Description
gan-style-harness is an agent skill from the ECC collection by affaan-m, listed on skills.sh. It sets up a multi-agent harness in which one agent generates work and a separate evaluator, instructed to be strict, critiques it, and the critique drives the next round. The idea follows the Generator and Evaluator pattern of GANs and is based on Anthropic's March 2026 write-up on harness design for long-running application development. The stated reason is that a single agent asked to judge its own work tends to praise it, while a separate evaluator is easier to make rigorous. The skill suits building a complete application from a one-line prompt, frontend work where visual quality matters, full-stack projects that need working features rather than just code, and any task where generic AI-looking output is unacceptable. The listing notes that such a run may cost $50 to $200 to produce production-quality output. A skill is a SKILL.md file of instructions that an AI agent loads when a task calls for it. It is free and open source, and the skills.sh listing shows all three security audits passing.…
Highlights
- Separate generator and evaluator agents
- Strict critique drives each iteration
- Builds applications from a one-line prompt
- Targets high visual and functional quality
Ratings & Reviews
No ratings yet
Be the first to rate gan-style-harness.
People also visit
gan-style-harness FAQ
What does the gan-style-harness skill do?
It runs a generator agent and a strict evaluator agent in a loop, so the evaluator's critique improves each round of the build.
Why separate the evaluator?
The skill says agents asked to evaluate their own work tend to be too positive, so a separate strict evaluator is more reliable.
What does it cost?
The listing says a serious run may cost roughly $50 to $200 to get production-quality output.
How do I install it?
Run: npx skills add https://github.com/affaan-m/ecc --skill gan-style-harness
Is it free?
The skill is free and open source. The model usage it drives is billed separately.
Are there security findings?
The skills.sh listing shows Gen Agent Trust Hub, Socket and Snyk all passing.
What are the best gan-style-harness alternatives?
Popular alternatives to gan-style-harness include agent-eval, blueprint and claude-devfleet. See all gan-style-harness alternatives