Vercel's playbook, packaged for your AI agent
A collection of instruction sets that teach coding agents to audit, deploy, and review like Vercel engineers.

What it does
This repo bundles Vercel’s internal expertise into “skills” — packaged instructions and optional scripts that extend AI coding agents. Install once via npx skills add and your agent gains the ability to optimize Vercel deployments, review React components against 40+ performance rules, check UI code for 100+ accessibility and UX violations, audit docs prose against Vercel’s writing handbook, deploy sites with claimable ownership, and more.
The interesting bit
The format matters more than the code. These follow the agentskills.io spec — essentially structured prompt engineering with version control. Vercel is treating institutional knowledge (“we don’t say easy or simple in docs”) as installable infrastructure, not blog posts.
Key highlights
vercel-optimize— collects metrics first, then investigates only the routes and files flagged by those metrics (avoids blind crawling)react-best-practices— 40+ rules prioritized by impact, from waterfall elimination to micro-optimizationswriting-guidelines— 80+ rules including banned words (easy/simple/quick), 80-column/25-line code limits, and AI workflow disclosure requirementsvercel-deploy-claimable— auto-detects 40+ frameworks, returns both preview URL and claim URL for ownership transferreact-view-transitions— covers Next.js App Router integration withtransitionTypesprop onnext/link
Caveats
- Most skills are instruction documents (
SKILL.md) plus optional helper scripts; heavy lifting still depends on your agent’s base capabilities - The
vercel-deploy-claimableskill is explicitly designed for Claude.ai and Claude Desktop; behavior with other agents is unclear
Verdict
Worth installing if you already use AI agents for Vercel/Next.js work and want consistent, opinionated output. Skip if you’re not on the Vercel stack or prefer your own linting and review rules.