What problem does it solve? AI models cannot reliably grade their own design output, so single-pass frontend builds ship with generic fonts, invented data, and visual slop. This Skill separates the builder from independent fresh-context critics and iterates against a checklist until the design passes. ## Core Features & Use Cases - Research-First Workflow: Scrapes 1-3 real reference sites with Firecrawl in branding format to extract typography, color tokens, and hierarchy before any design work begins. - Design System Codification: Defines deliberate font choices, full color scales (50-900), spacing bases, one consistent icon pack, and imagery libraries persisted in a DESIGN.md token file. - Multi-Critic Gauntlet Loop: Spins up 3+ independent critic agents (brief, system, craft/visual-impact) that grade the rendered artifact without seeing the builder's reasoning, looping 3-10 rounds until the checklist passes. - Use Case: When building a flagship landing page or reusable dashboard template, run the full gauntlet so fresh critics catch the 5 slop tells (typography, imagery, hierarchy, color, spacing) that the builder model would miss in self-review. ## Quick Start Use the design-gauntlet skill to build my landing page: research reference sites first, codify a design system, then iterate with independent critic agents until it passes the slop diagnostic.
