Back to home

GongYuanCaiJi

dsh-chinese-thinking

DeepSeek Harness 插件:让 agent 默认用中文思考与回复(移植自 superpowers-zh)

Stars
0
Language
JavaScript
Created
Aug 14, 2026
Updated
Aug 15, 2026

Introduction

dsh-chinese-thinking · 中文思考预设

🌐 简体中文 | English

让 DeepSeek Harness 的 agent 默认用中文思考、用中文回复。

Makes your DeepSeek Harness agent think and answer in Chinese by default.

MIT License dsh-plugin

功能 / Features

  • 默认中文:不指定语言时,agent 用中文思考、用中文输出;代码、命令、专有名词保留英文原文
  • 中文排版规范:中英文空格、全角标点、数字与单位、括号引号 —— 逐字移植自 superpowers-zh 的 chinese-documentation
  • 中英混排最佳实践:术语保留与翻译、首次出现标注、避免过度翻译
  • 代码注释语言:何时用中文 / 何时用英文(chinese-code-review)
  • 中文提交规范:type 保留英文、scope/description/body 用中文(chinese-commit-conventions)
  • 国内 Git 工作流:核心原则(chinese-git-workflow)

效果 / What it does

安装后,agent 的 system prompt 会多出「中文思考」一节:你在对话里用中文,它就全程用中文想、用中文答 —— 不再「聊着聊着吐英文」。这一节是可配置的预设:改 config.text 就能换成你自己的中文思考指令。

安装 / Installation

本插件尚未发布到 npm(dsh-chinese-thinking 这个名字已验证可注册),目前用本地路径安装。先克隆本仓库,再在仓库根目录执行:

npm install   # 先装依赖(@deepseek-ai/schemastery)
dsh plugin --profile <你的 profile> add .

装完重启 dsh,开一段新对话即可生效。

配置 / Configuration

默认配置(cordis.patch.yml 里的 config: {} 走 schema 默认值):

默认说明
enabledtrue是否注入「中文思考」system prompt 节
order60节的顺序(-100 身份、0 persona、100–199 工具指引)
textprompt.js提示词正文,可整体覆盖

覆盖示例(写在你自己 profile 的 cordis.patch.yml,patch 会整体替换该行的 config):

- insert:
    - id: chinese-thinking
      name: 'dsh-chinese-thinking'
      config:
        enabled: true
        order: 60
        text: |
          用中文思考,用中文回复。

移植来源 / Upstream

本插件是 superpowers-zh(MIT,9k+ 下载/月,7.6k★)的移植(port)。提示词正文逐字来自其 chinese-documentationchinese-code-reviewchinese-commit-conventionschinese-git-workflowusing-superpowers 五个 skill 正文,源文件 SHA-256 已钉在 THIRD_PARTY_NOTICES.md,可自行核验。

如果你觉得有用,请也给上游 superpowers-zh 一个 star

License

MIT。见 LICENSETHIRD_PARTY_NOTICES.md

授权/合规询问 → GitHub issues


dsh-chinese-thinking · Chinese Thinking Default

🌐 简体中文 | English

Makes your DeepSeek Harness agent think and answer in Chinese by default.

MIT License dsh-plugin

Features

  • Chinese by default: when no language is specified, the agent thinks and answers in Chinese; code, commands, and proper nouns stay in English
  • Chinese typography rules: spacing between CJK/Latin, full-width punctuation, numbers & units, brackets and quotes — verbatim from superpowers-zh's chinese-documentation
  • Bilingual writing best practices: which terms to keep in English vs. translate, first-mention annotations, avoiding over-translation
  • Code comment language: when to use Chinese vs. English (chinese-code-review)
  • Chinese commit conventions: English type, Chinese scope/description/body (chinese-commit-conventions)
  • Domestic Git workflow: core principle (chinese-git-workflow)

What it does

After installation, a "Chinese Thinking" section is added to the agent's system prompt: once you chat in Chinese, the agent thinks and answers in Chinese end to end — no more drifting into English mid-conversation. The section is a configurable preset: override config.text to ship your own Chinese-thinking instructions.

Installation

This plugin is not published to npm yet (the name dsh-chinese-thinking is verified available), so install from a local path. Clone this repo first, then run from the repo root:

npm install   # installs @deepseek-ai/schemastery first
dsh plugin --profile <your-profile> add .

Restart dsh and start a new conversation to activate.

Configuration

Defaults (the config: {} in cordis.patch.yml resolves to schema defaults):

KeyDefaultMeaning
enabledtrueWhether to inject the "Chinese Thinking" system-prompt section
order60Section order (-100 identity, 0 persona, 100–199 tool guidance)
textsee prompt.jsThe prompt body; replace wholesale to customize

Override example (in your profile's own cordis.patch.yml — a patch replaces the row's entire config):

- insert:
    - id: chinese-thinking
      name: 'dsh-chinese-thinking'
      config:
        enabled: true
        order: 60
        text: |
          Think in Chinese, answer in Chinese.

Upstream

This plugin is a port of superpowers-zh (MIT, 9k+ downloads/month, 7.6k★). The prompt body is copied verbatim from its chinese-documentation, chinese-code-review, chinese-commit-conventions, chinese-git-workflow, and using-superpowers skill bodies; source SHA-256 hashes are pinned in THIRD_PARTY_NOTICES.md for self-verification.

If you find this useful, please star the upstream superpowers-zh too

License

MIT. See LICENSE and THIRD_PARTY_NOTICES.md.

License/compliance inquiries → GitHub issues