Anthropic Agent Skills Collection
A public collection of Agent Skills covering documents, creative work, development, communication, and skill templates.
Quick start
/plugin marketplace add anthropics/skillsWhat it does
This repository demonstrates how self-contained skill folders can teach an agent a repeatable workflow. It includes examples, a starter template, and document-oriented skills used in production.
Best for
- Studying real-world
SKILL.mdpatterns - Adding document creation and editing workflows to Claude
- Starting a team-specific skill library from a tested structure
Before you use it
Licensing varies inside the collection. Some skills are open source while some document skills are source-available references, so check the notice for the specific folder you plan to reuse.
How to evaluate the collection
Open the folder for the individual skill and read its complete SKILL.md, referenced files, and scripts. Check that its trigger is narrow enough for your workflow and that any bundled commands are appropriate for your operating system and permissions. If you adapt a skill, keep the original license notice and document the changes your team makes.
Setup and operating notes
Install only the skills you need, then test them with representative inputs that do not contain confidential material. Confirm that generated files, citations, and formatting meet your acceptance criteria. For a shared installation, pin the reviewed revision or use a controlled update process so new upstream instructions do not reach production without review.
Alternatives
Use OpenAI Skill Creator when the goal is to author and validate a Codex-specific skill from a repeatable workflow. Use OpenAI Docs Skill when the task is specifically grounded research across current OpenAI documentation.
What to inspect in the repository
Treat the collection as a set of independent packages rather than one dependency with a single policy. Start in the folder for the skill you want, read its SKILL.md, and follow every referenced file. Check whether it includes scripts, templates, example assets, or environment assumptions. A document workflow may need local rendering tools while a creative workflow may rely mainly on instructions and examples.
The collection is also useful as a design reference. Notice how a focused description controls activation, how longer material is separated from the main workflow, and how output verification is made part of the instructions. Reuse the structural ideas without copying a folder whose license or platform assumptions do not fit your project.
Installation and team use
The marketplace command is convenient for Claude Code, but installation should follow review. Pin or record the source revision used by your team so a later upstream change does not silently alter a controlled workflow. Test the installed skill with non-sensitive sample files and verify where generated documents, temporary files, and logs are stored.
For shared use, document which folders are approved, who owns updates, and whether local modifications diverge from upstream. Review updates like dependency changes: compare the diff, rerun representative tasks, and confirm that required applications or network calls have not expanded.
Compatibility boundaries
The core SKILL.md pattern can inform skills on other agent hosts, but commands, plugin paths, tool names, and bundled capabilities are not automatically portable. When adapting a skill for Codex or another client, translate those integration points explicitly and re-test the trigger. If the goal is only to study portable authoring conventions, compare this collection with the OpenAI Skill Creator rather than assuming that one installation format works everywhere.
Keep exploring