vercel-labs/skills: The Open Agent Skills CLI

TL;DR: vercel-labs/skills is a TypeScript CLI published as npx skills that installs reusable agent skill packs into a project. It is MIT licensed, has 34,000 stars, and carries an unusually large 1,115 open issues for a tool of its size.

What it is

The repository description is one line: the open agent skills tool, distributed through npx. That economy tells you what kind of project this is. It is infrastructure, not a framework. Something you run once to pull capability packs into a repository so an agent can use them.

The hosted front door is skills.sh, which is the directory rather than the tool itself.

Written in TypeScript, created 2026-01-14, last pushed 2026-10-09.

$ curl -s https://api.github.com/repos/vercel-labs/skills \
  | jq '{stars: .stargazers_count, forks: .forks_count, open_issues: .open_issues_count, language, license: .license.spdx_id}'
{
  "stars": 34550,
  "forks": 2953,
  "open_issues": 1115,
  "language": "TypeScript",
  "license": "MIT"
}

Why a Vercel Labs tool keeps trending

Vercel Labs is a known incubator for experimental developer tooling, and the star count reflects that distribution as much as it reflects adoption. Seeing a Labs project at the top of trending usually means the underlying idea is being picked up faster than the tooling can stabilise.

The MIT license is worth noting alongside that. Compared to AGPL-3.0 projects in the same trend cycle, this one carries no copyleft obligation, so it is straightforward to adopt into commercial work.

The open issue problem

1,115 open issues is the number that deserves a second look. Against 2,953 forks, that is a ratio you do not see in a mature, well-staffed tool. It points to genuine, sustained demand colliding with a project that is still finding its shape.

Practically, that means two things:

• You are unlikely to get fast answers on issues, so budget for reading source rather than asking.
• The interface may still be shifting, which is a consideration if you script against it.

That is the honest tradeoff. You get a fast-moving tool with broad adoption and a permissive license, at the cost of some interface churn.

How to try it

Because it ships through npm, there is no build step to understand before you have a feel for it:

npx skills

Cloning for a closer look is equally direct:

git clone https://github.com/vercel-labs/skills.git
cd skills

When to use it

It makes sense if you are standardising how agent capabilities are distributed across a team or across several repositories. A single CLI that installs skill packs into a project removes the per-project hand-configuration that otherwise fragments quickly.

The permissive license makes it low-risk to adopt early. MIT means you can wrap it, extend it, and ship it commercially without negotiating terms.

When not to use it

If you need a stable, documented API surface with predictable behaviour, this is early for that. The issue volume is a fair proxy for how much the surface is still moving.

It is also worth being clear-eyed that a Labs project is a statement about experimentation, not long-term support guarantees. If you are building a production pipeline on it, pin versions and read the source before upgrades.

Source and Accuracy Notes

• Project page: https://skills.sh
• Source repository: vercel-labs/skills (verified via GitHub API html_url)
• License: MIT (verified via GitHub API license.spdx_id)
• Stars at time of writing: 34550