Introduction

DeepSeek Harness (DSH) uses a plugin-based architecture that allows users to extend functionality based on their needs. AI Input Enhancer is part of this ecosystem and is designed to address pain points in prompt writing and optimization. It adds a “Prompt Optimization” action to the chat input box, supports one-click revert and multiple optimization modes, and helps users quickly improve input quality.

Core Features

This plugin provides the following specific capabilities:

  1. Three Optimization Modes

    • Faithful Optimization: Improves clarity of expression without inventing requirements, facts, architecture, or scope.
    • Development-Focused Expression: Organizes the development information provided by the user into expressions more suitable for implementation, without choosing a tech stack or design solution on the user’s behalf.
    • Specification Expansion: Allows adding necessary default items, but they must be explicitly marked under ## Default assumptions for user review.

  2. Interaction and Revert

    • The optimization process can be canceled, and a loading state is displayed.
    • The optimization result is directly filled into the chat box.
    • One-click revert of the original prompt is supported.

  3. Context and Memory

    • Short-Term In-Session Memory: Optional; retains up to the latest three applied optimization results.
    • Current Session Context: Optional; reads the user/message and assistant/message text from the current session’s Surface and supports a configurable character budget.
    • Workspace Markdown Document Context: Optional; reads only Markdown files in the root directory of the current session’s workspace (up to 3 files), also subject to the character budget.

  4. Configuration and Privacy

    • The primary model, fallback model, timeout, and maximum output tokens can be configured.
    • API keys are not stored; only the current draft is sent to the DSH model router.

Installation and Activation

Install the plugin into the Web profile:

dsh plugin --profile web add zzy-dsh-prompt-optimizer

After installation, restart the existing DSH Web process and refresh the current URL.

Usage Steps

  1. Click the “Prompt Optimization” action in the chat input box.
  2. Select an optimization mode in the pop-up interface.
  3. The optimization result is automatically filled into the input box. If you are not satisfied with the result, click the “Revert” button to restore the original prompt.

Context and Privacy Notes

  • Data Flow: All additional context sources are disabled by default. Once enabled, the current draft is sent to the DSH model router.
  • Session Context: Once enabled, the plugin reads only the user/message and assistant/message text from the current session’s Surface.
  • Workspace Context: Once enabled, the plugin reads only Markdown files in the root directory of the current session’s workspace.
  • Fallback and Storage: Both context sources are subject to the character budget; if the service is unavailable or reading fails, the plugin automatically falls back. The plugin does not persist full drafts, session history, or document content.

Conclusion

AI Input Enhancer provides the ability to optimize prompts directly within the DSH environment. With clear modes and controllable context settings, it is suitable for scenarios where prompts need to be repeatedly refined during conversations.