dsh-plugin-image-gen

In DeepSeek Harness conversations, text models cannot produce images on their own, and external drawing APIs lack integration with the Harness workspace, credential service, and multimodal attachment pipeline. dsh-plugin-image-gen fills that gap: it exposes an OpenAI-compatible generate_image tool that makes a single synchronous POST /images/generations call, writes every returned image to the current session workspace, and—when conditions allow—stores an attachment reference so a vision-capable dialogue model can read the result directly.

The plugin supports declaring multiple provider routes, specifying a default model and an optional model allowlist, choosing the return encoding (png, jpeg, or webp), and setting per-call and per-image byte limits along with timeout margins. Credentials are referenced by POSIX identifier through the DSH credential service and resolved on every call without caching; with no providers configured the plugin stays dormant and registers no tool. File output uses Node local-filesystem exclusive creation for a balance of safety and simplicity.

It suits developers who have already built multimodal dialogue flows inside the DSH ecosystem, as well as teams that want reliable image generation behind a single minimal configuration in their own Harness profiles.