| name | review-image |
| description | Procedural guide for visually reviewing and evaluating a generated or downloaded image. Use after generating an image (e.g. via /chatgpt-image) or whenever Claude needs to inspect an image file to assess quality, composition, suitability, or correctness. Not invoked with arguments — the image path must already be known from prior context. |
Review Image
Procedure to visually inspect and evaluate an image file. Follow these steps in order.
1. Get image info and resize if needed
Run the bundled script (uses uv with inline Pillow dependency):
.claude/skills/review-image/scripts/resize-for-review.py <image-path>
The script prints file size, dimensions, aspect ratio, and a READ: line indicating which file to inspect. If the image is large (>2 MB or >960 px), it saves a resized preview to /tmp/review-preview.png.
2. Read and evaluate
Use the Read tool on the path from the READ: output. After viewing, assess:
- Composition: Is the layout balanced? Is there space for text overlay if needed?
- Color palette: Does it match the intended brand or mood?
- Clarity: Are elements sharp and recognizable, or muddy and generic?
- Relevance: Does the image actually communicate the intended subject?
- Text/artifacts: Any unwanted text, watermarks, or AI artifacts?
- Aspect ratio: Does it match the intended use (16:9 for presentations, 1:1 for social, etc.)?
Report findings honestly — flag problems and suggest a re-generation with an improved prompt if the image falls short.