niquedegraaff/dsh-chat-enhancements

dsh-chat-enhancements is a chat-experience umbrella for DeepSeek Harness: it adds a Gemini-style '+' composer menu with multi-file upload, global drag-and-drop of files and folders, paste-to-attach, Codex-style @path references and removable attachment cards, reads attachments from session-isolated .dsh-uploads storage with a paged read_document tool, and bundles the Microsoft MarkItDown TypeScript port for PDF, DOCX, PPTX, XLSX, HTML, CSV, JSON, XML, RSS, Atom, ZIP, Jupyter, image OCR and audio transcription with no Python and no downloads.

dsh-chat-enhancements is a chat-experience umbrella for DeepSeek Harness whose current feature is a Gemini-style '+' composer menu for attachments: multi-file upload, global drag-and-drop of files and folders, paste-to-attach, removable color-coded attachment cards and Codex-style @ mentions that pick an uploaded file by name rather than storage path. Small text files drop straight into the composer while larger ones show a preview snippet, documents live in session-isolated .dsh-uploads storage and are read by the agent with the paged read_document tool (offset/limit with a byte-budgeted LRU cache) when referenced by name, @, or prose like 'the file I just added'. The bundled MarkItDown engine converts 20+ formats to Markdown on demand with no external setup, and voice dictation, markdown rendering of your own messages and a redesigned composer bar are listed as planned.

Files & Data ★ 0
View on GitHub ↗

Install

dsh plugin --profile web add github:niquedegraaff/dsh-chat-enhancements

Copied verbatim from the README's Install block, which notes to restart dsh web and hard-refresh the browser; the README states the package is not yet on npm.

Compatibility

Uploads are loopback-only with sanitized file names, session-isolated storage under .dsh-uploads/<sessionId>, sha256 content dedup, bounded concurrency and a TTL sweep, and read_document reads through ctx.fs so it inherits sandbox and fs-observation policy. Document parsing is fully bundled - the MarkItDown TypeScript port (markitdown-node) is a regular dependency covering PDF, DOCX, PPTX, XLSX, HTML, CSV, JSON, XML, RSS, Atom, ZIP, Jupyter notebooks, image OCR via Tesseract (110+ languages) and audio transcription via an LLM that needs model credentials - so there is no Python, no pip, no downloads and no build-script approval; an existing MarkItDown CLI is preferred if present (adding EPUB and more). Images use the official read_image tool on multimodal routes, with a vision discovery chain (local Ollama then an OpenAI-compatible endpoint with a dsh-credentials key) for text-only routes.

Details

Alternatives

Aidenwu0209/dsh-Unlimited-OCR-Skill · amethystluna/logicprobe · bill9109/dsh-101

More plugins in Files & Data

Browse more in Files & Data

Guides for Files & Data plugins