A native DeepSeek Harness plugin that upgrades composer drafts with AI prompt enhancement and speech-to-text, plus a browser-free DSH restart.
DSH integration
Native runtime
Author-claimed
Safety audit
Unaudited
Last verified
2026-09-02
License
Not specified by the author
01What can it help you accomplish?
Turn a rough composer draft into a polished, model-ready prompt without leaving the DeepSeek Harness chat box
The current draft is replaced in place by an enhanced version (5 modes, optional memory chain), which you can keep optimizing or revert
DSH web users who want better prompts but don't want to hand-craft them
Dictate a message by voice and have it land in the composer draft, auto-stopped when you pause
Transcribed text (cloud Qwen3-ASR or offline SenseVoice), optionally cleaned by the base LLM, inserted into the draft
DSH web users who prefer speech input or hands-free drafting
02How to install into DeepSeek Harness
Prerequisites
- DeepSeek Harness (dsh) installed on the machine
- pnpm available in PATH
- DSH web profile running (restart with `dsh web`) so the ✨ button appears
- Official DSH web client, or any client implementing inputActions.setDraft, for voice input
- For offline voice: download the SenseVoice model (~228MB) once via Settings → Model config → 💬 Voice recognition
Installation steps
- 01
Run `dsh plugin --profile web add github:Fishsb/dsh-prompt-enhancer#v3.3.3`
$ dsh plugin --profile web add github:Fishsb/dsh-prompt-enhancer#v3.3.3
- 02
Restart DeepSeek Harness with `dsh web`
$ dsh web
- 03
Confirm the ✨ button now appears in the composer toolbar
- 04
For voice input, open Settings → Model config → 💬 Voice recognition and set engine to 本地 (local)
- 05
Click 下载模型 (Download model) in the 本地模型 area to fetch SenseVoice (~228MB) with progress
- 06
Optionally record a global hotkey for voice唤醒 under Settings → Voice recognition
Verify the integration
- After restart, the ✨ button shows in the composer toolbar
- Clicking ✨ rewrites the draft via a separate LLM call (revert with 可撤回)
- For voice: the 🎤 button appears; speaking then pausing auto-stops and fills the draft
Rollback
- Run `dsh plugin --profile web remove dsh-prompt-enhancer`
- Restart DSH (`dsh web`) to remove it from the running process
03DSH integration and capability boundaries
Native DSH plugin — installed via `dsh plugin --profile web add` and runs inside the DeepSeek Harness web composer (dsh-plugin).
One-click prompt enhancement
current composer draft text→rewritten draft (5 modes) via a separate LLM call, in place
replaces the current draft text in place; press 可撤回 to revertVoice recognition (dual engine)
microphone speech→transcribed text inserted into the draft (cloud Qwen3-ASR or offline SenseVoice), optionally LLM-cleaned
audio is held only in memory and never written to diskOne-click DSH service restart
user trigger (desktop shortcut or CLI)→DSH web service restarted / process-level fallback
restarts the dsh-web service and its processMulti-language UI
DSH interface language setting→buttons and copy follow the DSH locale (中文 / English)
04Who is it for? When not to use it?
Good for
- DSH web users who want better prompts but don't want to hand-craft them
- DSH web users who prefer speech input or hands-free drafting
Not for
- Voice input needs the DSH client to inject inputActions.setDraft (the official web client does). Third-party clients that implement the same contract can load it; otherwise voice input degrades (result appended to draft end) or is disabled with a notice.
05Compatibility, maintenance and safety notes
- Voice input needs the DSH client to inject inputActions.setDraft (the official web client does). Third-party clients that implement the same contract can load it; otherwise voice input degrades (result appended to draft end) or is disabled with a notice.
- The offline SenseVoice engine is 'framework + optional download' — the plugin does not bundle or pre-download the model; you must fetch SenseVoice (~228MB) once via Settings → Model config → 💬 Voice recognition → Download model.
- Requires DeepSeek Harness (dsh) installed on the machine and pnpm available in PATH; after install you must restart DSH for the ✨ button to appear.
Active — README install command pins v3.3.3
06Frequently asked questions
How do I install dsh-prompt-enhancer into DeepSeek Harness?
Run `dsh plugin --profile web add github:Fishsb/dsh-prompt-enhancer#v3.3.3`, then restart DSH with `dsh web`; the ✨ button in the composer toolbar confirms success. Requires DSH installed and pnpm in PATH.
Does voice input work on any DSH client?
Voice input needs the client to inject inputActions.setDraft (the official web client does). Third-party clients implementing that contract can load it; otherwise voice degrades (result appended to draft end) or is disabled with a notice.
Does the offline voice engine bundle the model?
No. The local SenseVoice engine is 'framework + optional download' — the plugin does not bundle or pre-download the model; you download SenseVoice (~228MB) once via Settings → Model config → 💬 Voice recognition → Download model.
What happens to my recordings and data?
Audio is held only in memory and not written to disk; the plugin states it does not record or report any data, though enhancement results come from an external LLM.
Can I restart DSH from the plugin if the web UI is down?
Yes — the one-click restart is independent of the browser/port: double-click the whale-icon desktop shortcut, or call the CLI executor with `--cli restart --service dsh-web --profile web`.
07Related DSH workflows
archify
by tt-a1i
Agent skill for beautiful, verifiable architecture, workflow, sequence, data-flow, and lifecycle diagrams—self-contained HTML with motion and crisp export.
openviking
by volcengine
Self-evolving Context Database for AI Agents. Unify Agent Memory, Knowledge RAG and Skills.
nocobase
by nocobase
NocoBase is an open-source AI + no-code platform for building business systems fast. Instead of generating everything from scratch, AI works on top of production-proven infrastructure and a WYSIWYG no-code interface, so you get both speed and reliability.
learn-harness-engineering
by walkinglabs
Harness engineering beginner tutorial, from 0 to 1
08Data and sources
DeepSeek Harness (DSH) 插件。**两大核心能力**：
dsh plugin --profile web add github:Fishsb/dsh-prompt-enhancer#v3.3.3
This page is generated from the project’s public documentation, repository metadata and a structured parse of DSH Plugins; last verified on 2026-09-02. Found an error? Submit a correction.
Best DeepSeek Harness Plugins
Twelve plugins worth installing first — picked from the whole catalog, across every category.