DSH Vision Toolkit
A vision toolkit and skill that gives text-only models running in DeepSeek Harness image understanding: pasted-image Q&A, long-screenshot OCR, UI restoration and pixel-level comparison.
DSH Vision Toolkit gives a text-only model running inside DeepSeek Harness the ability to inspect, locate, crop, trace, rebuild and verify visual work, rather than only produce a generic image caption. Pasting an image in DSH Web automatically switches the active model to its vision-toolkit variant, and a bundled skill tells the agent which tool to use for a given visual task, such as locating a UI element or reading an error message, and how to verify the result.
It is the native DeepSeek Harness integration of a separate agent-vision-toolkit project, and installs as a plugin into a chosen DSH profile. It works with DSH's Web and Headless profiles, sessions, Settings and Artifacts.