The shadcn/ui project released @shadcn/lint, an ESLint and Oxlint-based linter meant to catch AI coding agents breaking Tailwind design system rules, according to the project's GitHub repository. The project is MIT licensed.
Traditional linting tells an agent only that something is forbidden. This tool's rules, including no-restyle, no-raw-colors and no-arbitrary-values, generate error messages that explain why a change breaks the design system and suggest the specific component, variant or theme value to use instead, drawing those suggestions directly from the project's own components.
The project says testing across more than 150 agent task runs found that most tasks reached zero violations within one correction round once the linter's contextual guidance was in place, and that fixes made from the tool's feedback cost 10% to 48% less than fixes made from bare rule violations alone.
Linters have always told developers what not to do. Writing one that also tells an agent what to do instead is a small change that matters more for agents than people, since an agent has no teammate to ask and will otherwise just guess.