Preface

When developing agents in DeepSeek Harness (DSH), prompt precision often determines the final result. Developers frequently need to turn conversational ideas into structured task descriptions. This plugin provides one-click rewriting directly next to the input box.

Plugin Overview

zzy6-a/dsh-prompt-enhance is a model inference plugin. It reads the current draft, automatically completes objectives, scope, deliverables, constraints, and acceptance criteria, and replaces the entire draft with a clearer instruction. No extra configuration is required; by default, it uses the current session model.

Core Features

This plugin provides the following core capabilities:

  • One-click enhancement and undo: Provides a ✨ Enhance button and an ↩ Undo button beside the input box. The enhanced text replaces the entire draft and can be undone at any time before manual edits.
  • Five rewriting styles: Provides style options such as detail completion (default), more concise, more detailed, add constraints, and structured.
  • Model selection strategy: The rewriting model follows the current session by default; you can also pin a specific model in settings (for example, a cheaper model) to handle rewriting.
  • Chip protection: When a draft contains @file references, /commands, or skill chips, the Enhance button is disabled to prevent whole-draft replacement from breaking reference structures.
  • Chinese-English bilingual support: Interface text follows the DSH language setting.

Install and Enable

Install the latest published package via npm:

npm i /path/to/dsh-prompt-enhance-0.1.0.tgz

After installation, find the “Prompt Enhancement” section in DSH settings and make sure “Enable prompt enhancement button” is turned on.

Typical Usage

  1. Write your idea in the input box, for example Replace the time filter on the log page with a dropdown.
  2. Click the ✨ button to the left of the model selector.
  3. The button changes to ···; when finished, the draft is replaced with the rewritten prompt.
  4. If you are not satisfied, click ↩ next to it to restore the original text.

Use Cases and Cautions

This plugin is suitable for users who need to optimize prompts quickly and reduce time spent on prompt engineering.

Keep the following limitations in mind:
* Reference limitations: Drafts containing chips can only trigger the disabled state; the plugin currently does not support rewriting that preserves references, because whole-draft replacement will break references.
* Session limitations: New sessions (Hero) do not show the button.
* Routing limitations: HTTP API routes are served only from same-origin or loopback origins; other origins return 403.
* Privacy: The plugin does not save drafts, does not write conversation logs, and processes all requests via the hand-built approach, so it does not pollute the dialogue context.

Summary

With this plugin, you can quickly turn vague instructions into structured tasks without leaving the input box, lowering the barrier to prompt engineering.