DeepSeek Harness Plugin: dsh-image-auto-describe

The `dsh-image-auto-describe` plugin aims to address the issue that the primary text models in DeepSeek Harness (DSH) cannot directly process images. It intercepts messages containing pasted images, invokes a vision model to automatically transcribe the image content into text, and then injects that text for DeepSeek to process. During installation, the plugin must be added via the command line and a patch operation must be executed. Users need to specify the visual routing provider (such as SiliconFlow or Zhipu) and the model in the configuration. The plugin’s recognition process runs automatically; the resulting description is injected into the context as text, while the original image is used for display only. The plugin is open-sourced under the MIT license and is suitable for scenarios such as screenshots and logs.