Skip to content

About

Round-trip visual QA for AI-generated UI across design and implementation targets

Resources

Stars

1 star

Watchers

0 watching

Forks

Latest commit

 

History

4 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 

Repository files navigation

Design Roundtrip QA

Turn AI-generated UI screenshots, mockups, and prototype pages into target-ready design and implementation guidance.

This skill helps compare and rebuild user-owned or explicitly authorized material across Figma, SwiftUI, Web, React, and other product targets. It extracts a measurable design contract, preserves layout intent, and runs a visual QA loop after each correction.

Workflow

  1. Classify the source and target.
  2. Record the frame, grid, typography, spacing, imagery, and component contract.
  3. Rebuild with native, editable components for the selected target.
  4. Capture the rebuilt state at the matching viewport and scale.
  5. Produce a fresh overlay and pixel-difference image.
  6. Fix blocking drift and report pass, revise, or rework honestly.

Structural exploration and final delivery are separate outputs. Layout skeletons are useful for documenting design reasoning, but final iOS/iPadOS work requires the Apple UI Kit and native SwiftUI/UIKit components matching the target OS version. Final Android work requires the matching Material Design Kit, Material components, and Material Symbols. A project-specific UI kit is accepted only when it is a documented source of truth.

Install

npx skills add https://github.com/Jorgut/design-roundtrip-qa -a codex -g

The skill is also suitable for Claude Code, OpenCode, and other agents that support SKILL.md workflows.

Important boundary

Public screenshots may be used to study layout logic and test the QA process. Do not clone third-party brands, proprietary interfaces, or copyrighted assets. A pixel-level claim requires a fresh coded or native rebuild plus current overlay and diff evidence; a screenshot placed on the page is not a rebuild.

Reference evidence is separated into three layers: a complete source for exact QA, an optional selectively redacted presentation, and a final editable rebuild using original replacement content. A user-authorized complete screenshot may be shown as an annotated reference in a non-commercial GitHub case study when its source, purpose, rights ownership, and lack of affiliation are stated. Do not extract or package embedded third-party artwork as reusable assets. When redaction is used, keep it proportional: ordinary interface text and layout evidence remain visible while higher-risk or sensitive material is obscured.

Status

The workflow is production-oriented and continues to evolve through real Figma, SwiftUI, and Web validation cases. See SKILL.md for the complete operating rules and hard-stop criteria.

About

Round-trip visual QA for AI-generated UI across design and implementation targets

Resources

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors