dsh-prompt-polisher

When prompting DeepSeek models through DSH, a vague or imprecise instruction can noticeably degrade output quality, yet rephrasing by hand breaks your writing rhythm. dsh-prompt-polisher embeds a native Polish button right before the DSH send key, so the model already selected in your session refines your draft in one click—no tool-switching, no duplicate API-key setup. The result is always previewed first; you choose to apply, copy, or dismiss it. If you keep typing while the request is in flight, the plugin never silently overwrites your edits thanks to built-in draft-revision conflict detection.

Under the hood each session gets its own AbortSignal and generation guard, so a stale response can never bleed into another conversation. The panel supports cancel, retry, restore-original, copy, and force-apply flows, with a clear warning when the draft has drifted since the polish result was produced. The UI is bilingual (Chinese / English), keyboard-accessible, and respects reduced-motion preferences. By default only the current draft is sent to the model; an opt-in limited recent-context window can be enabled with configurable message count and character budget.

This plugin is built for developers and technical writers who use DeepSeek models daily inside DSH Web and want to sharpen prompts without interrupting their flow. It stores no API credentials, writes nothing to the chat transcript, and makes exactly one auxiliary LLM call when you press the button.