When developing model inference with DeepSeek Harness (DSH), developers often encounter an issue: the language of the AI’s “thinking process” (chain of thought) and its “final reply” may be inconsistent, or the AI may default to English. A common solution is to modify the system prompt to explicitly force the language, but manually managing these prompts can be cumbersome. The following DSH plugin uses system prompt injection to allow developers to independently control the AI’s reply language and thinking language.

This plugin is named @tirol/dsh-language-set and is maintained by Emck. Its core function is to use system prompt injection to separately control the language AI uses when generating replies and the language used in internal reasoning.

Core Features

  • Reply Language Control: Specify the language used in AI conversations.
  • Thinking Language Control: Specify the language used in the AI’s internal reasoning process.
  • Flexible Configuration: Supports multiple configuration options.
  • Command Support: Supports registering the language command.

Installation and Activation

Installing this plugin requires DSH’s command-line tools.

dsh plugin --profile web add @tirol/dsh-language-set

After installation, you must restart the DSH service for the plugin to take effect.

Typical Usage

After installing the plugin and restarting the service, the default language is Chinese. You can adjust the language settings by modifying the configuration file profiles/web/cordis.patch.yml (currently supports Chinese and English).

After enabling command support, you can use built-in commands to view the configuration and the injected prompts.

# 查看当前配置
> language config

# 查看注入的系统提示词
> language show

Add the following content to the configuration file to enable command support and set the default language to Chinese:

- id: dsh-language-set
  config:
    lang: "zh-CN"
    command: true

Applicable Scenarios and Notes

This plugin is suitable for scenarios that require strict control over AI output language, such as forcing the model to reason and reply in Chinese.

  • Permission Notice: The plugin runs with the current DSH process permissions.
  • Security Notice: Be sure to review the source code and license before installing.

Summary

By injecting system prompts, this plugin enables independent control over the reply language and thinking language, simplifying multilingual configuration in DSH environments.

Plugin directory: https://www.skillhub.cn/plugins/Emck/tirol-dsh-language-set
Source code: https://github.com/Emck/tirol-dsh-language-set