一个 AI Agent 的认知底座（Cognitive Base），让 Agent 自动发现不同学科之间的结构同构——表面不同但底层逻辑相同的模式。
不是搜文字，不是搜语义相似度，而是搜索想法之间的结构关系。
适用于任何 AI Agent 框架：Claude Code、Codex、Gemini、Cursor、Cline，或原始 system prompt。
人类知识最大的浪费不是信息不够，而是不同领域独立发现了本质相同的东西却彼此不知道。
经济学家说"均衡"，物理学家说"最小作用量"，生态学家说"稳态"，博弈论说"纳什均衡"——数学结构高度相关，但各学科用完全不同的语言理解它们。
Cross-Domain Connector 教 Agent 看穿领域词汇、看到抽象结构，然后在其他学科中找到同样的结构以及它们已有的解法。
- 带着具体问题做跨域扫描 — 你描述问题，Agent 从其他学科找结构相似的解法
- 诊断"卡住"的结构类型 — 你用了错误的思维框架？Agent 帮你识别并切换
- 模糊想法的跨域共振 — 你不知道该问什么，只是随便聊，Agent 主动从其他领域找到共振点
安装前（默认 Agent）：
"提高用户留存率的建议：1. 优化 onboarding 2. 添加互动钩子 3. 改善推送策略 4. 实施忠诚度计划 5. 收集用户反馈 6. A/B 测试不同流程。建议从数据分析入手。"
安装后（Cross-Domain Connector 生效）：
"你的留存曲线像是一个耗散问题——用户动力在系统中流失的速度快于补充速度。两个领域有成熟工具：
药理学：你的产品有一个'最低有效剂量'——低于某个使用量，用户感受不到价值。找到已转化用户的行为阈值，看流失用户是否大多没到达这个量。如果是，问题不在 onboarding，是试用期对他们的使用节奏太短。
化学：活化能太高。别优化'教用户怎么用'，而是降低'第一次获得价值所需的最低投入'。能不能预填数据？能不能 AI 自动完成初始配置？"
5 个认知转换：
| 默认模式 | 跨学科模式 |
|---|---|
| 领域锁定 — 在用户领域内回答 | 结构先行 — 提取抽象结构，跨域扫描 |
| 单一视角 — 用最相关的学科 | 多视角 — 检查 3-5 个远距离领域 |
| 词汇绑定 — 用源领域术语解释 | 翻译就绪 — 用用户已有概念重新表述 |
| 症状导向 — 在当前框架内给更多方案 | 框架诊断 — 先检查框架本身是否错配 |
| 被动 — 等用户主动问 | 主动浮现 — 高置信度时主动提一句 |
18 种结构模式索引：耗散、相变、局部最优陷阱、涌现、阻抗匹配、反馈回路、尺度崩坏、滞后、共振、梯度消失、生态位分化、熵/秩序维持、渗流、振荡、边界/膜、吸引子、瓶颈、对称性破缺。
8 个反模式门控：表面类比、装饰性引用、未翻译的导入、强制普适化、游客知识、缺失操作化、框架上瘾、重新包装。
# 克隆仓库
git clone https://github.com/d-wwei/cross-domain-connector.git
cd cross-domain-connector
# 复制 skill 文件（目录名用中文以支持 /跨学科思考 命令）
mkdir -p ~/.claude/skills/跨学科思考
cp SKILL.md anti-patterns.md examples.md ~/.claude/skills/跨学科思考/
# 注入核心规则（始终生效）
cp cognitive-protocol.md ~/.claude/cross-domain-connector.md
# 添加到 CLAUDE.md（已有则跳过）
grep -q "cross-domain-connector" ~/.claude/CLAUDE.md 2>/dev/null || echo '@~/.claude/cross-domain-connector.md' >> ~/.claude/CLAUDE.md安装后：
- 核心规则在每次对话中自动生效，无需手动调用
- 输入
/跨学科思考可调用完整框架进行深度跨域分析
- Codex：将
cognitive-protocol.md内容写入AGENTS.md头部 - Gemini：将
cognitive-protocol.md作为system_instruction传入 - 通用：注入到 system prompt 中。详见
install/generic.md
两者互补，作用于不同层面：
- Tacit Knowledge = HOW — 控制回答质量（判断优先、信息层级、内居视角）
- Cross-Domain Connector = WHERE — 扩展解题搜索空间（结构提取、跨域扫描、翻译）
同时安装即可，无冲突，自动叠加。
cross-domain-connector/
├── README.md ← 本文件
├── cognitive-protocol.md ← 核心规则（~30 行，始终在线）
├── SKILL.md ← 完整框架（~200 行，含结构索引和翻译协议）
├── anti-patterns.md ← 8 个反模式，含检测信号和修复示例
├── examples.md ← 3 个前后对比（产品/工程/写作）
└── install/
├── claude-code.md
├── codex.md
├── gemini.md
└── generic.md
A cognitive base (cognitive protocol) for AI agents that enables structural isomorphism detection across disciplines — seeing the abstract skeleton underneath any problem and finding where that same skeleton has already been solved in other fields.
Not searching text. Not searching semantic similarity. Searching structural relationships between ideas.
Works with any AI agent framework: Claude Code, Codex, Gemini, Cursor, Cline, or raw system prompts.
The biggest waste in human knowledge isn't lack of information — it's that different fields independently discover the same structures and never find out about each other.
Economists say "equilibrium." Physicists say "least action." Ecologists say "steady state." Game theorists say "Nash equilibrium." The mathematical structures are deeply related, but each field uses completely different language and intuitions.
Cross-Domain Connector teaches the agent to see past domain vocabulary, extract abstract structure, and find where other fields have already solved the same problem.
- Cross-domain scanning with a specific problem — Describe your problem; the agent finds structurally similar solutions from other fields
- Diagnosing why you're stuck — Wrong mental framework? The agent identifies the mismatch and suggests the right one
- Cross-domain resonance from vague ideas — You don't know what to ask; just describe what you're working on, and the agent finds connections from other fields
Before (default agent):
"To improve user retention: 1. Optimize onboarding 2. Add engagement hooks 3. Improve notifications 4. Loyalty program 5. Collect feedback 6. A/B test flows. Start with data analysis."
After (with Cross-Domain Connector):
"Your retention curve looks like a dissipation problem — user motivation leaks from the system faster than it's replenished. Two fields have mature tools:
Pharmacology: Your product has a 'minimum effective dose' — below a certain usage threshold, users perceive no value. Find converted users' behavior threshold and check if churning users never reached it. If so, the problem isn't onboarding — your trial period is shorter than your product's time-to-effect.
Chemistry: Activation energy is too high. Stop optimizing 'how to teach users' — instead, lower 'the minimum investment needed to get value for the first time.' Can you pre-fill data? Auto-complete initial setup with AI?"
5 Cognitive Shifts:
| Default Mode | Cross-Domain Mode |
|---|---|
| Domain-locked — answer within user's field | Structure-first — extract abstract structure, scan across fields |
| Single-lens — use most relevant discipline | Multi-lens — check 3-5 distant fields for the same structure |
| Vocabulary-bound — explain in source domain jargon | Translation-ready — re-express using user's existing concepts |
| Symptom-focused — more solutions in current framework | Framework-diagnostic — check if the framework itself is wrong |
| Passive — wait to be asked | Proactive — surface high-confidence parallels unprompted |
18 Structure Patterns Index: Dissipation, phase transition, local optimum trap, emergence, impedance mismatch, feedback loop, scaling breakdown, hysteresis, resonance, gradient loss, niche differentiation, entropy/order, percolation, oscillation, boundary/membrane, attractor, bottleneck, symmetry breaking.
8 Anti-Pattern Gate: Surface analogy, decoration citation, untranslated import, forced universalism, tourist knowledge, missing operationalization, framework addiction, repackaging.
git clone https://github.com/d-wwei/cross-domain-connector.git
cd cross-domain-connector
mkdir -p ~/.claude/skills/cross-domain-connector
cp SKILL.md anti-patterns.md examples.md ~/.claude/skills/cross-domain-connector/
cp cognitive-protocol.md ~/.claude/cross-domain-connector.md
grep -q "cross-domain-connector" ~/.claude/CLAUDE.md 2>/dev/null || echo '@~/.claude/cross-domain-connector.md' >> ~/.claude/CLAUDE.md- Codex: Prepend
cognitive-protocol.mdtoAGENTS.md - Gemini: Pass
cognitive-protocol.mdassystem_instruction - Generic: Inject into system prompt. See
install/generic.md
Complementary, different axis:
- Tacit Knowledge = HOW — response quality (judgment-first, hierarchy, indwelling)
- Cross-Domain Connector = WHERE — solution search space (structure extraction, scanning, translation)
Install both. No conflict. They stack automatically.
cross-domain-connector/
├── README.md ← This file
├── cognitive-protocol.md ← Core rules (~30 lines, always-on)
├── SKILL.md ← Full framework (~200 lines, with structure index & translation protocol)
├── anti-patterns.md ← 8 anti-patterns with detection & fixes
├── examples.md ← 3 before/after comparisons (product/engineering/writing)
└── install/
├── claude-code.md
├── codex.md
├── gemini.md
└── generic.md
Inspired by structural isomorphism — the mathematical insight that systems with different surfaces can share identical abstract structure. Douglas Hofstadter explored this in Gödel, Escher, Bach; category theory formalizes it. This cognitive base operationalizes that idea for AI agents — turning "interesting observation about knowledge" into "always-on reasoning capability."
Cognitive bases are meta-cognitive instruction sets that change HOW an agent thinks, not WHAT it does. Each one targets a different cognitive axis. Mix and match.
| Cognitive Base | What it changes |
|---|---|
| First Principles | Reason from verified foundations, not inherited conventions |
| Results-Driven | Require evidence for completion, not just activity |
| Tacit Knowledge | Think like an experienced practitioner |
| Attention Allocation | Find and concentrate on the ONE binding constraint |
| Bayesian Reasoning | Calibrated probability thinking, not binary judgments |
| Constraint as Catalyst | Turn constraints into innovation catalysts |
| Conviction Override | Override rational caution when obstacles are convention, not physics |
| Dialectical Thinking | Synthesize through contradictions (矛盾论) |
| Double-Loop Learning | Question the assumptions that produce errors |
| Frame Auditing | Detect and transcend invisible analytical frames |
| Interactive Cognition | Model others' cognition and manage information flow |
| Inversion Thinking | Map failure modes first, then avoid them |
| Motivation Audit | Audit motivational drivers before analysis (正心诚意) |
| Non-Attachment | Radical cognitive freedom — use frameworks without fusing |
| Principled Action | Unify knowing and doing through practice-theory spirals (知行合一) |
| Second-Order Thinking | Trace consequences beyond first-order effects |
| Systems Thinking | Feedback-driven structural analysis, not linear cause-effect |
| Temporal Wisdom | Make time your ally — compound effects and phase awareness |
| Cognitive Base Creator | Generate new cognitive bases from any thinking framework |
MIT