Claude Code Slash CommandsBuilt-In and Custom Commands

Slash commands are shortcuts you type at the start of a message, beginning with a forward slash, to control Claude Code or trigger a saved workflow. Type / in the prompt to see every command available to you. There are two kinds: built-in commands like /clear, /compact, and /init that are coded into the CLI, and custom commands you create yourself as reusable prompt templates. A command is only recognized at the start of your message; any text after the command name is passed to it as an argument.

Create a markdown file in a .claude/commands/ folder (for example .claude/commands/review.md) and the file becomes the command /review. The file's contents are the prompt Claude runs when you invoke it. As of the 2026 docs, custom commands have been merged into skills, so the recommended modern path is .claude/skills/review/SKILL.md instead, which produces the same /review command but adds support for a directory of supporting files and richer frontmatter. Both forms still work and you do not have to migrate existing command files.

Project commands live in .claude/commands/ or .claude/skills/ inside the repository, so they are committed to version control and shared with everyone on the team. Personal commands live in ~/.claude/commands/ or ~/.claude/skills/ in your home directory, so they follow you across every project on your machine. When a skill and a command share the same name, the skill takes precedence. When the same name exists at multiple scopes, personal overrides project.

Put the placeholder $ARGUMENTS in your command file and it gets replaced with whatever you type after the command name. For example, if /fix-issue contains 'Fix GitHub issue $ARGUMENTS', then running /fix-issue 123 sends 'Fix GitHub issue 123' to Claude. For individual positional arguments use $1, $2 (or $ARGUMENTS[0], $ARGUMENTS[1]), and wrap multi-word values in quotes. You can also declare named arguments in the frontmatter using the arguments field.

They overlap. A custom slash command is a markdown file you invoke by typing /name. A skill is a SKILL.md file in its own directory that you can also invoke with /name, but Claude can additionally load it automatically when your request matches its description. In the current Claude Code, custom commands have been folded into skills: a file at .claude/commands/deploy.md and a skill at .claude/skills/deploy/SKILL.md both create /deploy and behave the same way when you type the command. Skills add optional features like supporting files, invocation control, and subagent execution.

Yes, for skills with a clear description. When a skill's description matches what you are asking for, Claude can load and run it on its own. If you want a command to be manual-only, for example a deploy or commit workflow where timing matters, add disable-model-invocation: true to the frontmatter so only you can trigger it by typing /name.

Use the inline syntax with an exclamation mark and backticks, such as !`git diff HEAD`, on its own line in the command file. Claude Code runs that shell command before the prompt is sent and replaces the placeholder with the command's output, so the prompt arrives with live data already inlined. This is preprocessing, not something Claude decides to do. It is how you ground a command in your actual working tree, like injecting the current diff into a commit-message command.

Start with /clear to begin a fresh task while keeping project memory, /compact to shrink a long conversation that is filling the context window, /init to generate a starter CLAUDE.md, /plan to enter plan mode before a large change, /model to switch models, and /review or /security-review for a read-only pass before you ship. /context shows where your context window is going, and /agents and /mcp set up subagents and MCP servers. Type /help to see the full list available in your session.