dsh-read-image-jpeg-fallback: DSH plugin that converts read_image PNG/WebP to JPEG for LM Studio compatibility
This plugin addresses a compatibility issue when using DSH with LM Studio, where PNG/WebP images generated by `read_image` are rejected by the `openai-completions` route. It uses DSH's `tools/post-execute` hook to automatically convert PNG/WebP images to opaque sRGB JPEG in the background, while leaving JPEG and other formats unchanged, requiring no configuration. It requires DSH 0.1.1-rc.2, Node.js >= 22, and sharp 0.35.3, and can be installed and activated by restarting after running a PowerShell command. The plugin follows a Fail Open strategy: if conversion fails, it retains the original image and logs a warning, ensuring tool calls succeed.
Read Moredsh-image-preview: Inline Preview Providing the `read_image` Tool for DeepSeek Harness
The `dsh-image-preview` plugin takes over the tool rendering slot in DeepSeek Harness (DSH) to provide inline previews and image-copying support for the `read_image` tool. It supports direct and nested calls, such as Code Mode, converts image formats such as JPEG/WebP/GIF to PNG for copying, and uses authenticated attachment paths for secure rendering. Installation requires DSH 0.1.2-alpha.2. Once enabled, you can view images directly in the interface or copy them via a button. The plugin is open source under the MIT license and is suitable for developers who need to visually display and handle images in DSH.
Read Moredsh-read-image-view: Displaying Local Images Generated by read_image Directly in the DSH Web Interface
This plugin is for the DeepSeek Harness (DSH) web interface and directly renders images generated by `read_image` inline within the tool card. Core features include displaying formatted thumbnails, Lightbox full-screen viewing, error handling, and a purely client-side implementation, with support for PNG/JPEG/WebP/GIF. The install command is `dsh plugin --profile web add github:dblate/dsh-read-image-view`. The plugin is licensed under the MIT License, takes over the keyed `read_image` tool card, and relies on a specific upstream extension surface. It must be used with a routing model that supports image input.
Read Moredsh-image-reader: Image reading tool providing read_image for DeepSeek Harness Agents
`dsh-image-reader` is an MIT-licensed DeepSeek Harness plugin designed to provide image reading capabilities to text-only agents. By exposing the `read_image` tool, this plugin enables agents to utilize arbitrary OpenAI-compatible vision endpoints to parse screenshots, charts, or photos within the workspace and answer questions based on the image content. Key features include: support for inputting image paths and optional query questions; flexible configuration covering `baseUrl`, `model`, timeout settings, and allowed directories; and robust security mechanisms, such as dynamically reading API keys from environment variables, resolving paths via `realpath` to prevent symlink bypasses, and performing size and extension checks prior to upload. Usage.
Read More