Claude Code slash commands: the built-in ones and how to write your own

Before you start

You need Claude Code from lesson 1 and a project that uses git, because the sample command reads your uncommitted changes. Slash commands run inside a session. The shell commands from lesson 1, such as claude -p, run from the terminal instead.

Write your first custom command

Anthropic's skills page builds its first example in three moves, and this lesson follows it. The skill below summarizes your uncommitted changes. The line that starts with an exclamation mark runs git diff HEAD, and Claude Code puts the output into the skill before Claude reads it.

---
description: Summarizes uncommitted changes and flags anything risky. Use when the user asks what changed, wants a commit message, or asks to review their diff.
---

## Current changes

!`git diff HEAD`

## Instructions

Summarize the changes above in two or three bullet points, then list any risks you notice such as missing error handling, hardcoded values, or tests that need updating. If the diff is empty, say there are no uncommitted changes.

1. Make a folder for the command in your personal skills folder: mkdir -p ~/.claude/skills/summarize-changes. Personal skills work in all your projects. On Windows, create the same folder inside your user folder.
2. Save the text above as SKILL.md in that folder. The folder name becomes the command, so this one is /summarize-changes.
3. Open a git project, change one file and run claude.
4. Run the command directly with /summarize-changes. You can also ask "What did I change?" and let Claude load the skill because the description matches.
5. Edit the description or the instructions and run the command again. Claude Code picks up the change in the same session.

The built-in commands to learn first

The commands reference groups its commands by when you reach for them. In a new repository you run /init for a starter CLAUDE.md, then /memory to refine it. During a task, /plan and /model change how Claude works. When the conversation gets long, /context and /compact deal with the space.

Some entries in that list are bundled skills, such as /doctor and /code-review. They are prompts handed to Claude, and you run them the same way as the commands that execute fixed logic. Which commands you see depends on your platform, plan and environment.

Custom commands are skills now

The older way to make a command was a Markdown file in .claude/commands/. Anthropic's skills page now says custom commands have been merged into skills: a file at .claude/commands/deploy.md and a skill at .claude/skills/deploy/SKILL.md both create /deploy and work the same way. Existing command files keep working, and the docs prefer a skill for new work because skills also take supporting files.

If both exist under one name, the skill wins. A command file accepts the same frontmatter as a skill, except name and paths.

Where you save the file decides who gets the command. The Mindket guide to creating an Agent Skill goes deeper on the file itself.

Arguments, and who can run a command

Whatever you type after the command name replaces $ARGUMENTS. Run /fix-issue 123 against the skill below and Claude receives "Fix GitHub issue 123 following our coding standards" and the numbered steps. For several values, $ARGUMENTS[0] and $ARGUMENTS[1], or the shorter $0 and $1, pick them by position.

If the text has no placeholder, Claude Code adds a final line that starts with ARGUMENTS: and holds your input, so Claude still sees what you typed. A frontmatter argument-hint such as [issue-number] shows the expected input while you type.

Anthropic's own example of a command to keep away from Claude is /deploy. You do not want Claude deciding to deploy because the code looks ready. disable-model-invocation: true handles that, and user-invocable: false does the reverse.

---
name: fix-issue
description: Fix a GitHub issue
disable-model-invocation: true
---

Fix GitHub issue $ARGUMENTS following our coding standards.

1. Read the issue description
2. Understand the requirements
3. Implement the fix
4. Write tests
5. Create a commit

How Claude Code's creator turns daily work into commands

Boris Cherny created Claude Code at Anthropic. In January 2026 he wrote on X that he makes a slash command for every "inner loop" task he repeats many times a day. It saves him retyping prompts, and Claude can run the same workflow. His /commit-push-pr command runs dozens of times a day. It uses inline bash to collect git status before the model reads the prompt, which is the exclamation-mark line from your first command above. He kept his commands in .claude/commands/, checked into git, and a skill folder does the same job today.

In February 2026 he announced two bundled skills, /simplify and /batch, after using both daily. Today /simplify reviews the code you changed for cleanup and applies the fixes. /batch plans a large change, splits it into 5 to 30 independent units and hands each to a background subagent once you approve. Run /simplify after your next edit to see a bundled skill work like any other command.

Screenshots of X's own embeds of the posts, taken on 5 November 2026, with the time in UTC. X cuts both posts short with Show more. Select a screenshot to open the post on X.

Common mistakes

• Typing the command after other words. Claude Code recognizes a command only when it opens the message. After plain text, the name only gives Claude permission to run a skill, and nothing runs directly.
• Hitting Enter after a typo. Nothing is highlighted, Enter submits your text as typed and Claude Code reports "Unknown command". Pick a close match with Tab or the arrow keys. A command your platform or plan lacks is missing from the menu too.
• Misspelling a frontmatter field. A field name must match the docs exactly, hyphens included, and Claude Code ignores one it does not recognize without an error. The opening --- must also be the first line of the file.
• Naming a command after a built-in one. In a local terminal session, your skill replaces the built-in command of that name, though not its aliases.
• Writing new commands in .claude/commands/ because an older tutorial does. It still works. A skill does more, and a skill with the same name beats the command file.

Try it yourself

• Type / in a session and scroll the list. Filter it with a few letters, then run /help.
• Create /summarize-changes as above and run it twice: once with /summarize-changes and once by asking "What did I change?".
• Save the fix-issue text as a second skill and run /fix-issue 123. Then remove disable-model-invocation: true, which puts the description in Claude's context so Claude can load the skill itself.
• Make a command of your own for something you paste into chat every week, using the same two-part file: frontmatter, then instructions.

Next: hooks

A command runs when you ask. Lesson 4 covers hooks, which run on their own at fixed points, such as right before a tool call.

To share a command with other people, package it in a plugin, which our Claude Code plugins post covers.

Frequently asked questions

Sources

1. Claude Code docs: Commands (read November 4, 2026)https://code.claude.com/docs/en/commands
2. Claude Code docs: Extend Claude with skills (read November 4, 2026)https://code.claude.com/docs/en/skills
3. Claude Code changelog (read November 5, 2026)https://code.claude.com/docs/en/changelog
4. Boris Cherny on X: tip 7, slash commands, 2 January 2026 (read November 5, 2026)https://x.com/bcherny/status/2007179847949500714
5. Boris Cherny on X: /simplify and /batch, 28 February 2026 (read November 5, 2026)https://x.com/bcherny/status/2027534984534544489