What this prompt does
This prompt makes the model a senior incident commander writing a blameless postmortem. You supply the [incident_id], the affected [systems], the [severity], and the raw chat log, runbook edits, and timeline in [raw_notes], and it returns a structured document: a two-sentence summary, impact, a detection-to-resolution timeline, a 5-whys root cause with contributing factors, what went well, and action items with owners and due dates.
The structure works because it stays neutral and grounds every claim in the notes. The 5-whys analysis pulls the real cause out from under the symptom everyone fixated on, while the contributing-factors line acknowledges incidents rarely have a single cause. The [raw_notes] variable is the evidence base — the prompt is instructed not to speculate beyond it — and requiring every action item to carry an owner and date is what stops fixes from quietly never shipping.
The blameless framing is not just tone for its own sake. Postmortems that hunt for someone to blame stop people surfacing the truth, which is exactly what prevents the next incident. By grounding the timeline and impact in [raw_notes] and refusing to speculate past them, the document stays defensible when reviewed later. The what-went-well section is easy to skip but worth keeping, since reinforcing the parts of the response that worked is as useful as fixing the parts that did not.
When to use it
- An incident on
[systems]is resolved and you need a postmortem written up. - You have raw chat logs and a timeline but no clean narrative yet.
- You want a blameless analysis that names causes, not people.
- You need a 5-whys root cause rather than a surface symptom fix.
- You want action items with explicit owners and due dates.
- You are standardizing postmortems across incidents of varying
[severity].
Example output
Expect a structured postmortem: a two-sentence plain-language summary, an impact section (users affected, duration, and money if known), a detection-to-resolution timeline with timestamps drawn from [raw_notes], a 5-whys root-cause analysis plus contributing factors, a what-went-well section, and a list of action items each with an owner and due date followed by lessons learned. The tone is neutral, with no blame and no speculation beyond the notes.
Pro tips
- Paste the full raw timeline into
[raw_notes]; the postmortem can only be as accurate as the evidence you give it. - Set
[severity]and[systems]correctly so impact and scope are framed right. - Let the 5-whys run past the obvious symptom; the real cause usually sits a couple of layers down.
- Insist every action item has an owner and a date, since an unowned action item never ships.
- Keep the tone blameless on purpose; postmortems that hunt for someone to blame stop preventing repeats.
- If the notes are thin, the model should flag gaps rather than invent a narrative, so do not over-trust a sparse input.
