Claude Code has dozens of slash commands. A handful keep me in flow, so those come first. The rest are organized by the moment you need them, from setting up a project to shipping and automating the work.
My go-to five
These are the built-in commands and settings that show up most in my own Claude Code history since June 2026. Each one saves me a stop: an approval, a copy-paste or a restart.
| What I use | Why it stays in my routine |
|---|---|
/clear | I start fresh for each new task instead of letting unrelated work pile into one conversation |
! shell mode | I run a command myself and Claude reads the output, without me pasting anything |
/mcp | I connect the servers a project needs and turn off the ones it doesn't |
/plugin and /reload-skills | I build and edit skills often, and these load changes without restarting |
auto permission mode | Most of my sessions run in auto mode, so long tasks don't stop for every approval |
Most of what I build runs through custom skills, which you can make with the template in Repeat and automate.
Three prefixes to learn first
| Type | What happens | Example |
|---|---|---|
/ | Runs a command or skill. It only works at the start of a message | /context |
! | Runs a shell command yourself. The output goes into the conversation and Claude responds to it | ! npm test |
@ | Mentions a file with path autocomplete | explain @src/auth.ts |
Press ? on an empty prompt to see shortcuts. If you send a command while Claude is working, it waits until the turn ends. A few, like /usage, run right away.
Set up a project
| Command | What it does | Use it when |
|---|---|---|
/init | Generates a starter CLAUDE.md for the project | It's your first session in a repository |
/memory | Edits CLAUDE.md files and turns auto memory on or off | Claude keeps missing a project convention |
/permissions | Manages allow, ask and deny rules for tools | You approve the same safe command over and over |
/mcp | Connects, signs in to, enables or disables MCP servers | You add a server or want to trim unused ones |
/model | Switches the model and saves it as your default | A task needs more capability or more speed |
/import | Brings settings from Codex, Gemini CLI or Cursor | You're moving over from another coding agent |
Plan and run a task
| Command | What it does | Use it when |
|---|---|---|
/plan [task] | Enters plan mode so Claude investigates and proposes before editing | The change is large or touches unfamiliar code |
/goal <condition> | Keeps Claude working across turns until a separate model confirms the condition is met | The end state is verifiable, like "all tests pass" |
/effort <level> | Sets reasoning effort, from low to max | The task is harder or simpler than usual |
/fast | Toggles fast mode for quicker output | You're iterating on small changes |
Ctrl+B | Moves a running command to the background | A build or test run is blocking the session |
Permission modes decide what Claude does without asking. Press Shift+Tab to cycle through default, acceptEdits and plan.
| Mode | What runs without asking | Best for |
|---|---|---|
default | Reading files | Sensitive work where you review every action |
acceptEdits | File edits and common filesystem commands | Iterating on code you're watching |
plan | Reading and planning | Exploring before changing anything |
auto | Everything, with background safety checks | Long tasks with fewer interruptions |
dontAsk | Only pre-approved tools. Everything else is denied | Scripts and CI |
bypassPermissions | Everything | Isolated containers and VMs only |
Keep context clean
| Command | What it does | Use it when |
|---|---|---|
/context | Shows what's filling the context window, with suggestions | Responses slow down or quality drops |
/compact [focus] | Summarizes the conversation to free space. Add what the summary should keep | A long task still has work left |
/clear | Starts a fresh conversation. Project memory stays | You're switching to an unrelated task |
/btw <question> | Answers a side question from what's already in context, without adding to history | You need a quick answer mid-task |
/btw works even while Claude is busy, but it can't read files or run commands. To add instructions mid-task, type your message and press Enter. Claude Code queues it for Claude. Press Esc to interrupt instead.
Undo, branch and resume
| Command | What it does | Use it when |
|---|---|---|
/rewind or Esc Esc | Rolls code, conversation or both back to an earlier checkpoint | Claude went the wrong way |
/branch [name] | Branches the conversation at this point | You want to try another approach without losing this one |
/resume | Opens a picker of earlier sessions | You're returning to past work |
/rename [name] | Names the session so you can find it later | The session matters beyond today |
/export | Saves the conversation as plain text | You want a record outside Claude Code |
Checkpoints have limits. They don't track files changed by shell commands or edits made outside Claude Code. Commit to git before risky work.
From your terminal, claude -c continues the most recent conversation in the current folder, and claude -r "<name>" resumes a specific one.
Run work in parallel
| Command | What it does | Use it when |
|---|---|---|
/tasks | Lists background work in this session, including finished subagents | You want to check what's still running |
/subtask <task> | Starts a background subagent that inherits the full conversation | A side task needs the same context |
/fork [prompt] | Copies the conversation into a new background session | You want to explore two directions at once |
/background | Detaches the whole session to keep running and frees your terminal | A long task doesn't need you watching |
/batch <instruction> | Splits a codebase-wide change into parts that run in separate worktrees | A migration or refactor touches many files |
Starting Claude with claude -w puts the session in its own git worktree, so parallel sessions don't edit the same files.
Review before you ship
| Command | What it does | Use it when |
|---|---|---|
/diff | Shows changes in your working tree, including Claude's edits | Before you commit |
/code-review [level] [--fix] | Checks the diff for bugs and cleanups. --fix applies the findings. Pass a PR number to review a pull request | Before you open or merge a PR |
/security-review | Checks your branch's changes for security vulnerabilities | The change touches auth, input handling or data |
/simplify | Reviews changed code for cleanup and applies the fixes | The code works but feels heavier than it should |
/verify | Builds and runs your app to confirm the change actually works | Tests pass but you want proof in the real app |
/code-review ultra runs a deeper multi-agent review in the cloud.
Repeat and automate
| Command | What it does | Use it when |
|---|---|---|
/loop [interval] [prompt] | Repeats a prompt while the session stays open | You're watching a deploy or a flaky test |
/schedule | Creates cloud routines that run with no open session | Work should happen nightly or every morning |
/hooks | Shows hooks that run scripts on events, like after every edit | A rule must hold every time, not just when Claude remembers |
claude -p "prompt" | Runs one prompt without the interactive interface and exits | You're scripting Claude Code or running it in CI |
| Option | Runs on | Needs an open session | Shortest interval |
|---|---|---|---|
/loop | Your machine | Yes | 1 minute |
/schedule | The cloud, on a fresh clone | No | 1 hour |
Make your own command. Save instructions to .claude/skills/<name>/SKILL.md and they become /<name>. Older files in .claude/commands/ still work the same way.
---
name: pre-pr
description: Prepare this branch for a pull request.
---
1. Run the test suite and fix any failures.
2. Review the diff for bugs, missing tests and leftover debug code.
3. Draft a PR description: what changed, why, and how to test it.
Ask me before pushing or opening the PR.
Check usage and setup
| Command | What it does | Use it when |
|---|---|---|
/usage | Shows session cost, plan limits and activity | You're near a limit or tracking spend |
/insights | Generates a report on how you've used Claude Code recently | You want to find patterns in your sessions |
/skill-doctor | Shows each skill's context cost and how often it's used | Your setup feels bloated |
/doctor | Checks your installation and configuration and can fix issues | Something isn't loading or behaving right |
/release-notes | Shows what changed in each version | A command behaves differently than you remember |
Keyboard shortcuts
| Shortcut | Action |
|---|---|
Esc | Interrupt Claude or close a dialog |
Esc Esc | Clear your draft, or rewind |
Shift+Tab | Cycle permission modes |
Option+P / Alt+P | Switch model |
Option+T / Alt+T | Toggle extended thinking |
Ctrl+B | Send a running task to the background |
Ctrl+O | Open the transcript viewer |
Ctrl+R | Search your prompt history |
Ctrl+G | Write your prompt in your text editor |
Ctrl+S | Stash your prompt and restore it later |
Ctrl+V | Paste an image |
\ then Enter, or Ctrl+J | New line without sending |
On macOS, Option shortcuts require setting Option as Meta in your terminal.
Go deeper: Claude Code commands reference · MCP vs CLI · Claude, End to End · Routine vs Schedule vs Loop