dsh-thinking-language: Injecting Chinese Reasoning Instructions into DeepSeek Harness
liangzz1998/dsh-thinking-language
A DSH plugin that injects a system-prompt section so DeepSeek Harness agents think and reply in Simplified Chinese by default, with separately configurable thinking and answer languages across standard, PTC and custom presets.
dsh-thinking-language is a language-control plugin for DeepSeek Harness that tackles the annoyance of English-only reasoning traces. Instead of modifying any built-in preset, it appends a language requirement as a single system prompt section, so reasoning, analysis, plans and reflections are written in Simplified Chinese by default while replies follow the user's language. Thinking and replying can be toggled separately, extra text allows other languages, and it works in standard mode, PTC mode and sub-agents.
How to Install
dsh plugin --profile web add github:liangzz1998/dsh-thinking-language- Category
- Skills & Agents
- Platform
- DSH Plugin
- Author
- liangzz1998
- Distribution
- Plugin
dsh-thinking-language Key Features
dsh-thinking-language Repo Summary
What Does It Do?
Thinking Language is a DSH plugin for DeepSeek Harness that makes the agent think and reason in Simplified Chinese. It solves the problem of English-only reasoning traces by injecting a single system prompt section that instructs the model to reason in Chinese, without overriding any built-in preset, so it is unlikely to break when DSH is upgraded. Maintained by liangzz1998 under the MIT license, it relies only on the system prompt service provided by DSH and pulls in no external dependencies.
Core Features
- Chinese reasoning: by default the model's reasoning, analysis, plans, reflections and inner monologue are written in Simplified Chinese instead of English.
- Configurable reply language: replies follow the user's language by default, or can be pinned to Chinese, with thinking and answering controlled independently.
- Custom additions: an extra option appends arbitrary text to the system prompt, so you can ask the model to reason in another language such as Japanese.
- Broad compatibility: works in standard mode, PTC mode, custom presets, and sub-agents / Agent Teams.
- Non-invasive injection: the instruction is added as a system prompt section after the deployment persona, leaving built-in presets untouched so DSH upgrades are less likely to break it.
How to Use This Plugin?
Once enabled in DSH, new sessions reason in Chinese with no further action. To adjust behavior, edit the plugin config in your profile: thinking (zh-CN or auto), answer (auto or zh-CN), extra, and enabled, for example Chinese thinking with always-Chinese replies, or Chinese replies only without touching reasoning. Note that minimal mode uses a fixed system prompt with complete set, which suppresses extra sections, so the plugin does not take effect there; that is by design of minimal mode.
How to Troubleshoot This Plugin?
The plugin not taking effect in minimal mode is expected, because minimal mode uses a fixed complete system prompt that suppresses all extra sections; switch to standard or PTC mode instead. If English still appears in the reasoning, remember this is a language instruction rather than forced translation, so code, proper nouns and occasional English terms remaining in English is normal.
This page is an independent rewrite of the plugin's official README — for authoritative documentation and the latest changes, refer to the source: liangzz1998/dsh-thinking-language. The plugin is third-party code that runs on your machine once installed; inclusion does not imply endorsement — please review the source before installing.
DSH Plugin FAQ
