Vercel report: skills.sh hits one million agent skills in seven months
Vercel says its skills.sh registry reached one million agent skills and nearly 280 million installs seven months after launch.
Vercel's State of agent skills report, published September 25, says the skills.sh registry reached one million agent skills and nearly 280 million installs in seven months. If you package your own workflows as SKILL.md files, the demand data is useful. The scale is also a reminder that you are running a lot of other people's instructions inside your agent.
Vercel says installs are heavily concentrated. Nearly half of all skills were installed exactly once, and 375 skills (0.04%) account for 62% of installs. Software engineering leads demand at 18%, followed by agent workflows at about 15%. Business operations skills draw 74% more installs per listing than average. Cross-industry skills take 87.5% of installs in Vercel's classified sample. Vercel notes that install counts are aggregate counters, "not unique people or necessarily independent choices." No independent audit of the figures was found.
The report doesn't cover how skills are vetted at publish time, and that is still unclear. QASkills.sh says skills.sh docs describe routine audits but also say the directory can't guarantee the security of every skill. NVIDIA's SkillSpector README cites research finding that 26.1% of skills in an analyzed subset contain vulnerabilities and 5.2% show likely malicious intent. That underlying study wasn't independently checked.
There are several ways to lower the risk:
- Scan before you install. SkillSpector is an open-source scanner you can run on a skill before adding it.
- Prefer vendor-maintained repos. NVIDIA/skills says every skill it publishes ships with a verifiable signature.
- Write your own SKILL.md in-repo. This avoids third-party instructions entirely, though you maintain everything yourself.
Sources
- State of agent skills · Vercel
- skills.sh Explained: The Open Agent Skills Directory, npx skills add, and the Leaderboard | QASkills.sh · QASkills.sh
- GitHub - NVIDIA/skills: Agent Skills for NVIDIA products — install into Claude Code, Codex, and other coding agents to run Physical AI, robotics, simulation, CUDA, and RAG workflows end to end. · GitHub
- GitHub - NVIDIA/SkillSpector: Security scanner for AI agent skills. Detect vulnerabilities, malicious patterns, security risks, prompt injection, data exfiltration, and supply-chain risks in Claude Code, Codex, and MCP skills before you install them. · GitHub