dsh-image-bridge

Pasting a screenshot into DeepSeek Harness to ask a text-only model to read it usually ends with a flat refusal: the api-proxy rejects any image-bearing message when the model modality list is text-only. dsh-image-bridge closes exactly that gap, letting pure-text models inside the DeepSeek Harness still accept images and route them to your existing vision tools.

Under the hood the plugin patches three points. It wraps the model-info pre-check so image messages are no longer blocked at the proxy layer. It reads the image bytes via the attachments API and writes them into the session workspace .attachments/ directory. It then uses the DSH surface-replace mechanism to swap the image block in the model-visible transcript for a lightweight placeholder like [Image 1]:"" while the human-facing chat still renders the original thumbnail (clickable, zoomable) instead of a bare file path. The model can hand that placeholder to an existing vision or MCP tool—vision_glance, mcp__mcp-vision__analyze_image, mcp__mcp-vision__ocr_extract, etc.—for actual recognition. Models that natively support images are left completely untouched.

If your daily workflow is code review, doc drafting, or research with a DeepSeek-class text model and you occasionally need it to look at a screenshot, a diagram, or a PDF page, this plugin is the missing bridge. Two caveats: do not mix slash-commands and images in the same submit, and consider clearing or compressing chat history before switching back to a text-only model so leftover vision-model image blocks do not trip the adapter.