dsh-tools-plugins

When working with Agents in DeepSeek Harness, one recurring frustration is that vague, one-liner instructions often produce inconsistent or low-quality outputs. dsh-tools-plugins addresses this by adding the prompt-enhancer plugin, which intercepts a short or ambiguous instruction before it reaches the model and restructures it for better results. You no longer need to manually craft a detailed prompt for every task.

The core capability is the optimize_prompt tool. It takes a brief, vague instruction and expands it into a structured prompt with four explicit dimensions: Role, Task, Constraints, and Output Format. The rewriting is powered by your profile's default LLM, so it adapts to whatever model you have configured. The process is transparent to you: you can ask the Agent in the Web UI to rewrite an instruction, or let the Agent invoke the tool automatically whenever it detects a vague input.

This plugin is a great fit for anyone using DeepSeek Harness who wants more predictable Agent outputs without spending time designing a prompt from scratch each time. Especially useful for developers who drive their workflow with short, conversational instructions and would prefer the system to handle the structuring. Once installed into a profile, it is immediately available, and the underlying provider or model can be overridden per-profile via cordis.patch.yml if needed.