What is Agentic Vision in this context?

Agentic Vision enables the AI to autonomously crop, zoom, and annotate specific regions of an image to inspect dense diagrams or small labels that might be missed in a standard full-image view.

How does the Visual Verify scoring system work?

Visual Verify uses Gemini Vision to evaluate rendered outputs against a 9-point checklist of visual defects. A score of 0-10 is assigned, and the skill requires a minimum score of 9.5 to consider a task complete.

Can it automatically fix layout issues?

Yes. Once a defect is identified, the skill translates spatial feedback into code modifications, such as adjusting coordinates, padding, or font sizes, and re-renders the output for verification.

Which frameworks and languages are supported?

The skill includes specialized support for Typst (Fletcher/CeTZ), Python (Matplotlib/Seaborn), and general web UI screenshots, but can be adapted for any visual output that can be rendered to PNG.

Visual Verify

Name: Visual Verify
Author: edwinhu

byedwinhu

•

デザインツール

Automates the rendering, visual scoring, and iterative refinement of UI components, charts, and documents using Gemini Vision.

Visual Verify is a high-precision quality assurance skill for Claude Code that implements a rigorous 'render-vision-fix' loop to ensure visual artifacts are production-ready. While standard code verification checks for syntax, Visual Verify focuses on the final output, using agentic vision to detect subtle defects like text clipping, overlapping elements, and routing errors in complex diagrams. It enforces a strict scoring threshold of 9.5/10, iteratively modifying source code until the visual output is objectively correct, making it an indispensable tool for developers working on Typst slides, data visualizations, and frontend layouts.

主な機能

01Agentic vision routing for autonomous cropping and zooming into fine-grained layout details

02Closed-loop iteration that persists until visual thresholds are met or exceeded

035 GitHub stars

04Domain-specific support for Typst, Matplotlib charts, and web UI screenshots

05Intelligent translation of spatial pixel feedback into precise source code adjustments

06Automated 0-10 scoring system based on 9 specific visual defect categories

ユースケース

01Refining data visualizations to prevent text clipping and ensure axis readability

02Polishing complex architectural diagrams to ensure clean line routing and no overlapping labels

03Validating presentation slides for consistent spacing and professional composition

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add edwinhu/workflows visual-verify

For use in Claude.ai and ChatGPT

Download Skill