What it does
Vision Toolkit focuses on the awkward image jobs. It handles multi-image question answering, slices very long screenshots so text survives, restores a UI mockup into working front-end code, grounds a described element to coordinates, and diffs two renders. It is a good fit when a session regularly moves between a design and its implementation.
Key features
- Multi-image question answering in one call
- Long screenshot OCR that tiles instead of downscaling
- UI restoration from a screenshot into front-end code
- Grounding and pixel diff tools
- Artifacts saved into the workspace for later review
Install
dsh plugin --profile web add dsh-vision-toolkitCompatibility notes
Screenshot capture behaviour differs by platform; verify the capture path on your OS first.
