Introduction

If you need to fix the thinking, reasoning, and response language of the agent in a DeepSeek Harness (DSH) deployment, ayanJava111/dsh-thinking-language-zh provides a default Chinese single-section bundle. Below is an introduction to its positioning, installation method, and configuration considerations.

What It Is

It is a DeepSeek Harness single-section bundle maintained by ayanJava111, used to fix the thinking, reasoning, and response language of the agent. This plugin takes effect via the system-prompt section (thinking:language) and defaults to Chinese mode.

Core Features

  • Fix the agent’s thinking, reasoning, and response language.
  • Defaults to Chinese mode.
  • Takes effect via the system-prompt section (thinking:language).
  • Suitable for DeepSeek Harness Web UI, headless, and ACP deployments.
  • Does not add tools, does not access the shell, and does not modify the model.
  • Supports configuring the text and complete fields.
  • Supports overriding language or wording via a patch layer.

Installation and Activation

First, install it to the specified profile:

dsh plugin --profile <name> add dsh-thinking-language

Then start the corresponding profile:

dsh --profile <name>

You can also check if the configuration has been added without starting it first:

dsh --profile <name> --dump-config

Typical Usage

  1. After installation and startup, Chinese mode is used by default.

  2. If you want to change the language or wording, you can rewrite config.text for id: thinking-language in the profile’s cordis.patch.yml or --patch overlay.

  3. The complete field is a configurable item. If a preset persona in the deployment uses complete: true, it will suppress this section and other sections.

  4. It is a strong directive for the system prompt, not a wire-level hard guarantee, so non-Chinese fragments may occasionally still appear.

  5. The default text rendering cost is approximately 60 tokens; modifying the text will change the prompt prefix cache.

Dependencies and License

  • License: MIT.
  • peerDependencies include @deepseek-ai/cordis >=4.0.1 and @deepseek-ai/dsh-system-prompt ^0.1.0-rc.2.
  • The plugin runs with the current DSH process permissions; you should check the source code and license before installing.

Conclusion

This plugin is suitable for scenarios where you need to fix the agent’s output language in DSH Web UI, headless, or ACP deployments. It only contributes a system-prompt section and does not add tools, access the shell, or modify the model.

  • Directory Page: https://www.skillhub.cn/plugins/ayanJava111/dsh-thinking-language-zh
  • GitHub: https://github.com/ayanJava111/dsh-thinking-language-zh