Claude Skills for Claude Code

Claude Code Skills are modular workflows that add clear instructions and supporting files to Claude Code when needed.

The idea behind it:

Instead of teaching an AI agent everything at once (which would blow up the context window), you give it specialized instructions only when it needs them. This is called Progressive Disclosure.

You can think of Skills as "apps for your AI agent".

Just like you only install the apps you need on your smartphone, you give your AI agent only the skills relevant to your current project.

A skill is typically defined as a SKILL.md file. This file contains structured instructions telling the agent what to do, how to do it, and which tools it may use.

The format supports reusable workflows. Whether you may share or modify a particular skill depends on its license.

Skills come into play whenever your AI agent needs to handle a specialized task that goes beyond its standard capabilities.

Some examples:

• Create documents: Skills for Word documents, PDFs, Excel spreadsheets, or PowerPoint presentations. The agent then knows exactly how to create and format these file types.
• Development: Skills for testing, debugging, Git workflows, or specific frameworks. A "TDD skill" could teach your agent to always write tests first, for example.
• Automation: Skills for n8n workflows, CI/CD pipelines, or recurring tasks. The agent learns to build complex automations step by step.
• Security: Skills for penetration testing, code audits, or threat modeling. Particularly useful for security experts who want to automate recurring analyses.
• Research and analysis: Skills for data analysis, scientific work, or deep research. The agent gets a structured approach for complex investigations.

The table on this page lists only applications with verified native skill support.

Custom instructions, project rules, and slash commands are useful features, but they do not prove support for the SKILL.md format. The table reports them separately.

Each entry is checked against its official documentation.

The good news: You don't have to create skills from scratch. There are already hundreds of ready-made skills you can use directly.

The most important sources:

• This directory: Above you'll find the currently curated Claude Code Skills, filterable by category, source, and freshness.
• GitHub: Many skill collections are available as GitHub repositories. Search for SKILL.md, claude-skills, or agent-skills.
• Awesome lists: There are several curated lists like Anthropic's official Claude Code resources and community collections.

Read the linked source's license, requirements, and installation instructions first. A download from this directory supplies SKILL.md and, for ZIP files, license texts. It may not include scripts, references, templates, or other files that the skill needs. Get those from the linked source.

1. 1

   Create one folder per skill

   For a personal skill available across your local projects, put SKILL.md at ~/.claude/skills/my-skill/SKILL.md. For a skill shared with one project, use .claude/skills/my-skill/SKILL.md in that repository. Create the my-skill folder first, then move the downloaded SKILL.md into it or extract the ZIP there. Do not put SKILL.md directly in the parent skills directory.

   ~/.claude/skills/my-skill/
   ├── SKILL.md
   ├── reference.md
   └── scripts/
       └── helper.py

   This is an example layout. Keep included license and notice files. Add any required supporting files from the source repository, preserving their paths relative to SKILL.md. Follow the publisher's separate installation process for plugin skills.

2. 2

   Invoke the skill

   Invoke a locally installed skill by its name.

   /skill-name

   Plugin skills also include the plugin name in the command, for example /plugin-name:skill-name.

3. 3

   Check the requirements

   Check required programs, services, and permissions. Try the skill on a small task before using it for important work.

Claude Code documentation on skills and supporting files

Start with YAML frontmatter for the name and scope, followed by the instructions.

---
name: project-check
description: Checks changes for defects. Use when a code review is requested.
---

# Project Check

## Instructions
Step-by-step guide for the agent:
1. Analyze the request
2. Execute the action
3. Verify the result

## Examples
For a changed calculation, check the affected inputs and boundary cases.

A section listing tool names does not grant permissions. Your Claude Code settings and approvals still govern access.

Key tips:

• 1
  Be specific: The clearer the instructions, the better the result. Vague instructions lead to inconsistent outcomes.
• 2
  Use examples: Concrete examples help the agent understand the desired outcome.
• 3
  Define boundaries: Tell the agent what it should not do to prevent unwanted behavior.

Skills and MCP servers complement each other but are different concepts:

In short: Skills are knowledge (instructions, workflows, best practices) and MCP servers are tools (access to external systems and data). Ideally, you combine both.

A few tips to get the most out of Agent Skills:

• Review skills: Read the SKILL.md file before using it. Pay attention to which tools the skill is allowed to use and what permissions it requires. Be careful with skills that execute bash commands or delete files.
• Customize for your project: Adapt generic skills to your project. A "testing skill" should know your specific test framework and conventions, for example.
• Start small: Begin with a simple skill and expand it gradually. A skill that tries to do too much at once is harder to debug.
• Combine skills: Multiple specialized skills are often better than a single one that does everything. You can use a "research skill" and a "writing skill" sequentially, for example.