Skip to content

skillDevOps and cloud

Incident Postmortem

Writes a blameless postmortem from logs, chat and commits, with a verified timeline, a real root cause and follow-up actions that have owners.

npx workspacetemplates add incident-postmortem --tool claude-code

Writes 1 file to .claude/skills/incident-postmortem/

.claude/skills/incident-postmortem/SKILL.md27 lines · 1.5 KBFile
---
name: incident-postmortem
description: Writes a blameless postmortem from logs, chat and commits, with a verified timeline, a real root cause and follow-up actions that have owners.
---

# Incident postmortem

Write for a reader who was not there. Be factual, blameless and specific.

## Steps

1. Gather evidence: alert history, deploy and commit log around the time (`git log --since --until`), error logs, graphs, chat transcript. Ask the user for what you cannot read.
2. Build the timeline in UTC, one line per event with its source: first symptom, detection, escalation, each mitigation attempt, resolution. Mark each entry as confirmed or inferred.
3. Measure impact: duration, share of requests or users affected, data lost or delayed, money or SLA consequences.
4. Find the cause with "why" chains until you reach a process or design gap, not a person. Separate the trigger (what started it) from contributing factors (why it was possible, why detection was late, why recovery took long).
5. Note what went well, where luck helped, and what slowed recovery.
6. Write actions that change the system: each has an owner role, a due date, a priority and a success check. Prefer prevention and faster detection over "be more careful".

## Rules

- Never name an individual as the cause. Write "the deploy" or "the on-call runbook", not people.
- Do not guess; mark unknowns as open questions.
- Keep it under two pages.

## Report

Summary, impact, timeline, root cause and contributing factors, what went well, action items table, and open questions.

What it does

A postmortem is only useful if it changes something. Most of them list a timeline, blame the person who pushed, and end with actions nobody tracks. This skill collects the evidence first, separates what happened from why, and writes about systems and decisions rather than individuals.

Every action item is specific, has an owner and a due date, and says how you will know it worked.

Example

you › Write the postmortem for yesterday's 47 minute checkout outage

agent › Summary: a config deploy at 14:02 UTC set the payment timeout to 3 ms; checkout failed for 31 percent of orders until rollback at 14:49. Root cause: no validation of timeout units and no canary stage. Actions: add schema validation (platform, due Dec 2), canary config rollout (platform, Dec 16), alert on payment error ratio (checkout, Nov 25).

Questions

Why avoid names in the cause?
People hide mistakes when they expect blame. Naming the missing check or the unclear runbook finds problems that repeat, and the person who made the change tells you more.
Does it invent details it cannot find?
No. Gaps become open questions and uncertain timeline entries are marked inferred.
  • CI Pipeline Speedup

    SkillDevOps and cloud

    Free
  • Deployment Strategies

    SkillDevOps and cloud

    Free