dsh-plugin-vision-tool

Image recognition for text-only DeepSeek Harness models: registers the analyze_image tool, routing images to a dedicated vision model (default zhipu/glm-4.6v-flash) through the harness's own ctx.llm seam. 给纯文本模型（如 DeepSeek）加识图能力：文本模型会话也能'看图'。