shinegeer/dsh-vision-window

Lets a text-only DSH agent see a pasted image, and adds seven zero-config local tools for crop, diff, palette, SVG trace, OCR, screenshot and foreground extraction.

An image box on the composer takes a paste, a drag or a file pick, saves the image into the workspace and puts the file path at the front of your message so the draft stays clean. The agent can then call vision mid-task whenever it needs to look at a file, while vw_* handles the pixel work locally. A registered vision-fallback skill teaches when to use which and archives results into a recognised-images folder with markdown notes.

Vision & Multimodal ★ 2
View on GitHub ↗

Install

dsh plugin --profile web add github:shinegeer/dsh-vision-window

README install line for the web profile; the README also shows the headless profile. No npm package is published, so the repository is the install source. The vw_* tools run offline with no provider, key or Python.

Compatibility

The vision tool needs an OpenAI-compatible vision provider configured; the seven vw_* tools need nothing. The plugin is additive — it never replaces composer components or intercepts keyboard events — and its only network egress is the vision provider you configure, with vw_html_screenshot blocking network requests while rendering a local file.

Details

Alternatives

sprainJinyu/dsh-vision-link · Einskyle/dsh-llm-vision-bridge · 54xkeee/dsh-vision

More plugins in Vision & Multimodal

Browse more in Vision & Multimodal

Guides for Vision & Multimodal plugins