| name | postmortems |
| description | Use this skill to write blameless postmortems after an incident — building a timestamped timeline, identifying root cause and contributing factors, and producing specific, owned, tracked action items. |
Postmortems
A postmortem is the written record of an incident: what happened, why, and what
will change so it does not happen the same way again. Postmortems are
blameless — they explain outcomes through systems and conditions, never
through individual fault. That is what makes people comfortable being honest,
which is what makes the analysis useful.
Use this skill after an incident is resolved (the live-response side is covered
by the incident-response skill).
When to write one
Write a postmortem for any incident that caused user-visible impact, required
coordinated response, or surprised you. Near-misses that could easily have been
serious are worth writing up too. Trivial, well-understood, single-step issues
usually are not.
Principles
- Blameless. Describe what the system allowed to happen. Replace "X made a
mistake" with "the system permitted an unsafe action without a guardrail".
- Factual and timestamped. Build the narrative from evidence and a timeline,
not memory or impression.
- Action-oriented. The payoff is concrete, owned, tracked follow-up work —
not just a story.
- Honest about the unknown. If root cause is unconfirmed, say so and note
what would confirm it.
- Safe to publish. Keep credentials, personal data, and customer identifiers
out. Write as if it may be read widely.
Template
# Postmortem: <short incident title>
## Summary
2–3 sentences: what happened, the impact, and how it was resolved.
## Impact
- Severity:
- Duration: <detection → resolution>, plus time-to-detect and time-to-mitigate
- Who/what was affected and how severely
- Quantified where possible (error rate, requests failed, users affected)
## Timeline
All times in a single timezone (UTC recommended).
- T0 root-cause change or trigger
- T1 first symptom
- T2 detection (how was it noticed?)
- ... key investigation steps and actions
- Tn mitigation
- Tn+ full resolution
## Root cause
The technical chain of causation. What actually broke and why.
## Contributing factors
Conditions that made the incident possible, worse, or slower to detect/resolve
(e.g. missing alert, unclear runbook, insufficient capacity headroom, a fragile
dependency). These often matter more than the single "root cause".
## What went well
Detection, tooling, or response steps worth keeping and reinforcing.
## What went poorly / where we got lucky
Gaps, and places a slightly different situation would have been much worse.
## Action items
| Action | Type (detect / prevent / mitigate) | Owner | Tracking link |
|--------|-----------------------------------|-------|---------------|
| ... | ... | ... | ... |
## References
Dashboards, alerts, related changes, discussion threads (links only; no secrets).
Writing good action items
- Each action item is specific, owned, and tracked as a real work item —
not a vague aspiration.
- Prefer systemic fixes over "be more careful". Add a guardrail, an alert, a
test, a limit, or automation.
- Balance the three types: prevent the cause, detect it faster next time,
and mitigate faster when it recurs.
- Do not over-generate. A handful of high-value, actually-completed items beats a
long list that rots.
Timeline tips
- Anchor to the trigger (a change, a spike, an upstream event), not to when you
woke up.
- Capture time-to-detect and time-to-mitigate explicitly — they are the
numbers you most want to drive down.
- Note every mutating action during response, so the timeline explains recovery.