dsh-vision-primitives
Run the following command in DeepSeek Harness:
dsh plugin install zouyuanqing/dsh-vision-primitives
Paste the following prompt into your AI chat to install this plugin:
Run dsh plugin install zouyuanqing/dsh-vision-primitives in your DeepSeek Harness terminal to install the plugin (source: https://github.com/zouyuanqing/dsh-vision-primitives); the full vision_* toolset registers automatically after restarting DSH.
About this plugin
Text-only agents operating GUIs or analyzing screenshots often struggle with a persistent problem: vision models return vague coordinates that cannot be resolved to precise pixel positions. dsh-vision-primitives gives DeepSeek Harness agents a pixel-accurate set of eyes anchored on a Set-of-Mark numbered grid and deterministic coordinate math. A pure-JS kernel handles capture, lossless local zoom, geometric annotation, frame differencing, color segmentation, and native OCR entirely inside the DSH runtime with zero external MCP dependencies.
The plugin also integrates a MiMo V2.5 multimodal backend (vision_describe / vision_locate) and a Visual Evidence Protocol (VEP) that lets text-only models structurally interpret images. A chat-box image bridge ensures that pasting images while using a text-only model no longer raises an error; on send, images are cached as workspace files the model can read on demand.
Ideal for DSH developers building desktop automation, UI testing, or screenshot parsing agents that require precise visual-coordinate reasoning, especially teams who prefer a single-runtime, dependency-light setup over external MCP services.
Use Cases
- Precisely locating buttons and UI elements via a numbered grid when an agent drives a desktop GUI
- Injecting structured layout and element bounding boxes into a text-only model context after screenshot analysis
- Detecting inter-frame differences and localizing changed regions in automated UI regression testing
Best For
- Developers building desktop automation or UI-testing agents inside DeepSeek Harness
- Teams that need pixel-accurate visual coordinates without spinning up external MCP services
- Architects extending text-only models with deterministic, structured visual understanding
Related Plugins
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.
An armor-breaking plugin for DeepSeek: rewrites instruction priority, expands refusal-blocked phrases, persists across sessions, and shows a green active indicator.