Open Source · Agent Skills · Updated daily via GitHub Actions
A curated index of agent skills libraries and collections — folders of instructions and scripts that agents such as Claude Code and Codex load on demand — with live GitHub stats. Building an agent instead? See the frameworks index.
An agent skill is a folder containing a SKILL.md file — instructions, with a name and a short description at the top — plus any scripts and reference files the instructions need. The agent sees only the names and descriptions up front and loads a skill’s full contents when the task calls for it, which keeps context small while making a large library available.
A skills library is a collection of these folders: an official set from a vendor, a large community catalog, a focused set for one domain, or a curated list pointing elsewhere. The format is documented as an open specification and is read by more than one agent, so a good skill is portable. The flip side is that a skill can bundle executable scripts, so installing one is closer to installing software than to copying a prompt.
Every entry above is tagged by kind. Use the Topic filter to isolate any of these — or read on for what each one means.
The specification and documentation for the skill format itself — what a valid skill looks like.
Published by the vendor or team behind the product or agent, so it tracks that product and is accountable for it.
Large community libraries of many skills, often installable as plugins or bundles. Breadth varies; so does quality.
Curated “awesome” lists that point to skills elsewhere. Good for discovery; the skills themselves live in other repositories.
Focused libraries for one field — science, design, security, marketing, note-taking — with skills that share a point of view.
Installers, converters, scanners and optimizers around skills, rather than skills themselves.
Skills for security work, or tools that scan skills for risky content. Use security skills only on systems you are authorized to test.
Bundles that package skills together with commands, sub-agents or hooks so they install as one unit.
Libraries that encode a whole working method — planning, testing, review — rather than a single capability.
They solve different problems and are often used together.
Instructions and scripts loaded into the agent’s context on demand. A skill teaches the agent how to do something — a procedure, a house style, a workflow.
A live connection to a tool or data source. An MCP server gives the agent access to something — a database, a browser, an API.
The control layer that decides what the agent may actually do with either — permissions, budgets, approvals and verification.
Six checks worth making before you install skills into an agent that can run commands on your machine.
Some libraries target Claude Code, others Codex, GitHub Copilot or OpenClaw. The format is shared, but packaging and install steps differ, so check the README for your agent.
A skill may include scripts that execute on your machine. Read them before installing, prefer official or widely reviewed libraries, and consider running a scanner from the tooling group.
A small vetted library and a catalog of thousands serve different needs. Bigger is not better: a skill you never trigger is just clutter.
A skill loads only when its description matches the task. Overlapping or vague descriptions across many skills make an agent pick the wrong one or none at all.
MIT and Apache-2.0 are unrestricted for commercial use. Some collections mix content under different terms, so check before redistributing.
The Updated column shows the last push. Skills reference tools and APIs that change, so an untouched library goes stale quickly.
A folder with a SKILL.md file — a name, a description and instructions — plus optional scripts and reference files. An agent loads the full skill only when the task matches its description.
A skill teaches the agent a procedure and is loaded into its context on demand. An MCP server gives the agent live access to a tool or data source. Skills add know-how; MCP servers add reach.
Treat them like any code you run. A skill can include scripts, and its instructions steer what the agent does. Prefer official or widely reviewed libraries, read a skill before installing it, and scan third-party ones.
The format is documented as an open specification, and several agents load the same SKILL.md structure. This index includes collections aimed at Claude Code, Codex, GitHub Copilot and OpenClaw — check each library’s README for exactly what it supports.
It depends on the agent and the library. Some libraries ship as plugins, some as folders you copy into your agent’s skills directory, and there are installer tools listed under tooling. The library’s README is authoritative.
By hand: substantial collections and official libraries, plus the specification and tooling around them. Single-purpose one-off skills are out of scope. Each entry is verified against the live GitHub API when the list is reviewed.
Stars, forks and topics are refreshed daily via GitHub Actions — the freshness indicator in the nav bar shows the age of the current data pull. The list itself is reviewed by hand periodically.