What is the GAN-style harness for Claude Code?

It is a skill that organizes multiple Claude agents into an adversarial loop where a Generator builds and an Evaluator criticizes, ensuring output quality meets professional standards.

How does the Evaluator agent ensure quality?

The Evaluator uses the Playwright MCP to interact with the live application, clicking buttons and filling forms to test functionality and visual design rather than just reading the code.

Is this skill suitable for simple bug fixes?

No, this skill is designed for high-effort, long-running tasks. For quick fixes, standard Claude Code commands are more cost-effective and faster.

What models work best with this harness?

It is optimized for high-reasoning models like Claude 3.5 or 4.6 Opus to handle complex architectural planning and strict evaluation criteria.

Can I customize the evaluation criteria?

Yes, you can configure weights and pass thresholds for categories like design quality, originality, craftsmanship, and functionality via environment variables.

gan-style-harness: GAN-Inspired Claude Code Agent Skill

The gan-style-harness is a sophisticated multi-agent orchestration skill designed for Claude Code, inspired by GAN (Generative Adversarial Network) architectures. It separates the development process into three distinct roles: a Planner for product specification, a Generator for implementation, and a highly critical Evaluator that uses Playwright to live-test applications. This adversarial feedback loop pushes AI-generated code far beyond standard quality, focusing on craftsmanship, originality, and functional perfection, making it ideal for high-stakes, production-ready full-stack projects.

Key Features

01Sprint-based development management for complex app architectures

02Multi-agent workflow separating generation from rigorous evaluation

03Automated end-to-end testing via Playwright MCP integration

04Configurable scoring rubrics for design, craft, and functionality

05214,448 GitHub stars

06Structured feedback loops that force iterative quality improvements

Use Cases

01Building full-stack applications with professional-grade UI/UX from single prompts

02Developing high-reliability web platforms requiring extensive functional testing

03Refining AI-generated code to eliminate 'generic' aesthetics and edge-case errors

Tieline — product intent grounded in code

GAN-Style Agent Harness

Key Features

Use Cases

Healthcare CDSS Development Patterns

Git Commit Pr Writer

Vue.js Patterns & Architecture

Key Features

Use Cases