dsh-prompt-boost

When drafting prompts in the DSH web GUI, a rough sentence or two is often too vague and missing key details for the model to act on reliably. dsh-prompt-boost adds a sparkle button right beside the model selector that rewrites your current draft into a clear, structured, directly executable instruction in one click; if the result is not what you want, the button becomes an undo icon so you can restore the original text and iterate as many times as you like.

Rather than slapping on a fixed template, the optimizer is task-aware, detecting whether the work involves coding, data processing, writing, analysis, research, or ops and organizing the output accordingly. It is fidelity-first: every concrete detail such as file paths, numbers, names, and commands is preserved verbatim, and any missing critical information is flagged with a placeholder instead of being invented. Output language is locked to input language, and an anti-clobber safeguard ensures that if you edit the draft mid-optimization, the result is never written back over your changes.

It is well suited for DSH users who prefer typing short Chinese or English notes and want instantly structured, high-quality prompts, as well as developers who care about prompt engineering but would rather not hand-craft templates every time. The integration is lightweight with zero build step, ready to use from the composer tool row as soon as it is mounted.