If you've been using Claude Code for a while, you've probably copy-pasted the same multi-step instructions into the terminal over and over. "Always use TypeScript strict mode." "Follow TDD. Write the test first." "Review this PR and summarize the changes."

Claude Code Skills are the answer to all of that.

Skills are folders of instructions, scripts, and reference materials that Claude loads dynamically to improve its performance on specialized tasks. Instead of repeating yourself every session, you write the instructions once, store them in the right place, and Claude picks them up automatically — or on demand via a /skill-name slash command.

This guide covers everything: what Skills are, where to find thousands of them for free, how to install them, and how to build your own from scratch.

What Exactly Is a Skill?

At its core, a Skill is just a folder with a SKILL.md file inside. That file has two parts:

1. YAML frontmatter — metadata Claude uses to decide when to activate the skill
2. Markdown instructions — what Claude should actually do

Here's the minimal structure:

my-skill/
└── SKILL.md

And a minimal SKILL.md looks like this:

---
name: code-review
description: >
  Review code for bugs, style issues, and performance problems.
  Use when the user asks to review, check, or audit code.
---

# Code Review

## What to check
- Logic errors and off-by-one mistakes
- Performance bottlenecks
- Security vulnerabilities
- Naming consistency

## Output format
Return a bulleted list of issues grouped by severity: Critical, Warning, Suggestion.

That's it. Drop this folder into .claude/skills/ in your project (or ~/.claude/skills/ for global use), and Claude will use it automatically when the context matches — or you can invoke it explicitly with /code-review.

Skills can also go much further: they support bundled scripts (in any language), on-demand reference files, shell command substitution with the !command\` syntax, and even sub-agents. But the simple case is just a Markdown file.

Where Skills Live (Scope & Precedence)

Skills can be stored at three levels, and they override each other in a clear hierarchy:

When the same skill name exists at multiple levels, enterprise overrides personal, personal overrides project. A project skill with the name code-review completely replaces any bundled skill with the same name.

You can check what's loaded and whether any skills are being truncated by running:

plaintext
/doctor


The Open Standard: agentskills.io

The format behind SKILL.md files is an open standard maintained at agentskills.io. Anthropic originally developed it for Claude Code, but it's now supported across Claude Code, OpenAI Codex, Gemini CLI, Cursor, and many other agents. A skill you write for Claude Code can be adapted for other platforms with minimal changes — the frontmatter fields and Markdown structure are identical.

This portability matters if your team uses multiple AI coding tools.

Essential Resources

🏛️ Official: github.com/anthropics/skills

The canonical source. Anthropic's public Skills repository contains skills across four categories:

• Creative & Design — art, music, brand guidelines
• Development & Technical — webapp testing, MCP server generation, code review
• Enterprise & Communication — internal comms, runbooks, status reports
• Document Skills — the actual production skills behind Claude's pptx, docx, pdf, and xlsx capabilities

These document skills are source-available (not open source) and are a great reference for complex, real-world skill design.

Install from the marketplace:

shell
/plugin install document-skills@anthropic-agent-skills
/plugin install example-skills@anthropic-agent-skills


Or install a specific skill with:

shell
npx skills add https://github.com/anthropics/skills --skill doc-coauthoring


🌐 skills.sh

skills.sh is the community browsing layer for the Skills ecosystem. It auto-indexes public GitHub repositories and gives you per-skill detail pages, install commands, and quality indicators — all without needing to dig through raw GitHub repos.

Search for a skill, click through, and you get the npx skills add command ready to copy. It's the closest thing the ecosystem has to an npm registry, but for AI agent instructions.

🔍 claude-plugins.dev

claude-plugins.dev/skills is a community-maintained registry that auto-indexes all public SKILL.md files on GitHub. It covers Claude Code, Cursor, Codex, and more agents, and it's open-source. Every skill listed can be installed directly from within Claude Code — no browser switching needed.

📚 skillsdirectory.com

skillsdirectory.com curates security-tested skills. Every entry is scanned for malware, prompt injection, and credential theft before being listed. If you're installing skills on a work machine or in a shared team environment, this is the safer registry to browse first.

📦 claudskills.com

claudskills.com indexes over 67,000 open skills from public GitHub repositories. The free tier lets you search and browse everything. The Pro tier ($9/month) adds a multi-signal quality score and one-click install from the desktop app.

Best Community GitHub Repos

github.com/anthropics/skills (official)

The reference implementation. Start here.

github.com/travisvn/awesome-claude-skills

A curated list of the best skills, tools, and resources — the "awesome list" for the Skills ecosystem. Covers docx, pdf, pptx, webapp testing, MCP builder, brand guidelines, and more with quality notes on each.

github.com/obra/superpowers

One of the most starred behavioral skill collections. Encodes TDD enforcement, debugging workflows, and multi-hour autonomous coding patterns. The subagent-driven architecture prevents context drift on long tasks. Install with:

shell
/plugin marketplace add obra/superpowers-marketplace


github.com/alirezarezvani/claude-skills

337 skills across 13 platforms (Claude Code, Codex, Gemini CLI, Cursor, and more). Covers engineering, marketing, compliance, C-level advisory, and business operations. All 579 bundled Python CLI tools use the standard library — zero pip installs required.

`shell

For Claude Code

./scripts/install.sh --tool claude-code

For other tools

./scripts/convert.sh --tool all
./scripts/install.sh --tool
`

github.com/Jeffallan/claude-skills

66 skills for full-stack developers across 12 categories: languages, backend/frontend frameworks, infrastructure, APIs, testing, DevOps, security, data/ML, and platform specialists. Skills activate automatically based on your request context.

github.com/glebis/claude-skills

Community skills including a comprehensive Google Workspace skill (gws) for Gmail, Calendar, Drive, Sheets, Docs, Tasks, and Chat — all from the terminal. Also includes a deep-research skill and a TDD enforcer.

shell
npx skills add glebis/claude-skills --skill tdd
npx skills add glebis/claude-skills --skill deep-research


github.com/tech-leads-club/agent-skills

A security-first curated library. 100% open source, static analysis in CI/CD, and content hashing for integrity verification. The README notes that over 13% of marketplace skills in the wild contain critical vulnerabilities — this repo is built as the hardened alternative.

shell
npx @tech-leads-club/agent-skills install -s tlc-spec-driven
npx @tech-leads-club/agent-skills install -s coding-guidelines docs-writer


Installing Skills: Three Methods

Method 1: Plugin Marketplace (recommended)

`shell

Browse available plugins

/plugin marketplace

Install a specific skill collection

/plugin marketplace add anthropics/skills

Install a specific skill from a collection

/plugin install doc-coauthoring@anthropic-agent-skills
`

Method 2: npx skills CLI

`shell

Install a skill from any public GitHub repo

npx skills add https://github.com/anthropics/skills --skill webapp-testing

Install from a shorthand (if registered)

npx skills add obra/superpowers --skill tdd
`

Method 3: Manual copy

`shell

Personal (global) install

cp -r my-skill ~/.claude/skills/

Project install (check into git for team sharing)

cp -r my-skill .claude/skills/
`

Advanced: Shell Command Substitution

Skills can pull live data before Claude ever sees the prompt using the !`command`` syntax:

`yaml

name: pr-summary
description: Summarize changes in a pull request
context: fork
agent: Explore

allowed-tools: Bash(gh *)

Pull request context

• PR diff: !gh pr diff
• PR comments: !gh pr view --comments
• Changed files: !gh pr diff --name-only

Your task

Summarize this pull request in plain English. Highlight what changed and why.
`

When this skill activates, the !`...`` blocks run first, and their output replaces the placeholders. Claude receives the rendered prompt with actual PR data — not the command text.

This pattern works for anything: live test results, git log output, environment variables, API responses, database schema dumps.

Building Your Own Skill: A Practical Template

---
name: my-skill
description: >
  What this skill does and the specific phrases or contexts that should trigger it.
  Be concrete: "Use when the user asks to X, Y, or Z."
when_to_use: >
  Optional extra trigger guidance. Useful if your description is long.
---

# My Skill

## Context
[Brief background Claude needs to do this well]

## Instructions
[Step-by-step what Claude should do]

## Examples
- "Do X with Y" → expected behavior
- "Review Z" → expected behavior

## Output format
[Exactly what the response should look like]

Tips for good descriptions:

• Put the most important trigger phrase first — descriptions get truncated under context pressure
• Include both what the skill does and when to use it
• Be specific about inputs ("when the user provides a URL") not just intent ("when the user wants help")

Run /doctor to see if your skill's description is being shortened, and adjust accordingly.

Evaluating Skill Quality

The skill-creator skill (from the Anthropic repo) gives you a full eval loop:

• Generates should-trigger and should-not-trigger test prompts
• Measures the hit rate against your description
• Proposes edits when the skill fires on the wrong requests
• Opens an HTML report where you can record qualitative feedback

For the full eval format and iteration workflow, see agentskills.io.

Quick Reference: Resources at a Glance

Wrap Up

Claude Code Skills are one of those features where the investment pays off immediately. You spend 10 minutes writing a SKILL.md for your team's code review process, and from that point on everyone on the project gets consistent, repeatable behavior — no more pasting the same prompt into every session.

Start with the official Anthropic repo to understand the patterns, browse skills.sh or claude-plugins.dev to find community work that fits your stack, and build your own once you see what's possible.

The ecosystem is growing fast. Skills that work in Claude Code today are already portable to Codex, Gemini CLI, and Cursor via the same SKILL.md format — so anything you build now travels with you.

Have a favorite skill or repo not listed here? Drop it in the comments.