Browse / Software development / design-intelligence-system
design-intelligence-system
humanrace-ai
Multi-agent design system for Claude Code — research, competition, and continuous learning for world-class UI/UX
Open humanrace-ai/design-intelligence-system → On the boundary of the four rules.
Built for one agent
ships CLAUDE.md and nothing for other agents.
In its own words
A multi-agent design system for Claude Code that produces world-class UI/UX through research, competition, and continuous learning -- not single-shot generation.
The Problem
AI-generated frontend design converges on identical patterns: cards-in-grids, dark-mode-with-neon, Inter font, identical spacing, rounded corners with drop shadows. Every output looks like "AI made this." The Design Intelligence System fixes this through a 5-layer architecture that treats design quality as a search problem requiring exploration, competition, and human judgment.
5-Layer Architecture
Layer 1: Knowledge Arsenal (kb/)
A growing library of curated design knowledge that every agent consults before designing.
Layer 2: Continuous Research
The design-scout agent browses Awwwards, Dribbble, Mobbin, and design engineering blogs daily, producing trend digests that keep the Knowledge Arsenal current. The design-researcher agent performs deep domain research before each competition.
Layer 3: Competition Pipeline
Instead of generating one design, the system runs competitions:
Quoting 20% of the README. Read the rest on GitHub ↗
Read from the repository's git tree on 2026-08-17.