dsh-plugin-image-tools

DeepSeek Harness Web GUI had virtually no image-interaction capability before: a model could not let users pick an image from a set of options, could not inline images inside a reply, and had no way to handle image messages arriving at a text-only adapter. dsh-plugin-image-tools fills that gap as a pure plugin change to no core code, consumes zero tokens, and renders everything locally. It ships three tools, each targeting a distinct scenario: ask_user_choice attaches an image to every option in the choice card with a Lightbox enlarge view; show_images lets a model embed 1 to 9 captioned images directly in the reply body so they display inline with the text; save_received_images rewrites image blocks in user messages to text placeholders so text-only adapters no longer error, then lets the model call the tool to persist the image as a workspace file for further analysis. Image sources are uniform across all three - local paths (including ComfyUI output), HTTP/HTTPS URLs, and base64 data URIs - with no configuration, no API key, and no external service dependency. It is well suited to three groups of users: dsh users who want the model to present image options for selection; workflow authors who need to display generated images inside model replies; and developers pairing a text-only model (no vision input) with ComfyUI, Stable Diffusion, or similar tools who need to bring images into the conversation and inspect dimensions, pixel data, or hashes.