computer-user-vision
Run the following command in DeepSeek Harness:
dsh plugin install xie129716/computer-user-vision
Paste the following prompt into your AI chat to install this plugin:
Run dsh plugin install xie129716/computer-user-vision in your terminal to install the plugin, full source at https://github.com/xie129716/computer-user-vision
About this plugin
The upstream computer-user plugin silently vanishes on DSH 0.1.2 and later. A single missing named import fails the entire ES module at load time: no tools, no settings card, no error message. Beyond that, its design predates visual models -- a screenshot returns a file path and asks the model to hand it to a local OCR plugin. For a model whose inputModalities include image, that is an unnecessary detour.
This fork commits screenshots through the host attachment service as a native image block so the model can look directly. Windows UI Automation enumerates the controls in the focused window and generates refs; computer_click accepts a ref or a name and lands on the exact rectangle without pixel arithmetic. Window rectangles come from the DWM visible frame, and PerMonitorV2 DPI awareness eliminates the systematic offset on scaled displays. Input tools accept an optional expect_window guard: if the foreground window does not match, the call is refused before any keystroke is sent. While operating, a pulsing gradient traces all four screen edges with a floating banner; Ctrl+Alt+Esc or the Stop button hard-stops every computer tool until the user re-approves.
Built for developers and automation authors running a visual model on DSH with Windows who need the AI to read the screen, click buttons, fill forms, and switch windows with precise coordinates, controllable focus, and an interruptible safety net.
Use Cases
- A visual model reads the Windows screen and performs clicks, typing, and window switching
- An AI assistant auto-fills forms, clicks buttons, and works across multiple windows
- Desktop automation workflows requiring precise coordinates, focus safety, and a hard stop
Best For
- Developers running visual models on DSH to drive the Windows desktop
- Automation engineers building Windows desktop AI agents
- Teams needing precise coordinates, controllable focus, and interruptible computer-use workflows
Related Plugins
A unified suite combining hot runtime injection, task-aware thinking-mode routing, and a graded session protocol with red-team gates to sustain model diligence across long-horizon inference.
ModLens is a vision plugin for DeepSeek Harness that gives text-only models sight by reading images pasted directly into chat, with zero-config setup and multiple vision engines.
On-demand vision for text-only DeepSeek Harness agents: built-in free keyless vision chain and 14 vision tools, routing image turns as tool calls to vision models with pixel fidelity, no Python needed, one-command install.
Give text-only models in DeepSeek Harness eyes, enabling image Q&A, long-screenshot OCR, UI restoration, and GUI visual tasks.