The Best skills.sh Alternative for Agencies in 2026

skills.sh is Vercel's open directory and leaderboard for AI agent skills, launched in January 2026. Skills are modular SKILL.md packages — Markdown instructions plus optional scripts, templates and assets — that teach agents like Claude Code procedural knowledge without bloating the context window. Discovery is by install count (Hot, Trending, Top tabs) with hourly deduplication, and installation is a one-liner: npx skills add owner/repo. The CLI detects which agent you are using and drops the skill into the right directory, and the same skill works across roughly 20 supported agents including Claude Code, Cursor, Codex, GitHub Copilot, Windsurf, Gemini and Zed (Aider is not on the list). Every indexed skill is scanned by Gen Agent Trust Hub, Socket and Snyk, and a skill that fails every partner audit is excluded; the Official section lists 6,588 skills from 102 organisations. At 83,627 skills with 8M+ total installs it is smaller than SkillsMP but has better signal on what people actually use. It is free infrastructure, not a hosted service: no account is needed to install, and the CLI is MIT-licensed open source at vercel-labs/skills.

pricing 0

• + Free, open and fast to try — no sign-in needed to install, and the CLI is MIT-licensed open source
• + Cross-agent portability: the same skill installs into roughly 20 coding agents, avoiding vendor lock-in
• + Install-count leaderboard with deduplication gives real adoption signal, and the Official section (6,588 skills from 102 organisations) is a better starting point than trending tabs
• + Indexed skills are routinely audited by Gen Agent Trust Hub, Socket and Snyk

• − Everything is public — no natural home for internal conventions, runbooks or client-specific skills an agency would not want on the open internet
• − No frozen versions: installs point at whatever the author's repo contains right now, so there is no meaningful rollback or diff between what was reviewed and what was installed
• − No review gate or team constructs — no organizations, member roles, access control or audit trail
• − skills.sh states it cannot guarantee the quality or security of every listed skill; popularity based on anonymous install telemetry does not prove correctness, maintenance quality or fit

https://skills.sh

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. Agent skills are everywhere, but nobody can tell which ones actually help and which are risky: published research found ~26% of skills carry at least one vulnerability, and Snyk's ToxicSkills research detected prompt injection in 36% of skills tested. SkillzBench is building around five 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 — toggle skills and tools per agent and the agent only sees what was chosen), and Stay current (track upstream changes, re-scan, re-benchmark and notify). A Trust score from 0-100 combines security, quality, license, maintenance and community signals; Lift is always shown separately from ratings, and money never touches scores. Everything on the platform is free — the directory, the MCP connector, benchmarks and alerts — funded by advertising around blog articles, never by charging users, and ad money never influences scores or listings. The product is pre-launch and actively being built; the waitlist is live.

pricing 0

• + Completely free — directory, MCP connector, benchmarks and alerts, with no user-facing charges at any tier
• + Independent verification addresses the trust gap: commit-SHA pinning, security scans and license checks counter the ~26% vulnerability rate in published research
• + Sandbox benchmarks publish Lift separately from ratings, so agencies can see what a skill actually contributes before installing it
• + Per-agent loadouts over a single free MCP server give agencies control over exactly what each client agent sees

• − Pre-launch: the product is not yet public — the waitlist/newsletter is live while benchmarking, scanning and the MCP are still being built
• − Focuses on verifying and equipping public skills rather than private, org-internal skill distribution
• − No published benchmark results or live skill counts yet, so claims rest on the methodology being built

https://skillzbench.com

localskills.sh is a full agent skills registry rather than a directory — the strongest skills.sh alternative for teams that need private distribution with team controls, according to its own honest comparison of the market. You publish skills as versioned packages with a root SKILL.md, and packages can be multi-file folders up to 100 MB and 500 files per version. One 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, and one install can target several tools at once. Versioning includes rollback, plus public, private and unlisted visibility and built-in download analytics. Org structure covers folders, namespaces, teams, member roles, custom roles and folder-level access restrictions. Enterprise plumbing includes SAML SSO with setup templates for Okta, Cloudflare Access and Authentik, SCIM directory sync, audit logs and verified-domain auto-join. Bidirectional GitHub sync mirrors an org's skill library to a repo so review happens in normal pull requests, and an MCP server with OAuth and scoped access lets agents search and install skills without a human in the loop.

pricing See pricing

• + Real registry: private skills, versioning with rollback, team roles and access control — the constructs skills.sh lacks
• + Multi-tool output from one source of truth, including tool-native formats per agent
• + Enterprise plumbing (SSO, SCIM, audit logs, verified-domain auto-join) where a platform team needs it
• + GitHub sync lets review happen in normal pull requests

• − It is a registry, so it asks you to publish rather than just point at a repo — overhead if you never need private distribution
• − The publish step is extra workflow a public directory does not have

https://localskills.sh

Agentman's Agent Skills (myAgentSkills.ai) is a hosted skills library that gives agencies the team constructs skills.sh has no answer for, on self-serve plans. The free plan stores, versions and shares skills for up to 10 private skills and 3 members with no credit card, usable from Claude, ChatGPT, Cursor and your own agents over a single MCP connection. Team is $10/seat/mo with a 3-seat minimum ($30/mo entry) adding unlimited members and unlimited private skills with email support and a 14-day trial. Business is $25/seat/mo with a 5-seat minimum ($125/mo entry) adding per-skill access control, use-only sharing (run a skill without seeing its instructions), approvals to review changes before they publish, and access/activity visibility — exactly the review gate agencies need. Beyond self-serve, a custom Enterprise track covers private and on-premise deployment with custom security and compliance scoping, and a Partner track is built for agencies, integrators and skill authors delivering for clients: multi-tenant workspaces with a separate workspace and library per client, custom per-user pricing, on-premise hosting and Slack support. Agentman's library is practitioner-authored and reviewed, with 90 production skills that are spec-compliant and curated.

pricing 0 · free trial

• + Genuine free tier (10 private skills, 3 members) with a clear upgrade path
• + Approvals and use-only sharing give agencies a real review gate without running infrastructure
• + Partner plan is purpose-built for agencies: multi-tenant client workspaces and on-premise hosting
• + Practitioner-authored, reviewed, spec-compliant library of 90 production skills

• − Seat minimums raise the real entry cost: $30/mo for 3 seats on Team, $125/mo for 5 seats on Business
• − Smaller catalog than open directories — curation trades off against volume

https://agentman.ai

ClawHub grew out of the OpenClaw ecosystem and is the most curated of the three dominant agent skills marketplaces, built by OpenClaw creator Peter Steinberger as the 'quality over quantity' alternative. It sits at roughly 3,286 skills with 1.5M+ downloads — by far the smallest catalog compared to SkillsMP's 351,349 and skills.sh's 83,627, but the only one of the three with proper versioning and rollback support. Discovery uses vector-powered semantic search, and installation runs through its own CLI: npx clawhub@latest install skill-slug. Agent support centers on OpenClaw plus bridges to other tools, so it is narrower than skills.sh's ~20-agent compatibility. For agencies, ClawHub's appeal is the curation model: versioned installs with rollback mean a skill update that degrades agent behavior can be pinned or rolled back instead of hunted through git history — one of the core evaluation criteria for a team-ready registry. The trade-off is ecosystem breadth: the catalog is small, and if your team is not on OpenClaw, the bridges determine how usable it is.

pricing 0

• + Most curated of the major marketplaces — quality over quantity, with versioning and rollback
• + Semantic search makes the small catalog navigable
• + Free to use with a simple CLI install

• − Smallest catalog by far (~3,286 skills vs 83,627 on skills.sh and 351,349 on SkillsMP)
• − Agent support centers on OpenClaw plus bridges — narrower than skills.sh's ~20 agents

https://clawhub.com

SkillsMP is the volume leader among agent skills marketplaces: a community-run discovery platform that crawls GitHub for SKILL.md files and indexes them with AI-powered semantic search. Its catalog exploded from a few thousand skills in December 2025 to 351,349 by early March 2026 (with later reporting citing ~1.9 million public skills scraped from GitHub), organized in a deep category taxonomy — 89K tools skills, 70K development skills, 60K business skills. It supports Claude Code, Codex CLI and ChatGPT. Think of it as the npm-registry equivalent of the skills world: massive catalog, deep taxonomy, but no quality gate beyond GitHub stars and no security review — the guidance is to inspect before installing. For agencies, SkillsMP works as a breadth-first discovery layer: when you need to find whether a skill exists for an obscure framework or business workflow, its semantic search over the largest catalog is the place to look. It is not, however, a standardization layer: no versioning, no team constructs, no review workflow, so anything found there still needs the verification and pinning that a registry or a verification service provides.

pricing 0

• + Biggest catalog in the market — the best place to check whether a skill exists at all
• + Semantic search over a deep taxonomy makes breadth usable
• + Free, community-run, no account friction for browsing

• − No quality gate beyond GitHub stars and no security review — inspect before installing
• − Closer to skills.sh than to a replacement: no versioning, rollback, private skills or team constructs
• − Supports only 3 agents

https://skillsmp.com

SkillHub is a mid-size agent skills directory that occupies the middle ground between the giant open catalogs and the small curated registries: 7,000+ skills with an AI-evaluated curation model and automated scoring. In the Agentman ecosystem report's comparison of major directories, SkillHub is one of the few community directories with any automated evaluation layer at all — unlike SkillsMP (scraped from public GitHub with no review) or skills.sh (builder-side auditing), every listed skill passes through automated scoring before it reaches the listing. For agencies, that automated scoring is a useful pre-filter: it does not replace a human review gate or a sandbox benchmark, but it raises the floor above raw GitHub scraping in a market where a security audit of 22,511 skills found 140,963 issues and Snyk's ToxicSkills research detected prompt injection in 36% of skills tested. The catalog is smaller than skills.sh's 83,627, so discovery breadth is limited, and the research does not document team constructs, private skills or versioning — treat it as a safer-than-average discovery surface rather than a full standardization layer.

pricing 0

• + Automated scoring raises the floor above raw GitHub-scraped catalogs
• + Mid-size catalog is more navigable than the 350K+ giants
• + Free directory access

• − Smaller catalog than skills.sh (7,000+ vs 83,627)
• − No documented team constructs, private skills or versioning in the research
• − Automated scoring is a pre-filter, not a human review gate or sandbox benchmark

https://skillhub.com

The zero-infrastructure option: Anthropic maintains an official skills repo, and community 'awesome' lists function as a de facto Claude skills directory in README form. The workflow is simple — clone or copy the skill folder into .claude/skills/, done. It is free, fully transparent, and easy to fork, and review happens naturally because you read the file before copying it — which, in a market where a skill is instructions your agent will follow and often scripts it will execute, is a genuine security advantage over one-liner installs from strangers' repos. For agencies with a platform team, a private GitHub repo extends the same pattern to internal skills: versioning comes via git, review comes via pull requests, and access control comes via repo permissions — all constructs skills.sh lacks. The limits are the flip side: installs are manual, there is no multi-tool translation (a Claude-format skill does not automatically become a Cursor- or Windsurf-native format), there is no discovery layer or analytics, and keeping parallel copies per tool is exactly the maintenance burden registries exist to remove.

pricing 0

• + Free, fully transparent, easy to fork
• + Natural human review before install — a real trust advantage over blind one-liner installs
• + Private repos + PR review + git versioning give agencies basic team constructs with no new tooling

• − Installs are manual — no CLI, no discovery layer, no analytics
• − No multi-tool translation: a Claude-format skill does not become a Cursor- or Windsurf-native format
• − No frozen package versions or rollback distinct from git history

https://github.com/anthropics/skills