Introduction

The design philosophy of DeepSeek Harness (DSH) is “everything is a plugin.” For agent developers, prompt iteration often requires repeated adjustments and validation. This plugin aims to address the need to optimize prompts directly in the DSH Web UI by adding an action entry point in the input box, providing a visual optimization workflow and the ability to roll back results.

Plugin Introduction

This plugin is named zzy-dsh-prompt-optimizer (AI Input Enhancer) and is maintained by user zhengzeyong9527-droid. It is mounted into the DSH Web process via a patch, adding a “Prompt Optimization” action to the chat input box. After optimization, the result is filled directly into the input box, and you can roll back to the original prompt at any time.

Core Features

The plugin mainly includes the following core capabilities:

  1. Three optimization modes
    - Fidelity Optimization: Improves expression clarity without inventing requirements, facts, or architecture.
    - Developer-oriented Expression: Organizes the provided development information into a more implementation-friendly format without choosing a tech stack for you.
    - Spec Expansion: Adds necessary default items, but they must be clearly marked under ## Default assumptions.

  2. Interaction and Experience
    - Supports cancellation during optimization and displays a loading state.
    - The optimization result is filled directly into the chat box, with one-click rollback to the original prompt.

  3. Context and Memory
    - In-session short-term memory: Optionally retains the most recent three applied optimization results.
    - Session context: Optionally reads the user/message and assistant/message text from the current session Surface, and supports setting a character budget.
    - Workspace documentation context: Optionally reads Markdown files in the current session workspace root directory, and supports setting a character budget.

  4. Configuration and Privacy
    - Supports configuring the primary model, fallback model, timeout, and maximum output tokens.
    - The plugin does not store API keys.

Installation and Activation

Install to the Web profile as an npm package:

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

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

Typical Usage

  1. Find the “Prompt Optimization” action entry in the chat input box.
  2. Select the optimization mode in advanced settings, and configure whether to enable session context, workspace documentation context, and the character budget.
  3. Click Optimize. The plugin will fill the result into the input box.
  4. If you are not satisfied, you can roll back to the original prompt and edit it at any time.

Applicable Scenarios and Notes

  • Applicable Scenarios: Suitable for developers who need to frequently adjust prompts, optimize instruction clarity, or organize development specifications.
  • Privacy Notes: The current draft will be sent to the DSH model routing you select. All additional context is disabled by default. The plugin does not persist complete drafts, session history, or document content. It is recommended to review the source code and license before installation.

Conclusion

By embedding optimization capability into the native DSH interface, this plugin reduces the operational cost of prompt iteration. You can view plugin details or submit feedback via the following links: