Most AI coding agents are generalists that re-derive the same conventions from scratch on every prompt. This collection flips that: domain expertise ships as installable, version-controlled units the agent loads only when a matching task appears—so a Vercel cost audit or an accessibility pass runs against a fixed, reviewable rule set instead of whatever the model happens to recall that day.
What Sets It Apart
- Skills are plain files (a
SKILL.mdplus optional scripts and docs), so they live in your repo and travel through code review like any other source, rather than hiding inside a vendor product. - Each one is narrow and deep:
react-best-practicescarries 40+ rules across 8 categories,web-design-guidelineschecks 100+ accessibility and UX points, andwriting-guidelinesenforces 80+ voice rules—depth a general system prompt can't hold. - Some skills act rather than advise:
vercel-deploy-claimableships an app straight to Vercel from the conversation, andvercel-optimizeaudits a live project for cost, caching, and function-usage wins. - Maintained by Vercel itself, so the platform-specific skills reflect how Vercel actually wants projects deployed and tuned.
Who It's For
Great fit if you already run an agent like Claude Code on a Vercel or React stack and want its guidance to be consistent, auditable, and tuned to your platform. Look elsewhere if you're off these tools or prefer one hand-written system prompt—most skills here assume the Vercel/React ecosystem, and their value drops sharply outside it.
