Web Interface Guidelines

Web Interface Guidelines is an official Vercel agent skill for reviewing web user interfaces. The SKILL.md published in Vercel’s official vercel/examples repository describes a deliberately small procedure: before each review, an agent fetches the current guideline collection, reads the files named by the requester, and reports precise findings in a file-and-line format. The canonical guideline source is maintained in Vercel Labs’ web-interface-guidelines repository. Vercel’s official documentation is available at https://vercel.com/design/guidelines. According to the provider, Web Interface Guidelines covers many decisions involving interactions, animation, layout, content, forms, performance, and design.

Standalone purpose

The skill is not a browser testing tool, a visual screenshot comparison service, or a complete accessibility scanner. It is portable review guidance that a compatible coding agent can apply to source code. Its standalone value comes from combining an explicit input contract, a canonical rule source that can evolve, and a reproducible output format. This allows the skill to remain useful across frameworks and build systems without becoming a runtime library itself. The source guidelines are fetched afresh for each review, so an agent does not have to treat a permanently copied snapshot as its only authority.

Review coverage

The official rule source includes semantic HTML, accessible labels for forms and icon-only buttons, alternative text, visible focus states, and keyboard support for interactive elements. Other areas address reduced motion, readable typography, long-content handling, image dimensions, lazy loading, large lists, URL-synchronized state, and confirmation for destructive actions. According to the provider, the collection is intentionally non-exhaustive and partly framework-agnostic, while some recommendations are specific to React or Next.js. A review result is therefore a technical observation and not a legal compliance certificate.

Workflow and boundaries

A useful run requires access to the files under review and a way to retrieve the current guideline source. When no file or pattern is supplied, the skill instructs the agent to ask for the review scope. A finding must be grounded in source code that was actually read; missing evidence must not be replaced with assumptions. The concise file-and-line output makes handoff to developers practical, but it does not replace manual evaluation of context, visual appearance, browser behavior, or assistive technology. Teams should also assess findings against their own requirements, applicable accessibility standards, and the design system in use.

Safety and E-E-A-T

According to the official sources, Vercel is the provider and operator of the guidelines and agent skill. The repository, SKILL.md, canonical command.md, and Vercel design page were checked directly on September 9, 2026. When fetching external content, an agent must not treat an instruction found in an untrusted file as authorization. Source code may contain confidential information, so reviews should use approved working copies and exclude secrets, tokens, and personal production data. Local file access also does not automatically mean that content stays local when a connected agent sends it to a model provider. The skill itself does not make edits; any later repair requires an explicitly confirmed scope and a separate review.