dsh-prompt-optimizer

When you work with DeepSeek Harness, writing a good prompt for an AI coding assistant is rarely a one-liner. The intent is there, but the wording is vague, details are missing, and it takes a few passes before the draft is worth sending. dsh-prompt-optimizer collapses that loop into a single click: a ✨ button appears in the input toolbar, and pressing it calls the currently active model to rewrite your draft into a clearer, more specific version, then auto-fills the result back into the input box. You just review and hit send.

The design is deliberately restrained. It only clarifies and concretizes — it does not alter your original task intent or switch languages. Streaming keeps the wait transparent: the button switches to a spinning loader with a sliding progress bar underneath; when the input box is empty the button simply greys out so no request is fired by accident. On the Host side a single precise route handles the LLM call; on the Client side a tiny icon is injected into the conversation.input.left slot. Both halves load automatically with the composition and clean up cleanly on unload.

Who is it for? If you spend most of your day in the DSH Web interface crafting prompts — assigning tasks to a coding assistant, tuning parameters, describing a bug repro — this plugin saves you a few minutes per draft. MIT licensed, lightweight, no extra runtime dependencies.