Take A Coffee

Claude Code auto-wake

Wake schedules are useful. They are not a host proof.

If a Mac wakes at 5am, starts Claude Code, then sleeps again, verify the separate laptop layer before trusting the run.

Wake

Prove it woke

Write a timestamp before launching the agent and again after the first real tool call. A scheduled wake is not proof that work started.

Run

Cap the session

Use a narrow prompt, a stop time, and a receipt path. Morning runs should leave evidence, not just an open terminal.

Sleep

Restore defaults

After the run, return the Mac to normal sleep behavior unless another local task explicitly needs awake mode.

Receipt command

Give the run a tiny heartbeat file.

Use this alongside any Claude Code auto-wake script so you can tell whether the host actually stayed alive during the window.

while true; do date >> ~/claude-code-host-awake.log; sleep 30; done
Use before scheduling

Copy an auto-wake safety prompt.

Paste this into Claude Code before relying on a morning wake script.

Audit this Claude Code auto-wake setup before I trust it. Separate scheduled wake, agent start, host heartbeat, work receipt, and restored sleep. Verify the task is atomic, has a stop time, has no expected approval/login prompts, writes a timestamp outside the agent, records battery/heat/network state, and restores normal sleep settings. If any check is missing, stop and list the missing check instead of running overnight.

The minimum host checklist.

Want Codex to run the host mode?

Take A Coffee gives Codex a temporary awake mode for local macOS and Windows agent work, then restores normal sleep.

Run it with Codex