Agent Skills Marketplace: Where to Find Claude Skills

If you have started using Agent Skills, you have probably hit the obvious next question: where do you actually get more of them, and is there an agent skills marketplace you can browse like an app store? You build one or two yourself, they save you real time, and then you start wondering whether someone has already packaged up the exact workflow you keep repeating by hand.

The short version is that there is no single official “app store” the way there is for phones — but there is a fast-growing ecosystem of places to find, browse, and install skills. This guide walks through the whole landscape: the sources Anthropic maintains, the community marketplaces that aggregate thousands of skills, the GitHub collections worth your time, and — just as important — how to vet a skill before you let it run on your machine.

Is there an agent skills marketplace?

Yes — there are several, but “marketplace” mostly means a searchable directory that points to open-source skill files on GitHub, not a paid app store with a checkout.

An Agent Skill is just a folder with a SKILL.md file (plus any helper scripts), so distributing one is as simple as sharing a folder. That low bar is exactly why the ecosystem exploded: anyone can publish, and dozens of sites have sprung up to index what people publish. When someone says “agent skills marketplace,” they usually mean one of these aggregators that crawl public repositories, tag the skills, and give you a clean way to search.

The trade-off is that quality varies wildly. A directory listing thousands of skills is not the same as a curated shelf — most entries are someone’s personal experiment. So the right mental model is less “trusted store” and more “search engine for skill files.” That makes the safety section later in this guide non-optional. If you are still fuzzy on how skills differ from related features, the breakdown in claude skills vs mcp vs subagents is worth a detour before you start installing things.

The official sources

Start with Anthropic’s own materials and the built-in plugin marketplace inside Claude Code — these are the only sources Anthropic itself stands behind.

There are three official starting points worth knowing:

The Claude docs. The canonical reference for what skills are, how the SKILL.md format works, and how Claude loads them lives at https://docs.claude.com/en/docs/claude-code. Read this first; everything else in the ecosystem is built on top of the format it describes.
Anthropic’s pre-built skills. Anthropic ships a set of first-party skills for common document work — creating and editing PowerPoint, Excel, Word, and PDF files. These are available across the surfaces where Claude runs, and because Anthropic maintains them, they are the safest skills you can use.
The plugin marketplace inside Claude Code. Claude Code has a built-in plugin system, and skills can be distributed as part of a plugin from a marketplace you add by its repository address. This is the closest thing to a native install flow: you point Claude Code at a marketplace, browse what is there, and install with a command rather than copying files by hand.

If you have not set the tool up yet, the install claude code walkthrough gets you to a working command line, and the broader claude code guide covers how plugins and skills fit into day-to-day use. Non-developers should not feel shut out here — claude code for non-coders exists precisely because a lot of skill users never write code themselves.

Community collections on GitHub

The richest source of real, battle-tested skills is GitHub itself — both curated “awesome” lists and the third-party marketplace sites that simply index those same repositories.

Because a skill is just a folder, GitHub is the native home of the ecosystem. You will find two flavors:

Curated collections and “awesome” lists. These are human-maintained repositories that link out to skills worth trying, often grouped by job — writing, data analysis, DevOps, design. The advantage is curation: a maintainer has at least glanced at each entry. Search GitHub for terms like “awesome claude skills” or “agent skills collection” and sort by recent activity, since this space moves fast.
Marketplace and directory sites. A wave of third-party sites now crawls public GitHub repos and presents them with search, tags, and previews. SkillsMP, SkillHub, claudeskills.info, MCP Market, LobeHub, and similar directories all do a version of this. They are genuinely useful for discovery — you can search by keyword and compare a few options side by side — but remember that almost none of them author the skills. They are a window onto GitHub, so the trust still bottoms out at the original repository.

When you find a candidate, click through to the actual source repo and read the SKILL.md before anything else. A good skill is usually a small, focused folder you can read top to bottom in a couple of minutes. If you want to understand what “good” looks like from the inside, how to create an agent skill shows the anatomy, and claude skills examples collects patterns you will recognize once you start browsing.

What to watch out for (security)

A skill can include scripts that run on your machine, so treat installing one with the same caution you would treat running any code you found online — read it before you trust it.

This is the part most directory sites gloss over. Skills are not sandboxed magic; many of them ship helper scripts that Claude can execute, and those scripts run with whatever permissions you have given the tool. A malicious or careless skill could read files it should not, make network calls, or run commands you did not intend. The convenience of “just install it” is exactly what makes a quick safety habit worth building.

A practical checklist before you install:

Read the SKILL.md and any scripts. If a skill is small enough to be useful, it is small enough to skim. Look for anything that touches the network, reads broadly across your filesystem, or executes shell commands you do not understand.
Check the source, not the directory. Marketplace listings are convenience layers. Open the underlying GitHub repo, look at who maintains it, when it was last updated, and whether anyone else uses it.
Watch for prompt-injection risk. Skills work by feeding instructions into Claude’s context. A poorly written or hostile skill can try to steer the model in ways you did not ask for. The same instincts that apply to mcp security apply here — untrusted instructions are untrusted instructions, wherever they come from.
Prefer least privilege. Don’t grant a document-formatting skill broad system access. Keep Claude’s permissions scoped to what the task actually needs.

If you are weighing skills against connecting external tools, the same threat model shows up in claude code mcp, where an MCP server is another piece of third-party code you are inviting in. Skills and MCP servers are different mechanisms, but the “would I run this code?” question is identical.

How to find the right skill (a short roadmap)

Define the job first, search the directories second, and read the source last — in that order — so you end up with a skill that fits the work rather than the most-starred one you found.

Here is a sequence that keeps you out of trouble:

Name the repetitive task. “Format my release notes,” “summarize a CSV,” “draft an email in our house style.” A skill is worth installing only if it removes a recurring chore. If the task is one-off, a plain prompt is faster.
Search a directory by that job. Use one of the marketplace sites — SkillsMP, SkillHub, MCP Market — and search the verb, not the brand. You want the skill that does the thing, not the most popular skill in an adjacent category.
Open the source repo. Click through to GitHub and read the SKILL.md. Check the last commit date and whether the scope matches your task. Abandon anything that does far more than you need.
Run the safety pass. Skim the scripts using the checklist above. If anything is unclear, it is fine to walk away — there is almost always another skill for the same job.
Install and test on something harmless. Try the skill on a sample file or a throwaway task before you point it at anything important.

When you cannot find a good match, building your own is genuinely approachable — that is the whole point of the format. how to create an agent skill takes you from empty folder to working skill, and for higher-level orchestration, claude code subagents and claude code workflow show how skills slot into bigger automations.

FAQ

Is there an official Claude skills marketplace?

There is no single official store, but Anthropic maintains pre-built skills for document tasks and supports a plugin marketplace inside Claude Code that you add by repository address. The third-party “marketplaces” you find by searching are community directories, not official Anthropic products. Start with the official docs and the built-in plugin system before relying on outside sites.

Are agent skills free?

The vast majority are free and open source, because a skill is just a SKILL.md folder published on GitHub. Some creators may sell premium bundles or templates, but the core ecosystem — including Anthropic’s own pre-built skills — costs nothing beyond your normal Claude usage. The real “cost” is the time it takes to vet a skill before you trust it.

Are agent skills safe to install?

They are as safe as the code inside them, which means you should read the SKILL.md and any scripts before installing. Skills can run commands on your machine, so a hostile or sloppy one carries real risk. Stick to maintained repositories, prefer least privilege, and apply the same caution you would to any third-party code — the principles in mcp security carry over directly.

What is the difference between a skill, an MCP server, and a subagent?

A skill is packaged knowledge and instructions Claude loads on demand; an MCP server connects Claude to an external tool or data source; a subagent is a separate Claude instance you delegate a task to. They solve different problems and often combine. The full comparison lives in claude skills vs mcp vs subagents.

Where do I find new skills as the ecosystem grows?

Watch GitHub directly — search for “awesome claude skills” lists and sort by recent activity — and check the aggregator directories like SkillsMP, SkillHub, and MCP Market for keyword search. Because these directories index public repos, new skills show up there shortly after they are published. Following a few active maintainers on GitHub is the fastest way to catch quality additions early.

Conclusion

The “agent skills marketplace” is less a single destination and more a layered ecosystem: Anthropic’s official docs and pre-built skills at the trusted core, the built-in plugin marketplace inside Claude Code for native installs, GitHub as the real home of the code, and a swarm of third-party directories that make all of it searchable. Each layer is useful, but the trust always bottoms out at the source repository.

If you remember one thing, make it this: a skill is code, and you read code before you run it. Name the job you want done, search the directories by that job, open the source, run a quick safety pass, and test on something harmless. Do that consistently and the ecosystem becomes a genuine force multiplier rather than a security headache. When you are ready to go further, how to create an agent skill and the broader claude code guide are the natural next steps.