impeccable

I want to create a sophisticated system called 'Impeccable' that acts as a design language and toolkit for Large Language Models (LLMs), enabling them to generate significantly higher quality and more opinionated frontend designs. The core problem is that LLMs often produce generic or flawed UI without deep, explicit design guidance, falling into common patterns like predictable fonts, colors, or component structures.

My architectural intent is to build a system that fights these inherent LLM biases by providing expert design knowledge and control. This project should function as an expanded "skill" that AI agents can leverage, offering a comprehensive vocabulary for design.

Key components of this system should include:

1.  **An expanded design skill:** This will be a collection of deeply specialized reference files, each focusing on a critical aspect of frontend design like typography (modular scales, OpenType), color theory (OKLCH, accessibility), spatial design (grids, hierarchy), motion design (easing, staggering), interaction design, responsive design, and UX writing. These references should provide opinionated guidance to counteract generic LLM output.

2.  **A suite of specialized commands:** The system should expose an API or command structure that allows an AI agent to invoke specific design actions. Examples include commands to `craft` a full UI flow, `critique` UX design, `audit` for technical quality (accessibility, performance), `polish` for final refinement, or even commands to adjust design intensity like `bolder` or `quieter`. The architecture needs to be extensible for many such commands.

3.  **Curated anti-patterns:** Crucially, the system will explicitly define what *not* to do in design, guiding the AI away from common mistakes and predictable patterns.

From a technology perspective, I envision this as a JavaScript/TypeScript project, ideally leveraging Bun for efficient development and runtime, given its modern ecosystem. It needs to integrate with various AI SDKs such as Anthropic, OpenAI, and Google GenAI, as it's designed to augment the capabilities of agents built on these platforms. The deployment target could be serverless functions or an edge platform, potentially using Cloudflare Wrangler, to serve these design capabilities to AI agents. The output of this system should range from refined design suggestions and code snippets to automatically generated design documentation.

The goal is to create a powerful, opinionated design co-pilot that transforms generic LLM design output into truly impeccable frontend experiences.