Incident Postmortem Writer

After an incident, the evidence is scattered: pager alerts in UTC, chat in local time, log findings in someone's notes. Writing the postmortem takes hours, and it's easy to slip into blaming a person instead of fixing the system.

This free agent kit turns that material into a blameless postmortem draft. It converts every source to UTC and checks the conversion against events that appear twice, builds a timeline with sources, and computes time to detect, acknowledge, mitigate and resolve. It traces the root cause with 5 Whys, backing each step with evidence or labeling it a hypothesis, then lists contributing factors, what was ruled out, what went well, what went poorly and where you got lucky. Action items get a type, a priority, an owner (only if someone agreed) and a link to what they fix. Missing numbers and owners become open questions.

Inside: the agent instructions, a postmortem template with rules for each section, a severity rubric you customize, a blameless writing guide with the 5 Whys method, conversation starters, a Claude Code subagent that searches log files and config in your repo, and a worked example.

Limitations: it can't query your monitoring, pager or chat tools, and the Claude Code version doesn't run commands. Hypotheses need confirming by the system owners. It drafts; your team reviews before sharing.