Skip to content

Build1 publisher2 min readPublished

375 skills draw 62% of installs on skills.sh, Vercel's million-skill agent registry

Vercel says skills.sh reached one million agent skills and nearly 280 million installs in seven months, with 375 skills taking 62% of installs. Outside the top 1.2%, skills average about 17 installs each, so the million mostly measures how cheap a skill file is to publish.

The Engineer · Build desk

Illustration accompanying 375 skills draw 62% of installs on skills.sh, Vercel's million-skill agent registry

What happened

  • Anthropic introduced Agent Skills in October 2025, and Vercel launched the skills.sh registry three months later.
  • Nearly half of all skills listed on skills.sh have been installed exactly once, according to Vercel's registry data.
  • Software engineering makes up about a quarter of listings and 18% of installs, the biggest share in a spread where no category tops a fifth.
  • Agent workflow and automation skills take 14.8% of installs and form the largest category among skills with at least 100,000 installs.

Compiled by The EngineerSomething wrong?How this is made

Why it matters

  • exposure A skill from the long tail arrives with little or no install history to judge it by, so reviewing its instructions falls entirely on the team that adopts it.
  • decision For an author choosing what to write, a business operations skill averages 74% more installs than the norm, so written-down operations know-how reaches more agents per file than another coding skill.
  • constraint Counts from one registry cannot show how much team-specific process has moved into skill files, so the growth curve does not settle whether skills are becoming the default way to package know-how.

A skill is a set of reusable instructions for one job, and it can be as simple as a file written in ordinary language [2]. The agent and its tools already supply the capability [5]. The skill adds the judgment specific to the job: the steps to follow, what good looks like, and how to evaluate the result [5]. Once written, the same file can be distributed and installed without being rewritten for each agent [7].

In my view, that design explains most of the growth curve. Vercel runs skills.sh and built the report from its own aggregate registry data [17]. It sets the million-skill mark beside three older counts: GitHub took 27 months to reach a million repositories, the App Store just over five years to reach a million apps, and npm more than nine years to reach a million packages [4]. For that comparison to measure adoption, a skill, a repository, an app and a package would need to cost about the same effort to publish. Vercel's own report says they do not. It calls skills faster and easier to create than conventional software [5], so a count of skills mostly tracks how cheap one is to publish. "Far more people can describe how work should be done than program a computer to do it," Vercel wrote [6].

The report does not define an install. Divided out, the totals still show the shape. Nearly 280 million installs across a million skills is a mean of about 280 each [1]. If 375 skills hold 62% of installs [15], those 375 average roughly 460,000 installs apiece [2]. The top 1.2% of the registry, about 12,000 skills, holds 94% [16][3]. The other 988,000 or so share the remaining 6%, about 17 installs per skill [3].

The strongest part of the report is its category work. Vercel treated published listings as supply and installs as demand, classified the most-installed skills, a set covering more than four-fifths of installs, and compared each category's installs per listing with the average [8]. The numbers hold together. Software engineering's 18% of installs divided by its roughly 25% of listings is 0.72, about 28% below par [4]. The report's own figure is about 30% fewer installs per software engineering skill [12]. Research skills run 55% below the average, and education and productivity skills 39% below [12].

What to watch

  • A definition from Vercel of what counts as an install, separating unique users from repeat or automated installs.
  • Whether the top 1.2% of skills keeps its 94% share of installs as non-engineering categories gain more listings.
  • Whether business operations and writing skills keep their installs-per-listing premium once more authors publish in those categories.
Loading claim ledger
Loading source directory links
Loading share composer
Loading topic controls
Loading related stories