Module 10 · Project and quiz
A Routine in Production for 7 Days
Put one scheduled routine into real use in your business, either the morning brief or the weekly report, and let it run for seven days with every run recorded. The goal is not a perfect routine on day one. It is a routine you trust by day seven, with evidence.
1Choose the routine: morning brief or weekly report.
2Write the self-contained prompt file and test it by hand.
3Run it headless with --permission-mode dontAsk and a minimal allowlist.
4Add the audit log hook and the risky-command block hook, and test that the block fires.
5Schedule it with a cloud routine, launchd or cron on Mac, or Task Scheduler on Windows.
6Log every run in the Routine Run Log for seven days.
7At the end, ask Claude to review the log and outputs and improve the prompt.
Deliverable
A scheduled routine running in your business, protected by hooks, with a seven-day run log showing status, minutes saved and issues found.
Done when
- The routine ran on schedule for at least five of seven days.
- It never had permission to send, delete or pay.
- You tested that your block hook actually blocks.
- The run log has one line per run with minutes saved.
- You made at least one improvement to the prompt based on the log.
What does claude -p do?
In dontAsk permission mode, what happens to an action that would normally ask for approval?
Which hook event can block an action before it happens?
Why do scheduled prompts need to be fully self-contained?
How do you keep a client-facing custom command from running on the agent's own initiative?
