Image Analysis Router

中文文档 | English

🎯 This skill does not treat every image like the same homework. It decides how the image should be read first, then returns a targeted critique and a practical study plan.

✨ Features

🚀 Quick Start

Prerequisites

# 1. Python for the local routing script
py --version

# 2. An AI agent environment that can read local skills
# For example: an environment that can load SKILL.md, references/, scripts/, and image input

# 3. Image input
# Local images, screenshots, filenames, OCR text, or image-heavy requests with context

Usage

🧩 This is a local skill for OpenAI- and Claude-based agent environments. Put the whole folder into your skill directory:

mkdir -p "$CODEX_HOME/skills"
cp -r ./image-analysis-router "$CODEX_HOME/skills/image-analysis-router"

📁 If your agent uses a different skill folder, such as .agents/skills/, replace the target path and keep the same folder structure.

💬 Then call the skill with a real image request, for example:

"Use image-analysis-router to review this poster. Focus on hierarchy and typography."

"Break down this batch of interior renders and tell me what to practice next."

"Figure out which route fits this slide first, then give me a study report."

🔎 If you only have text clues, filenames, or OCR, run the routing script first:

py .\scripts\route_image_request.py --prompt "Analyze this tower facade and check how it meets the street" --file "tower-facade-render.jpg"

🧭 How Routing Works

🛣️ The skill answers one question before anything else: what is the right way to read this image?

Image request
    ↓
[1️⃣ Text and filename pass] ──→ route_image_request.py generates a routing prior
    ↓
[2️⃣ Visual confirmation] ──→ confirm the main route or split the batch
    ↓
[3️⃣ Method selection] ──→ choose 1 of 17 specialized routes
    ↓
[4️⃣ Final output] ──→ Analysis Report + Study Report

Route Confidence

🧠 The skill marks how sure it is about the route before moving on:

🗂️ Route Coverage

🖍️ The skill currently ships with 17 routes for common image-reading jobs.

Design and communication

• graphic-design: posters, brand visuals, packaging, UI screenshots, ad creatives
• infographic-diagram: charts, maps, process diagrams, information graphics
• typography-lettering: type posters, lettering, logotypes, calligraphy, letterform study
• presentation-document: slides, report pages, proposal pages, document spreads

Image and narrative

• photography: documentary, portrait, street, editorial, commercial photography
• film-frame: movie stills, animation frames, storyboard shots, cinematic compositions
• comics-sequential: comic pages, manga, strips, webtoon panels, sequence storytelling
• game-visual-design: game UI, HUD, level screenshots, character panels

Art and space

• painting-illustration: paintings, illustrations, concept art, stylized image work
• interior-design: interior renders, room photos, material and furniture studies
• architecture-urban: facades, street views, public space, urban and site relationships
• sculpture-installation-craft: sculpture, installation, ceramics, craft-based 3D work

Objects and specialist imagery

• product-industrial-design: products, prototypes, object form, packaging structure
• fashion-styling: outfits, silhouettes, accessories, lookbooks, styling visuals
• scientific-medical-imaging: medical scans, microscopy, technical and research imagery

Fallback routes

• generic-mixed: mixed batches that need grouping before critique
• universal-fallback: images that do not fit cleanly anywhere else but still need a structured read

🧠 What You Get

📌 Every run starts with a route decision: which route won, how confident the skill is, and why that route fits better than the alternatives.

📝 The Analysis Report gives a quick read, a route-specific breakdown, and a final judgment about what the image is trying to do and where it actually lands.

📚 The Study Report turns that into next steps: what to learn now, what drills to run, what to watch next time, and what comparisons are worth making.

📁 Project Structure

image-analysis-router/
├── SKILL.md
├── README.md
├── README.zh-CN.md
├── agents/
│   └── openai.yaml
├── scripts/
│   └── route_image_request.py
└── references/
    ├── route-matrix.md
    ├── output-contract.md
    └── method-*.md

⚙️ Configuration

🛠️ The skill does not require a separate config file for the default workflow.

🔤 If you only want a local routing guess before the full image review, you can call the script directly:

py .\scripts\route_image_request.py --prompt "<user goal>" --file "<file name or path>" --hint "<OCR or extra clue>"

📎 Parameters:

• --prompt: the user's goal or question
• --file: image file name or path, repeatable
• --hint: OCR text, title, note, or any extra clue, repeatable

📄 Output

📦 A normal run gives you two core outputs:

🧪 For batches, the skill can also comment on set-level consistency, before/after changes, and whether the images should be split into subgroups first.

⚠️ Ground Rules

🧱 The script output is a starting point, not the final answer.

👀 Important judgments have to go back to visible evidence.

📐 Different image types should not be judged with the same standard.

🤝 When the route is unclear, the skill should say so instead of bluffing.

✅ Supported Environments

💻 This skill works best in AI agent environments that can load local skill folders and accept image input.

🔗 Related Files

📚 Key files in this repo:

• README.zh-CN.md
• SKILL.md
• references/route-matrix.md
• references/output-contract.md
• scripts/route_image_request.py