用 Opus 5.5 将六顶思考帽转为 Claude Code 子智能体
Six Thinking Hats came out 40 years ago, and plenty of companies still run meeti…
提供了可直接复用的 Claude Code 子智能体配置代码与调用逻辑，结构清晰且包含实测数据，对开发者有参考价值。
Six Thinking Hats came out 40 years ago, and plenty of companies still run meetings with it. I used Opus 5.5 to turn it into 𝟲 𝗖𝗹𝗮𝘂𝗱𝗲 𝗖𝗼𝗱𝗲 𝘀𝘂𝗯𝗮𝗴𝗲𝗻𝘁𝘀, basically my own team of advisors.
《六顶思考帽》问世已有40年，仍有大量公司用它来开会。我使用 Opus 5.5 将其转化为 𝟲 𝗖𝗹𝗮𝘂𝗱𝗲 𝗖𝗼𝗱𝗲 子智能体（subagents），本质上是我自己的顾问团队。
Edward de Bono once wrote, "The main difficulty of thinking is confusion. We try to do too much at once. Emotions, information, logic, hope, and creativity all crowd in on us. It is like juggling with too many balls." His fix was six hats in six colors. Wear the white hat and you only talk facts. Wear the black hat and you only look for risks.
爱德华·德·博诺曾写道："思考的主要困难在于混乱。我们试图同时做太多事情。情绪、信息、逻辑、希望和创造力全都涌向我们。这就像同时抛接太多的球。"他的解决方案是六种颜色的六顶帽子。戴上白帽，你只谈论事实。戴上黑帽，你只寻找风险。
⚪ white-hat (medium): the facts ↓ same facts to all four, at the same time 🔴 red-hat (low): gut call 🟡 yellow-hat (medium): the case for it ⚫ black-hat (high): risks 🟢 green-hat (high): options, tried in its own worktree ↓ all five outputs 🔵 blue-hat (xhigh): one plan
⚪ 白帽（中等）：事实 ↓ 所有四人同时获取相同的事实 🔴 红帽（低）：直觉判断 🟡 黄帽（中等）：支持的理由 ⚫ 黑帽（高）：风险 🟢 绿帽（高）：选项，在其独立的工作树中尝试 ↓ 所有五个输出 🔵 蓝帽（极高）：一个计划
To set up the hats, send this straight to Claude Code 👇
要设置这些帽子，直接将以下内容发送给 Claude Code 👇
"Set up Edward de Bono's Six Thinking Hats as six subagents in .claude/agents/. Create them, then stop. Don't use them yet.
"在 .claude/agents/ 中将爱德华·德·博诺的六顶思考帽设置为六个子智能体。创建它们，然后停止。暂时不要使用它们。
- white-hat (effort: medium, read-only): facts only. List what we know and what's missing, and mark each point verified or guessed. - red-hat (effort: low, read-only): gut reactions in a few lines. No justification needed. - yellow-hat (effort: medium, read-only): the strongest case for doing it and the value it creates. - black-hat (effort: high, read-only): risks and what could break. Point them out, don't fix anything. - green-hat (effort: high, isolation: worktree): alternatives and new options. It may try them in its own worktree. Keep scratch files inside it, and end by listing the files it changed and where the worktree is. - blue-hat (effort: xhigh, read-only): reads everyone's output and returns one plan: the decision, next steps, open questions.
- white-hat（工作量：中等，只读）：仅事实。列出我们知道什么和缺少什么，并标记每一点是已验证还是猜测。 - red-hat（工作量：低，只读）：用几行文字表达直觉反应。无需提供理由。 - yellow-hat（工作量：中等，只读）：做这件事的最有力理由及其创造的价值。 - black-hat（工作量：高，只读）：风险和可能出错的地方。指出它们，不要修复任何东西。 - green-hat（工作量：高，隔离：工作树）：替代方案和全新选项。它可能会在自己的工作树中尝试这些方案。将临时文件保留在其中，最后列出它更改的文件以及工作树的位置。 - blue-hat（工作量：极高，只读）：读取所有人的输出，并返回一个计划：决策、下一步行动、待解决问题。
Give each hat its matching color where Claude Code has one. If this project had no .claude/agents/ folder before, remind me to restart Claude Code when you're done."
为每个帽子分配其对应的颜色，如果 Claude Code 中有对应颜色的话。如果此项目之前没有 .claude/agents/ 文件夹，请在完成后提醒我重启 Claude Code。"
To use them 👇
要使用它们 👇
"Use the six hat subagents in .claude/agents/ on the decision below. Call each hat by its agent type. If a type isn't available, stop and tell me to restart Claude Code. Don't fall back to general-purpose. Run white-hat first. Then run red-hat, yellow-hat, black-hat and green-hat in parallel, each with white-hat's facts. Last, give blue-hat the full output from all five. Show me the plan before changing anything on the main branch.
"在以下决策中使用 .claude/agents/ 中的六顶帽子子智能体。通过其智能体类型调用每个帽子。如果某种类型不可用，请停止并告诉我重启 Claude Code。不要回退到通用模式。先运行 white-hat。然后并行运行 red-hat、yellow-hat、black-hat 和 green-hat，每个都使用 white-hat 的事实。最后，将所有五个帽子的完整输出交给 blue-hat。在修改主分支上的任何内容之前，向我展示该计划。
Decision: [one sentence on what you're stuck on]"
决策：[关于你卡住的一句话]"
Once they're split into subagents, four hats think at the same time. A full round on Opus 5.5 took 𝟴.𝟱 𝗺𝗶𝗻𝘂𝘁𝗲𝘀. Really efficient.
一旦拆分为子代理，四顶帽子会同时思考。在 Opus 5.5 上完成一轮需要 𝟴.𝟱 𝗺𝗶𝗻𝘂𝘁𝗲𝘀。效率非常高。
更进一步：量化金融体系
看懂新闻只是起点——沿量化金融路径，把它变成能交付的工程能力