vercel-labs / agent-eval

Those three skills — frontend-design, vercel-react-best-practices, and web-design-guidelines — are all classified as needing no account, key, or local tool. By their own descriptions: frontend-design is a general skill for building "distinctive, production-grade" web UI that avoids generic AI-generated aesthetics; vercel-react-best-practices is Vercel Engineering's own React and Next.js performance guidance, triggered on data fetching, bundle size, and component-pattern work; web-design-guidelines reviews UI code against a "Web Interface Guidelines" accessibility and UX standard. None of the three is mentioned anywhere in the roughly 12,000 characters of README we read — the quick start, full CLI reference, eval-authoring guide, and configuration reference for the @vercel/agent-eval testing framework itself.

This is for anyone writing or reviewing React, Next.js, or general web frontend code who wants these three guidance skills on their own merits. The eval framework the README actually documents is a separate, much larger tool for framework authors running controlled experiments — pass rates, agentic LLM judges, multi-model comparisons — and most people who install this package for its skills will never touch that CLI.

READ THE FULL ANALYSIS＋

We don't know why they're bundled together, and we're not guessing. Nothing in the material we have — the readme excerpt or the three skill descriptions — says whether these skills are fixtures for agent-eval's own test suite, unrelated add-ons published under the same npm scope, or something else.

There's more README we haven't read. The full file is about 27,500 characters and we captured the first 12,288; the missing back half could in principle explain the skills, though nothing in the table of contents implied by what we did read suggests it does.