loudMore/dsh-drop-to-path
dsh-drop-to-path lets you drop or paste images, PDFs, office docs, zips, videos, and audio into DeepSeek Harness so they reach your text-only model as workspace file paths: images keep the native attachment UI (preview/remove), everything else shows as square chips in the attachment rail, and paths are appended automatically when you hit send. The image admission preflight is bypassed — no image block is ever sent, the model only sees text. Files are uploaded to the workspace .drops/ directory and the message becomes plain text paths on send; upload failure is never silent (a visible notice explains what happened before falling back to the native send path). Supports paste and drag-and-drop, mixed batches, and ordered upload as one text block. The README also documents pairing with dsh-vision-toolkit for full vision capability, and the comparison with dsh-drag-and-drop (which inserts raw paths without copying files).
Install
⚠️ Upstream documents a local-checkout install — the command it gives is a path on your own machine, so there is nothing pasteable here. See the README for its exact steps.
Compatibility
DeepSeek Harness Web profile. Node.js and dsh CLI required. Bypasses the host image-admission preflight: no image block is ever sent, the model only sees text paths — built for text-only models (e.g. deepseek-v4-flash) that reject native image attachments. Images (png/jpg/jpeg/webp/gif, 30MB) keep the native attachment card; documents (pdf/doc/docx/xls/xlsx/ppt/pptx/txt/md/csv/json), archives (zip), video (mp4/mov/webm/mkv/avi), and audio (mp3/wav/flac/m4a) up to 100MB show as square chips.
Details
- Repo: loudMore/dsh-drop-to-path
- Category: Other
- Stars: 11
- Version: Git-source plugin (local path install); MIT; DSH Web profile
- Last push: 2026-09-18
- First seen: 2026-08-13
Recent updates
The current README documents the demo workflow, the why (text-only models rejecting native attachments), the file-type table with limits, the two install options, the dsh-vision-toolkit pairing, and the comparison with dsh-drag-and-drop.
FAQ
- Does it work with multimodal models?
- It is built for text-only models that reject native image attachments; images and files arrive as workspace paths the model can read with vision/document tools.
- What is the file size limit?
- Images up to 30MB; documents, archives, video, and audio up to 100MB.
- How does it differ from dsh-drag-and-drop?
- dsh-drag-and-drop inserts the raw path with no file copying; this plugin copies files to the workspace and converts them to paths on send, keeps the native attachment UI, and supports paste.
Alternatives
a903067276-rgb/dsh-file-mentions · bill9109/dsh-drag-and-drop · poiuyjie/dsh-vision-opencode