Changelog from raw commit history
writingGroups commits into user-facing changes, without inventing anything that is not in the log.
Incident postmortem from a raw timeline
Turns logs and a chat transcript into a blameless postmortem with contributing factors, not a culprit.
When to use it
The day after an incident, with a messy timeline and no one volunteering to write it up.
The prompt
Copy into your assistant
Write a blameless postmortem from the timeline below.
Structure:
1. What happened, in two sentences a non-engineer can follow.
2. Impact: who was affected, for how long, and what they could not do.
3. Timeline: one line per event, with timestamps.
4. Contributing factors: at least three, each phrased as a system property rather than a person's mistake ("the deploy had no canary", not "Sam deployed badly").
5. What would have caught this earlier, ranked by cost to implement.
Rules:
- No names of individuals anywhere in the document.
- Do not invent events not present in the timeline; mark gaps as "unknown".
- Do not write "human error" as a cause. It is the beginning of the analysis, not the end.
Timeline:
<timeline>
Systems involved:
<systems>What to change
If the timeline is chat logs, strip the jokes and side threads before pasting — they get summarised as facts otherwise.
Why it works
The prohibition on names and on "human error" is what makes a postmortem useful. Both are the default output of an unconstrained model, and both end the investigation early.
More writing prompts
- Audit interface copy for clarityRewrites button labels, errors, and empty states so they say what happened and what to do.
- An email that says no, without burning the bridgeDeclines a request clearly, names the constraint, and offers a smaller yes.
- Changelog from raw commit historyGroups commits into user-facing changes, without inventing anything that is not in the log.
Going deeper
- Prompts that survive contact with realityUsing agents · Level 2
- Writing a skill that changes behaviorSkills and prompts · Level 2
- The last hour before launchRunning agents in production · Level 3