Anthropic's Official Agent Skills: How to Use Them
The `anthropics/skills` repository run by Anthropic is the official place to load Claude's Skills system. It covers the SKILL.md format and installation commands.
Claude's skills repo is trending
Anthropic's anthropics/skills repository is climbing GitHub's trending charts and drawing attention. At the time of writing it has recorded roughly 174,000 stars and over 20,000 forks. It is the official implementation of the Skills system Claude uses, and it serves as a reference for the pattern of bundling repetitive work into a single folder so a task can run through one invocation.
Skills, put simply, are folders. They hold a prompt, scripts, and resources; when Claude starts a task it dynamically loads the folder to raise its performance on specialized work. The key point is that you don't need to rewrite a prompt every single time.
The concept of Skills that the repo explains
Skills make repeatable work possible, such as creating on-brand company documents or handling data in a specific way matched to a particular workflow. The repo README demonstrates this idea with example skills across four areas: creation & design, development & technical, business & communication, and document work.
The four building blocks
This repository is broadly split into 4 folders.
skills/— includes example skills that actually run.spec/— the Agent Skills standard document. It points to the official standard site (agentskills.io).template/— a minimal template to use as a starting point when creating a new skill.README.md— shows installation and usage guidance.
Most examples are published under Apache 2.0. That said, the document creation and editing skills (docx, pdf, pptx, xlsx) are not fully open source; they are offered in a source-available form that carries license terms. If you plan to use them in production, you need to check this part first.
SKILL.md — the standard format that starts with two fields
The identity of a skill is actually simple: inside the folder, the SKILL.md file is everything. Fill just two fields in the YAML frontmatter — name and description — and you have a skill. Below that, you write in markdown the instructions, examples, and guidelines that Claude follows whenever the skill is activated.
For instance, the template suggests an advisory structure after name and description, presenting real instructions, examples, and guideline items. Thanks to this minimal schema, anyone can create and share their own skill in just a few minutes.
Installing and using it in Claude Code right away
This repository can be registered as a plugin marketplace in Claude Code. The command below is officially documented in the repo README.
/plugin marketplace add anthropics/skills
Next, pick the skill bundle you want (such as document skills or example skills) from the list and install it. After installation, simply mentioning the skill in your prompt is enough to activate it.
For example, if you installed the document skills, you could say "extract the form fields of that file with the PDF skill." There is also a way to directly name and install a plugin such as document-skills or example-skills.
On Claude.ai's paid plan, example skills are already provided. With the Claude API, you can use pre-made skills or upload skills you have built yourself.
Interest measured by numbers
The repo's popularity shows in its scale too. It is recording 174.7k stars, 20.7k forks, and 1.1k watchers, with only 17 contributors. The repository's language makeup is Python 85.6%, HTML 11.4%, Shell 1.8%, and JavaScript 1.2%. There are no releases yet, so in practice the README, the example skills, and the standard document are effectively the whole repository.
Why it is useful for developers
The value of this repository can be seen in two ways. First, you can learn the skill pattern — fixing a repetitive task into a single command — alongside the standard document and real code. Second, best-practice examples for tool-specific skills (such as the Notion skill) are also open, similar to partner examples.
If your team wants to create and standardize its own skills, starting from the template/ folder is a good approach. One caution the README emphasizes: the public examples are intended for demo and educational purposes. In a real operating environment, you should verify them yourself first.
Reference links
Related posts
스타 24만 오픈소스 AI 에이전트 '헤르메스'
Nous Research가 만든 오픈소스 AI 에이전트 '헤르메스(Hermes)'를 분석했어요. CLI와 메신저 두 갈래로 쓰는 방식, 도구·스킬·메모리·MCP로 이어지는 확장 구조, 그리고 설치부터 첫 대화까지 2분이면 충분한 시작법까지 정리했어요.
Open Source & Dev ToolsThe open source tool teaching AI agents to 'write less', ponytail
ponytail is an open source tool that teaches AI coding agents YAGNI-based minimalism. Benchmarks show 54% fewer LOC, 22% fewer tokens, and maintained safety, with install guides and tunable commands.
fmtlib/fmt: A 25k-star modern C++ formatting library
fmtlib/fmt improves on printf with type safety, compile-time checks, Unicode support, faster builds, and smaller binaries.
Curated, fact-checked, and edited by a single operator before publishing.