_postmortem(incident_id)

Read the current postmortem for an incident. Returns markdown content, generated_at, and updated_at timestamps. Returns status: "not_found" if no postmortem exists yet. Only the latest version is returned; historical versions are browsable in the UI but not via this tool.

save_postmortem(incident_id, content)

Write or update a postmortem. Only available during the "Generate Postmortem" action — not during RCA. Each save automatically snapshots the previous content as a version for history tracking. Content must be complete markdown — partial updates are not supported. Max 100,000 characters.

Workflow

1. Call get_postmortem — if a prior version exists, use it as a baseline to preserve structure and confirmed facts
2. Gather context from RCA summary and connected communication platforms (Slack, etc.)
3. Write structured markdown following the output format below
4. Call save_postmortem

Output Format

Always generate postmortems with these sections in order:

# Postmortem: <Incident Title>

**Date:** YYYY-MM-DD HH:MM UTC
**Duration:** Xh Ym
**Severity:** critical/high/medium/low
**Service:** <service name>

## Summary
2-3 sentences describing what happened.

## Timeline
- **HH:MM UTC** - Event description
- **HH:MM UTC** - Event description

## Root Cause
Technical explanation of what failed and why.

## Impact
Services, users, SLAs, and data affected.

## Contributing Factors
Human/process factors: deployment pressure, alert fatigue, communication gaps, handoff confusion.
Only include if evidence exists from conversations.

## Resolution
How the incident was resolved or mitigated.

## Action Items
- [ ] Concrete follow-up item
- [ ] Another follow-up item

## Lessons Learned
What can prevent similar incidents in the future.

Guidelines

• Professional, factual tone
• Do not speculate beyond what data supports
• Incorporate human context from Slack/communication tools when available
• If Slack tools are available, search relevant channels in the incident time window before writing
• Keep the document concise but thorough
• Use checkboxes for action items so they can be tracked