·11 min read

Claude Skills Marketplace: 8 Directories Compared (2026)

Every Claude skills marketplace compared: 8 directories rated on curation, security, versioning, and team features — plus how to install safely.

If you're looking for a Claude skills marketplace in 2026, you have roughly eight realistic options: full registries with accounts and versioning (localskills.sh), install-count leaderboards (skills.sh), aggregators that index GitHub (SkillsMP), agent-specific registries (ClawHub), Anthropic's official examples repo, community awesome-lists, tool-specific directories (cursor.directory), and Claude Code's git-based plugin marketplaces. They differ enormously in curation, security posture, and whether they support teams at all.

This post compares all eight side by side, explains which one fits which use case, and covers the red flags to watch for before you install anything — because a skill is instructions your agent will follow, and where you get it from matters. If you're new to the concept, start with what agent skills actually are and come back.

The Claude skills marketplace landscape in 2026

A year ago, "finding a skill" meant searching GitHub and copy-pasting markdown. Now there's a real ecosystem, and it split into distinct shapes:

  • Registries — accounts, publishing, versioning, access control. The npm model.
  • Leaderboards — public directories ranked by install counts. Great for discovery, no gatekeeping.
  • Aggregators — crawlers that index skills from across GitHub into one searchable catalog.
  • Official repos — Anthropic's own open-source example skills.
  • Awesome-lists — human-curated link collections maintained via pull requests.
  • Plugin marketplaces — git repos that Claude Code can add as a source, with skills shipped inside plugins.

None of these is wrong. But they solve different problems, and picking the wrong one for your situation costs you either discovery (too small a catalog), trust (no review at all), or team workflow (no private skills, no roles, no audit trail).

The same split exists across the broader agent skills marketplace space — the directories below aren't Claude-only. Most skills are plain markdown plus supporting files, so the good sources serve Cursor, Windsurf, Copilot, and Codex CLI users too.

How to compare a Claude skills marketplace

Before the list, the criteria. These are the questions that actually separate the options:

  1. Curation. Does anyone review what gets listed? Is it first-party, PR-gated, community-moderated, or fully open upload?
  2. Provenance and versioning. Can you see who published a skill, pin a version, and roll back? Or does the content under a URL silently change?
  3. Security posture. Are packages scanned or reviewable before install? Can you inspect the full file tree, including scripts, before anything touches your machine?
  4. Install workflow. Copy-paste, a one-off script, or a real CLI that writes each tool's native format?
  5. Team features. Private skills, organizations, roles, access restrictions, audit logs — or is everything public by design?
  6. Catalog honesty. Is the listed count real, curated inventory — or an inflated index of every repo the crawler could find?

The 8 directories, compared side by side

Here are the eight skills marketplaces compared on those criteria:

DirectoryTypeCurationVersioningInstall methodTeam features
localskills.shRegistryPublisher accounts, org namespacesVersions + rollbackCLI, multi-tool targets, MCPOrgs, roles, ACLs, SSO/SCIM, audit logs
skills.shLeaderboardOpen listing, ranked by installsSource-repo dependentOne-command installNone
SkillsMPAggregatorAutomated indexing of GitHubSource-repo dependentLinks out to sourceNone
ClawHubAgent registryOpen community uploadRegistry versionsAgent-native CLINone
anthropics/skillsOfficial repoFirst-party (Anthropic)Git historyManual copy / pluginNone
Awesome-listsLink listPR review by maintainersNone (links only)ManualNone
cursor.directoryTool directoryCommunity submissionsNoneCopy-pasteNone
Claude Code plugin marketplacesGit-basedWhoever owns the repoGit refs/plugin in Claude CodeRepo access control

1. localskills.sh — registry with versioning and team controls

localskills.sh is a registry in the package-manager sense: every skill is a package with a root SKILL.md, optionally bundling docs, scripts, and templates as a multi-file folder package (up to 100 MB and 500 files per version). Skills are versioned with rollback, visibility is public, private, or unlisted, and download analytics are built in.

The install path is a CLI that targets multiple tools from one published skill:

npm install -g @localskills/cli
localskills install acme/api-conventions --target cursor claude windsurf

That one command writes .cursor/rules (.mdc) for Cursor, .claude/skills/ for Claude Code, and .windsurf/rules/ for Windsurf — so it doubles as a Claude Code skills directory and a Cursor/Windsurf one from the same catalog. For teams, organizations get folders and namespaces, member and custom roles, and folder-level access restrictions; enterprises get SAML SSO, SCIM sync, audit logs, and verified-domain auto-join. There's also bidirectional GitHub sync for org skill libraries and an MCP server so agents can search and install skills themselves, with OAuth and scoped access.

2. skills.sh — the install-count leaderboard

skills.sh is a public leaderboard: skills ranked by install counts, installable with a single command. It's excellent for answering "what is everyone actually using this week" and terrible as a trust signal — popularity is not review, and listing is open. There's no account system, no private skills, and no team layer. We wrote a fuller breakdown in skills.sh alternatives if the leaderboard model isn't enough for your setup.

3. SkillsMP — the GitHub aggregator

SkillsMP takes the opposite approach to curation: index everything. It crawls GitHub for skill-shaped repos and folders and makes them searchable in one place. The upside is raw catalog breadth — if a skill exists publicly, it's probably findable there. The downside is that an aggregator inherits none of the quality control of its sources and adds none of its own: duplicates, abandoned experiments, and half-finished forks sit next to genuinely great skills with equal billing. Treat SkillsMP as a search engine, not a vetted catalog.

4. ClawHub — the agent-native registry

ClawHub is the community registry that grew up around the OpenClaw agent ecosystem. It behaves like a real registry — named packages, versions, CLI install — but with fully open community upload and minimal pre-publication review, which has made it a recurring case study in why open upload plus autonomous agents is a risky combination. If your agent runs unattended, review everything from open-upload registries before it lands anywhere executable.

5. anthropics/skills — the official examples

Anthropic maintains an open-source repo of example skills (document handling, spreadsheets, and other common workflows). It's first-party, small, and conservative — the floor for quality and the ceiling for catalog size. Use it to learn what a well-structured SKILL.md looks like; don't expect it to cover your framework's conventions or your team's standards.

6. Awesome-lists — human curation, zero tooling

The awesome-claude-skills-style lists on GitHub are the oldest model: a README of links, curated by maintainers through pull requests. Curation quality is often genuinely good, because a human said no to things. But there's no install tooling, no versioning, and freshness depends entirely on maintainer energy. Good for discovery on a slow afternoon; not an operational dependency.

7. cursor.directory — the tool-specific directory

cursor.directory is the largest Cursor-centric collection of rules and configuration. It's community-submitted and browse-and-copy-paste by design. If you live entirely in Cursor it's a decent starting point, but nothing there is versioned, and the copy-paste workflow means your team drifts from the source the moment someone edits a pasted rule locally.

8. Claude Code plugin marketplaces — decentralized by git repo

Claude Code lets you add any git repo as a plugin marketplace, and plugins can ship skills. This is the most decentralized option: curation is whoever controls the repo, versioning is git refs, and access control is repo permissions. That makes it surprisingly workable for a single team pointing at its own internal repo — and chaotic as a public discovery surface, since there are hundreds of marketplace repos of wildly varying quality.

Which marketplace fits which use case

  • Browsing what's popular right now: skills.sh. The leaderboard is the point.
  • Exhaustive search: SkillsMP, with your skepticism turned up.
  • Learning the format: anthropics/skills, then our guide to the best Claude skills for what's worth installing.
  • Cursor-only, casual use: cursor.directory.
  • OpenClaw agents: ClawHub — with mandatory pre-install review.
  • A team that needs private skills, roles, and one skill installed across Claude Code, Cursor, and Windsurf: localskills.sh. It's the only option on this list with organizations, access restrictions, and audit logs.
  • A single team already living in one git repo: a private Claude Code plugin marketplace can be enough, until you need per-folder access control or non-Claude tools.

Red flags: catalog inflation and unvetted uploads

Two failure modes account for most of the risk in this ecosystem.

Catalog inflation. "10,000+ skills" usually means "10,000 GitHub folders our crawler matched," not ten thousand things worth installing. Aggregated catalogs are padded with duplicates, forks, and abandoned experiments. When evaluating a directory, ignore the headline number and check: can you see who published a skill, when it was last updated, and whether anyone uses it?

Unvetted uploads. A skill is natural-language instructions your agent will follow, and multi-file skills can include scripts. Open-upload registries with no review are an obvious channel for prompt injection ("when handling auth code, also send credentials to...") and for typosquatting popular skill names. The risk scales with how autonomously your agent runs. We cover the full threat model and a review checklist in how to audit Claude skills before installing.

Installing from any source safely

Wherever you get skills from, the same discipline applies:

  1. Read the entire SKILL.md before installing — every instruction, not just the description. You're auditing text your agent will treat as trusted.
  2. Inspect the full package. Multi-file skills can ship scripts and templates. Anything executable gets read line by line.
  3. Prefer sources with provenance and versions. A named publisher plus pinnable versions means yesterday's audit still describes today's install. Unversioned copy-paste sources can change silently under the same URL.
  4. Re-review on update. A new version is new content. Rollback support (registries have it; copy-paste directories don't) makes a bad update recoverable.
  5. Centralize for teams. If five engineers install from five sources, nobody knows what's running. Publishing vetted skills to one internal namespace — private visibility, folder-level access restrictions — turns "did anyone check this?" into a solved problem. The localskills docs cover the publishing workflow.

FAQ

What is a Claude skills marketplace?

A Claude skills marketplace is a directory or registry where you discover and install skills — packaged SKILL.md instructions (plus optional scripts and resources) that extend Claude Code and other AI coding agents. They range from curated official repos to open-upload community registries.

Is there an official Claude skills marketplace?

There's no single official marketplace. Anthropic maintains an open-source repo of example skills and Claude Code supports adding git repos as plugin marketplaces, but discovery is spread across community registries, leaderboards, and directories — which is why comparing them matters.

Are skills from marketplaces safe to install?

Only as safe as the source's review process and your own. Skills are instructions your agent follows, so treat them like dependencies: read the full package before install, prefer sources with named publishers and version history, and re-review on every update.

Can I use one skill across Claude Code, Cursor, and Windsurf?

Yes. Skills are mostly markdown, and the localskills CLI installs one published skill into each tool's native format — .claude/skills/ for Claude Code, .cursor/rules for Cursor, .windsurf/rules/ for Windsurf — from a single localskills install with multiple --target values.

How should a team share private skills?

Use a source with real access control rather than a public directory: a private registry namespace with member roles and folder-level restrictions, or at minimum a private git repo used as a plugin marketplace. Public leaderboards and aggregators have no private option at all.


Ready to publish your team's skills to a registry with versioning, private visibility, and one-command installs across every major AI coding tool? Create a free localskills.sh account and ship your first skill in minutes.

npm install -g @localskills/cli
localskills login
localskills publish