Direct answer ⏱️ 5 min read

Can Claude Code Run a Chore Without Me Watching It?

Yes. Claude Code has a headless mode that runs with no one answering prompts, and a permission setting that decides how much it does on its own. Pairing that with your computer's own scheduler is what actually starts it while you are away. Some steps should still stop and ask.

TL;DR

💤 Can Claude Code run a chore without me watching it?

Yes. Claude Code's headless mode runs a task start to finish with nobody at the keyboard to answer its questions, and a permission setting controls how much it is allowed to do before it would normally stop and ask.

What actually starts it at 3am is not Claude Code, it is your own operating system's scheduler.

This post itself was written that way. The job that produced this page ran headless, with no window on screen and no one clicking anything, under a written rule that everything it starts gets stopped and confirmed gone before it finishes. That is the same shape as any unattended chore: a clear job, a permission setting, and a cleanup step.

🔌 How it actually runs while you're not there

Three separate things have to line up. Miss one and the chore either never starts, or stops and waits for a person who is not there.

PieceWhat it doesWhere it comes from
Headless modeRuns the instruction with no one answering its promptsClaude Code's own -p flag
Permission modeDecides how much it does before it would stop and askA setting you choose per run
The schedulerStarts the run at a time, on a repeat, with no one clicking anythingYour computer's own cron or Task Scheduler, not Claude Code

Claude Code documents the headless flag and its options on its own site, and documents the permission settings separately: the safest of them only reads and proposes, the boldest auto-approves almost everything.

🎚️ The supervision dial

Click a level to see what it means for a real chore, like sorting invoices into folders or drafting the same weekly report.

How much supervision does the chore get?
Low risk

You run each step yourself and approve every action as it comes up. Nothing happens you did not just watch happen. This is where you start with any new chore.

Think of the dial like cruise control, not autopilot. Higher settings take more of the small, repeated decisions off your plate. None of them take the car off the road for you if something unusual shows up ahead.

🚫 What should never run fully unattended

The dial only measures how much typing and clicking you get back. It says nothing about how expensive a wrong move is.

Safe to leave running

  • Reading, sorting, renaming, tagging files you can always re-derive
  • Drafting a report, email, or summary a person still reads before it goes anywhere
  • Anything already backed up, so a bad run is a revert, not a loss
  • A chore you have watched succeed the same way several times

Keep a human in the loop

  • A real payment, refund, or anything that moves money
  • Deleting files with no backup
  • Anything sent straight to a customer with no review step
  • A judgment call with no clear right answer written down

🗝️ Setting one up

Five steps, in order, each one a gate before the next.

  1. Write the chore down as a plain-English instruction, the same way you would tell a new hire.
  2. Run it once yourself, watching every step, to see where it actually needs a decision.
  3. Move it to a permission setting that lets it edit or act without asking, but still logs what it did.
  4. Point a scheduler at it, cron or Task Scheduler, for the time it should run.
  5. Check the log the next morning before you trust the second run.

Not sure the chore is even describable enough to reach step one? The free Chore Audit asks five questions about one task you do by hand and hands back the hours it costs a year, plus a starter SKILL.md written from your own answers.

Get the unattended version built for you

A written scope first, so you know exactly what it will and will not do on its own before anything is built.


Get a written scope → Or run the free chore audit → The Chore Audit runs in your browser. Nothing you type is sent anywhere.

❓ Quick answers

Can Claude Code run a chore without me watching it?

Yes. Headless mode runs with no one answering its prompts, and a permission setting decides how much it does on its own. Your computer's scheduler is what starts it while you're away.

What is Claude Code's headless mode?

Passing -p with your instruction runs Claude Code non-interactively, so it does not stop to wait for a person at the keyboard.

What are Claude Code's permission modes?

Plan mode only reads and proposes, it changes nothing. Accept-edits applies file edits without asking but still checks other actions. Bypass mode auto-approves almost everything and needs the most trust.

What should never run fully unattended?

Anything hard to undo: a real payment, deleting files with no backup, sending something to a customer, or a judgment call with no clear right answer.

How do I know if my chore is describable enough to automate?

The free Chore Audit asks five questions about one task you do by hand and returns the hours it costs a year plus a starter SKILL.md, so you can see the shape of the job first.

Sources for the facts above: the headless -p flag is documented on Claude Code's own headless-mode docs. The permission modes (plan, accept-edits, bypass) are documented on Claude Code's permissions docs. The Chore Audit's five questions, its "hours a year" and "times a year" figures, and its "free, no account, nothing leaves this tab" line are all on the Chore Audit page itself. The $250 fixed price, 5-day delivery and free-fix-until-it-matches terms are on the custom build page. The claim that this post itself ran headless is a first-hand account of how it was produced, not a measurement about anyone else's setup.

Need a Claude Code skill built? Free scope in 24 hours, $250 fixed →

Related: is Claude Code worth it for a small business and how many hours a skill actually saves.