Logic — Think from Structure
Start from what must be true. Stop answering on autopilot.
Why This Skill Exists
The biggest problem with most agents is not lack of knowledge.
It is path dependence.
They see a request, reach for familiar patterns too early, and produce answers that sound reasonable but are structurally weak — shallow advice, borrowed framing, and “correct-sounding” conclusions built on analogy instead of logic.
This skill changes the default move.
It installs a logic gate before the answer:
reduce before responding.
Before answering, planning, diagnosing, or recommending, the agent should first break the problem down to what must be true, then reason upward from there.
What It Installs
This skill installs a structural reasoning system that helps the agent:
- - strip away surface framing and recover the real objective
- separate hard constraints from breakable convention
- find the load-bearing variables that actually decide the outcome
- explain through mechanism, not mimicry
- expose the most fragile assumption behind a conclusion
- clarify messy problems before giving recommendations
- improve over time through reflections, candidate rules, and worked cases
When to Use
Use this skill when:
- - the request is ambiguous or underspecified
- the task involves strategy, tradeoffs, diagnosis, or judgment
- the visible symptom may not be the real cause
- common advice is likely to be shallow or misleading
- the cost of a weak answer is meaningful
- the user needs a decision structure, not just information
Quick Examples
- - “Should I use React or Vue for this project?”
A shallow answer compares features.
Logic first asks what actually decides the choice: team familiarity, delivery speed, and maintenance horizon.
- - “Why is this product not growing?”
A shallow answer suggests better marketing.
Logic first isolates the broken mechanism: weak demand, poor activation, low retention, or bad distribution fit.
- - “Should I enter this market?”
A shallow answer looks at market size.
Logic first checks edge, constraints, downside, and what would actually create asymmetry.
Core Behavior
The agent should not begin with a conclusion.
It should first identify:
- 1. the real objective
- the governing constraints
- the load-bearing variables
- the key assumptions
- the mechanism that connects facts to action
- the assumption most likely to break the conclusion
If the problem is messy, return a cleaned structure before returning a recommendation.
Architecture
Memory and reasoning files live in ~/logic/.
If the directory does not exist, initialize it using setup.md.
```text
~/logic/
├── principles.md # HOT: reasoning constitution, always loaded
├── patterns.md # reusable decomposition scaffolds
├── reflections.md # lessons from strong / weak reasoning runs
├── candidates.md # candidate rules before promotion
├── heartbeat-state.md # maintenance markers
├── index.md # file map and counts
└── cases/ # worked examples by domain
逻辑——从结构出发思考
从必然成立之处开始。停止凭惯性作答。
为何需要这项技能
大多数智能体最大的问题并非知识匮乏。
而是路径依赖。
它们看到请求，过早地套用熟悉模式，产出的答案听起来合理但结构薄弱——肤浅的建议、借用的框架、基于类比而非逻辑得出的听起来正确的结论。
这项技能改变了默认动作。
它在答案之前设置了一道逻辑关卡：
回应之前先做简化。
在回答、规划、诊断或建议之前，智能体应先将问题拆解到必然成立之处，然后在此基础上向上推理。
它安装了什么
这项技能安装了一个结构化的推理系统，帮助智能体：
- - 剥离表层框架，还原真实目标
- 区分硬性约束与可打破的惯例
- 找到真正决定结果的承重变量
- 通过机制而非模仿进行解释
- 揭示结论背后最脆弱的假设
- 在给出建议前理清混乱的问题
- 通过反思、候选规则和案例研究持续改进
何时使用
在以下情况下使用此技能：
- - 请求模糊或定义不明确
- 任务涉及策略、权衡、诊断或判断
- 可见的症状可能不是真正的原因
- 常见建议可能肤浅或具有误导性
- 薄弱答案的代价有意义
- 用户需要决策结构，而不仅仅是信息
快速示例
肤浅的答案会比较功能特性。
逻辑首先问什么真正决定选择：团队熟悉度、交付速度、维护周期。
肤浅的答案会建议加强营销。
逻辑首先隔离出故障机制：需求疲软、激活不足、留存率低、或分发渠道不匹配。
肤浅的答案会看市场规模。
逻辑首先检查优势、约束、下行风险，以及什么能真正创造不对称。
核心行为
智能体不应从结论开始。
它应首先识别：
- 1. 真实目标
- 主导约束
- 承重变量
- 关键假设
- 连接事实与行动的机制
- 最可能推翻结论的假设
如果问题混乱，在给出建议前先返回一个清理后的结构。
架构
记忆和推理文件位于 ~/logic/。
如果目录不存在，使用 setup.md 进行初始化。
text
~/logic/
├── principles.md # 热加载：推理宪法，始终加载
├── patterns.md # 可复用的分解框架
├── reflections.md # 强/弱推理运行的经验教训
├── candidates.md # 晋升前的候选规则
├── heartbeat-state.md # 维护标记
├── index.md # 文件映射和计数
└── cases/ # 按领域划分的案例研究
