Gan Style Harness

This is Anthropic's March 2026 research on adversarial agent loops turned into a practical harness. You get three agents: a Planner that expands your one-liner into a full product spec, a Generator that builds it, and an Evaluator that runs Playwright tests against the live app and scores it on design, originality, craft, and functionality. The evaluator is engineered to be ruthlessly strict because agents are terrible at critiquing their own work. It's expensive (budget $50-200) and overkill for simple tasks, but if you're building a complete frontend or full-stack app where "AI slop" aesthetics aren't acceptable, this feedback loop can push quality way beyond what a single agent produces.

skills.sh installs: 3.7K
GitHub stars: 229.1K
last updated: Jul 14, 2026

View on GitHub

Install

$ npx -y skills add affaan-m/everything-claude-code --skill gan-style-harness --agent claude-code

Installs into .claude/skills of the current project.

Files

GAN-Style Harness Skill

Inspired by Anthropic's Harness Design for Long-Running Application Development (March 24, 2026)

A multi-agent harness that separates generation from evaluation, creating an adversarial feedback loop that drives quality far beyond what a single agent can achieve.

Core Insight

When asked to evaluate their own work, agents are pathological optimists — they praise mediocre output and talk themselves out of legitimate issues. But engineering a separate evaluator to be ruthlessly strict is far more tractable than teaching a generator to self-critique.

This is the same dynamic as GANs (Generative Adversarial Networks): the Generator produces, the Evaluator critiques, and that feedback drives the next iteration.