Most coding agents produce functional but generic UIs — predictable, repetitive, and visually ‘sloppy’. This project packages higher-level design guidance as portable agent skills so an LLM or coding agent can prefer clearer hierarchy, smarter spacing, and richer visual direction instead of boilerplate output. That shift makes the agent a collaborator on visual taste, not just on scaffolding.
What Sets It Apart
- Portable SKILL files for agent consumption: the repo exposes instruction-level skills (both implementation and image-generation) that agents can load and use as modular capabilities — so agents apply consistent design constraints across outputs.
- Explicit design dials (VARIANCE / MOTION / DENSITY): these tune layout experimentation, animation intensity, and visual density, letting teams trade exploration vs. consistency without rewriting prompts.
- Image-first and image→code workflows: includes image-generation skills for reference boards and pipelines that encourage agents to generate frames first, then implement them — so visual direction and code are aligned.
- Agent-agnostic integration: documentation and examples target multiple coding agents (Codex-style, Cursor, Claude Code, etc.), making the skills reusable across agent toolchains rather than tied to one framework.
Who It's For and Trade-offs
Great fit if you use code-generating agents to scaffold frontends and want outputs that look intentionally designed rather than generic. Useful for teams that run agent-led design→implement loops, designers who want reproducible visual rules, and devs who pair image generation with code agents.
Look elsewhere if you need a component library or production-ready UI kit—this repo supplies instruction/skill layers and visual guidance, not a shipped component system. Also note v2 is described as experimental: expect active iteration and opinionated anti-slop rules that may require workflow adjustments.
