← ClaudeAtlas

postmortemlisted

Write a blameless incident postmortem from the raw material: logs, alerts, deploy history, commits and chat threads. Builds a timestamped timeline, separates the trigger from the root cause and the things that made it worse, measures impact, and turns lessons into owned, checkable actions. Use for "write a postmortem", "incident report", "what happened with the outage", "RCA", or after any production incident or near miss.
ychampion/ystack · ★ 4 · AI & Automation · score 77
Install: claude install-skill ychampion/ystack
# Postmortem The value of a postmortem is the actions that stop the next one. A good one is built from evidence, names systems instead of people, and ends with a short list of fixes someone owns. ## Procedure ### 1. Gather the evidence Collect, with timestamps: alerts and pages, logs and error rates around the window, deploys and config changes (`git log --since --until`, the deploy tool's history), the chat or incident channel, support tickets. Note the source of every fact. Normalize all times to one timezone, stated once. ### 2. Build the timeline ``` | Time (UTC) | Event | Source | | 14:02 | deploy of v4.12 starts | deploy log | | 14:09 | error rate on /checkout rises to 18% | dashboard | | 14:21 | first alert fires | pager | | 14:35 | rollback started | chat | | 14:41 | error rate back to baseline | dashboard | ``` Mark four moments: started, detected, mitigated, resolved. The gaps between them are findings in themselves (why did detection take 12 minutes?). ### 3. Separate the causes - **Trigger:** the change or event that started it. - **Root cause:** why the trigger could do damage. Keep asking why until the answer is a missing safeguard, check or design property, not a person's slip. - **Contributing factors:** what made it bigger or longer: slow detection, missing runbook, noisy alerts, hard rollback. If the evidence doesn't settle the root cause, say so and list what would. ### 4. Measure impact Duration, users or requests affected, errors, data los