dsh-prompt-enhance

Writing prompts for AI coding agents is rarely about eloquence. The real gap is completeness: what to change, what to leave alone, what the deliverable looks like, and how done is defined. A one-liner in your head often covers only the first 20 percent of that, and manually filling in the rest on every single draft is slow and error-prone.

dsh-prompt-enhance collapses that ritual into a single click next to the input box. Press the ✨ button and the plugin reads your draft, makes one isolated model call using the same model in your current session, and rewrites the text into a goal-scoped task specification with explicit deliverables, constraints, and acceptance criteria. The result replaces the draft in place, and the ↩ button restores your original the instant you are not happy. Five rewriting styles (fill in details, tighten, expand, add constraints, structure) cover different needs; you can pin a cheaper model route for rewriting if you prefer. When the draft contains @file references, /commands, or skill chips, the button greys out with a clear explanation, because a whole-text replacement would silently downgrade those structured tokens.

It is built for developers who write agent prompts in DSH Composer many times a day and want the clarifying step handled by the model rather than by muscle memory. The plugin writes no session log, stores no draft on disk, and keeps the one model call out of your conversation context. The only network action it performs is sending your draft to the model route you have already configured, with same-origin enforcement, a 12 000-character cap, and a 90-second timeout.