imlishiyuan/deepseek-harness-zh-cn
40Last commit Aug 14, 2026
deepseek-harness-zh-cn DSH plugin
This plugin encourages the DeepSeek Harness model to produce reasoning and final answers in Simplified Chinese. It offers two setup methods: copying an AGENTS.md file into the DSH home directory, or installing the npm package and mounting it in the profile. The plugin works by injecting a language directive into each turn, increasing the probability of Chinese output.
How to install the deepseek-harness-zh-cn DSH plugin
dsh plugin --profile web add deepseek-harness-zh-cnCopying does not run this command. Review the repository and version before installing the deepseek-harness-zh-cn DSH plugin.
deepseek-harness-zh-cn DSH plugin data source
deepseek-harness-zh-cn DSH plugin snapshot date: Aug 16, 2026
discovered
What the deepseek-harness-zh-cn DSH plugin can do
- Prompts the model to reason and answer in Simplified Chinese by default
- Supports two installation methods: AGENTS.md copy or npm plugin mount
- Reinforces the Chinese language directive once per turn via the agent/pre-step waterfall
- Does not modify model weights or inference logic; only injects a guideline
Where the deepseek-harness-zh-cn DSH plugin fits
- Chinese-speaking users who want to read the model's reasoning chain in Chinese
- Avoiding the need to manually translate or prompt for Chinese output
- Localizing the DSH interface for teams or projects that operate in Chinese
- Ensuring consistent language behavior across multiple DSH sessions
Who the deepseek-harness-zh-cn DSH plugin is for
- Chinese-speaking DSH users who prefer Simplified Chinese output
- Developers integrating DSH into Chinese-language applications or workflows
deepseek-harness-zh-cn DSH plugin limitations
- On technical tasks, the model's reasoning_content may still be English despite the directive
- A fully Chinese prompt maximizes the probability but does not guarantee Chinese output
- The plugin only injects a language guideline; actual output language depends on the selected model's behavior
deepseek-harness-zh-cn DSH plugin: from the repository README
Quoted from the imlishiyuan/deepseek-harness-zh-cn README, the upstream source of the deepseek-harness-zh-cn DSH plugin. Copyright remains with the original authors.
**English** | [简体中文](README.zh.md)    A Chinese-first plugin that makes [DeepSeek Harness](https://github.com/deepseek-ai/deepseek-harness) reason (`reasoning`) and answer in Simplified Chinese by default. ## Overview This project offers two ways (**choose one; the first is recommended**) to turn your DSH's reasoning and output into Simplified Chinese, so Chinese-speaking users can read the trajectory and processing directly without translation overhead: - **Method 1 (recommended, simplest)**: copy `AGENTS.md.example` to `~/.dsh/AGENTS.md`. - **Method 2 (plugin)**: install this package from npm and mount it in the profile. ## Screenshot <div align="center"> <img src="images/1.png" alt="Screenshot 1" width="48%" /> <img src="images/2.png" alt="Screenshot 2" width="48%" /> </div> ## Quick Start ### Method 1: Copy AGENTS.md (recommended) Copy `AGENTS.md.example` to the DSH home directory as `AGENTS.md`: - **Linux / macOS**: `~/.dsh/AGENTS.md` - **Windows**: `C:\Users\<you>\.dsh
Read the full READMERepository license: Apache-2.0
deepseek-harness-zh-cn DSH plugin questions
How do I install deepseek-harness-zh-cn?
You have two options. The recommended method is to copy AGENTS.md.example to ~/.dsh/AGENTS.md. Alternatively, install the npm package with `dsh plugin --profile web add deepseek-harness-zh-cn` (requires pnpm) or manually run `npm install deepseek-harness-zh-cn --save` inside your DSH profile directory, then mount it in cordis.patch.yml.
Does the plugin guarantee the model will output in Chinese?
No, it does not. The plugin injects a language directive, which increases the probability of Chinese output, but the actual language depends on the model's behavior. On technical tasks, the reasoning may still be in English. Using a fully Chinese prompt helps but is not a guarantee.
How can I verify that the plugin is mounted correctly?
Run `npx @deepseek-ai/dsh web --dump-config` and look for the line containing 'dsh-zh'. If you see it, the plugin is active. Then start a new session and the model should reason and answer in Simplified Chinese.
Does this plugin modify the model or require API keys?
No, it only injects a system reminder into the request. It does not modify model weights, inference logic, or require any additional API keys. You still need your own DeepSeek API key for DSH to work.
What if the model still outputs in English after installation?
Check that the plugin is mounted correctly via the dump-config command. Also note that the directive is a guidance, not a command. If the model consistently outputs English, try using a more explicit Chinese prompt in your session. The plugin works best when combined with a Chinese user message.