4 Claude Code Commands I Use Every Day: The Cheat Sheet
Get Claude Code to learn your project, check things on a timer, tidy its own code and stop asking about the safe stuff.
Back to all free guidesSo in the reel I showed the four Claude Code commands I use literally every day: /init, /loop, /simplify and /fewer-permission-prompts. Here's the little cheat sheet I promised, with what each one does and how I actually use it. They all come built into Claude Code, so there's nothing to install, you just type them in like a normal message.
Before you start
- Claude Code. Type
claude --versionin your terminal to check you've got it. If a command doesn't come up when you type/, runclaude updatefirst. - Start it in your project folder. Go into the folder in your terminal and type
claude. /init and /fewer-permission-prompts both work on the project you start it in. - Keep the session open for /loop. It runs inside your Claude Code session, so leave that window open while it does its checks.
/init, so Claude knows your project
- What it does. Claude scans your project and writes a CLAUDE.md file, basically a cheat sheet on how it works, with the commands to build and test it and how the code fits together. Claude reads that file at the start of every session, so it's not starting from scratch each time you open it.
- When I use it. The first time I open Claude Code in a project. If there's already a CLAUDE.md it suggests improvements instead of writing over it, so you can run it again whenever the project's moved on.
- What to type. Just this:
/init- Then add what it can't see. Type
/memory, pick the CLAUDE.md and add anything Claude wouldn't work out from the code on its own. Anthropic's advice is to keep the file under 200 lines. - Want it to ask you questions first? Start Claude Code with
CLAUDE_CODE_NEW_INIT=1 claudeinstead. Then /init asks what you want set up (CLAUDE.md files, skills and hooks), asks you follow-up questions and shows you a proposal before it writes anything.
/loop, to check things on a timer
- What it does. It runs a prompt over and over on a timer while your session's open. The first run happens straight away, then it repeats on the schedule.
- When I use it. I have it check a deploy every five minutes while I crack on with something else.
- What to type. /loop, then how often, then what you want checked. Use m for minutes, h for hours or d for days.
/loop 5m check if the deploy finished and tell me what happened- Make it yours. Change the 5m to how often you want it, and everything after it to whatever you want checked. Putting "every 2 hours" on the end works too.
- Leave the time off and Claude picks the gap itself, anything from a minute to an hour. It checks more often while something's still running and backs off when it's all quiet.
- To stop it, press Esc if Claude's pacing the loop itself. If it's on a fixed timer, just ask Claude to "cancel the deploy check job".
/simplify, a second pair of eyes before you ship
- What it does. It goes back over the code that's changed and tidies it up. Four review agents run at once and each one checks something different: existing code it could reuse instead of rewriting, anything that could be simpler, anything wasteful, and whether the change sits at the right level. It won't make a fix that changes what the code does, and at the end you get a summary of what it fixed and what it skipped.
- When I use it. After Claude's written a load of code, right before I ship it.
- What to type. On its own, it looks at everything that's changed:
/simplify- Point it at one thing. Add a file, folder or pull request after it, like
/simplify src/checkout, to review just that. - Hunting for bugs? That's a different command,
/code-review. /simplify is purely about tidying.
/fewer-permission-prompts, so it stops asking every two seconds
- What it does. It looks back through your recent Claude Code sessions for the commands and tools Claude keeps using. It keeps the read-only ones, the ones that just look at things without changing anything, and adds them to an allow list in your project's
.claude/settings.json, so Claude stops asking you about them. - What it leaves out. Anything that writes, deletes, pushes or installs, anything that could run whatever code it likes, like
pythonornpx, and commands Claude Code already runs without asking, likelsorgit status. - What you get. A ranked table of up to about 20 commands and how often each one came up. Then it updates
.claude/settings.jsonand tells you what it added and what it skipped. - When I use it. Whenever I catch myself approving the same thing for the tenth time. To be honest, it's my favourite of the four.
- What to type. Just this, in the project you want it for:
/fewer-permission-prompts- See what it added. Type
/permissionsto look through the rules and remove any you don't want.
The cheat sheet
All four in one place, and every other built-in command is in Anthropic's commands reference. Send each one as its own message, because Claude Code only picks up a command at the start of a message.
/init
/loop 5m check if the deploy finished and tell me what happened
/simplify
/fewer-permission-promptsRun /init on whatever you're working on today, it takes a minute and it's pretty sweet having Claude actually know the project from then on.