Claude Code Commands: The Complete Cheat Sheet (November 2026)

Claude Code commands come in four kinds: slash commands you type inside a session (/clear, /model, /code-review), CLI commands and flags you pass when you start claude from your shell (claude -p, claude -c, --model), keyboard shortcuts and input prefixes (Esc Esc, Shift+Tab, !, @), and custom commands you write yourself as Markdown files. This page lists all of them in scannable tables first, then shows how to build your own, with seven complete custom commands you can paste into a repo today. Everything here is checked against the Claude Code docs in November 2026 (up to Claude Code v2.1.319), so it also flags what changed: custom commands merged into skills, /vim and /pr-comments were removed, and the # memory shortcut is gone.

What are Claude Code commands?

A Claude Code command is anything you type that controls the session instead of talking to the model. The commands reference splits slash commands into two families:

• Built-in commands run fixed logic coded into the CLI. /clear, /model and /permissions are examples.
• Bundled skills are prompts that ship with Claude Code and hand Claude a detailed procedure. /code-review, /batch, /debug, /loop and /simplify are bundled skills. They look and feel like commands, and Claude can also load some of them on its own when relevant.

A few rules apply to all of them:

• A command only runs when it is at the start of your message. Text after the name becomes its arguments, as in /compact focus on the auth module.
• If you send a command while Claude is working, it is queued and runs after the current turn. A few, such as /status, /tasks and /usage, run right away.
• Not every command shows for every user. Availability depends on platform, plan and provider. /desktop, for example, needs macOS or x64 Windows and a Claude subscription.
• Typing / mid-prompt (after a space) offers completions too, but the command only runs if it starts the message.

If you are new to the tool, start with How to use Claude Code, then come back here as a reference.

Claude Code slash commands: the full list by purpose

Below is every current command from the official commands table, grouped by what you use it for. Skill marks a bundled skill. Aliases are in parentheses. <arg> is required and [arg] is optional.

Session and context

Model, settings and project config

Account, usage and updates

For what plans and API usage cost, see Claude Code pricing.

Code review and git

Agents, background work and cloud

Tools, MCP, plugins and skills

Help and diagnostics

Also in the menu, used less often: /recap (one-line session summary), /focus (focus view in fullscreen), /scroll-speed, /powerup (interactive feature lessons), /heapdump (memory snapshot, hidden until typed in full), /remote-env (default cloud environment), /web-setup (GitHub for cloud sessions), /install-slack-app, /artifact-capabilities (skill), /design consent and /design revoke, /passes (share a free week), /stickers and /radio.

Claude Code CLI commands and flags

The second half of "Claude Code commands" is what you type in your shell. The CLI reference lists these subcommands:

The flags worth memorizing

Here is how these combine in real scripts:

# One-shot question about the repo, plain text out
claude -p "Summarize what changed in the last 5 commits" 

# Read-only review in CI: no edits, git read commands pre-approved, JSON out, capped
git diff origin/main...HEAD | claude -p "Review this diff for bugs. Be terse." \
  --permission-mode plan \
  --allowedTools "Bash(git log *)" "Bash(git diff *)" "Read" \
  --output-format json --max-turns 5 --max-budget-usd 2.00

# Structured output you can parse
claude -p "List the TODO comments in src/" \
  --json-schema '{"type":"object","properties":{"todos":{"type":"array","items":{"type":"string"}}},"required":["todos"]}'

# Run one of your custom commands (skills) headless
claude -p "/review-diff"

# Pick up yesterday's named session and give it the next step
claude -r "auth-refactor" "Now write the migration"

Claude Code shortcuts and input prefixes

These come from the interactive mode docs. On macOS, the Alt word-editing shortcuts need "Option as Meta" turned on in your terminal.

Input prefixes change what a message is:

You can remap most shortcuts in ~/.claude/keybindings.json; run /keybindings to open it. The word-editing shortcuts such as Alt+B, Alt+F and Ctrl+W can't be remapped.

How do custom slash commands work in Claude Code?

Claude Code custom commands and skills are now the same thing. The skills docs say it plainly: a file at .claude/commands/deploy.md and a skill at .claude/skills/deploy/SKILL.md both create /deploy and work the same way. The merge landed in v2.1.3. Your existing command files keep working; skills add a folder for supporting files and let Claude load them on its own when the description matches.

1. v2.0.70The # quick-memory shortcut is removed.
2. v2.1.3Slash commands and skills merge, with no change in behavior.
3. v2.1.91/pr-comments is removed.
4. v2.1.92/vim is removed; Vim mode moves to /config.
5. v2.1.223/review becomes an alias of /code-review.

Where to put a custom command

If two share a name, enterprise beats personal and personal beats project. A skill beats a .claude/commands/ file with the same name. A skill named like a built-in (say usage) replaces the built-in in local sessions, but not its aliases. Claude Code watches skill folders, so edits apply without a restart; run /reload-skills if you created a brand-new skills folder mid-session.

Arguments: $ARGUMENTS, $0, $1 and named arguments

Frontmatter fields you'll actually use

All fields are optional. The ones that matter for commands, from the frontmatter reference:

Two more features make commands much better:

• Live shell output. A line like !`git diff --cached` runs before Claude sees the prompt and is replaced by the output. Use a fenced block opened with ```! for multi-line scripts.
• The permission catch. Injected commands never prompt. Outside auto mode, a command that your rules don't already allow aborts the whole skill. So list every injected command in allowed-tools, as the examples below do.

7 custom Claude Code commands you can paste

Each one is a project skill. Create the folder, paste the file, then type the command. They are written for git and the GitHub CLI (gh); swap in your own tools where needed.

1. /commit: write the commit message from staged changes

.claude/skills/commit/SKILL.md

---
description: Write a conventional commit message for the staged changes and commit them.
disable-model-invocation: true
model: haiku
allowed-tools: Bash(git status *) Bash(git diff *) Bash(git commit *)
---

Staged changes:
!`git status --short`
!`git diff --cached`

Task:
1. If nothing is staged, say so and stop.
2. Write one commit message in Conventional Commits format: `type(scope): summary` under 72 characters, a blank line, then 1-4 bullets on why.
3. Extra guidance from me, if any: $ARGUMENTS
4. Run `git commit -m` with that message. Do not stage other files and do not push.

Use it as /commit or /commit mention the ticket ABC-123. The model: haiku line makes this cheap; delete it to keep your session model.

2. /pr-description: draft the pull request body

.claude/skills/pr-description/SKILL.md

---
description: Draft a pull request title and description for the current branch.
disable-model-invocation: true
argument-hint: [base-branch]
allowed-tools: Bash(git log *) Bash(git diff *) Bash(git branch *)
---

Base branch: $ARGUMENTS (use `main` if empty).

1. Run `git log --oneline <base>..HEAD` and `git diff <base>...HEAD --stat`, then read the full diff of the important files.
2. Output, in Markdown:
   - A title under 70 characters
   - **What**: 2-4 bullets
   - **Why**: one short paragraph
   - **How to test**: numbered steps
   - **Risks**: anything a reviewer should look at closely
3. Do not create the PR. Print the text so I can paste it.

3. /fix-issue: fix GitHub issue #N

.claude/skills/fix-issue/SKILL.md

---
description: Fix a GitHub issue by number, with a test.
disable-model-invocation: true
argument-hint: [issue-number]
allowed-tools: Bash(gh issue view *)
---

Fix GitHub issue #$0.

1. Run `gh issue view $0 --comments` and restate the problem in two sentences.
2. Find the relevant code. If the cause is unclear, tell me what you'd check before editing.
3. Write a failing test that reproduces the bug, then make it pass.
4. Run the test suite for the files you touched.
5. Summarize the change and suggest a commit message that ends with `Fixes #$0`.

Run /fix-issue 482. Because gh issue view is pre-approved, Claude reads the issue without stopping to ask.

4. /explain: explain a file or folder

This one uses the legacy single-file format to show it still works.

.claude/commands/explain.md

---
description: Explain what a file or directory does and how it fits the codebase.
argument-hint: [path]
allowed-tools: Read Grep Glob
---

Explain $ARGUMENTS to a developer who is new to this repo.

- Start with one sentence on its job.
- Then: main exports or entry points, what calls it, what it depends on, and any side effects (network, disk, database).
- Point out anything surprising or risky.
- Keep it under 300 words. Use file:line references.

5. /write-tests: add tests for a file

.claude/skills/write-tests/SKILL.md

---
description: Write unit tests for a file using the project's existing test setup. Use when asked to add or improve tests.
argument-hint: [file]
arguments: [file]
---

Write tests for $file.

1. Find the test framework, config and an existing test file near $file. Copy its style, naming and helpers.
2. Cover the happy path, edge cases (empty, null, large input) and error paths. Don't test private details.
3. Put the tests where this repo keeps them.
4. Run only the new tests and fix failures in the tests, not in $file. If $file itself looks wrong, tell me instead of changing it.

Since disable-model-invocation isn't set, Claude can also load this skill when you just say "add tests for the parser".

6. /review-diff: review uncommitted changes

.claude/skills/review-diff/SKILL.md

---
description: Review my uncommitted changes before I commit.
disable-model-invocation: true
context: fork
agent: Explore
allowed-tools: Bash(git diff *) Bash(git status *)
---

Changes:
!`git status --short`
!`git diff HEAD`

Review. List problems in this order, each with file:line and a one-line fix:
1. Bugs and broken logic
2. Security issues (secrets, injection, missing auth checks)
3. Missing or weak tests
4. Naming and readability, only if important

If everything looks fine, say so in one line. Do not edit files.

context: fork runs the review in a separate Explore subagent that reports back when done, so the diff doesn't fill your main context. For a deeper review that can also fix things, the bundled /code-review is the better tool; this is the fast, read-only version.

7. /changelog: release notes since the last tag

.claude/skills/changelog/SKILL.md

---
description: Write user-facing release notes from git history.
disable-model-invocation: true
argument-hint: [from-ref] [to-ref]
arguments: [from, to]
allowed-tools: Bash(git describe *) Bash(git log *) Bash(git tag *)
---

From: $from (if empty, use the latest tag from `git describe --tags --abbrev=0`)
To: $to (if empty, use HEAD)

1. Run `git log --no-merges --pretty=format:'%h %s' <from>..<to>`.
2. Group entries under **New**, **Improved**, **Fixed** and **Breaking changes**. Drop chores, CI and refactors users won't notice.
3. Rewrite each line for users, not developers. One line each.
4. Output Markdown only, ready to paste into a GitHub release.

Run /changelog for everything since the last tag, or /changelog v1.4.0 v1.5.0 for a range. It uses named arguments because an unfilled $0 or $1 stays in the prompt as literal text, while an unfilled named argument becomes empty. It isn't called /release-notes because a project skill with a built-in command's name replaces that command. Want more ideas? The agent skills directory lists community skills you can install.

MCP prompts as slash commands

MCP servers can expose prompts, and Claude Code turns them into commands. They appear in the / menu as /servername:promptname (MCP), and typing /mcp__servername__promptname also runs them. Arguments are space-separated, per the MCP docs:

/mcp__github__list_prs
/mcp__github__pr_review 456
/mcp__jira__create_issue login-bug high

Prompts are discovered from connected servers, so what you see depends on what you've added. Find servers in the MCP server directory, for example MCP servers for GitHub or for Jira. If a server's prompts are missing, check /mcp for its connection status, or run /mcp reconnect all.

Plugin commands

Plugins bundle skills, agents, hooks and MCP servers. Every skill in a plugin becomes a namespaced command, /<plugin>:<skill>, so it can't clash with yours. The plugin install docs use Anthropic's commit-commands plugin as the example:

/plugin install commit-commands@claude-plugins-official
/commit-commands:commit

In your shell, the same install is claude plugin install commit-commands@claude-plugins-official. You can also type just the bare skill name: /deploy finds a plugin skill named myplugin:deploy-app in autocomplete. Plugins can still ship a legacy commands/ folder (commands/db/migrate.md becomes /my-plugin:db:migrate), but the docs recommend skills for new work. Keep in mind that the descriptions of every enabled plugin's skills that Claude can invoke on its own sit in context on every turn, so disable plugins you don't use.

Which Claude Code commands should you learn first?

To compare Claude Code's command set with other agents, see the AI coding agents comparison.

FAQ

How do I see all Claude Code commands?

Type / in a Claude Code session. The menu lists built-in commands, bundled skills, your own skills and command files, plugin commands and MCP prompts, and filters as you type. /help shows the same list. A few commands, such as /heapdump, stay hidden until you type the full name.

How do I use commands in Claude Code?

Type the command at the start of your message, followed by any arguments, for example /compact keep the API decisions. Commands typed later in a message don't run directly. From your shell, use CLI commands such as claude -p "query" or claude -c instead.

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

Create .claude/skills/<name>/SKILL.md in your repo (or ~/.claude/skills/ for every project), add a description in YAML frontmatter, and write the prompt below it. It runs as /<name>. A single Markdown file in .claude/commands/ also works, because custom commands have been merged into skills.

What is the difference between Claude Code commands and skills?

There is no longer a real difference for your own files: both create a /name command and accept the same frontmatter. Skills add a folder for scripts and reference files, the name and paths fields, and automatic loading by Claude when the description matches. Built-in commands like /clear are different: they run fixed code, not a prompt.

What happened to the # memory shortcut in Claude Code?

It was removed in v2.0.70. To save something to memory, ask Claude to add it to CLAUDE.md or open the files with /memory.

How do I run Claude Code commands from the command line?

Use print mode: claude -p "/review-diff" runs a custom command non-interactively and prints the result. Built-in commands that need a dialog won't work there, but several support -p, including /config key=value, /model <model>, /rename and /mcp.

What is the difference between /clear and /compact?

/clear starts a fresh conversation with empty context; the old one stays available in /resume. /compact keeps the same conversation but replaces its history with a summary, and you can tell it what to keep: /compact focus on the database schema.