围绕一个理念构建的不断增长的Claude技能集合：停止提示AI，开始用它构建系统。

这个仓库里的每项技能都来自我实际使用或教授的工作流程。每个流程都设计成安装一次，永久使用。随着我制作新内容，我会在这里添加新技能。

新来的？先从三件事开始

如果我今天要用Claude重新开始，/start是我会先构建的三个东西：一个重写自己规则的Claude，一个能分裂成一群代理的提示，以及一个在你的机器上运行并给你发短信的代理。每个代理都链接到一个可用的起始者。

想自己打造吗？docs/make-your-skill.md 会一步步指导你。

这是给谁的

任何使用 Claude 并想从中获得更多内容的人。不需要编程经验。

里面是什么

思考

当你需要先理解问题再尝试解决问题时。

• 信念更新器——维护三个相互竞争的假说，并随着新证据的出现进行更新
• 盲点查找器——完全找到你思维中缺失的内容，不是对你所说内容的批判，而是你未说的地图

Q Claude Skills

A growing collection of Claude skills built around one idea: stop prompting AI and start building systems with it.

Every skill in this repo comes from a real workflow I use or teach. Each one is designed to be installed once and used forever. As I make new content I add new skills here.

New here? Start with three things

If I were starting over with Claude today, /start is the three things I’d build first: a Claude that rewrites its own rules, one prompt that splits into a fleet of agents, and an agent that runs on your machine and texts you. Each links to a working starter.

Want to build your own? docs/make-your-own-skill.md walks you through it.

Want it to work in Codex, Cursor and Copilot too? docs/write-a-skill.md covers the open format and where each agent looks.

Who this is for

Anyone who uses Claude and wants to get more out of it. No coding experience needed.

What’s inside

Thinking

For when you need to understand a problem before you try to solve it.

• Belief Updater — maintains three competing hypotheses and updates them as new evidence comes in
• Blind Spot Finder — finds what is missing from your thinking entirely, not a critique of what you said but a map of what you did not say
• Complexity Mapper — maps all the moving parts of a problem and finds the single lever that creates the most change
• The Five Chains — five four-message sequences for getting Claude past its first answer, every message written out ready to paste
• Five 2026 techniques: Fail, Taste, True, Flip and Bio, five three-prompt chains for finding the failure already underway, your unwritten rules, the assumption a plan rests on, the weaker side of an argument, and who your last month says you are

The thinking skill works in Claude Code, Codex, and Cursor.

Decision Making

For when you need to choose a path and you are too close to see it clearly.

• Parallel Conversations — runs the same idea through three completely different perspectives simultaneously
• Assumption Audit — surfaces every assumption your thinking is based on and stress tests each one
• Time Machine — shows you what your future self would think about this decision in 6 months, 1 year, and 5 years

Research

For when you need to validate an idea or understand a space before committing to it.

• Adversarial Room — builds the strongest possible case against your idea so you find the fatal flaw before it costs you
• Premortem — assumes the project already failed and works backwards to explain why
• Steelman — builds the strongest possible version of any argument so you engage with what is actually true

Building

For when you are ready to execute and need a system not just a plan.

• Weekly Planning — builds a complete weekly plan from your actual context, not generic time blocks