skills.sh — The Marketplace That Made AI Agent Skills Mainstream

Launched by Vercel on January 20, 2026, skills.sh ranks community skills by actual install counts. Six months later, it hosts 34,000+ skills and drives most of the ecosystem adoption. Here is why it matters.

Before skills.sh, finding and evaluating AI agent skills meant manually searching GitHub, reading SKILL.md files in random repositories and hoping the maintainer was trustworthy. The skills standard was solid but discovery was painful — and adoption suffered accordingly.

Skills.sh changed this in one move. It became the central directory, the leaderboard, the discovery layer and the trust signal all at once.

The Install-Count Leaderboard

The single most important feature of skills.sh is that it ranks skills by actual install counts rather than stars, downloads or other vanity metrics. This matters because install count is the most honest signal of usefulness — installs require trust, and the trust is given by tens of thousands of developers voting with their machines.

A skill at 133,000 weekly installs (the current leader, web-design-guidelines) has been audited and used successfully by a substantial fraction of the entire skills-using developer community. Skills with hundreds of weekly installs are reliable and battle-tested. Skills with single-digit installs are unproven and require personal due diligence.

The 19 Compatible Agents

Skills.sh supports 19 different AI agents at the time of writing: amp, antigravity, claude-code, clawdbot, codex, cursor, droid, gemini, gemini-cli, github-copilot, goose, kilo, kiro-cli, opencode, roo, trae, windsurf and more. New agents are added as they adopt the standard.

The practical implication for developers: install a skill once via skills.sh and it works in every AI agent you use that supports the standard. No vendor lock-in, no parallel implementations, no rewriting.

The npx skills Command

Skills.sh is more than a website — it ships with an open-source CLI that handles installation, updates and removal across every supported agent at once. The command is invoked through npx (no global install needed):

npx skills find                              # browse interactively
npx skills add vercel-labs/agent-skills      # install
npx skills list                              # see what is installed
npx skills update                            # update everything
npx skills remove [skill-name]               # remove

The CLI auto-detects which AI agents you have installed and routes skills to the correct directories. Skills installed via the CLI appear automatically in Claude Code, Cursor, Codex CLI and any other compatible agents you have set up — no per-agent configuration needed.

How to Submit Your Skills

Once you have built a skill worth sharing, submission to skills.sh is straightforward: publish to a public GitHub repository with SKILL.md following the specification, ensure you have a README and LICENSE, then submit a pull request to the skills.sh registry. Approved skills are discoverable through npx skills find immediately, with install counts tracked over time.