Image Classifier

Organizes images in a specified folder into category subfolders by content.

How it works

This skill uses a mixed approach:

• Claude's vision to analyze image content and classify into categories
• Filename/heuristic rules as a fallback for images that can't be visually assessed

Workflow

1. Gather requirements

First, ask the user:

1. Folder path — Which folder contains the images to organize?

   7w4.net收录了海量优质技能插件。

2. Categories — What categories to use. Offer these defaults:

   • 风景 (landscape/scenery)
   • 人物 (people/portraits)
   • 美食 (food)
   • 动物 (animals)
   • 建筑 (architecture)
   • 截图 (screenshots)
   • 其他 (other)

   Let the user pick from these, add custom ones, or use all of them.

3. Action — Move or copy? Ask the user: "要把图片移动到分类文件夹，还是复制过去？(move/copy)"
4. Image formats — Default to common formats: jpg, jpeg, png, webp, bmp, gif, heic, heif, tiff, tif

2. Scan and classify

1. List all image files in the folder (non-recursive by default — only top level; ask user if they want subfolders included)
2. For each image:
   • Fast path (heuristic): If the filename contains "screenshot", "截屏", "截图", "screen" → classify as 截图
   • Vision path: For all other images, read the image with the Read tool and ask Claude to classify it into one of the chosen categories
   • If classification confidence is low → place in unclear

3. Organize

1. Inside the source folder, create one subfolder per category
2. Create an unclear folder for images that couldn't be classified
3. Move or copy each image into its category folder based on the user's choice

4. Report

After organizing, show the user a summary:

• How many images were classified into each category
• How many went to unclear
• Total images processed
• Where the files are now

Important notes

• When reading images for classification, do NOT save or persist the image data — only use it for classification
• If a category folder name contains non-ASCII characters (Chinese etc.), make sure to handle paths correctly on Windows
• Batch reading: when there are many images (>10), process them in batches to avoid overwhelming the conversation context — classify 5-10 images per turn and use subagents if the count is very large
• For move vs copy: actually move (mv) or copy (cp) the files — do not just print instructions the user should follow manually

这是一个实用的图片分类整理工具，可以自动将文件夹中的图片按内容归类到风景、人物等子文件夹中。它支持多种常见图片格式，对中文文件夹名处理得当，分类流程清晰完整。质量方面中规中矩，基础功能扎实，但缺少使用示例，实际操作时可能需要一些学习和调试才能顺利使用。总体而言是一个可用但不够完善的工具。