Claude Code commands: 2026 cheat sheet of slash commands, flags, keys

Claude Code commands come in three kinds: slash commands you type inside a session (like /clear or /plan), CLI commands and flags you pass when you launch claude, and keyboard shortcuts. This cheat sheet covers the ones worth learning, when to reach for each, which old commands were renamed or removed, and how to build your own custom commands.

What are Claude Code commands?

Claude Code commands are the controls you use to steer Anthropic's coding agent, and they come in three kinds: slash commands typed inside a session, CLI commands and flags passed when you start claude from your shell, and keyboard shortcuts. Knowing which kind does what saves you from hunting through menus mid-task.

Type / at the prompt to see the commands available to you; letters after the slash filter the list. Per the official commands reference, a command is only recognized at the start of a message, and any text after the name becomes its arguments. If you send a command while Claude is still responding, it is queued until the turn ends, except for a few like /status, /tasks and /usage that run immediately.

If you have not installed Claude Code yet, start with our Claude Code tutorial and come back.

Which slash commands should you learn first?

Eight slash commands cover most daily work: /init, /plan, /context, /compact, /clear, /rewind, /diff and /code-review. The table below groups the most useful ones by when you need them in a session.

A few of these take arguments that save a step. /plan fix the auth bug enters plan mode and starts on the task in one go. /compact focus on the API changes tells the summary what to keep. /code-review high 1234 reviews pull request 1234 at high effort, and /review is an alias for the same command.

Commands for parallel and background work

Claude Code can now run work beyond the current conversation, and a handful of commands control that:

• /tasks lists background work in the session, including finished subagents.
• /subtask <task> hands a side task to a background subagent that reports back here.
• /fork [prompt] copies the conversation into a new background session while you keep working.
• /branch [name] creates a branch of the conversation and switches you into it.
• /background detaches the whole session so it keeps running and frees your terminal.
• /batch <instruction> splits a large change into 5 to 30 units and runs each in its own git worktree.
• /loop [interval] [prompt] reruns a prompt while the session stays open, for example /loop 5m check if the deploy finished.

/clear vs /compact vs /rewind vs /btw: which one do you need?

Use /clear for a new task, /compact to continue the same task with a shorter history, /rewind to undo a wrong turn, and /btw for a question that shouldn't touch the conversation. People mix these up more than any other commands, and picking the wrong one either throws away useful context or keeps stale context that makes answers worse.

The interactive mode docs note that /btw answers only from what is already in the conversation, works while Claude is still responding, and shows the answer in an overlay you can dismiss. Our guide to the context window explains why a cleaner context produces better answers.

Which Claude Code keyboard shortcuts are worth memorizing?

Ten shortcuts do most of the work: Shift+Tab, Esc, Esc Esc, Ctrl+C, Ctrl+O, Ctrl+R, Ctrl+B, Ctrl+G, Ctrl+S and Ctrl+J. Press ? on an empty prompt to see the full shortcut panel in your own terminal.

For multiline prompts, \ then Enter works in every terminal and Ctrl+J works without configuration. Shift+Enter works natively in iTerm2, WezTerm, Ghostty, Kitty, Warp, Apple Terminal and Windows Terminal; for VS Code, Cursor, Alacritty or Zed, run /terminal-setup once to install it.

Three prompt prefixes act like shortcuts too:

• ! at the start runs a shell command directly, such as ! npm test, and adds its output to the conversation.
• @ triggers file path autocomplete so you can point Claude at exact files.
• / opens the command menu, and also completes commands partway through a prompt.

Which Claude Code CLI commands and flags matter most?

At the shell, claude starts a session, claude -c continues the last one, claude -r <name> resumes a named one, and claude -p "<task>" runs a task without the interactive UI. Flags set the model, permission mode and working directories before the first prompt.

Combined, these cover most scripted and parallel work:

# Pick up yesterday's work by nameclaude -n "auth-refactor"claude -r "auth-refactor" "Finish the remaining tests"# Start a careful session that can't edit until you approve a planclaude --permission-mode plan# One-shot task for a script or CI jobcat build.log | claude -p "Explain why this build failed" --output-format json --max-turns 3# Two features in parallel, each in its own worktreeclaude -w feature-loginclaude -w feature-billing

The permission modes page recommends --dangerously-skip-permissions (the same as --permission-mode bypassPermissions) only inside isolated containers and VMs. For more on -p, output formats and CI, read our guide to Claude Code headless mode.

Background sessions started with claude --bg or /background still run on your own computer, so they pause when it sleeps or shuts down. If you want agents to keep going overnight, Orkio rents an always-on, isolated cloud box where Claude Code runs on your own Claude account, and you check on it from your phone.

How do you create a custom slash command in Claude Code?

To create a custom command, add a SKILL.md file in a folder under .claude/skills/; the folder name becomes the command. The skills docs say custom commands have been merged into skills: .claude/commands/deploy.md and .claude/skills/deploy/SKILL.md both create /deploy, and existing command files keep working.

1. Pick the scope. A project command goes in .claude/skills/<name>/ and is shared through git. A personal one goes in ~/.claude/skills/<name>/ and works in every project.
2. Create the folder, for example mkdir -p .claude/skills/fix-issue.
3. Write SKILL.md with YAML front matter and the instructions.
4. Use $ARGUMENTS for everything typed after the command, or $0, $1 and so on for individual arguments.
5. Add disable-model-invocation: true if only you should trigger it, for anything with side effects such as deploys or commits.
6. Run /reload-skills if the new command doesn't appear in the / menu, then try it.

---description: Fix a GitHub issue by numberargument-hint: "[issue-number]"disable-model-invocation: trueallowed-tools: Bash(git status *) Bash(gh issue view *)---## Working tree right now!`git status --short`## Your taskFix GitHub issue #$0 following the conventions in CLAUDE.md.1. Read the issue with `gh issue view $0` and find the root cause before editing.2. Write a failing test, then make it pass.3. Run the full test suite and summarize what you changed.

Running /fix-issue 123 replaces $0 with 123. The !`git status --short` line runs before Claude sees the prompt, so the current state of the working tree arrives already inlined, and allowed-tools pre-approves both commands so neither stops for a permission prompt. That dynamic context is what makes a custom command better than a pasted prompt.

A file in a subfolder of .claude/commands/, such as frontend/component.md, becomes /frontend:component. Plugin skills appear as /plugin-name:skill-name.

For supporting files, subagent execution and sharing skills across a team, see our full guide to Claude Code skills.

Which Claude Code commands were renamed or removed?

As of November 2026, several commands that older cheat sheets still list have changed: some are now aliases, and a few are gone. If a tutorial tells you to run one of these, use the replacement.

Aliases still worth knowing: /new and /reset for /clear, /undo and /checkpoint for /rewind, /continue for /resume, /settings for /config, and /bg for /background.

A copyable Claude Code command cheat sheet

Keep this block in a note or pin it next to your terminal. It holds the commands from this guide that you will use every week.

START  claude                      new session  claude -c                   continue last conversation here  claude -r <name>            resume a named session  claude -p "task"            run once, print, exit  claude -w <name>            new session in a git worktree  claude --permission-mode planIN A SESSION  /init        create CLAUDE.md          /memory      edit memory  /plan task   plan before editing       /model       switch model  /context     what fills context        /compact     summarize, keep going  /clear       fresh conversation        /resume      reopen old one  /btw q       side question             /rewind      undo to checkpoint  /diff        see changes               /code-review review the diff  /usage       cost and limits           /doctor      fix setup problemsKEYS  Shift+Tab  permission modes    Esc       stop Claude  Esc Esc    rewind menu         Ctrl+O    full transcript  Ctrl+R     search history      Ctrl+B    background a task  Ctrl+G     edit in $EDITOR     Ctrl+J    new line  !cmd       run shell command   @file     mention a file

How to learn Claude Code commands this week

Learn commands by attaching them to moments in your workflow rather than memorizing the list. A week of this is enough for them to become habits.

1. Day one: run /init in your main repository, then /memory to trim what it wrote. Our CLAUDE.md best practices show what belongs there.
2. Start every multi-file change with /plan or Shift+Tab into plan mode.
3. Run /context once a day to see what fills the window, and /clear whenever you change tasks.
4. Use Esc the moment Claude heads the wrong way instead of waiting, and Esc Esc to rewind.
5. End each task with /diff and /code-review before you commit.
6. When you paste the same prompt a third time, turn it into a skill in .claude/skills/.
7. Once a week, run /release-notes to see what changed, since commands are added and retired often.

Frequently asked questions