Anthropic Skills: The Official Repo and Docs, Explained

Search for Anthropic skills and you land on three different things: a GitHub repo (anthropics/skills), a section of official documentation, and a feature that shows up across Claude Code, the Claude apps, and the Claude API. They're all facets of the same system. A skill is a folder with a SKILL.md file at its root — instructions, plus optional scripts and reference docs — that Claude loads on demand when a task matches the skill's description. Anthropic maintains an official set of these skills in the open, and it's the single best reference for how good skills are written.

This guide walks through the anthropics/skills GitHub repo folder by folder: the document skills that power Claude's file handling (docx, pdf, pptx, xlsx), example skills like skill-creator that are worth reading even if you never install them, what the licensing actually lets you reuse, and where community-built skills fit alongside the official set.

What "Anthropic skills" refers to

When people say "Anthropic skills," they usually mean one of three surfaces:

1. The repo: github.com/anthropics/skills is where Anthropic publishes its official skills in source form — both the production document skills and a set of example skills you can install or fork.
2. The docs: the Anthropic Agent Skills documentation covers the format spec (frontmatter fields, folder layout, progressive disclosure) and how skills behave in each product. The Claude Code docs cover the developer-facing side: personal skills, project skills, and plugin-distributed skills.
3. The product surfaces: skills run in Claude Code (from .claude/skills/ directories or plugins), in the Claude apps (where document skills power file creation), and via the Claude API alongside the code execution tool.

The mechanism underneath all three is the same, and it's worth internalizing because it explains almost every design choice in the official repo: progressive disclosure. At startup, Claude only sees each skill's name and description — a few dozen tokens per skill. When a task matches a description, Claude reads the full SKILL.md body. Only if the body points to additional files (reference docs, scripts, templates) does Claude load or execute those. Skills scale because most of their content stays out of the context window until it's needed.

Anthropic shipped Agent Skills in late 2025, and the format has since spread well beyond Claude — the same SKILL.md convention now gets read by other agents and registries too. If you want the full field-by-field spec, see our SKILL.md format deep dive.

A tour of the anthropics/skills GitHub repo

The repo is small enough to read in an afternoon. At a high level it splits into two groups (exact layout shifts over time, so treat this as a map rather than a snapshot):

anthropics/skills
├── document-skills/     # production skills: docx, pdf, pptx, xlsx
│   ├── docx/
│   ├── pdf/
│   ├── pptx/
│   └── xlsx/
└── example skills       # skill-creator, mcp-builder, webapp-testing,
                         # artifacts-builder, brand-guidelines, and more

Every skill in the repo follows the same anatomy:

skill-name/
├── SKILL.md          # required: YAML frontmatter + instructions
├── scripts/          # optional: executable helpers Claude can run
├── references/       # optional: docs loaded only when needed
└── assets/           # optional: templates, fonts, examples

And every SKILL.md opens with minimal frontmatter:

---
name: pdf
description: Comprehensive PDF manipulation for extracting text and tables,
  creating new PDFs, merging documents, and filling forms.
---

That description field is doing the heavy lifting: it's the only part of the skill Claude sees before deciding whether to load it. Notice how the official descriptions name concrete verbs and triggers ("extracting text and tables", "filling forms") rather than vague summaries. If you've ever wondered why your own skill never fires, compare your description against these.

The repo also doubles as a Claude Code plugin marketplace, so you can install the official skills directly:

/plugin marketplace add anthropics/skills

Or copy any skill folder straight into .claude/skills/ in a project. Both paths are covered in our Claude Code skills guide.

The official document skills: docx, pdf, pptx, xlsx

The document-skills/ folder is the most interesting part of the repo because these aren't demos — they're the same skills that power document creation and editing in the Claude apps. When Claude produces a Word document or an Excel spreadsheet in production, these instructions are what it's following.

Each one handles a file family:

Two design lessons stand out when you read them:

Scripts over generation. The document skills lean heavily on bundled scripts instead of asking the model to emit raw file bytes. Office files are zipped XML; generating them token-by-token is slow and error-prone. So the skills ship deterministic code for the mechanical parts and reserve the model's judgment for content and structure. If your own skill involves any repeatable transformation, this is the pattern to copy.

Reference files keep SKILL.md short. Rather than inlining every OOXML detail, the skills split deep documentation into separate files that Claude reads only when a task requires them. The SKILL.md body stays a concise routing layer: what the skill does, when to use which script, where to look for edge cases.

skill-creator and the example skills worth studying

Beyond the document skills, the repo carries a set of example skills. Some are practical tools, some are teaching material, and a few are both.

skill-creator is the one to start with. It's a skill for writing skills: it interviews you about the workflow you want to capture, scaffolds the folder structure, drafts the frontmatter and body, and helps you package the result. Using skill-creator once teaches you more about the format than any spec document — it encodes Anthropic's own opinions about what a good description looks like, when to split content into reference files, and how much detail belongs in the body. If you'd rather build one by hand, our guide on how to create a Claude skill walks the same ground step by step.

Other examples worth reading:

• mcp-builder — guides Claude through building an MCP server. A good example of a skill that encodes a multi-step engineering workflow rather than a single transformation.
• webapp-testing — drives browser automation for testing web apps. Shows how a skill can orchestrate external tooling.
• artifacts-builder — conventions for building richer claude.ai artifacts. Demonstrates skills that encode platform-specific constraints.
• brand-guidelines — a template for encoding a company's visual and voice standards. Of all the examples, this one looks most like what teams actually need day to day: not a general capability, but your conventions made loadable.

The pattern across all of them: one focused job per skill, a description that names its triggers, and progressive disclosure for anything bulky.

Anthropic skills licensing: what you can reuse

The short version: the repo is published so you can read, learn from, install, and adapt these skills — but the terms are not uniform across the whole repo, so check before you redistribute.

The example skills are published for exactly the kind of reuse you'd expect: install them, fork them, adapt them into internal variants. The document skills have carried their own notices — they're published as a source-available reference for how Claude's production document handling works, which is not automatically the same as a blanket grant to rehost them.

Practical guidance:

• Reading and learning from any skill in the repo is the primary intended use. Copy the patterns freely — folder anatomy, description style, script-first design.
• Installing and adapting for your own use is the normal path for the example skills.
• Redistributing or republishing — vendoring a skill into a product, mirroring it into your own public registry — is where you should read the LICENSE file at the repo root and any per-folder notices first, because terms differ and have changed over time.

When in doubt, write your own skill that borrows the structure rather than the text. The format is the valuable part, and the format is free.

Where community skills fit alongside the official set

The official Claude skills are deliberately generic. Anthropic ships skills for problems everyone has — documents, spreadsheets, building MCP servers — and stops there. Everything specific to your stack, your conventions, and your team is left as an exercise: how you structure API routes, what your review checklist enforces, which migration patterns your codebase uses.

That's the gap community and team skills fill, and it's most of the value in practice. Our roundup of the best Claude skills covers what's worth installing beyond the official set.

It's also where a registry earns its keep. On localskills.sh, every skill is a package with a root SKILL.md — the same format the anthropics/skills repo uses — and packages can bundle docs, scripts, and templates up to 100 MB and 500 files per version. The difference is distribution: skills are versioned with rollback, can be public, private, or unlisted, and one published skill installs into multiple tools at once. The CLI writes each tool's native format from the same source — .claude/skills/ for Claude Code, .cursor/rules for Cursor, .windsurf/rules/ for Windsurf:

localskills install acme/api-conventions --target claude cursor windsurf

So the workflow that works well in mid-2026: study the official repo to learn the format, use skill-creator (or write by hand) to capture your own workflows, and publish the results somewhere your whole team — and all their tools — can install from.

FAQ: Anthropic skills

Are the skills in anthropics/skills free to use?

You can read, install, and adapt them; that's what the repo is for. The example skills are meant for reuse, while the document skills are published more as a source-available reference. Before redistributing anything from the repo, check the root LICENSE and per-folder notices, since terms differ between the two groups.

How do I install the official Claude skills in Claude Code?

Two ways: add the repo as a plugin marketplace with /plugin marketplace add anthropics/skills and install the skills you want, or copy a skill folder directly into your project's .claude/skills/ directory. Project-level skills are checked into your repo; personal skills live in ~/.claude/skills/.

Do Anthropic agent skills work with tools other than Claude?

The SKILL.md format is plain markdown with YAML frontmatter, and it has become a de facto convention beyond Claude. Some tools read it natively; for the rest, a registry like localskills.sh converts one published skill into each tool's own format — Cursor .mdc rules, Windsurf rules files, and so on — so a skill written once isn't locked to a single agent.

What's the difference between the document skills and the example skills?

The document skills (docx, pdf, pptx, xlsx) are production code — the actual skills behind file creation in the Claude apps — published for transparency and reference. The example skills (skill-creator, mcp-builder, webapp-testing, and others) exist to be installed, forked, and studied as templates for writing your own.

---

Ready to publish your own skills in the same SKILL.md format the official repo uses? Create a free localskills.sh account and share your first skill with your team in minutes.

npm install -g @localskills/cli
localskills login
localskills publish