dsh-prompt-workshop (Prompt Workshop): Expanding a one-sentence idea into a usable prompt for target generative models

dsh-prompt-workshop (Prompt Workshop) is a native plugin for DeepSeek Harness, maintained by 18518182595-cell, licensed under MIT, and focused on categorized model inference. It addresses the pain point in generative Agents where "users provide only a single idea, but the model is unaware of the target platform's prompt specifications." By selecting a target generator—featuring built-in syntax rules, parameter suggestions, and positive/negative prompt specifications for 19 models covering image/video/text models such as Midjourney, SDXL, FLUX, GPT-Image, Doubao Seedream, Kling, Hailuo, Sora 2, Veo 3.1, Runway Gen-4, Doubao Seedance, etc.—and inputting a single sentence, users receive the main prompt, negative prompt, parameter suggestions, variants, and gap analysis. The plugin features a dual-side architecture of Host and Browser. The Browser side provides an input workspace, while the Host side offers three tools—prompt_enhance, prompt_score, and prompt_targets—along with an HTTP API that can be directly called by Agents. The enhancement model reuses the DeepSeek and OpenAI-compatible endpoints already configured in DSH, without storing additional API keys. Prompt scoring employs a 10-dimensional rubric covering composition, lighting, motion sequence, audio dialogue, and copyright compliance. Additional features include image-to-prompt inference and an auto-updating knowledge base (default every 5 days) that fetches data online, prioritizing DeepSeek search and falling back to Bing RSS and official documentation seed pages if credentials are unavailable. Settings and history are stored in `~/.dsh/prompt-workshop.json`. Installation supports the plugin marketplace, the `dsh plugin add github:owner/repo` command, and local linking methods, requiring Node.js >= 22.13.0 and DeepSeek Harness >= 0.2.1-rc.1. Note that the API only accepts requests from localhost. The plugin runs with the permissions of the current DeepSeek Harness process; please check the source code and license before installing.