The 5 Best skills.sh Alternatives in 2026

skills.sh is Vercel's open directory and package manager for AI agent skills, launched in January 2026. It works like npm for procedural knowledge: developers browse a leaderboard of skills ranked by install count and trending momentum, then run `npx skills add <owner/repo>` to install a skill into whichever agent they use. The CLI detects the agent (Claude Code, Cursor, Codex, GitHub Copilot, Windsurf, Zed and more — 20 agents in total) and writes the skill into the right directory. The directory had grown to 83,627 skills with 8M+ total installs, including an Official section listing 6,588 skills from 102 organizations. Every indexed skill is scanned by Gen Agent Trust Hub, Socket and Snyk, and skills that fail every partner audit are excluded. The CLI is MIT-licensed open source. The trade-off: it's a public discovery surface — no private skills, no version pinning, no team constructs.

pricing 0

• + Completely free — no sign-in needed to install and the CLI is MIT-licensed open source
• + Works across 20 coding agents, so skills are portable if you switch tools
• + Install counts and trending data provide real adoption signal (8M+ total installs)
• + Partner security scans by Gen Agent Trust Hub, Socket and Snyk, with failing skills excluded
• + Official section with 6,588 skills from 102 organizations offers a vetted starting point

• − Everything is public — no home for internal conventions, runbooks, or proprietary skills
• − No version pinning or meaningful rollback: installs point at whatever the author's repo contains right now
• − No review gate before a skill reaches your machine — installing is a supply-chain trust decision
• − No team constructs: no organizations, roles, access control, or audit trail
• − skills.sh itself states it cannot guarantee the quality or security of every listed skill

https://skills.sh

Agentman's Agent Skills is a hosted, curated skills library built for teams rather than solo browsing. It holds 90 production skills that are practitioner-authored, reviewed, and spec-compliant, and one MCP connection serves Claude, Claude Code, ChatGPT, Cursor, and Agentman agents from a single library. Every skill has version history with one-click restore, and the Business tier adds the controls a platform team wants: per-skill access control, use-only sharing (run a skill without seeing its instructions), approvals that review changes before they publish, and access and activity visibility. The free plan covers up to 10 private skills and 3 members with no credit card, which makes it an easy first step for a small business that has outgrown a public directory. Team and Business plans bill per seat with minimums, and Enterprise adds private or on-premise deployment.

pricing 0/month · free trial

• + Genuine free tier: 10 private skills and 3 members with no credit card
• + Real team controls — access levels, approvals, and activity visibility on Business
• + Use-only sharing lets you distribute a skill without exposing its instructions
• + Curated, reviewed library rather than an open firehose
• + Private and on-premise deployment available on Enterprise

• − Seat minimums mean the real entry cost is $30/mo (Team) or $125/mo (Business), not the headline per-seat price
• − Smaller catalog (90 production skills) than the open directories
• − Free tier caps private skills at 10, with an upgrade prompt at the limit
• − Tied to Agentman's ecosystem for the deepest features

https://agentman.ai/agentskills/pricing

localskills.sh is a full agent skills registry rather than a directory: teams publish skills as versioned packages and install them into whichever tools the team uses. A single published skill installs into Claude Code, Cursor, Windsurf, GitHub Copilot, Cline, Codex CLI, Aider and more — the CLI writes each tool-native format (.cursor/rules .mdc for Cursor, .claude/skills/ for Claude Code, .windsurf/rules/ for Windsurf) from one source of truth. Packages can be multi-file folders up to 100 MB and 500 files per version, with public, private, and unlisted visibility, versioning with rollback, and built-in download analytics. On the org side it offers folders and namespaces, teams, member roles and custom roles, plus enterprise plumbing: SAML SSO with setup templates for Okta, Cloudflare Access and Authentik, SCIM directory sync, audit logs, and verified-domain auto-join. Bidirectional GitHub sync lets review happen in normal pull requests, and an MCP server with OAuth lets agents search and install skills without a human in the loop.

pricing See pricing

• + Private, unlisted, and versioned skills — the gaps that push teams off skills.sh
• + Deepest enterprise feature set of the alternatives: SSO, SCIM, audit logs, custom roles
• + Multi-tool install from one source of truth eliminates parallel copies
• + GitHub sync means review can happen in normal pull requests
• + Download analytics show which skills earn their maintenance cost

• − Requires a publish step — overhead if you never need private distribution or versioning
• − Newer, smaller ecosystem than skills.sh's 83,627-skill directory
• − Less suited to solo developers who just want to browse community skills

https://localskills.sh

ClawHub grew out of the OpenClaw ecosystem and positions itself as the quality-over-quantity alternative to the giant directories. It sits at roughly 3,200 skills with 1.5M+ downloads — by far the smallest of the three major marketplaces, but the most curated. It has its own CLI (`npx clawhub@latest install skill-slug`), vector-powered semantic search, and proper versioning with rollback support, which neither skills.sh's install-count ranking nor SkillsMP's star-based sorting matches. Peter Steinberger, the OpenClaw creator, built it as the deliberate counterweight to the volume-first registries that exploded in early 2026. Agent support centers on OpenClaw with bridges to other tools, so it's narrower than skills.sh's 20-agent compatibility list. For teams that value curation and version control over raw catalog size, it's a credible middle path between a public firehose and a full private registry.

pricing See pricing

• + Most curated of the major marketplaces — quality-over-quantity approach
• + Proper versioning with rollback, unlike skills.sh's live-repo installs
• + Vector-powered semantic search beats browse-only discovery
• + 1.5M+ downloads show real usage despite the small catalog

• − Smallest catalog by far (~3,200 skills vs 83,627 on skills.sh)
• − Agent support centers on OpenClaw with bridges — narrower than skills.sh's 20 agents
• − Younger ecosystem with less community momentum than Vercel-backed skills.sh

https://clawhub.com

SkillsMP is the volume leader of the agent skills ecosystem: a community-run discovery platform that crawls GitHub for SKILL.md files and indexes them with AI-powered semantic search. Its catalog crossed 351,000 skills by early March 2026 and has since grown to roughly 1.9 million public skills scraped from GitHub — the npm-registry equivalent of the skills world, with a deep category taxonomy (89K tools skills, 70K development skills, 60K business skills). It supports Claude Code, Codex CLI, and ChatGPT, but its quality control is limited to GitHub stars and date, with no security review — the ecosystem report is blunt that you must inspect before installing. For a small business that wants maximum choice for browsing business- or operations-oriented skills, the sheer taxonomy is unmatched; for anyone who cares about vetting, it's a starting point for discovery rather than a distribution layer you standardize on.

pricing 0

• + Unmatched catalog size for discovery — roughly 1.9 million indexed skills
• + Semantic search and rich category taxonomy make browsing fast
• + Strong coverage of business and operations skill categories
• + Free to browse

• − No quality gate beyond GitHub stars — the ecosystem report says inspect before installing
• − No security review workflow
• − Supports only 3 agents versus skills.sh's 20
• − Browse-and-manual install rather than a universal one-command CLI

https://skillsmp.com

Anthropic maintains an official skills repository, and community "awesome" lists function as a de facto Claude skills directory in README form. This is the zero-infrastructure option: clone the repo, copy the folder into .claude/skills/, and you're done. It's free, fully transparent, easy to fork, and review happens naturally because you read the file before copying it. Anthropic also runs a small, manually curated directory inside Claude Code and a partner directory with skills from Atlassian, Canva, Cloudflare, Figma, Notion, Ramp, Sentry, Stripe, and Zapier — everything there is reviewed. The limits are structural: installs are manual, there's no multi-tool translation (a Claude-format skill doesn't become a Cursor rule automatically), no versioning or rollback, and no analytics. For a small business that already lives in Claude and wants a handful of trusted, reviewed skills with zero new tooling, it's the simplest possible answer.

pricing 0

• + Free and fully transparent — you read every file before it reaches your agent
• + Reviewed and verified by Anthropic, unlike open directories
• + Zero new infrastructure to adopt or maintain
• + Trusted first-party and partner publishers (Stripe, Sentry, Notion and more)

• − Installs are entirely manual
• − No multi-tool translation — a Claude-format skill doesn't become a Cursor or Windsurf rule
• − No versioning, rollback, or usage analytics
• − Small catalog compared to the community directories

https://github.com/anthropics/skills

SkillzBench is AIQ Labs' free, independent directory of AI agent skills, plus a free MCP tool that connects any agent to the skills you choose — currently in pre-launch with a live waitlist. It exists to answer the question the directories don't: which skills actually help, and which are risky. Published research found roughly 26% of skills carry at least one vulnerability, and SkillzBench is building a pipeline to address exactly that. Its pillars: Discover (continuously crawl and dedupe skills from GitHub), Verify (pin every version to a commit SHA, validate, license-check, security-scan, and quality-judge), Benchmark (test every skill in a sandbox with the same model, with and without the skill, and publish the Lift in percentage points), Equip (one free MCP server plus per-agent loadouts — the agent only sees what was chosen), and Stay current (track upstream changes, re-scan, re-benchmark, notify). A 0-100 trust score combines security, quality, license, maintenance and community signals; Lift is always shown separately from ratings, and money never touches scores. Everything on the platform — directory, MCP connector, benchmarks, alerts — is free, funded by advertising around blog articles that never influences scores or listings.

pricing 0

• + Everything is free — directory, MCP connector, benchmarks and alerts
• + Directly addresses the security problem: research found ~26% of skills carry at least one vulnerability
• + Commit-SHA pinning means you always know exactly what you reviewed versus what you installed
• + Benchmarked Lift shows whether a skill measurably helps, separately from ratings — money never touches scores
• + Per-agent loadouts let you control exactly what each agent sees

• − Pre-launch — the product is actively being built and not yet public, so you can't evaluate live benchmarks today
• − A verification-and-benchmarking layer adds rigor that pure-browse users may not need
• − Newer entrant without skills.sh's install-count adoption signal

https://skillzbench.com