dsh-reasoning-cn
Run the following command in DeepSeek Harness:
dsh plugin install William123666/dsh-reasoning-cn
Paste the following prompt into your AI chat to install this plugin:
Run dsh plugin install William123666/dsh-reasoning-cn in DeepSeek Harness to install this plugin, with the source at https://github.com/William123666/dsh-reasoning-cn
About this plugin
The Think dropdown in DeepSeek Harness Web UI typically streams reasoning text in English, yet no major model exposes an API parameter to pin the thinking language, and system-prompt instructions are often ignored in agent mode. dsh-reasoning-cn addresses this with two complementary layers: a zero-cost steering layer that appends a Simplified Chinese instruction to the system prompt and re-asserts it at the first step of every turn, and an optional translation layer that hooks the response stream, buffers a reasoning block when English, Japanese, or Traditional Chinese characters are detected, and calls a cheap non-thinking model to convert the block into Simplified Chinese. Translation is off by default, and every failure, timeout, or oversized block fails open with the original text passed through unchanged so no thinking content is ever lost.
Because translation happens before the agent loop persists session logs, a successful conversion propagates consistently into the session log, the Web replay, and the reasoning_content field returned by tool calls; a degraded path keeps all three in the original language. The plugin validates translated output for non-emptiness, length expansion, code fences, and Chinese-character ratio, rolling back to the original text if any check fails. A WeakSet marker prevents the translation request from being recursively processed by its own converter.
This plugin is built for Simplified Chinese speakers who use DeepSeek Harness Web UI daily and want to follow the reasoning process in their own language. If you only need a language nudge, the zero-cost steering layer alone suffices and sends no extra LLM requests. Enable the translation layer only when you want a hard display guarantee, and make sure your chosen provider's data-retention policy, access scope, and per-call cost are acceptable. The current version targets Simplified Chinese exclusively, does not translate final answers, and does not retain the English source text, keeping the data boundary simple and easy to audit.
Use Cases
- Reading the Think dropdown reasoning text in the Web UI and wanting it displayed in Simplified Chinese
- Making the model natively think in Chinese during agent-mode runs to reduce English reasoning noise
- Converting English reasoning to Simplified Chinese with consistent output in session logs and Web replay
Best For
- Simplified Chinese users who use the DeepSeek Harness Web UI daily
- Developers who want to lower the friction of reading English reasoning
- Technical teams that need a unified language across session logs and replays
Related Plugins
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.
An armor-breaking plugin for DeepSeek: rewrites instruction priority, expands refusal-blocked phrases, persists across sessions, and shows a green active indicator.