Fishsb/dsh-prompt-enhancer
One-click prompt enhancement: an independent LLM call rewrites your rough draft in the composer, fully undoable.
dsh-prompt-enhancer adds two core capabilities to DeepSeek Harness: prompt enhancement (the ✨ button in the composer toolbar triggers an independent LLM call that rewrites the current draft in place — fully undoable, cancellable, with 5 optimization modes: Basic, Lite, Standard, Expert, and One-click Publish; an optional memory switch accumulates optimize→edit→re-optimize rounds into a memory chain; a model chain tries multiple models in order with thinking toggle and inline connectivity tests) and voice recognition (the 🎤 button starts listening with cloud Qwen3-ASR / local offline SenseVoice dual engines, stops automatically on silence via VAD, supports a recordable global hotkey, optionally refines the transcript to remove filler words, and can auto-trigger enhancement). A bundled standalone utility provides one-click DSH service restart even when the web UI is down.
Install
dsh plugin --profile web add github:Fishsb/dsh-prompt-enhancer#mainGitHub source install per README.en.md (no npm package; dsh-prompt-enhancer returns registry 404 as of 2026-09-18): dsh plugin --profile web add github:Fishsb/dsh-prompt-enhancer#main, then restart DSH (dsh web). Requires DeepSeek Harness installed locally and pnpm in PATH. The ✨ button appears in the composer toolbar after install. Re-verified 2026-09-18: README install line = dsh plugin --profile web add github:Fishsb/dsh-prompt-enhancer#main; entry had pinned #v3.3.2 (no longer in the README). Newest release tag v3.3.3.
Compatibility
DeepSeek Harness web profile, pnpm in PATH. Voice recognition relies on the client injecting inputActions.setDraft (the official web client satisfies this); third-party clients implementing the same contract can load it — when the capability set differs, voice input degrades gracefully. i18n follows the DSH interface language (中文 / English). The bundled one-click restart utility works even when the web UI is down (desktop shortcut or CLI, service-managed and process-level fallback).
Details
- Repo: Fishsb/dsh-prompt-enhancer
- Category: Tools & Capabilities
- Stars: 48
- Version: GitHub source install github:Fishsb/dsh-prompt-enhancer#main (README ref, re-verified 2026-09-18; no npm package — dsh-prompt-enhancer returns registry 404)
- Last push: 2026-08-24
- First seen: 2026-08-13
Recent updates
The current English README (README.en.md) documents: the two core features (enhance modes, memory chain, model chain; voice dual engines, VAD silence stop, hotkey, transcript refinement, auto-enhance), i18n, the standalone one-click restart utility, install (github:#v3.3.2), restart requirement, and client compatibility notes for voice input.
FAQ
- How does the voice input stop?
- It stops automatically on silence — VAD silence detection means no manual stop is needed. Audio stays in memory only and is never written to disk.
- Can I undo an enhancement?
- Yes — enhancement is fully undoable, you can keep refining, undo anytime, or cancel while enhancing.
- Does voice input need a cloud API?
- It ships dual engines: cloud Qwen3-ASR and a local offline SenseVoice engine. The local engine keeps audio processing offline when selected.
Alternatives
SongMiao-tech/dsh-prompt-optimizer · seven282/oss-prompt-optimizer · stardustlc666/dsh-ppt