Yu-369/VibeCurb

What it solves

AI agents often generate generic, "average" website designs because they are trained on common web patterns. VibeCurb prevents this "drift toward the mean" by imposing strict design constraints and quality gates, forcing agents to produce high-end, production-ready UIs with professional typography, complex grids, and deep contrast instead of safe, generic layouts.

How it works

VibeCurb provides a set of "skills" (structured markdown protocols) that agents must follow. Every skill implements a five-step pipeline:

1. The Design Read: The agent extracts design signals (typography, color, spacing) from a reference or brief before coding.
2. Quality Gate 1: The agent must prove it understands the design direction before any code is generated.
3. Precise Build: The agent builds the UI based strictly on the extracted signals.
4. Visual Diff: The output is verified against the reference using PASS/FAIL tables for composition and motion.
5. Drift Rejection: The system identifies and rejects common AI defaults like generic purple gradients or placeholder patterns.

Who it’s for

Developers and designers using AI coding agents (such as Cursor, Claude Code, v0, or ChatGPT) who want to create award-winning, high-fidelity frontend interfaces without the AI defaulting to generic templates.

Highlights

• Specialized Skill Pipelines: Includes dedicated protocols for hero sections (awwwards-hero), visual redesigns (visual-redesign), pixel-perfect replicas (pixel-perfect), and kinetic animations (awwwards-motion).
• Agent Integration: Can be injected via a CLI tool or manually placed in agent rule folders (e.g., .cursor/rules/).
• Design-First Workflow: Mandates a "Design Read" phase to ensure the agent understands the aesthetic before writing code.
• Brand Generation: Includes skills for creating cohesive brandkits and editorial-grade imagery to serve as references for UI builds.