← LeaderboardThe Skill Leaderboard

postmortem

A blameless incident postmortem: summary with impact numbers, a timeline, root cause and contributing factors, what went well and badly, and action items with owners.

Published by feimacode ↗feimacode/open-design-agent-kit

Try your first task

A source-grounded starter task has not been verified for this skill yet.

SKILL.md

name: postmortemzh_name: "事故复盘"en_name: "Postmortem"emoji: "🩺"description: "A blameless incident postmortem: summary with impact numbers, a timeline, root cause and contributing factors, what went well and badly, and action items with owners."en_description: "A blameless incident postmortem: summary with impact numbers, a timeline, root cause and contributing factors, what went well and badly, and action items with owners."category: docscenario: engineeringtags: ["postmortem", "incident", "retrospective", "sre", "outage", "事故复盘"]triggers:   - "postmortem"  - "post-mortem"  - "incident review"  - "incident report"  - "outage report"  - "事故复盘"od:   mode: prototype  craft:     requires:       - typography      - typography-hierarchy      - color      - accessibility-baseline  platform: desktop  scenario: engineering  preview:     type: html    entry: index.html  design_system:     requires: false  example_prompt: "Write a postmortem for this incident from the notes and timeline I paste: impact, timeline, root cause, contributing factors, what went well and badly, and action items with owners."

Postmortem

Intent. A blameless account of an incident that helps the team learn and prevents a repeat. It describes systems and decisions, never blames people.

Ground it

Write only what the sources say. Gather them yourself before writing a word:

  • the incident notes, chat or pager timeline, and dashboards the user provides. Ask for them if they're missing: never invent an incident, a timestamp or an impact number;
  • the commits that caused and fixed it (git log --since … --until …, git show <sha>), and the code paths involved.

Keep the list of files, commits and refs you read: they go in a Sources section at the bottom of the page and in sources when you register (register_open_design_artifact with kind html). If something the document needs isn't in the sources, ask the user, or mark it plainly as an open question. Never invent numbers, dates, names or quotes.

Structure

  1. Header: title, incident id if any, date, a severity pill (SEV1–SEV4), status (data-od-status="draft|reviewed|closed"), authors.
  2. Summary: what happened, who was affected and for how long, in three sentences.
  3. Impact: a row of 2–4 big numbers (duration, users or requests affected, error budget used), each with its source.
  4. Timeline: a table of timestamps (with the timezone) and events: detection, escalation, mitigation, resolution.
  5. Root cause: the technical chain, step by step. Link the commits.
  6. Contributing factors: what made it worse or slower to fix.
  7. What went well and What went badly: two short lists.
  8. Action items: a table of action, owner, priority and due date, each tied to a cause above.
  9. Sources.

Use blameless language: "the deploy pipeline allowed…", not "X forgot…".

Design

  • One readable column (max-width 720–780px), body 16–17px with line-height 1.55–1.65, generous space between sections. It must read well on a phone: tables sit in an overflow-x: auto wrapper, nothing has a fixed width wider than the screen.
  • Style from the active design system (--bg, --surface, --fg, --muted, --border, --accent); status colours are the only extra colours (green for accepted/done, amber for proposed/in progress, red for rejected/high severity, grey for superseded).
  • Headings are short labels; the content does the talking. Code, paths and commands in a monospace face, with overflow-wrap: anywhere so long paths wrap on phones.
  • Finish with check_open_design_artifact and fix every error at desktop and mobile width.