dsh-prompt-optimizer

Writing prompts in the DSH Web GUI often hits a wall right after the draft is done: the wording is close but not quite right. The old workaround was copying text into another window for a polish pass and pasting it back, or just sending it and letting context quality quietly degrade.

dsh-prompt-optimizer adds an optimize button and an auto-toggle directly to the composer tool row, right beside the model switcher. A manual click runs one optimization pass on the current draft; with auto mode on, pressing Enter or clicking send first intercepts the action, optimizes the text, and presents a read-only original alongside an editable optimized version in a side-by-side diff dialog. You can adopt and send, fill the input only, fall back to the original, or discard. On any failure or timeout, auto mode sends the original seamlessly, manual mode surfaces the error and allows a retry, and the conversation never stalls. Two backends are supported: the DSH built-in route (no API key needed) and any OpenAI-compatible endpoint (DeepSeek, OpenAI, or custom), with keys stored only in a local host-side file and masked in the browser.

It is well suited for developers who write system prompts, role definitions, or multi-turn dialogue instructions in DSH daily and would rather not juggle an extra window, hand-build API calls, or lose a full context block because one optimization pass failed. At roughly 55 KB gzipped to 12 KB and registered through the official slot system, it behaves more like a small toolbar attachment than a standalone workbench.