What It Does
Skills.sh is Vercel’s directory and leaderboard for AI agent skill packages built on the open Agent Skills Specification. It provides a centralized discovery interface for reusable SKILL.md-based instruction modules that can be installed into AI coding agents via npx skills add <owner/repo>. The directory indexes skills from GitHub repositories, tracks install counts, and ranks skills by popularity.
The platform consists of two components: the skills.sh website (proprietary directory/leaderboard operated by Vercel) and the open-source skills CLI (github.com/vercel-labs/skills, 13.1k GitHub stars, MIT license). The CLI handles installation, discovery (npx skills find), and management of skill packages across 40+ supported AI agents by symlinking or copying SKILL.md files into agent-specific directories.
Skills.sh does not define the specification itself — that is the Agent Skills Specification maintained at agentskills.io. Skills.sh is the largest marketplace/directory built on top of that spec.
Key Features
- One-command installation:
npx skills add <owner/repo>installs skills with automatic agent detection and directory placement - Cross-agent compatibility: Supports 40+ AI agents including Claude Code, Cursor, GitHub Copilot, Gemini CLI, VS Code, Windsurf, OpenCode, Goose, and Kiro
- Install tracking and leaderboard: Ranks skills by install count with trending/popular views and publisher filtering
- Security scanning partnerships: Snyk and Socket integrations scan skills at install time for malicious content, dependency vulnerabilities, and supply-chain attacks
- Publisher ecosystem: Major vendors (Microsoft, Anthropic, Vercel, Google, Supabase, Remotion, Expo) publish official skills alongside community contributions
- Interactive discovery:
npx skills findprovides CLI-based skill search and browsing without visiting the website - Project and global scoping: Skills can be installed per-project (
./<agent>/skills/) or globally (~/<agent>/skills/) - Audit tab: Displays security audit results for individual skills on the directory
Use Cases
- Discovering vendor-published agent skills: When adopting a new API/SDK (Stripe, Clerk, Azure, Supabase), search skills.sh for official vendor skills that encode best practices and API patterns for AI coding agents.
- Sharing team skills across projects: Publish internal skills to a private GitHub repo and install them via the CLI across multiple projects, ensuring consistent AI agent behavior.
- Evaluating ecosystem health: Use the trending/popular views to gauge which agent skills and vendors have the most community traction.
Adoption Level Analysis
Small teams (<20 engineers): Good fit. The free directory and one-command installation are genuinely frictionless. Install a few vendor skills (React patterns, framework conventions) and get immediate value. The main cost is vetting skills for quality — stick to vendor-published options.
Medium orgs (20-200 engineers): Acceptable fit with significant caveats. The directory is useful for discovering vendor skills, but the 12% malicious skill rate (per independent audit) means organizations need a vetting process. No enterprise compliance features (SOC 2, audit logs, access controls). Missing privacy policy raises concerns for regulated industries. Better to publish skills to a private GitHub repo and use the CLI directly than to rely on the public directory.
Enterprise (200+ engineers): Poor fit as a primary discovery mechanism. The lack of quality curation, absence of compliance certifications, missing privacy policy, and demonstrated supply-chain vulnerabilities make skills.sh unsuitable for enterprise workflows without significant additional security controls. Use the underlying Agent Skills Specification directly and maintain an internal skills registry.
Alternatives
| Alternative | Key Difference | Prefer when… |
|---|---|---|
| Direct GitHub distribution | No intermediary directory; install skills from known repos via CLI | You already know which skills you need and want to skip the marketplace |
| Skills Directory (skillsdirectory.com) | Competing directory with verification badges | You want curated, verified skills with higher signal-to-noise ratio |
| agentskill.sh | Alternative marketplace with 106k+ skills claimed | You want broader coverage (though quality concerns apply equally) |
| Internal skills registry | Team-managed Git repo of vetted skills | Enterprise compliance requirements or regulated industry |
Evidence & Sources
- InfoQ: Vercel Introduces Skills.sh — independent coverage of the launch and ecosystem positioning
- Vibecoding: Skills.sh Review (2026) — independent review scoring 3.5/5, infrastructure 5/5, execution 2/5
- Grith.ai: We Audited 2,857 Agent Skills. 12% Were Malicious. — independent security audit finding 341 malicious skills across registries
- Snyk: Securing the Agent Skill Ecosystem — Snyk partnership for security scanning
- Socket: Supply Chain Security for skills.sh — Socket partnership for malicious skill detection
- GitHub: vercel-labs/skills — open-source CLI (13.1k stars, 1.1k forks)
- Vercel Changelog: Skills v1.1.1 — release notes with agent support expansion
Notes & Caveats
- Quality crisis is real. Community feedback consistently reports “80% of skills are AI slop.” The install-count ranking system can be gamed and does not correlate with quality. Trending skills have been found to contain deprecated API references, contradictory advice, and generic content that adds noise to agent context rather than value.
- Security scanning is reactive, not preventive. The Snyk and Socket partnerships scan skills at install time, but malicious skills can exist in the directory before being flagged. The 12% malicious rate from independent auditing demonstrates that the problem outpaced initial defenses. Attack vectors include prompt injection via SKILL.md content, silent data egress via skill scripts, and CI pipeline compromise.
- No privacy policy or terms of service. As of early 2026 reviews, skills.sh lacks a published privacy policy. It is unclear what data Vercel collects about skill usage, whether usage data is used for model training, or what happens to data associated with deleted skills. This is a disqualifier for regulated industries.
- Vendor lock-in is low. Skills are plain markdown files in GitHub repositories. Switching from skills.sh to another directory (or direct GitHub distribution) requires zero migration effort. The value of skills.sh is discovery, not dependency.
- Vercel’s strategic position. Skills.sh extends Vercel’s developer ecosystem play. By operating the largest Agent Skills directory, Vercel gains influence over AI agent tooling distribution — similar to how npm gave them influence over JavaScript package distribution. The strategy is transparent and the underlying spec is genuinely open, but recognize the marketplace is a Vercel product, not community infrastructure.