Skills

Reusable prompt workflows you can invoke with a slash command. Create custom skills for your team or project, or use the built-in ones that ship with Claude Code.

What are skills?

A skill is a named, reusable prompt workflow stored as a Markdown file. When you type /skill-name in a session, Claude Code loads the skill’s instructions and executes them. Skills are useful for:

• Repeatable tasks you run often (e.g. /review, /document, /refactor)
• Team conventions you want to encode once and reuse everywhere
• Complex workflows that would be tedious to type out each time

Built-in skills

Claude Code ships with a set of built-in skills that are available in every session:

Skill	Description
/verify	Verify that the work done matches the original request
/debug	Systematic debugging workflow
/simplify	Simplify complex code or text
/remember	Store information in persistent memory
/skillify	Convert a workflow into a reusable skill
/batch	Run a workflow across multiple items
/stuck	Get unstuck when progress stalls
/update-config	Update Claude Code configuration
/keybindings	Set up terminal key bindings

Built-in plugins can provide additional skills. See the Plugins page for details.

The /skills command

Run /skills in a session to open the skills browser. From there you can:

• See all available skills (built-in, user, and project)
• View a skill’s description and usage hints
• Invoke a skill directly from the list

Skill directories

Claude Code loads skills from the following locations, in priority order:

Scope	Path
Managed (policy)	<managed-path>/.claude/skills/
User	~/.claude/skills/
Project	.claude/skills/ (walked up to home)

Skills closer to the project root take precedence over user-level skills with the same name.

Claude Code also supports a legacy ~/.claude/commands/ directory for backward compatibility. New skills should use the skills/ directory format.

Creating a custom skill

Each skill lives in its own subdirectory under a skills/ folder. The directory name becomes the slash command name.

1. Create the skill directory

   mkdir -p .claude/skills/my-skill
   
2. Create the SKILL.md file

   Create .claude/skills/my-skill/SKILL.md with your skill’s instructions:

   ---
   description: Review the code for security issues and suggest fixes
   argument-hint: "[file or description]"
   allowed-tools: Read, Grep, Glob
   ---
   
   Review the following code for security vulnerabilities.
   
   Focus on:
   - SQL injection
   - XSS vulnerabilities
   - Insecure dependencies
   - Hardcoded secrets
   
   For each issue found, explain the risk and suggest a concrete fix.
   
   $ARGUMENTS
   
3. Invoke the skill

   In a Claude Code session, type:

   /my-skill src/api/auth.ts
   

Frontmatter reference

Skills support YAML frontmatter at the top of SKILL.md to control their behavior:

description · string

Short description shown in /skills and autocomplete.

argument-hint · string

Hint shown next to the skill name when typing (e.g. "[file]").

allowed-tools · string

Comma-separated list of tools this skill is allowed to use (e.g. "Read, Write, Bash").

model · string

Override the model used for this skill (e.g. "claude-sonnet-4-6"). Use "inherit" to use the session default.

when_to_use · string

Guidance for Claude on when to automatically invoke this skill.

user-invocable · boolean

Set to false to hide the skill from user-facing autocomplete. It can still be invoked by the model internally.

context · string

Set to "fork" to run the skill in an isolated context fork.

effort · string

Control the reasoning effort level. Accepts a named level or an integer.

paths · string

Gitignore-style path patterns. The skill is only activated when Claude operates on matching files (e.g. "src/**/*.ts"). Skills with paths are loaded lazily and activated on first file match.

hooks · object

Hooks scoped to this skill. Use the same format as global hooks in settings.json.

Example skill definitions

Code reviewer with path scoping

---
description: Review TypeScript files for best practices
paths: "**/*.ts"
allowed-tools: Read, Grep
---

Review this TypeScript file for:
- Type safety issues
- Missing error handling
- Performance anti-patterns

$ARGUMENTS

Multi-step workflow

---
description: Create a GitHub PR with tests and docs
allowed-tools: Bash, Read, Write, Glob
argument-hint: "[branch description]"
---

Create a pull request for the current branch:

1. Run the test suite and confirm all tests pass
2. Check for missing documentation on any new public APIs
3. Write a clear PR description summarizing the changes
4. Create the PR with `gh pr create`

Branch context: $ARGUMENTS

Dynamic skill discovery

Claude Code automatically discovers skills inside subdirectories as you work. When you read or write a file, Claude Code walks up the directory tree looking for .claude/skills/ directories. Newly discovered skills become available immediately without restarting the session.

Skills with a paths frontmatter are stored as conditional skills and only activated when Claude operates on a file matching the pattern. This keeps the skill list focused and avoids context bloat.