dsh-drop-in

DeepSeek Harness Desktop already supports native drag-and-drop for images and folders: images can enter the multimodal vision attachment pipeline, while folders can be adopted into the workspace through the desktop bridge. Many everyday files, however, have no official drag-and-drop path. PDFs, archives, source files, audio, installers, and other local resources cannot be turned into a path the agent can access. In browser environments, the security model often exposes only the file name and contents, not the local absolute path, so the agent cannot locate the file on disk.

dsh-drop-in completes the drag-and-drop workflow by turning dropped content into readable file references. Local files are saved into a session-isolated directory, and the composer receives a markdown reference whose path can be read directly by the agent's filesystem tools. Web images and URLs are downloaded on the host side with redirect following, automatic retry, and a 2 GiB cap. Short selected text is inserted into the input field, while longer selections are saved as a .md file. Pure local images continue to use the native vision attachment pipeline, and folders are resolved to their original disk locations for direct referencing without unnecessary copying.

If you regularly hand local documents, code files, web screenshots, or browser resources to the agent, and you would rather drop them in than type paths or re-save files, dsh-drop-in removes a common friction point. It is especially useful for frequent DSH Desktop users who want to bring file context into conversations naturally and keep agent access to local or web resources direct, clear, and path-accurate.