A meta-skill for AI agents: give it any thinking framework, and it generates a complete cognitive base skill — ready to install on Claude Code, Codex, Gemini, Cursor, or any LLM-based agent.
The AI agent skill ecosystem is booming — design skills, coding skills, writing skills. But they all solve one problem: what to do. Nobody is solving the other problem: how to think.
A cognitive base skill is a different category of skill. It doesn't teach the agent to do a specific task — it changes the agent's reasoning process, improving the quality of all tasks.
| Domain Skill | Cognitive Base Skill | |
|---|---|---|
| Analogy | A recipe | Knife technique |
| Layer | Operational (what to do) | Meta-cognitive (how to think) |
| Scope | One task type | All tasks |
| Effect | Directly determines output | Changes reasoning, indirectly improves output |
This Creator automates the generation of cognitive base skills from any thinking framework.
Give the Creator a framework name:
"Generate a cognitive base from first principles thinking"
It produces a complete, installable package:
first-principles/
├── README.md ← Project documentation
├── cognitive-protocol.md ← Core rules (~30 lines, always-on)
├── SKILL.md ← Full framework (~120 lines, reference)
├── anti-patterns.md ← Framework-specific failure modes
├── examples.md ← 3 before/after comparisons
└── install/
├── claude-code.md ← Claude Code installation
├── codex.md ← Codex installation
├── gemini.md ← Gemini installation
└── generic.md ← Any other agent
- Framework Analysis — Extract core principles, map cognitive shifts, identify characteristic failure modes
- File Generation — Produce all 6 files following strict quality specs
- Self-Review — Verify: no theory names in core rules, all instructions are actionable, anti-patterns are unique
The Creator enforces these design principles:
- Meta-cognitive, not operational — "Before answering, list three constraints" (meta) vs "Use 8px grid" (operational)
- No theory in core rules —
cognitive-protocol.mdreads as pure instructions. If you can identify the theory from the rules alone, it's too theoretical. - Framework-specific anti-patterns — Each framework has unique failure modes, not generic quality issues
- Stackable — Every cognitive base defines its relationship to Tacit Knowledge (the foundational base layer) and to domain skills
This skill works on any AI agent platform.
mkdir -p ~/.claude/skills/cognitive-base-creator
cp SKILL.md ~/.claude/skills/cognitive-base-creator/Add the contents of SKILL.md to your AGENTS.md or system prompt.
Add the contents of SKILL.md to your system instructions (API system_instruction field or AI Studio).
Paste the contents of SKILL.md into whatever file your agent reads as instructions on startup.
Using Tacit Knowledge as a reference — the first cognitive base skill built with this approach:
Core rules (cognitive-protocol.md, 30 lines):
## Lead with judgment
- First sentence must be a directional judgment, not a preamble.
- When uncertain: "My judgment leans toward X because…" — not a pros/cons list.
## Information hierarchy
- One core judgment first (focal point), then 2-3 supporting signals.
- Proactively surface what experienced practitioners know but rarely say.
Before (default agent):
"Microservices and monoliths each have pros and cons. The choice depends on your team size and business complexity."
After (with cognitive base):
"Stay on the monolith. With 8 people, the coordination cost of microservices will eat all the gains. Fix your module boundaries first — that's the actual problem."
Tacit Knowledge is the first and foundational cognitive base skill. It provides general reasoning quality improvements (commitment, hierarchy, indwelling, holism).
This Creator generates additional cognitive bases that stack on top of Tacit Knowledge. Each generated base:
- Defines whether it's complementary, reinforcing, or specialized
- Specifies stacking order and conflict resolution
- Has anti-patterns that don't overlap with Tacit Knowledge's 8 patterns
Think of it this way:
- Tacit Knowledge = the base operating system
- Generated cognitive bases = kernel modules that extend it
- Domain skills = applications that run on top
📖 What is a Cognitive Base? (中文) — A detailed article explaining the concept, with 3 real before/after scenarios and design principles.
cognitive-base-creator/
├── README.md ← You are here
├── SKILL.md ← The Creator skill (the only file you need)
└── docs/
├── article-zh.md ← 认知基座详解文章（中文）
└── images/ ← Article illustrations
MIT
一个 AI Agent 的元技能：给它任意一个思维框架，它会自动生成一套完整的认知底座 Skill —— 可直接安装在 Claude Code、Codex、Gemini、Cursor 或任何基于 LLM 的 Agent 上。
AI Agent 的 Skill 生态正在快速膨胀 —— 设计 Skill、编程 Skill、写作 Skill。但它们全部在解决同一个问题：做什么。没有人在解决另一个问题：怎么想。
认知底座 Skill 是一个不同类别的 Skill。它不教 Agent 做某个任务 —— 它改变 Agent 的推理过程，提升所有任务的质量。
| 领域 Skill | 认知底座 Skill | |
|---|---|---|
| 类比 | 一道菜的菜谱 | 拿刀的方式 |
| 作用层面 | 操作层（做什么） | 元认知层（怎么想） |
| 绑定范围 | 特定任务类型 | 所有任务 |
| 影响路径 | 直接决定输出 | 改变推理过程，间接提升输出 |
这个 Creator 能从任何思维框架自动生成认知底座 Skill。
给 Creator 一个框架名称：
"用第一性原理帮我创建一个认知底座 Skill"
它会生成完整的、可安装的文件包：
first-principles/
├── README.md ← 项目说明
├── cognitive-protocol.md ← 核心规则（~30 行，始终加载）
├── SKILL.md ← 完整框架（~120 行，按需参考）
├── anti-patterns.md ← 该框架特有的思维反模式
├── examples.md ← 3 个 Before/After 对比
└── install/
├── claude-code.md ← Claude Code 安装指南
├── codex.md ← Codex 安装指南
├── gemini.md ← Gemini 安装指南
└── generic.md ← 其他 Agent 通用安装
- 框架分析 —— 提取核心原则，映射认知转换，识别特征性失败模式
- 文件生成 —— 按严格质量规格生成全部 6 个文件
- 自检 —— 验证：核心规则不含理论名字、所有指令可执行、反模式不重复
Creator 强制执行以下设计原则：
- 元认知而非操作层 —— "回答前先列出三个约束条件"（元认知） vs "使用 8px 网格"（操作层）
- 核心规则不提理论 ——
cognitive-protocol.md是纯操作指令。如果能从规则猜出理论来源，说明太理论化了 - 框架特有的反模式 —— 每个框架有独特的失败模式，不是通用质量问题
- 可叠加 —— 每个认知底座都定义与 Tacit Knowledge（基础底座）及领域 Skill 的关系
这个 Skill 适用于任何 AI Agent 平台。
mkdir -p ~/.claude/skills/cognitive-base-creator
cp SKILL.md ~/.claude/skills/cognitive-base-creator/将 SKILL.md 的内容添加到 AGENTS.md 或 system prompt 中。
将 SKILL.md 的内容添加到 system instructions（API 的 system_instruction 字段或 AI Studio）。
将 SKILL.md 的内容粘贴到你的 Agent 启动时读取的指令文件中。
📖 所有 Agent 都缺了 Skill 无法解决的一层：认知基座 — 详细解释认知基座概念，包含 3 个真实 Before/After 场景对比和设计原则。
Tacit Knowledge 是第一个也是基础的认知底座 Skill，提供通用推理质量提升（承诺、层级、内居、整体）。
这个 Creator 生成额外的认知底座，叠加在 Tacit Knowledge 之上。每个生成的底座：
- 定义与 Tacit Knowledge 的关系（互补、增强、特化）
- 指定叠加顺序和冲突处理
- 反模式不与 Tacit Knowledge 的 8 个反模式重复
架构关系：
- Tacit Knowledge = 基础操作系统
- 生成的认知底座 = 扩展内核模块
- 领域 Skill = 运行在上层的应用
MIT
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 |
| Cross-Domain Connector | Detect structural isomorphisms across disciplines |
| 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 |