version: "1.0.0" name: postmortem-report description: How the SRE team writes a postmortem PDF. Use this whenever asked for a postmortem, an incident report, a write-up to share, or a PDF of what happened.

Postmortem report

You build the PDF yourself, in your sandbox. There is no template. Design it so a director understands the event in one minute and an engineer can check every claim.

What it must contain

1. A headline and a summary of three or four sentences.
2. A table of pages: page id, service, root cause. Pages that share a root cause are one incident; mark the others as symptoms.
3. At least one chart drawn from real data. Compute it in your script from /mnt/session/uploads/app.log (for example p99 latency or error rate per minute for each paged service) and mark the deploys on it.
4. For each root cause: a timeline in UTC, the evidence that proves it, what was ruled out, and the fix.
5. Action items, each with an owner team and a priority.

How to build it

• Python with matplotlib and pandas is installed. matplotlib.backends.backend_pdf.PdfPages writes multi-page PDFs. pandoc and LaTeX are there too. Use whichever you can do fastest.
• The investigators' findings are in /workspace/findings/*.json. Start from those instead of investigating again.
• Save the result to /mnt/session/outputs/postmortem.pdf. Only files under /mnt/session/outputs/ reach the user.
• Be quick: write one script, run it, and fix it only if it errors. Two to four pages. Do not render the PDF to images to inspect it.
• Keep everything inside the page margins: wrap the title and any long table cell instead of letting it run off the edge.
• Follow any postmortem conventions you find in team memory.

When it is built, tell the user the file name in one or two sentences. Do not paste the report into the chat.