Claude Skills are folders of instructions, scripts, and reference files that Claude loads only when a task calls for them. Anthropic launched them on October 16, 2025, and in December published the format as an open standard called Agent Skills (Anthropic, agentskills.io). The same skill now works in Claude Code, the Claude apps, the Claude API, and other coding tools like OpenAI Codex, GitHub Copilot, and Cursor.
This guide covers what a skill is, how to install one anywhere you use Claude, and the 10 best skills for vibe coding, with links to each one's source and free downloads where the license allows it.
Key takeaways
- A skill is a folder with a SKILL.md file: a name, a description of when to use it, and instructions.
- Claude reads a skill's full instructions only when your request matches its description, so you can install many without slowing it down.
- In Claude Code, install skills with /plugin install, or drop a folder into ~/.claude/skills.
- In the Claude apps, turn on code execution, then add skills under Customize > Skills.
- Treat a skill like software. Skills can run code, so install from sources you trust and read them first.
What's inside a skill
Every skill is a folder with one required file, SKILL.md. The top of the file names the skill and describes when to use it. The rest is plain instructions:
---
name: release-notes
description: Writes release notes from merged pull requests. Use when the user asks for release notes or a changelog.
---
# Release notes
1. List the pull requests merged since the last release tag.
2. Group them into Features, Fixes, and Internal.
3. Write one plain sentence per change, in the past tense.The name can be up to 64 characters of lowercase letters, numbers, and hyphens. The description can be up to 1,024 characters and should say both what the skill does and when to use it. A skill can also carry reference documents, a scripts/ folder, and assets like templates (Claude docs).
How Claude decides when to use a skill
Skills load in three steps, so installing a lot of them stays cheap (Claude docs):
- At startup, Claude reads only each skill's name and description, about 100 tokens per skill.
- When your request matches a description, Claude reads that skill's full instructions.
- When the instructions point to other files, Claude opens them or runs the scripts. Only a script's output enters the conversation, not its code.
That's why the description matters most. It's the only part Claude sees until the skill is needed.
How to install and use skills
In Claude Code
- From Anthropic's official marketplace, which Claude Code adds on first launch:
/plugin install frontend-design@claude-plugins-official - From any other marketplace, add it first, then install:
/plugin marketplace add anthropics/skills, then/plugin install example-skills@anthropic-agent-skills - Browse what's available with
/plugin, and list what's installed with/skills. - Your own skills go in a folder:
~/.claude/skills/<name>/SKILL.mdfor all your projects, or.claude/skills/<name>/SKILL.mdfor one project. Commit the project folder and your whole team gets the skill.
Claude uses a skill on its own when it's relevant, or you can call one directly by typing / and its name (Claude Code docs, plugins).
In the Claude apps (web and desktop)
Skills are available on the Free, Pro, Max, Team, and Enterprise plans (Claude help center).
- Go to Settings > Capabilities and turn on Code execution and file creation. Skills need it.
- Go to Customize > Skills and switch on the skills you want.
- To add your own, go to Customize > Skills, click +, then Create skill > Upload a skill, and upload a ZIP file. The folder inside the ZIP must have the same name as the skill.
On Team and Enterprise plans, an owner turns skills on first, under Organization settings > Plugins & skills.
In the Claude API
Developers attach skills to a request through the code execution tool, and upload their own through the Skills API (Claude docs).
In other coding tools
Because Agent Skills is an open standard, the same folder works elsewhere: OpenAI Codex reads .agents/skills (OpenAI), GitHub Copilot reads .github/skills, .claude/skills, or .agents/skills (GitHub), and Cursor reads .cursor/skills or .agents/skills (Cursor).
Where to find skills
- Anthropic's skills repository, the official examples. Most are open source under Apache 2.0. The Word, Excel, PowerPoint, and PDF skills are source-available only, so they can be used but not copied or shared.
- Anthropic's official plugin marketplace, which you can browse with
/pluginin Claude Code or on the Claude plugin directory. - skills.sh, Vercel's skills directory and install leaderboard. It installs skills with
npx skills add <owner/repo>. - Community lists like awesome-agent-skills and awesome-claude-code.
The 10 best Claude skills for vibe coding
Vibe coding means building software by describing what you want and letting AI write most of the code. It's fast, and it has the same weak spots every time: no plan, generic-looking design, no tests, and no security review. These ten skills cover those gaps.
We picked skills that are actively maintained (all updated within the last week), clearly licensed, and focused on building software. Star counts are from GitHub on October 8, 2026.
About the downloads: four of Anthropic's Apache 2.0 skills are available here as ZIP files. Each one is an unmodified copy from anthropics/skills (commit 683bc88, October 5, 2026), includes its license file, and is ready to upload to the Claude apps or drop into ~/.claude/skills. Anthropic doesn't endorse this page, and the originals on GitHub are always the latest version. The other skills install as plugins, so we link to the source.
| Skill | Best for | Get it |
|---|---|---|
| Superpowers | Planning before building | Plugin |
| Frontend Design | Design that doesn't look generic | Plugin or ZIP |
| Matt Pocock's skills | Getting the plan right | Plugin |
| Addy Osmani's Agent Skills | Spec to ship, step by step | Plugin |
| Web App Testing | Checking work in a real browser | Plugin or ZIP |
| Trail of Bits security | Review before you ship | Plugin |
| Web Artifacts Builder | Apps with no setup | Plugin |
| MCP Builder | Connecting outside tools | Plugin or ZIP |
| Claude API | Adding AI features | Built into Claude Code |
| Skill Creator | Making your own skills | Plugin or ZIP |
1. Superpowers
What it does: turns a vague idea into a finished feature in stages. Claude asks questions to refine the idea, writes a plan broken into small tasks, then builds it task by task with test-driven development, code review, and a final check.
Why it helps: it adds the planning and testing that vibe coding usually skips.
Install: /plugin install superpowers@claude-plugins-official
obra/superpowers · MIT license · 296,000+ stars
2. Frontend Design
What it does: pushes Claude toward a deliberate visual direction, with real choices about typography, color, and layout, instead of the template look most AI-built apps share.
Why it helps: it's the fastest fix for the "this looks AI-generated" problem.
Install: /plugin install frontend-design@claude-plugins-official, or download the ZIP
Source on GitHub · Anthropic · Apache 2.0
3. Matt Pocock's skills
What it does: a set of small engineering skills. /grill-me has Claude question you about your plan before it writes code, and others cover test-driven development (/tdd), debugging (/diagnosing-bugs), and cleaning up a codebase's structure (/improve-codebase-architecture).
Why it helps: most bad AI code starts with a vague request. This gets you and Claude on the same page first.
Install: /plugin install mattpocock-skills@claude-plugins-official, then run /setup-matt-pocock-skills once in each project.
mattpocock/skills · MIT license · 280,000+ stars
4. Agent Skills by Addy Osmani
What it does: a full development workflow as commands: /spec, /plan, /build, /test, /review, and /ship, backed by 25 skills.
Why it helps: a guided path from idea to release, with a quality check at each step.
Install: /plugin marketplace add addyosmani/agent-skills, then /plugin install agent-skills@addy-agent-skills
addyosmani/agent-skills · MIT license · 103,000+ stars
5. Web App Testing
What it does: Claude writes and runs Playwright scripts against your app on your own machine, takes screenshots, and reads the logs to check what it built. It needs Python and Playwright installed.
Why it helps: instead of telling you the feature works, Claude opens the app and checks.
Install: /plugin marketplace add anthropics/skills, then /plugin install example-skills@anthropic-agent-skills, or download the ZIP
Source on GitHub · Anthropic · Apache 2.0
6. Trail of Bits security skills
What it does: security reviews from a well-known security firm. Good starting points: insecure-defaults finds hardcoded passwords and weak settings, differential-review reviews code changes for security problems, and supply-chain-risk-auditor checks your npm, PyPI, and Go dependencies.
Why it helps: AI-written code ships fast, and it can ship secrets and risky defaults just as fast. Run these before anything goes live.
Install: /plugin marketplace add trailofbits/skills, then /plugin install insecure-defaults@trailofbits (or browse the rest with /plugin)
trailofbits/skills · CC BY-SA 4.0 · 7,400+ stars
7. Web Artifacts Builder
What it does: builds React apps with Tailwind and shadcn/ui and bundles them into a single HTML file, so a working app can live right inside a Claude chat.
Why it helps: you can prototype an idea with nothing installed on your computer.
Install: included in example-skills (see Web App Testing above)
Source on GitHub · Anthropic · Apache 2.0, with MIT-licensed shadcn/ui parts
8. MCP Builder
What it does: a step-by-step guide, with test scripts, for building MCP servers in Python or TypeScript. MCP servers are how Claude connects to outside tools and data.
Why it helps: when your app needs to talk to another service, this keeps Claude on a proven pattern.
Install: included in example-skills, or download the ZIP. Anthropic's official mcp-server-dev plugin covers similar ground: /plugin install mcp-server-dev@claude-plugins-official
Source on GitHub · Anthropic · Apache 2.0
9. Claude API
What it does: up-to-date reference for the Claude API and its official SDKs in eight languages.
Why it helps: when you add AI features to your app, Claude uses current model names and parameters instead of outdated ones from its training data.
Install: built into Claude Code (type /claude-api), or /plugin install claude-api@anthropic-agent-skills
Source on GitHub · Anthropic · Apache 2.0
10. Skill Creator
What it does: helps you write your own skills, then tests them by comparing Claude's results with and without the skill.
Why it helps: any instruction you keep repeating to Claude can become a skill it follows every time.
Install: /plugin install skill-creator@claude-plugins-official, or download the ZIP
Source on GitHub · Anthropic · Apache 2.0
Also worth a look
- gstack by Garry Tan: role-based commands like
/office-hours,/review,/qa, and/shipfor solo founders. It's a heavier install that downloads its own browser, so read the README first. - Vercel's agent skills: React and Next.js performance rules and UI audits, if that's your stack.
- Supabase agent skills: Postgres and Supabase best practices, including row-level security, if Supabase is your backend.
How to use skills safely
Skills can run code and change how Claude behaves, so treat installing one like installing software. Anthropic's own guidance is to "use Skills only from trusted sources: those you created yourself or obtained from Anthropic" (Claude docs).
- Read before you install. Open the
SKILL.mdand any scripts. Be especially careful with skills that fetch content from outside websites. - Plugins run with your permissions. Claude Code plugins can run code as you, outside the sandbox, and an update can change files you already reviewed (Claude Code docs).
- Prefer well-maintained sources. Every skill on this list has a clear license and was updated within the last week.
Frequently asked questions
What's the difference between a Claude Skill and an MCP server? An MCP server connects Claude to outside tools and data, like a database or another app. A skill teaches Claude how to do a task, with instructions and sometimes scripts. They work well together: a skill can tell Claude how and when to use the tools an MCP server provides.
Are Claude Skills free? Skills work on every Claude plan, including Free, as long as code execution is turned on. Every skill on this list is free to use. Check each license before you copy or share one.
Do Claude Skills work in other tools?
Yes. Agent Skills is an open standard, and OpenAI Codex, GitHub Copilot, and Cursor all read the same SKILL.md format.
How do I make my own skill?
Create a folder in ~/.claude/skills with a SKILL.md inside: a name, a description of when to use it, and your instructions. Or install Skill Creator and let Claude help you write and test it.
How many skills can I install? Plenty. Claude only reads each skill's name and description until one is needed. If you're calling the API, the limit is 20 skills per request.
We use skills to package how a business works, its processes, rules, and playbooks, into the private AI systems we build for clients. If you want AI that already knows how your business runs, see what we build or book a call.