dsh-vimina

Driving Windows GUI from DSH has traditionally relied on screenshot-plus-OCR loops that are slow, brittle, and costly in vision tokens. dsh-vimina plugs Vimina's UIA text-scanning engine directly into DSH's native tool layer, letting the model read control names and coordinates as structured text and then perform clicks, typing, scrolling, and more without invoking any vision model. A single scan call returns the full control tree with human-readable labels, making the loop far faster and cheaper than repeated screenshot-and-read cycles.

The plugin exposes 26 tools spanning window listing, control scanning and lookup, coordinate and label-based clicking, mouse drag, keyboard combinations, a full VMA scripting engine (pause, resume, log), and region screenshots. Under the hood it drives a standalone Vimina.exe service (.NET 8) over stdio JSON-RPC; the child process is started lazily on first use and torn down automatically when the plugin unloads, requiring no extra ports or auth tokens.

It is ideal for DSH profile authors on Windows who need to automate desktop applications, drive Chromium browser pages, or batch-run VMA scripts, and especially for setups where per-interaction screenshot token cost is a real concern.