Vercel Skills — Agent Skill Ecosystem & CLI

The problem Vercel Skills solves

Every AI coding agent — Claude Code, Cursor, Codex, Windsurf, Cline — loads context differently. Same instruction set needs to live in .claude/skills/, .cursor/rules/, .codex/, AGENTS.md. Sharing a reusable workflow across tools meant copy-pasting 30 different times. Vercel's Skills CLI normalizes it: one package, any agent.

Install anything in one line

npx skills install react-best-practices
npx skills install web-design-audit
npx skills install api-security

The CLI detects your agent from project files (.claude/, .cursor/, .vscode/, etc.) and installs the SKILL.md into the correct location with the correct naming convention.

What makes it different from awesome-lists

The SKILL.md format

A Skill is just a markdown file with frontmatter:

---
name: api-security
description: Security checklist before shipping any API endpoint
version: 1.2.0
triggers: [touches api/, touches /routes/]
---

# API Security Checklist

...instructions...

Agents load this into their system prompt based on triggers. No build step.

Stats that matter

• 13,000+ GitHub stars in 9 months
• 420+ public skills in the default registry
• 30+ compatible agents including Claude Code, Cursor, Codex, Zed, Windsurf, Cline, Aider
• Native support for both personal (~/.skills/) and project-level installs

Typical workflows

Team-wide shared skills

# Install a curated set into your repo
npx skills install @mycompany/security-review
npx skills install @mycompany/commit-conventions
git add .claude/ .cursor/ && git commit -m "team skills"

Now every teammate cloning the repo has identical agent behavior.

Publishing your own skill

npx skills init my-skill
# edits SKILL.md
npx skills publish

Registry at skills.vercel.com/registry.

Integration with Vercel's broader AI stack

When paired with v0 (Vercel's AI UI generator) and Vercel AI SDK, Skills provide consistent agent behavior across design → code → deploy. This is the "agent operating system" narrative Vercel has been pushing since late 2025.

Limitations to know

1. Not all skills work in all agents. Triggers are interpreted per-agent; some skills need agent-specific forks.
2. Registry is still sparse — fewer skills than npm has packages, and quality varies.
3. No revenue model for authors yet. Skills are free. If you're building enterprise skills, consider private registries.