Tostoevsky
TsienHsueShen
钱学森《工程控制论》全书蒸馏的 DeepSeek Harness 方法论技能插件 | A DeepSeek Harness skill plugin distilled from Qian Xuesen's Engineering Cybernetics
- Stars
- 0
- Language
- JavaScript
- Created
- Aug 14, 2026
- Updated
- Aug 14, 2026
Introduction
钱学森 · TsienHsueShen
A DeepSeek Harness skill inspired by Qian Xuesen (Tsien Hsue-Shen)
不是替 Agent 增加更多知识,而是给它一套面对复杂项目时的控制论。 Not more knowledge for the agent — a cybernetics for getting complex work under control.
🌟 致敬钱学森(Qian Xuesen, 1911–2009)
他是中国航天事业与"两弹一星"的奠基人之一,是工程控制论的奠基人。
1954 年,在争取回国屡遭阻挠、被长期监视、几近失去自由的岁月里,他在美国写下 Engineering Cybernetics(《工程控制论》英文初版),开创性地将控制理论从具体的工程技术问题中系统抽象出来,建立了工程控制论的统一分析框架。1980–1981 年,他与宋健合作完成修订版,由科学出版社分上、下两册出版(共二十一章),将大系统、自适应、容错、信息论纳入了同一个框架。
这个插件,是对《工程控制论》(修订版)全书的一次蒸馏(distillation)——剥离掉具体的工程实例,只留下其中高度凝练的、可以迁移到任何项目 / 工程 / 学术研究的方法论与哲学思想。
谨以此插件,向新中国科学先驱们致敬。
📖 这是什么
一个为 DeepSeek Harness(dsh) 打造的 Skill 插件。
安装后,你的 agent 会话里会多出一个名为 tsien-hsue-shen 的技能——模型在规划、执行或复盘任何项目时,都可以随时调用它,获得钱学森工程控制论式的方法论指导。
🎯 它会改变什么
当一个项目开始失控时,tsien-hsue-shen 不会只问"下一步做什么",而会先问:
目标是否定义清楚?状态是否可观测?关键变量是否可控制?反馈闭环是否存在?不确定性是否被显式处理?什么时候应该停止,而不是继续投入?
它不替 agent 解决领域问题,而是帮它重新定义问题、识别杠杆、建立反馈,并知道何时停止。
💬 调用示例
不需要理解 Skill API,直接对 agent 说:
用 tsien-hsue-shen 审查这个研究计划,先检查可观测性和可控制性。
这个项目已经迭代三轮仍然没有收敛。按工程控制论的方法重新定义问题。
不要继续优化方案。先判断我们是否已经触发停止规则,以及最低可交付结果是什么。
✨ 技能正文一览
技能正文共 13 个主题章节 + 1 个"第一原理"序章(编号 0–13),是全书方法论骨架的浓缩版:
| 章节 | 内容 |
|---|---|
| 0 | 第一原理:系统观、过程观、实践循环、分析与综合 |
| 1–2 | 问题的提法 与 模型观(一切模型是有条件的近似) |
| 3–4 | 状态 / 观测 / 杠杆三元结构,能控性与能观测性审查,反馈闭环 |
| 5–7 | 不变性分级、不确定性的统计化与探索、最优性原理与多目标权衡 |
| 8–9 | 大系统的分散—协同、复杂性规律;可靠性来自结构而非元件 |
| 10–11 | 信息观(信息 = 不确定性的减少)、人机关系与主体性 |
| 12–13 | 停止规则、降级交付、终点复盘;最小执行清单 |
🚀 快速安装
第一步:安装并启动 DeepSeek Harness(需要 Node.js ≥ 22.19)
npx @deepseek-ai/dsh web
第二步:把本插件装进你的 profile
dsh plugin --profile <你的profile名> add github:Tostoevsky/TsienHsueShen
纯 Markdown + 少量 JavaScript,无需构建步骤,可直接从 GitHub 安装。
第三步:验证安装
dsh --profile <你的profile名> --dump-config # 应看到 "# == dsh-tsien-hsue-shen" 层
启动会话后,模型会在技能目录中看到 tsien-hsue-shen 并按需调用。
🗑️ 卸载
dsh plugin --profile <你的profile名> remove dsh-tsien-hsue-shen
🔧 兼容性
DeepSeek Harness 目前处于 Developer Preview 阶段,官方明确说明可能存在不兼容的破坏性变更。本插件基于当前 dsh CLI 的 dsh plugin --profile ... add <package> 用法与 npx @deepseek-ai/dsh web 启动方式测试,会随 Harness 的更新持续维护。
📂 仓库结构
TsienHsueShen/
├── package.json # dsh.bundle 清单
├── cordis.patch.yml # 安装时应用的配置层
├── index.js # 注册 skill 的插件入口(纯 JS,无需构建)
└── skills/
└── tsien-hsue-shen/
└── SKILL.md # 技能正文:钱学森《工程控制论》全书蒸馏
📜 许可与说明
代码以 MIT 许可发布。
技能正文是对钱学森、宋健《工程控制论》(修订版,1980–1981,科学出版社)的学习性蒸馏与再创作,不含原书原文大段引用;版权归原作者与出版方所有。若涉及版权问题请提交 Issue 联系删除。
In Honor of Qian Xuesen(钱学森, 1911–2009)
A DeepSeek Harness skill inspired by Qian Xuesen (Tsien Hsue-Shen)
Not more knowledge for the agent — a cybernetics for getting complex work under control.
🌟 In Honor of Qian Xuesen
A founding father of China's aerospace program and a founder of engineering cybernetics.
In 1954, during the years in which he was blocked from returning to China and kept under prolonged surveillance, he wrote Engineering Cybernetics in the United States — a pioneering work that systematically abstracted control theory out of concrete engineering problems, establishing a unified analytical framework for the discipline. In 1980–1981, together with Song Jian, he completed the revised Chinese edition, published by Science Press in two volumes (twenty-one chapters total), unifying large-scale systems theory, adaptation, fault tolerance, and information theory into one framework.
Everything in this plugin is a distillation of the revised Engineering Cybernetics — stripped of domain-specific engineering examples, keeping only the condensed, transferable methodology and philosophy applicable to any project, engineering effort, or research endeavor.
Dedicated, with deepest respect, to the pioneering scientists of the new born of People's Republic of China.
📖 What Is This
A Skill plugin for DeepSeek Harness (dsh).
Once installed, a skill named tsien-hsue-shen appears in the agent's skill catalog, giving the model cybernetic-methodology guidance whenever it plans, executes, or reviews any project.
🎯 What It Changes
When a project starts to spiral, tsien-hsue-shen doesn't just ask "what's next." It asks first:
Is the goal well-defined? Is the state observable? Are the key variables controllable? Does a feedback loop exist? Is uncertainty handled explicitly? When should we stop, instead of pushing forward?
It doesn't solve your domain problem for you — it helps the agent redefine the problem, identify leverage points, build feedback, and know when to stop.
💬 Example Prompts
No need to understand the Skill API — just tell the agent:
Use tsien-hsue-shen to review this research plan — start with observability and controllability.
This project has iterated three times without converging. Reframe the problem using engineering-cybernetics methodology.
Stop optimizing the current approach. First determine whether we've hit a stopping condition, and what the minimum viable deliverable is.
✨ What's Inside
13 topic chapters + 1 "first principles" prologue (numbered 0–13) — the condensed methodological backbone of the whole book: first principles, problem framing, the state/observation/leverage triad, controllability and observability, uncertainty and optimality, large-system reliability, information theory, and stopping rules for delivery.
🚀 Install
# 1. Install & run DeepSeek Harness (requires Node.js ≥ 22.19)
npx @deepseek-ai/dsh web
# 2. Add this plugin to a profile — no build step needed
dsh plugin --profile <your-profile> add github:Tostoevsky/TsienHsueShen
# 3. Verify — expect a "# == dsh-tsien-hsue-shen" layer
dsh --profile <your-profile> --dump-config
🗑️ Uninstall
dsh plugin --profile <your-profile> remove dsh-tsien-hsue-shen
🔧 Compatibility
DeepSeek Harness is currently in Developer Preview, and the maintainers note there will be compatibility-breaking changes. This plugin is tested against the current dsh CLI's dsh plugin --profile ... add <package> usage and the npx @deepseek-ai/dsh web launch flow, and will be maintained as Harness evolves.
📜 License & Attribution
Code is released under MIT.
The skill body is a study-derived distillation of Engineering Cybernetics (revised ed., 1980–1981, Qian Xuesen & Song Jian, Science Press). It contains no extended verbatim excerpts; all rights in the original work belong to its authors and publisher. Please open an Issue for any copyright concern.