AI-ready design system: a machine-readable source of truth

If a coding agent is drifting off-brand, you do not need a new component kit. You need the system you already ship to be readable by Cursor, Claude, and other LLMs.

Start with how we structure design systems for AI if you want the definition and the failure mode: stateless sessions that infer tokens from nearby code. Jump to the five-tier structure for the folder itself (foundations, tokens, atoms, molecules, organisms). Use the design system audit when the repo already has components. Design tokens for Tailwind v4 is the runtime wiring. Component specs are the contract an LLM can follow without re-reading the file. When the design system and the code disagree is the reconciliation. The Claude Code skill is how the pattern loads without stuffing CLAUDE.md.

DESIGN.md is the editorial vocabulary at the root of that folder, not a substitute for the folder. Claude Code design skills is the installable skill write-up. Generate design specs from code is the reverse-specs workflow when the frontend already ships. AI UI brand governance is the process after the file exists: stop reactive per-output fixes, install a root system, then choose scaffold, audit, or partner. designgov.md is the governance contract above DESIGN.md: behavior, accessibility tiers, and multi-brand rules agents must not violate. None of those pages is this series. Use them when you already know the job and need the adjacent artifact.

This hub is not a component library to install, a Figma styleguide pitch, or a DESIGN.md tutorial. It is not a design-system-audit offer. You keep shipping the components you have. You add a parallel structured layer (tokens, spec stubs, docs) so a machine-readable design system for an LLM is sitting next to the visual system your team already uses.