skills.rest
yekinya
yekinyaCOMMUNITY
Explore all skills by this author →
GitHub

moai-workflow-gan-loop

Implements Builder-Evaluator GAN loop for iterative design quality scoring and escalation.

2|Updated May 27, 2026
Search Tags:#quality assurance#scoring rubric#evaluator#iterative review#design quality#gan loop#sprint contract
One-click install
npx skills add https://github.com/yekinya/moai-novel --skill moai-workflow-gan-loop-yekinya
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: moai-workflow-gan-loop
Source: https://github.com/yekinya/moai-novel/tree/main/moai-novel/.claude/skills/moai-workflow-gan-loop
Command: npx skills add https://github.com/yekinya/moai-novel --skill moai-workflow-gan-loop-yekinya

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Iterative design work often stalls without objective quality gates, leading to subjective reviews, score inflation, and unclear pass/fail decisions. This Skill enforces a structured Builder-Evaluator loop with measurable scoring, stagnation detection, and escalation rules. ## Core Features & Use Cases - 4-Dimension Scoring: Evaluates Design Quality, Originality, Completeness, and Functionality with weighted averages and configurable pass thresholds read from design.yaml. - Sprint Contract Negotiation: Generates per-iteration acceptance checklists, test scenarios, and pass conditions that constrain both Builder and Evaluator. - Leniency Prevention: Applies rubric anchoring, must-pass firewalls, anti-pattern penalties, evidence requirements, and regression baselines to prevent score inflation. - Use Case: A frontend team building a landing page runs the loop so the Evaluator scores each Builder iteration against the Sprint Contract, escalating to the user after 3 failed iterations or detected stagnation. ## Quick Start Run the GAN loop on my current design iteration using the parameters in .moai/config/sections/design.yaml and score it against the four quality dimensions.

Frequently Asked Questions about moai-workflow-gan-loop

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How does the Builder-Evaluator GAN loop work for design quality?▼

The Builder implements a design iteration, then the Evaluator scores it across Design Quality, Originality, Completeness, and Functionality using a weighted average. If the score meets the pass threshold the loop exits; otherwise feedback returns to the Builder for the next iteration.

How do I configure pass thresholds and max iterations for the GAN loop?▼

All loop parameters are read from .moai/config/sections/design.yaml under design.gan_loop, including max_iterations, pass_threshold, escalation_after, and improvement_threshold. Thresholds should not be hardcoded in prompts or code.

What is a Sprint Contract in iterative design review?▼

A Sprint Contract is a JSON document the Evaluator generates before each iteration containing an acceptance checklist, priority dimension, test scenarios, and pass conditions. The Builder can accept it or negotiate adjustments within a limited number of rounds.

Does the GAN loop support Playwright for automated design testing?▼

Yes, when Playwright or claude-in-chrome MCP is available the Evaluator captures desktop and mobile screenshots, tests CTA interactions, and runs accessibility scans. Without these tools it falls back to static code analysis and notes the limitation.

What causes an automatic FAIL in the evaluation scoring?▼

The must-pass firewall triggers an immediate FAIL when copy differs from the contracted copy.json, AI-slop purple gradient patterns appear, the 375px mobile viewport breaks, interactive elements return 404, or Lighthouse Accessibility falls below 80.

When does the GAN loop escalate to the user instead of iterating?▼

Escalation occurs when iterations reach max_iterations, when escalation_after iterations pass without a passing score, or when score improvement stays below improvement_threshold for two consecutive iterations. The user then chooses to continue, adjust criteria, or abort.

Related Skills

Explore More Design & Creative Skills→
AppsVortexAppsVortex
31

arn-spark-style-explore

Define and document visual style direction for Arness projects.

Official
Advanced
packetalienpacketalien

smoke-check

Run automated smoke tests and generate a PASS/FAIL report for QA hand-off.

Community
Advanced
Hack23Hack23
9

playwright-testing

Automate end-to-end browser testing for static websites with Playwright.

Official
Advanced
slowman2084slowman2084
6

meta-eval-judge

Score agent outputs against YAML rubrics or gold references with verified calculations.

Community
Advanced