dsh-plugin-image-tools: Three tools to add image capabilities to DSH Web GUI
# Introduction When using the Web GUI of DeepSeek Harness (DSH) to create agents, there are three image-related scenarios that have always been difficult to handle: 1. The model wants you to select an image from a set of candidates (e.g., choosing a book cover), but images cannot be placed in the tabs; the model can only print the file path for you to open manually. 2. The model generates an image and wants to mix it with text in the response body, but there is no channel for this either. 3. You send an image into the chat, and a text-only model adapter directly reports `UNSUPPORTED_CONTENT`, interrupting the turn. dsh-plugin-image-tools is a DSH plugin designed for these three scenarios: image selection card, inline images in the response, and blind model receiving images. Three tools, each handling a specific task, with zero configuration.
Read Moredsh-image-tools: Unified Multi-Provider Image Generation and Editing for DeepSeek Harness
Agent development within DeepSeek Harness (hereinafter referred to as DSH) inevitably involves requirements for image generation or editing. The challenge lies in the inconsistency of interfaces across different service providers: OpenAI-compatible services utilize the Images API, while Google Gemini uses the Interactions API, with distinct authentication methods and parameter systems. If an agent is directly bound to a single provider, switching service providers or implementing failover later would require significant rework. The approach of `dsh-image-tools` is to consolidate this functionality into two tools: `image-generate` and `image-edit`. It exposes provider-agnostic, unified parameters to the model, allowing for the attachment of multiple services in the background.
Read Morexiaoxianyu-office/dsh-image-tools: Add image recognition capability to DeepSeek Harness text sessions
`xiaoxianyu-office/dsh-image-tools` is a DSH plugin designed to equip text-only models like `deepseek-v4-pro` with image recognition capabilities. The plugin works by automatically saving chat images to the workspace, dynamically disabling the native `read_image` tool, and guiding the model to use the `image_recognize` tool to delegate visual sub-agents (such as `xiaomi/mimo-v2.5`) for image recognition. It supports various preset sessions like `standard` and `code`, dynamically determines interception or bridging strategies based on current model capabilities, and includes a first-time setup wizard and fault self-diagnosis functions. Image recognition output follows strict specifications, requiring pixel coordinates and `#RRGGBB` color values, and addresses items not appearing in the image or...
Read More