Autopilot Readiness Checklist

• I know the exact trigger, schedule, or input pattern.
• The task has a single owner and an on-call fallback.
• Logs are written somewhere I can actually inspect quickly.
• Failure notifications reach a human, not just a log file.
• Secrets, tokens, and permissions are limited to the minimum required.
• There is a documented rollback or disable switch.
• The job is idempotent or explicitly protected from duplicates.
• I know what happens if the job runs twice, late, or partially.
• Upstream and downstream systems are rate-limit safe.
• There is a manual test case that proves the critical path.
• I have a clear stop condition for noisy or dangerous failures.
• The first 24 hours include active observation, not blind trust.

This page is versioned, specific, and built to be printed or copied into an SOP without cleanup. No generic AI filler, no vague advice, and no hidden dependencies.