evaluate-diagram

Score AI-generated diagrams against human references using a VLM-as-Judge mechanism.

1|1|Updated Feb 21, 2026

Search Tags:#evaluation #comparison #scoring #quality assessment #diagram #vlm

Ready to run

npx skills add https://github.com/dtsong/claude-code-windows-setup --skill evaluate-diagram

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: evaluate-diagram
Source: https://github.com/dtsong/claude-code-windows-setup/tree/main/skills/evaluate-diagram
Command: npx skills add https://github.com/dtsong/claude-code-windows-setup --skill evaluate-diagram

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) components.

What problem does it solve?

This Skill provides an objective way to score and compare AI-generated diagrams against human-created references, ensuring accuracy and quality.

Core Features & Use Cases

Diagram Evaluation: Uses a VLM-as-Judge approach for scoring.
Quality Assessment: Compares generated diagrams against a reference image based on faithfulness, conciseness, readability, and aesthetics.
Use Case: After an AI generates a system architecture diagram, use this Skill to evaluate how well it matches a hand-drawn diagram provided by a senior engineer.

Quick Start

Use the evaluate-diagram skill to score the diagram 'output.png' against 'reference.png'.

Frequently Asked Questions about evaluate-diagram

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

FAQPage Schema

How do I evaluate AI-generated diagrams against human references?▼

You can evaluate AI-generated diagrams against human references using a VLM-as-Judge scoring mechanism that compares the generated image to a reference image based on faithfulness, conciseness, readability, and aesthetics.

Can I score diagram quality without a reference image?▼

No, scoring diagram quality requires both a generated diagram and a human-created reference image to perform the comparison and calculate scores across faithfulness, conciseness, readability, and aesthetics.

Do I need a specific tool to perform VLM diagram quality assessments?▼

Yes, diagram quality assessments require the paperbanana MCP tool or a local paperbanana CLI installation to execute the image analysis and scoring processes.

What is the best way to compare a generated system architecture diagram with a hand-drawn one?▼

The best way to compare a generated system architecture diagram with a hand-drawn one is to use a VLM-as-Judge approach that scores faithfulness, conciseness, readability, and aesthetics against the provided reference image.

What dimensions are scored when comparing generated diagrams to references?▼

When comparing generated diagrams to references, the evaluation scores four dimensions: faithfulness, conciseness, readability, and aesthetics to ensure accuracy and quality.

Why should I use an automated approach for diagram evaluation instead of manual review?▼

Automated diagram evaluation provides an objective way to score and compare AI-generated diagrams against human-created references, ensuring consistent accuracy and quality across faithfulness, conciseness, readability, and aesthetics.