Design.MD vs SkillClaw

Design.MD is a collection of Markdown files that encode brand visual languages in a format AI coding agents actually understand. Drop a DESIGN.md file into your project and your AI coding agent — Cursor, Claude Code, Lovable, v0, Bolt — generates UI that matches the target brand instead of defaulting to "the AI beige" of generic Tailwind defaults. The library ships with 60+ ready-made design system files covering popular brands like Stripe, Notion, Linear, and Vercel, encoding their exact color palettes, typography scales, spacing systems, component patterns, and motion guidelines. Files include Tailwind configurations, CSS variables, and component-level patterns — not just vibe words. If a brand isn't available, there's a custom generation flow and a request system. This is a deceptively simple idea with real product leverage. AI agents are excellent at building functional UIs but terrible at design consistency without explicit constraints. DESIGN.md files act as a persistent design brief that the agent can read every time it touches the front end. For indie builders, agencies, and rapid prototypers, this solves a real and recurring problem — free and open, which removes any friction to adoption.

SkillClaw is a research framework from Alibaba's AMAP-ML team that enables collective skill evolution for LLM agent systems deployed at scale. The core idea: instead of each user's agent interactions existing in isolation, SkillClaw aggregates anonymized skill-improvement signals across all users to continuously refine a shared library of reusable agent skills — without requiring centralized fine-tuning. The framework introduces a three-component architecture: a Skill Extractor that identifies and catalogs atomic capabilities from interactions, a Skill Evolver that proposes improvements based on aggregate feedback, and a Skill Selector that routes tasks to the best-available skill version per user context. Published on April 9 and hitting #1 on Hugging Face trending papers this week with 277 upvotes, the paper reports significant improvements over per-user baselines on complex multi-step agentic tasks. This matters especially for production agent deployments where cold-start problems are severe — a new user's agent immediately benefits from millions of prior interactions. It's a fundamentally different model of agent improvement than either fine-tuning (expensive, periodic) or RAG (retrieval-only, no learning).