Postmortem Format Selector Prompt
Decide whether this specific incident is best told as a narrative, a strict timeline, a 5-whys, or a contributing-factors analysis — before you start writing the document.
- Target user
- SRE / incident commander writing the post-incident review
- Difficulty
- Intermediate
- Tools
- Claude, ChatGPT, Cursor
The prompt
You are a staff SRE who has edited hundreds of postmortems and knows that the wrong structure buries the real lesson. There is no single "correct" postmortem template — the right shape depends on the failure mode. I will paste a short description of the incident and what we already know: [INCIDENT SUMMARY: trigger, blast radius, duration, how it was detected and mitigated] [WHAT IS STILL UNCERTAIN: anything we have not yet confirmed] [AUDIENCE: who will read this — eng team, leadership, customers, regulators] Do the following: 1. Classify the failure shape: single linear cause, multiple converging contributing factors, slow-burn/degradation, human-in-the-loop decision under uncertainty, or external dependency. State which fits and why. 2. Recommend ONE primary format — narrative, strict timeline, 5-whys, or contributing-factors map — and explain in 2-3 sentences why it surfaces the lesson better than the alternatives for THIS incident. 3. Warn me where the recommended format tends to mislead (e.g. 5-whys collapsing parallel causes into one chain) and what to add to compensate. 4. Suggest a second, supporting section if the primary format alone would hide something important. 5. Give a one-line skeleton outline of section headings for the chosen format. Output format: a short recommendation block (Primary format / Why / Failure mode / Watch-outs), then the heading skeleton as a bulleted list. Guardrails: keep all language blameless — describe systems and conditions, not people's intent. Mark anything I have not confirmed as [UNVERIFIED]. You are recommending a structure only; I own the final document and the decision on which format to use.
Why this prompt works
Most teams reach for whatever template lives in their wiki, then spend the writing session fighting the structure. A converging multi-factor outage crammed into a 5-whys chain loses every parallel cause except the one the author happened to follow; a simple config typo dressed up as a contributing-factors map reads as padding. Choosing the shape first, against the actual failure mode, makes the document do its job.
The prompt forces an explicit classification step before any prose is written, which is exactly where experienced reviewers add value — they recognize the failure shape and reach for the structure that exposes it. By naming the format’s blind spots and suggesting a supporting section, it guards against the common trap of believing the template is neutral. Templates are not neutral; they decide what gets emphasized and what disappears.
Critically, the format is a scaffold, not the analysis. The prompt keeps the AI in an advisory role, tags unverified claims, and leaves authorship with the human. The reviewer still owns the lesson, the tone, and the call on whether the recommended shape actually fits.