| name | incident-postmortem |
| disabled | true |
| description | Use when assembling, updating, or reviewing an incident postmortem, and the repository has no postmortem skill of its own. |
Incident Postmortem
Assemble an evidence-backed, blameless postmortem. Preserve human responder context, fill gaps with telemetry, and identify actionable repair items.
Inputs
Accept any of:
- GitHub issue URL or
owner/repo#number
- Datadog incident URL or ID
- Incident document, timeline, Slack excerpt, alert issue, dashboard, or freeform notes
If no source is provided, ask whether to update an existing postmortem or draft a new one.
Workflow
- Read the source. Gather issue body, comments, incident timeline, linked alerts, dashboards, runbooks, and responder notes.
- Identify known facts and gaps. Preserve filled-in human-written sections. Mark missing impact, timeline, mitigation, root cause, response effectiveness, or repair items.
- Set the investigation window. Use two hours before the first known symptom through two hours after recovery unless the data suggests a wider window.
- Gather evidence.
- Alerts/events: state changes, pages, deploys, config changes, mitigations.
- Metrics: user impact, service health, saturation, dependency health, recovery confirmation.
- Logs/traces: error propagation, latency paths, exceptions, failed dependencies.
- Code: relevant paths for root cause and repair items.
- Build the timeline. Use UTC unless the existing template says otherwise. Cross-check timestamps across sources and flag discrepancies.
- Analyze cause. Separate trigger, proximate cause, systemic cause, and recovery factors.
- Assess impact. State customer/user impact first. If there was no impact, explain what contained it.
- Create repair items. Each item should have a concrete owner-ready outcome, acceptance criteria, and evidence link.
- Draft the update. Supplement existing content instead of overwriting firsthand responder notes.
- Review before posting. Present the proposed issue/comment update to the user before applying it.
Optional Datadog notebook
Create a Datadog notebook when the incident has meaningful telemetry evidence or when the user asks for one. Include:
- incident summary and time window;
- key metrics and log queries;
- trace examples or links;
- timeline and recovery evidence;
- query documentation sufficient for another engineer to reproduce the analysis.
Link the notebook from the postmortem draft.
Postmortem structure
# Incident Postmortem: <title>
## Impact / Executive Summary
<Who was affected, for how long, severity, current state.>
## Timeline
| Time (UTC) | Event | Source |
|---|---|---|
| <time> | <event> | <link/query> |
## Mitigation and Recovery
<What stopped the impact and how recovery was verified.>
## Root Cause Analysis
**Trigger:** <external event/change>
**Proximate cause:** <immediate failure mode>
**Systemic cause:** <design/process/monitoring gap>
## Response Effectiveness
**Went well:** <evidence-backed notes>
**Could improve:** <evidence-backed notes>
**Where we got lucky:** <optional>
## Repair Items
| Item | Outcome | Evidence |
|---|---|---|
| <title> | <acceptance criteria> | <link/query/code> |
Guidelines
- Be blameless: describe system behavior, not personal fault.
- Every factual claim needs a source link, query, trace ID, log sample, dashboard, issue, or code reference.
- Do not overwrite human-written incident responder content unless the user explicitly asks.
- Distinguish confirmed facts from hypotheses.
- If posting to GitHub, follow the GitHub Posting Protocol in the global instructions.