AI Agent Skills
Directory.

Claude Agent Skills are modular capabilities that extend Claude. Each skill is a folder that packages instructions, metadata and optional resources such as scripts or templates, and Claude uses it automatically when a request matches what the skill is for. Anthropic ships four pre-built skills (PowerPoint, Excel, Word and PDF) and publishes more example skills in the anthropics/skills repository.

SKILL.md is the entry point of every skill. It starts with YAML frontmatter containing a name (up to 64 characters, lowercase letters, numbers and hyphens, and it may not contain "anthropic" or "claude") and a description (up to 1,024 characters) that says what the skill does and when to use it. The rest of the file holds the instructions; other files in the folder can be scripts, references or templates.

Skills load progressively. At startup Claude only sees each skill's name and description (roughly 100 tokens per skill). When a request matches, it reads the SKILL.md body. Extra files and scripts are read or executed only if the instructions call for them, so installed skills cost almost nothing until they're used.

It depends on where you use Claude. In Claude Code, add the anthropics/skills plugin marketplace or put a skill folder in ~/.claude/skills (personal) or .claude/skills (project). On claude.ai, upload a custom skill as a ZIP in settings. On the Claude API, upload it through the /v1/skills endpoints and reference it in the container parameter with the code execution tool. The install section on this page has the exact commands.

No. Anthropic's documentation states that custom skills don't sync across surfaces: a skill uploaded to claude.ai must be uploaded separately to the API, and Claude Code skills live on the filesystem, separate from both.

The skills listed here are publicly readable on GitHub. Most skills in anthropics/skills are Apache 2.0, while the document skills (docx, pdf, pptx, xlsx) are source-available rather than open source. Using skills on claude.ai requires a Pro, Max, Team or Enterprise plan with code execution enabled, and API usage is billed as normal.

MCP (Model Context Protocol) connects Claude to external tools and data through servers: it gives Claude new actions. A skill packages know-how, meaning instructions, workflows and scripts for doing a kind of task well. They complement each other: a skill can describe how to use an MCP server, and the mcp-builder skill helps you build one.

A prompt applies to one conversation and you have to supply it each time. A skill is installed once, sits dormant until its description matches a request, and can bundle scripts and reference files alongside its instructions.

Codex uses the same folder-plus-SKILL.md idea. OpenAI now distributes Codex skills inside plugins: the openai/plugins repository powers the "Codex official" plugin marketplace, and installing a plugin makes its bundled skills available in your next session. Some plugins are published by OpenAI and others by partners such as Figma, Stripe or Vercel. The older openai/skills repository is marked deprecated. For your own skills, Codex loads them from .agents/skills in your repository and $HOME/.agents/skills.

Yes. OpenAI Help lists Skills for eligible ChatGPT Business, Enterprise, Healthcare and Edu workspaces. You can create a skill with chat, with the skills editor, or by uploading one, and every workspace with Skills includes the default skill-creator skill.

This directory currently covers skills published by Anthropic, OpenAI, Microsoft, Google, AWS, Cloudflare, Vercel and HashiCorp in their own repositories, plus partners who publish plugins in OpenAI's official Codex marketplace (for example Twilio, Shopify, Figma and Stripe). Each publisher page lists that company's skills and its own install instructions.

A skill is labelled Official only when the company that maintains it publishes it in its own repository (for example Microsoft in microsoft/skills), or when it is a built-in or default skill from Anthropic or OpenAI. Skills that partners publish through OpenAI's Codex marketplace are labelled "Codex marketplace" with the partner's name, and plugins from individual developers are labelled "Independent developer". Skills that only remain in a deprecated repository are labelled Legacy. Status badges (Example, Preview, Deprecated) quote what the publisher itself says.

Check each skill's license first. Every entry shows the license declared by the skill, its plugin or its repository: many are MIT or Apache-2.0, some are proprietary (for example Anthropic's document skills and OpenAI's own Codex plugins), and some repositories declare no license. This directory itself only shows names, short descriptions and links; it never republishes skill files.

Skills can contain executable code, so only install ones from sources you trust. Anthropic recommends using skills you created yourself or obtained from Anthropic, and OpenAI recommends reviewing a skill before uploading it. Every entry here links to its source so you can read it first.

A sync script clones each publisher's official skills repository (Anthropic, OpenAI, Microsoft, Google, AWS, Cloudflare, Vercel and HashiCorp), finds every SKILL.md, reads its description, de-duplicates skills that appear in more than one official location, and logs every addition, removal or change in the public changelog with the commit it was checked at. The Claude pre-built skills and the ChatGPT default skill aren't listed in a repository, so they're checked by hand against the official docs. Repositories outside that list, and three Codex marketplace plugins hosted elsewhere, aren't covered yet.