Anthropic's skills repository: a working reference for Claude's Agent Skills
Anthropic's public collection of Agent Skills — folders of instructions, scripts, and resources Claude loads dynamically to perform specialized tasks repeatably.
At a glance
- What is it?
- A look at anthropics/skills, the public home of Claude's Agent Skills examples, the SKILL.md format, and the document skills behind Claude's file creation. The repository is a demonstration set, not a production library, and that distinction matters.
- Who is it for?
- Adopt this repository if you are building skills for Claude and want concrete, working examples of the SKILL.md format, especially the document skills for DOCX, PDF, PPTX, and XLSX. Do not adopt it as a drop-in production toolkit: the skills are provided for demonstration and education, and the implementations may differ from what Claude actually runs.
- Can I use it commercially?
- Not without permission. GitHub finds no licence file in the repository, and without a licence all rights are reserved by default: you may read the code but not reuse it. Check the README, or ask the authors, before using it.
- Is it still maintained?
- Yes. The repository last received commits 5 days ago.
- What is it written in?
- Mainly Python, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on September 15, 2026, and from our analysis. They are not legal advice.
DEEP OPEN-SOURCE ANALYSIS
What the repository actually contains
Anthropic's skills repository is a folder of example skills for Claude, each self-contained in its own directory with a SKILL.md file. The README groups them into creative applications, technical tasks, and enterprise workflows. The repository also holds the Agent Skills specification under ./spec and a skill template under ./template. The most notable contents are the document skills in skills/docx, skills/pdf, skills/pptx, and skills/xlsx. The README states these power Claude's document capabilities under the hood and are included as a reference for complex skills used in a production AI application. That is a different status from the rest of the repo, and it affects how you should treat them.
The SKILL.md format is the core mechanism
A skill is a folder with a SKILL.md file containing YAML frontmatter and markdown instructions. The frontmatter requires exactly two fields: name, a lowercase identifier with hyphens for spaces, and description, a complete statement of what the skill does and when to use it. The markdown body holds the instructions, examples, and guidelines that Claude follows when the skill is active. The template-skill in the repository shows the minimal shape: a name, a description, then sections like Examples and Guidelines. This is a simple contract. The complexity lives in the instructions themselves, not in the packaging. The README gives no detail on how the model selects or loads a skill beyond the description field, so the matching logic remains opaque from this material.
Installation paths: plugin marketplace, Claude.ai, and the API
The README documents three ways to use these skills. In Claude Code, you register the repository as a plugin marketplace with the command /plugin marketplace add anthropics/skills. Then you browse and install either the document-skills or example-skills plugin, or run /plugin install document-skills@anthropic-agent-skills or /plugin install example-skills@anthropic-agent-skills. After installation, you invoke a skill by mentioning it, for example asking Claude Code to use the PDF skill to extract form fields from a file. On Claude.ai, the example skills are already available to paid plans, and you can upload custom skills following the linked support article. For the Claude API, the README points to a Skills API Quickstart for pre-built and custom skills. The plugin route is the most concrete, and it treats the repository as a distributable bundle rather than a source tree you copy.
The document skills are a separate licensing case
Most skills in the repository are open source under Apache 2.0. The document creation and editing skills in the docx, pdf, pptx, and xlsx folders are not. The README calls them source-available, not open source. They are shared as a reference for complex skills used in a production AI application, but you cannot treat them like the Apache-2.0 examples. If you plan to build on those document skills, check the license file inside each folder before copying code. The README does not give a specific license identifier for them, only the source-available label. That is a real constraint for teams that need permissive redistribution. The distinction also signals that these skills are closer to production code than the demo skills, which means they carry more integration complexity and more risk if you modify them.
Limitations: demonstration status and behavioral drift
The README contains a disclaimer that the skills are provided for demonstration and educational purposes only. It states that implementations and behaviors you receive from Claude may differ from what is shown in these skills. This is the central limitation. The repository is not a guarantee of behavior, and it is not a maintained product. If you rely on a skill for a critical task, the README explicitly says to test it thoroughly in your own environment first. Another limitation is that the repository does not include versioned releases, and the last push date is unknown. There is no release history to track changes against. For a production deployment, that means you are pinning to a moving main branch or forking the repository yourself. The wrong tool would be a team expecting stable, versioned, production-ready skills with support guarantees.
Alternatives: build your own skills or use Claude's built-in set
The obvious alternative is to write your own skills instead of using these examples. The repository itself encourages this by providing the template-skill and pointing to Anthropic's guide on creating custom skills. The difference in approach is control: with your own SKILL.md, you define the exact instructions, examples, and guidelines for your domain, and you avoid any behavior drift from Anthropic's examples. The trade-off is effort. You must write and test the instructions yourself, and you lose the reference implementations for complex formats like DOCX or PDF. Another alternative is to rely on the skills already available in Claude.ai or the API without pulling from this repository. That route gives you Anthropic-managed behavior, but you cannot inspect or modify the underlying instructions. The repository sits between those two: it gives you inspectable code that you can adapt, but with the caveat that it is not the exact code Claude runs.
Maintenance and upgrade cost
The repository has no listed releases and no release notes, so there is no structured upgrade path. Changes land on the main branch, and you will need to track them manually or fork. The README does not describe a contribution process or a changelog. For the document skills, the maintenance cost is higher because they are tied to Claude's production document capabilities; if Anthropic updates those internally, the public reference may lag or diverge. The Apache-2.0 example skills are simpler to maintain because they are small and self-contained, but they still require you to test against Claude's current behavior. The licensing split adds another maintenance burden: you must track which folders are Apache-2.0 and which are source-available, and ensure your use complies with each. There is no single license file covering the whole repository, so the cost of compliance is per-skill.
Editorial conclusion
Adopt this repository if you are building skills for Claude and want concrete, working examples of the SKILL.md format, especially the document skills for DOCX, PDF, PPTX, and XLSX. Do not adopt it as a drop-in production toolkit: the skills are provided for demonstration and education, and the implementations may differ from what Claude actually runs. Before relying on any skill, verify the license for each folder, since the document skills are source-available but not open source, and test the skill's output against your own files and workflows. The repository is a reference and a template source, not a maintained application.
Community notes