Code
Write on-call runbooks
One page per alert: what it means, how to check, what to do.
Fill it in
Point at the infra repo, or describe it.
What currently pages someone.
Who to wake, and when it is worth it.
Your prompt
Write runbooks for [the system]. Alerts: [the alerts] One page per alert, written for someone woken at 3am who did not build this. That reader cannot infer anything, so every step is a literal command with the expected output next to it. Start each with what the alert MEANS in user terms. "Queue depth above 10,000" tells a sleepy person nothing; "customers' exports are not being produced" tells them whether to keep going. Include how to confirm it is real, because a meaningful share of pages are the monitoring, and the runbook should say how to tell within two minutes. Say who to escalate to and when it is worth waking them. A runbook without that leaves the on-call person guessing at the worst possible moment. End each with how to verify it is actually fixed.
Use Write on-call runbooksOpens with everything above already filled in.
Why this works
A runbook is read by someone woken at 3am who did not build the thing, so it cannot ask them to infer anything. Every step is a literal command with the expected output, and each page opens with what the alert means in user terms.