Claude Code slash commands: the cheat sheet, and how to write your own

How slash commands work

Type / in Claude Code to see every command available to you, and keep typing to filter the list. A command only counts at the start of a message; whatever follows its name becomes its arguments.

• Built-in commands are coded into the CLI, such as /clear, /model and /compact.
• Bundled skills ship with Claude Code and work like prompts handed to Claude, such as /simplify, /loop and /batch.
• Custom commands are the ones you write: skills in .claude/skills/ or older files in .claude/commands/. Plugins and MCP servers can add more.
• Send a command while Claude is still answering and it runs after the current turn; a few, such as /status and /usage, run straight away.
• Not every command shows for everyone: some depend on your platform, plan or organization's policy.

Claude Code slash commands cheat sheet

The commands most people use, in the order a session tends to need them.

How to create a custom slash command

Custom commands have been merged into skills. A file at .claude/commands/fix-issue.md and a skill at .claude/skills/fix-issue/SKILL.md both create /fix-issue and work the same way; the skill format is preferred for new commands because it can hold supporting files.

1. Create the folder and SKILL.md

   For a command the whole team gets, create .claude/skills/<name>/SKILL.md in the repository and commit it. For one only you use in every project, put it in ~/.claude/skills/<name>/SKILL.md. The folder name becomes the command: fix-issue gives /fix-issue.

2. Write the frontmatter

   Between --- lines, add a description (Claude uses it to decide when the skill applies), argument-hint to show expected arguments in the menu, disable-model-invocation: true if only you should trigger it (for anything with side effects, such as deploys), and allowed-tools to pre-approve tools for that turn.

3. Write the prompt, with arguments

   Below the frontmatter, write the instructions as Markdown. $ARGUMENTS is replaced by everything typed after the command; $0, $1 and so on (or $ARGUMENTS[0]) pick single arguments. A line with ` !command ` runs that shell command first and pastes its output into the prompt.

   ---
   description: Fix a GitHub issue by number, with a test
   argument-hint: [issue-number]
   disable-model-invocation: true
   allowed-tools: Bash(gh issue view *)
   ---
   
   Issue: !`gh issue view $ARGUMENTS`
   
   Fix this issue following our conventions in CLAUDE.md.
   1. Write a failing test that reproduces it.
   2. Make the smallest change that passes.
   3. Run the test suite and summarize what changed.

4. Run it

   Type /fix-issue 123. Skills you add or change during a session load on their own; if one doesn't show up, run /reload-skills. Without disable-model-invocation, Claude can also run the skill by itself when your request matches its description.

5. Use positional arguments for multi-part commands

   Running /migrate-component SearchBar React Vue fills $0, $1 and $2 in order.

   ---
   description: Migrate a component between frameworks
   argument-hint: [component] [from] [to]
   ---
   
   Migrate the $0 component from $1 to $2.
   Keep every existing behavior and test passing.

Source: Claude Code docs, Commands and Skills

Where custom slash commands live

Slash commands vs skills

Since the merge, the question is less "which one" than "which file format". If a skill and a command file share a name, the skill wins.

Tips and gotchas

• At the terminal prompt you can stack skills: /write-tests /fix-issue 123 loads both and passes 123 to each, up to six. In claude -p and the Agent SDK only the first one loads.
• Built-in commands you remember may be gone: /pr-comments, /vim and /ultraplan have been removed. Check /help or the docs.
• allowed-tools grants permission only for the turn that invokes the skill, and repository skills can grant themselves tools, so read them before running Claude Code in an unfamiliar repo.
• Every skill's description sits in context; /skill-doctor shows what each costs and how often it's used. Long procedures belong in skills, not in CLAUDE.md.
• /loop repeats a prompt only while your session is open. To have work start on its own, see how to make Claude Code proactive; to keep it going overnight, see how to run Claude Code 24/7.

Slash commands make you faster. Crewmark makes Claude Code start without you.

Every slash command still needs you at the keyboard to type it. Crewmark turns a Claude Code, Codex, DeepSeek Harness or OpenCode session into an AI employee with a role, a goal and working hours, and wakes it on a heartbeat to take the next issue from a board, using the same skills and commands you've set up.

1. Keep your skills in the repository

   Your .claude/skills/ and CLAUDE.md load in the employee's session like in yours, so /fix-issue and your conventions come along.

2. Let the heartbeat start the work

   During its working hours the employee wakes every 15, 30 or 60 minutes in the same conversation and takes the next issue. Several can share a repository, each in its own worktree.

3. Assign issues and answer its messages

   Put work on the board; the employee comments as it goes and messages you when it needs a decision. Read what an AI employee is.

Claude Code slash commands FAQ