dsh-zh-thinking
Run the following command in DeepSeek Harness:
dsh plugin install yingjian666/dsh-zh-thinking
Paste the following prompt into your AI chat to install this plugin:
In DSH Desktop 4.1+, go to Settings → Plugins → Install from other source, enter https://github.com/yingjian666/dsh-zh-thinking as the source, check the advanced plugin option, confirm full access, then install and restart DSH.
About this plugin
DeepSeek inference models drive their internal reasoning from a single system prompt. When tool results return English content — scraped pages, documents, error logs — the chain of thought tends to drift toward English, making subsequent reasoning and the final answer linguistically inconsistent. dsh-zh-thinking tackles exactly this problem.
On startup it registers an "always think in Simplified Chinese" rule into the system prompt, positioned right after the persona section. The rule covers chain-of-thought, step-by-step planning, pre- and post-tool-call reasoning, and internal self-review. The plugin has zero external dependencies, zero configuration, and is a single file of a few dozen lines; registered sections are cleanly reclaimed on uninstall or context teardown.
It is ideal for anyone who wants the reasoning trace of a DeepSeek Harness agent to stay in Chinese, whether for debugging, team collaboration, or any workflow that demands linguistic consistency in the model's thinking. No npm account is required, and it can be installed from GitHub, a local directory, or a .tgz package with zero setup.
Use Cases
- Prevent chain-of-thought drift to English when tools return English content
- Audit and debug agent internal reasoning in Chinese during team workflows
- Keep model thinking and final output language-consistent in inference pipelines
Best For
- Developers running Chinese-language inference in DeepSeek Harness
- Teams that require strict language consistency in agent reasoning traces
- Users who prefer zero-dependency, single-file, plug-and-play lightweight plugins
Related Plugins
A unified suite combining hot runtime injection, task-aware thinking-mode routing, and a graded session protocol with red-team gates to sustain model diligence across long-horizon inference.
ModLens is a vision plugin for DeepSeek Harness that gives text-only models sight by reading images pasted directly into chat, with zero-config setup and multiple vision engines.
On-demand vision for text-only DeepSeek Harness agents: built-in free keyless vision chain and 14 vision tools, routing image turns as tool calls to vision models with pixel fidelity, no Python needed, one-command install.
Give text-only models in DeepSeek Harness eyes, enabling image Q&A, long-screenshot OCR, UI restoration, and GUI visual tasks.