The Runbook Writer

Write an operational runbook for this agent system.

System: [DESCRIBE]
Who operates it: [DESCRIBE THEIR CONTEXT]
Known failure modes: [LIST]

Produce:
1. What this system does, in operator terms
2. Normal behaviour: what healthy looks like in the dashboards
3. Alerts: what fires, what each means, what to do
4. For each known failure mode: symptom, diagnosis, fix
5. How to turn it off, quickly and safely
6. How to turn it back on
7. How to investigate a specific user's failed run
8. How to roll back a prompt or configuration change
9. Escalation: when to wake someone, and who
10. What NOT to do — the interventions that make things worse

Then:
11. The three most likely 3am pages, and the exact steps for each

Write it for someone who did not build this and is reading it under
pressure.