Vercel has published the first market report on agent skills, and the numbers are large enough to need the comparisons it gives them. In State of agent skills, Amelia Charles, Andrew Qu, and Jonathan Hefner report that the skills.sh registry Vercel operates reached one million skills in seven months, against 27 months for GitHub to reach a million repositories, 63 for the App Store, and 117 for npm, with 280 million installs recorded along the way. The explanation is the thesis: “Far more people can describe how work should be done than program a computer to do it,” and “skills pair a much larger pool of creators with software’s capacity for reuse.”
The figures are Vercel’s, from a registry Vercel runs, and the report says its classifications are samples subject to revision, which is the right caveat. The distribution is the part that survives. “The top 375 skills account for 62% of installs,” 0.04% of the registry, and the top 1.2% account for 94%. “Nearly half of all skills were installed exactly once.” Software engineering skills lead at 18% of installs but are oversupplied, drawing 30% fewer installs than average, while business operations, writing, and cloud infrastructure have more demand than supply. Cross-industry skills are 66% of listings and 87.5% of installs. The closing line sets up the next report: “The first million skills taught agents what everyone knows. The next million will teach your agents what only you know.”
The catalog can read that line back at the company that wrote it. The Vercel provider page lists 6 API pages, and on the Agent Readiness score the MCP server, the event surface, and the full identity layer are lit, delegated identity, protected resource metadata, and dynamic client registration. The agent skills dimension is unlit. The operator of a registry holding a million skills has not, as far as the catalog can find, published a skill that teaches an agent to use Vercel’s own API. The agentic access profile maps 6 operations, 4 of them acting.
The Kin Score is 59.0, strong band. Access clarity carries it at 68.4, with contract quality and developer ergonomics at 63.1, discoverability at 61.7, operational transparency at 52.6, and contract governance at 13.6. The Agent Readiness score is 45.7, agent-ready. OpenAPI examples, error semantics, idempotency, and reversibility are unlit alongside the skills. The report’s own finding is that demand exceeds supply for cloud infrastructure skills, and that the next million will be specific rather than general. A skill for deploying to Vercel through Vercel’s API is both of those things, and it is the one skill the registry’s operator could write without asking anyone. The catalog is waiting for it.