Three Claude Code commands make Claude finish the whole job on its own, so you stop being the bottleneck. Here’s the cheat sheet, with copy-paste examples.
Most people babysit Claude. They give it a task, then sit there nudging it and retyping "keep going" until it finishes. That turns you into the bottleneck.
Three commands fix it. /goal keeps Claude working until a finish line you set is true. /loop repeats a task on a timer while your session is open. /schedule runs a task in the cloud on a cadence, even when your laptop is closed. All three live in Claude Code, which is included with Claude Pro and up. (/goal docs, /loop docs, /schedule docs)
This is the big one. Type /goal and a finish line, and Claude starts working right away. After every turn, a second model checks your finish line. If it isn’t met, Claude keeps going on its own. Type /goal alone to see the status and /goal clear to cancel.
One catch: in the normal permission mode, Claude still stops to ask before actions you haven’t allowed. To actually walk away, run it in auto mode.
The One Rule That Makes It Work
Give it one measurable finish line. The checker only reads what Claude shows in the chat, so your finish line has to be something Claude can prove there. "Plan my upcoming trip" never ends. "Plan my 5-day trip with a morning, afternoon and dinner plan for every day, nothing blank, and show me all 5 days before you stop" has a clear end. Name the thing that has to exist at the end, and tell Claude to show it to you before stopping.
01
Plan a trip with nothing left blank
For a complete itinerary instead of a rough sketch you have to finish.
02
Get your inbox to zero
For email that’s actually handled. Connect Gmail first.
03
Finish a research deliverable
For a finished comparison with no missing pieces.
04
Organize a cluttered folder
For a folder that’s actually sorted, with proof.
Where /goal works toward a finish line, /loop runs the same task on a timer. Type the interval, then the task. Leave the interval off and Claude picks its own pace, anywhere from a minute to an hour.
01
Watch for something to finish
Runs every 5 minutes until you cancel it.
02
Repeat a routine check
Runs every hour. Needs the form, sheet or channel connected.
Good To Know
/loop runs in your open session, so keep it open. To stop a timed loop like /loop 5m, ask Claude to cancel it. Esc only stops a loop where Claude picks the pace. Repeating loops delete themselves after 7 days.
This is the hands-off one. /schedule saves a task that runs in Anthropic’s cloud on a schedule, so it keeps running with your laptop closed. Describe the timing in plain English and Claude asks a few questions, then saves it. Type /schedule list to see your tasks and /schedule run to start one now. You can also manage them at claude.ai/code/routines.
01
A morning triage
Uses your connected Gmail and Slack.
02
A weekly file cleanup
Cloud tasks can’t see the files on your computer, so point this at a Google Drive folder.
Be Honest About This One
Scheduled tasks are on Pro ($20 a month, or $17 billed yearly), Max, Ultra, Team and Enterprise. The fastest cadence is once an hour. Because they run with nobody watching and can use your connectors, never schedule anything that sends or deletes until you’ve tested it a few times by hand. Tip: pick a few minutes past the hour, like 8:07, because runs set exactly on the hour can start late.
The moment you can write a clear finish line, you can walk away and let Claude do the work. That’s the difference between using AI as a chat box and using it as a teammate. If you don’t see one of these commands, update Claude Code.