dsh-prompt-refine
Run the following command in DeepSeek Harness:
dsh plugin install sojo-negai/dsh-prompt-refine
Paste the following prompt into your AI chat to install this plugin:
Run dsh plugin install sojo-negai/dsh-prompt-refine in the DeepSeek Harness terminal; the source repository is at https://github.com/sojo-negai/dsh-prompt-refine. Restart DSH after installation to activate the refine button next to the send key.
About this plugin
You write a prompt, sense something is off, hit send anyway, and the model misses the point. dsh-prompt-refine exists to give you a quick thirty-second checkup before that moment happens.
What it does is intentionally restrained. Click the sparkle button next to Send and the plugin reads your current draft plus the last few turns of conversation context, calls the agent-default-model you already have in settings.yaml with zero extra configuration, and returns three suggestions in a problem-plus-patch format. The philosophy is patch, not rewrite: your original wording stays intact while the model flags only what is missing and offers minimal fill-in text. All three suggestions can be accepted simultaneously rather than forcing a single choice, and a live preview in the input box lets you see the merged result before you commit. You always keep the final send action; Esc, a backdrop click, or the Skip button dismisses the panel instantly without ever overriding you.
It is built for anyone who writes prompts frequently and iterates across multiple turns, especially when you want to preserve your own phrasing while making sure no critical constraint slips through. Version 0.1 works out of the box, inheriting your existing agent-default-model from settings.yaml with no per-plugin model configuration required.
Screenshots
Use Cases
- Your multi-turn prompt is getting scattered; use a quick check to fill in constraints you forgot before sending
- You wrote a complex instruction and are unsure whether every critical condition is covered - get a concise gap list from a second-pass model
- A final sanity check before hitting send to stop the model from heading in the wrong direction due to missing context
Best For
- Developers who iterate on multi-turn LLM conversations and need fast prompt refinement
- Prompt writers who value their own phrasing and reject full-rewrite auto-completion
- Users who want zero extra model config and simply reuse what is already in settings.yaml
Related Plugins
A unified suite combining hot runtime injection, task-aware thinking-mode routing, and a graded session protocol with red-team gates to sustain model diligence across long-horizon inference.
ModLens is a vision plugin for DeepSeek Harness that gives text-only models sight by reading images pasted directly into chat, with zero-config setup and multiple vision engines.
On-demand vision for text-only DeepSeek Harness agents: built-in free keyless vision chain and 14 vision tools, routing image turns as tool calls to vision models with pixel fidelity, no Python needed, one-command install.
Give text-only models in DeepSeek Harness eyes, enabling image Q&A, long-screenshot OCR, UI restoration, and GUI visual tasks.