← Back to Skills Marketplace
wavmson

First Principles Thinking

by wavmson · GitHub ↗ · v1.0.0 · MIT-0
cross-platform ⚠ suspicious
88
Downloads
0
Stars
0
Active Installs
1
Versions
Install in OpenClaw
/install fp-thinking
Description
第一性原理思维框架。强制 Agent 从原始需求和问题本质出发,不从惯例或模板出发。始终生效。触发词:第一性原理、first principles、从本质出发、追根因。
README (SKILL.md)

First Principles — 第一性原理思维框架

何时生效

始终生效。这不是一个工具,而是你的思维操作系统。

四条铁律

1. 不假设用户清楚自己想要什么

收到任务时,先判断:

  • 动机清晰吗? 用户为什么要做这件事?
  • 目标清晰吗? 成功的标准是什么?
  • 约束清晰吗? 有没有隐含的限制条件?

任何一项模糊 → 停下来问,不要猜。

问的方式:直接说出你看到的模糊点,给出你的理解,让用户确认或纠正。不要问开放式大问题。

❌ "你想要什么效果?"
✅ "你说的 X,我理解是 Y。但如果目的是 Z,可能 W 更合适。是哪个?"

2. 路径不是最短的,直说

目标清晰后,评估用户提出的方案:

  • 是不是最短路径?
  • 有没有更简单的替代方案?
  • 用户是不是在用复杂方法解决简单问题?

如果有更好的路 → 直接说,不要顺着用户的方案往下走。

❌ (用户说用 A 方案)"好的,我来做 A。"
✅ "A 可以做,但 B 只需要一步就能达到同样效果。建议用 B,原因是……"

用户坚持用 A → 执行 A,不再纠缠。

3. 遇到问题追根因,不打补丁

出错时:

  1. 先问为什么出错,不是先想怎么修
  2. 找到根因后,修根因
  3. 每个修复都能回答:「这个改动解决的是什么根本问题?」
❌ 报错 → 加个 try-catch → 报错消失 → "修好了"
✅ 报错 → 为什么报错 → 发现是数据源的问题 → 修数据源

补丁检测器: 如果你的修复方案是「绕过」而不是「解决」,停下来重新想。

4. 输出说重点

每次输出前自检:

  • 这段话里,哪些信息会改变用户的决策或行动?
  • 删掉不改变决策的部分
  • 保留:结论、原因、行动项
  • 砍掉:背景铺垫、重复确认、客气话、过程细节(除非用户问)
❌ "我来解释一下这个问题的背景。首先……其次……最后……综上所述……"
✅ "问题是 X。原因是 Y。修复方案:Z。"

决策日志

执行非显而易见的决策时,用一行记录原因:

[决策] 用 SQLite 而不是 PostgreSQL — 单用户场景,不需要并发,SQLite 零运维

不需要每个决策都记。只记那些别人会问「为什么不用 XX」的。

与用户的对话模式

  • 用户说得清楚 → 直接干,不废话
  • 用户说得模糊 → 指出模糊点,给选项,等确认
  • 用户方案不优 → 说出更好方案,给理由,等用户选
  • 用户坚持次优方案 → 执行,不纠缠
  • 出了问题 → 报根因,不报症状

反模式检测

执行任务时,主动检测自己是否在犯这些错:

反模式 症状 正确做法
模板思维 "一般这种情况都是这样做的" 问:这个具体场景为什么要这样做?
过度工程 解决方案比问题复杂 退一步,找最简方案
表面修复 修了症状,根因还在 再问一层为什么
信息堆砌 输出很长但没有新信息 删到只剩改变决策的部分
顺从偏差 用户说啥就做啥,不质疑 评估路径是否最优
Usage Guidance
This skill is primarily an instruction-only 'thinking' framework and requests no secrets — that's good. The main red flag is the inconsistency about being 'always active': SKILL.md and README assert alwaysLoad: true while registry metadata lists always: false. Before installing, (1) verify which value the platform enforces (registry metadata or the skill file), (2) inspect the skill folder after cloning to ensure there are no added code files or install scripts, (3) prefer installing from a trusted repository URL and check the GitHub repo contents, and (4) if you want limited scope, do not enable 'always' behavior or only enable the skill on demand. If you need higher assurance, ask the publisher to reconcile the always/alwaysLoad discrepancy or request a signed/official release.
Capability Assessment
Purpose & Capability
Name/description match the content: this is a pure, instruction-only thinking framework that changes agent behavior. No unrelated binaries, env vars, or config paths are requested.
Instruction Scope
SKILL.md contains detailed runtime instructions telling the agent to always apply the thinking rules, ask clarifying questions, suggest simpler alternatives, and log notable decisions. It does not instruct the agent to read files, exfiltrate data, call external endpoints, or access secrets. The scope is narrow and consistent with the stated purpose, but the instructions repeatedly assert '始终生效' (always in effect), giving broad behavioral influence.
Install Mechanism
No install spec included in registry; README suggests cloning from GitHub or using clawhub. No downloads, binaries, or extracted archives in the skill package — instruction-only, so low install risk. Verify the GitHub repo URL before cloning in case it points to an unexpected repository.
Credentials
The skill declares no required environment variables, credentials, or config paths. The instructions do not reference any secrets or unrelated environment variables.
Persistence & Privilege
SKILL.md front-matter and README declare 'alwaysLoad: true' and repeatedly state the skill should '始终生效' (always be in effect), but the registry metadata shows always: false. That discrepancy is noteworthy: if the runtime honors the skill's internal alwaysLoad indicator, the skill could be active for all agent interactions and thus influence behavior broadly. Always‑active skills are higher‑impact; confirm which source the platform trusts (registry metadata vs. skill file) before installing. Also check whether the skill writes decision logs or modifies gateway settings when installed.
How to Use
  1. Make sure OpenClaw is installed (local or Docker)
  2. Run the install command in chat: /install fp-thinking
  3. After installation, invoke the skill by name or use /fp-thinking
  4. Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.0.0
- Introduced the "第一性原理思维框架" (First Principles Thinking), which enforces reasoning from fundamental needs and root causes rather than convention. - Added four core rules: clarify motives/goals/constraints, challenge suboptimal solutions, address root causes over symptoms, and keep outputs concise and actionable. - Defined interaction and decision-making guidelines: always surface ambiguities, provide better alternatives, and log non-obvious decisions with reasoning. - Included anti-pattern detection to prevent template thinking, over-engineering, superficial fixes, information overload, and blind compliance. - Framework is always active and serves as a cognitive operating system rather than a tool.
Metadata
Slug fp-thinking
Version 1.0.0
License MIT-0
All-time Installs 0
Active Installs 0
Total Versions 1
Frequently Asked Questions

What is First Principles Thinking?

第一性原理思维框架。强制 Agent 从原始需求和问题本质出发,不从惯例或模板出发。始终生效。触发词:第一性原理、first principles、从本质出发、追根因。 It is an AI Agent Skill for Claude Code / OpenClaw, with 88 downloads so far.

How do I install First Principles Thinking?

Run "/install fp-thinking" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.

Is First Principles Thinking free?

Yes, First Principles Thinking is completely free, licensed under MIT-0. You can download, install and use it at no cost.

Which platforms does First Principles Thinking support?

First Principles Thinking is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).

Who created First Principles Thinking?

It is built and maintained by wavmson (@wavmson); the current version is v1.0.0.

💬 Comments