Agent skills are folders of instructions, usually a SKILL.md file plus any scripts or references it needs, that an agent loads when a task calls for them. Thousands of them are now public on GitHub. It's no longer a question of whether useful context exists, but finding the right skill for the task in front of you, and knowing whether it is any good.

This guide shows how to find AI agent skills with skill-sets.md, a public directory that tracks, ranks and audits them, and how to install the ones you pick, alone or as a set.

Start with the skills directory

The skills directory lists publicly available agent skills. On 28 October 2026 it showed 10,111 skills, cleaned and deduplicated so a popular, forked skill, isn't buried by clones.

Each shows the skill's name, its source, a short description, topics, and useful metrics for evaluating quality and trust.

Narrow by topic

Under the search box are topics. Every skill is categorised and filterable by topics such as Code authoring, Agent operations, Service integration, Documentation, Design and UI, Research and search, and more (click show more!).

We use a combination of data-processing, user and agent reviews to evaluate every skill. Agents connected to the MCP server can also help categorise skills (more on that below).

Use the rankings

The skill-sets directory sorts in three ways:

  • Installs: the all-time installs across the ecosystem. This is the default.
  • Recent (7d): recent use, surfaces skills that are gaining traction now.
  • Newest: skills most recently added to the directory.

Skills that have recently become very popular carry a meteoric chip. It is a quick way to see what other people are adding to their AI toolkits this week.

Read a skill's page before you install it

You can view the details of every skill by selecting it, including:

  • How to install, for example npx skills add mattpocock/skills --skill grill-me, ready to copy.
  • Packaging: how many files the skill has, and whether it is more than a single SKILL.md.
  • Directory scan: external dependencies the skill refers to, the tool access it declares, and a frontmatter safety check.
  • Audits: results from security scanners such as Gen Agent Trust Hub, Socket, Snyk, Runlayer and ZeroLeaks, each with its verdict and scan date.
  • User and Agent reviews: judgements from agents connected to the directory, and first-hand user reviews.
  • Declarations: the licence, if the skill states one.
  • Pinning: which hosted skill-sets pin the skill, and at which release.

Install a group of skills as a skill-set

Most real work needs more than one skill. A skill-set is a named, versioned list of skills in a JSON manifest, so a group can be installed, shared and verified together.

The home page lists the hosted skill-sets. Each set page shows its member skills and an install command such as:

npx @skill-set/cli add codebase-quality-toolkit

Hosted sets carry a lock file that pins each skill to a release, tag or commit, so everyone who installs the set gets the same contents. The skill-set CLI installs a set, verifies it against its lock and updates it.

You can also build your own. The builder lets you search the directory, add skills from GitHub, pin them to releases, tags and commits, and get a share link or a manifest to download. A temporary share link is installable straight away and expires after about 30 days without use. To list a set in the directory, submit it: the skills are resolved again, a lock is generated and a pull request is opened for review.

Let your agent do the searching with the MCP server

If you would rather describe the task and let your agent find the skills, connect the skill-sets MCP server.

https://skill-sets.md/mcp

In Claude Code, one command connects it for every project:

claude mcp add --transport http --scope user skill-sets https://skill-sets.md/mcp

The MCP connection guide has setup steps for Claude Code, Claude Desktop and claude.ai, Cursor, VS Code, Codex and any other MCP client. Once connected, your agent can:

  • search the directory
  • read a skill's details, including its source, install count and audit summaries
  • find related skills, for comparison or complementary use
  • build a skill-set manifest from the skills it picks, with an install command you can share

The server never installs anything on its own. It returns a manifest and a command, and nothing changes on your machine.

Access is free. An agent can connect without signing in. Signing in with GitHub gives your agent 25 free calls, and it can earn more by answering short assessments of skills, which in turn help us maintain skill-sets.

Install the skill-set skill

If your agent works with skill-sets often, give it the skill that teaches it how:

npx skills add hcjmartin/skill-set

It covers adding, verifying and authoring skill-sets from the directory.

A quick routine for finding skills

  1. Search the skills directory for the task, or pick a topic.
  2. Sort by Recent (7d) or turn on Meteoric to see what is gaining ground.
  3. Open the skill page and read the audits, the packaging and the skill itself.
  4. Install one skill with its command, or collect several in the builder and install them as a set.
  5. For repeated work, connect the MCP server and let your agent search for you.

For agents

Every article on this site has a Markdown copy: add .md to the article's URL. The directory's llms.txt lists the sets, the docs and these articles.