|

课后实操指南
把第二节课变成你自己的能力Post-Class Practice Guide
Turn Session 2 Into Real Skill

第二节课你看到的是：从空目录建一个知识库，再用它服务客户、出一份 PPT。 但 看懂 ≠ 会用。真正的能力，来自你把这套流程搬到一个你自己有知识的领域，亲手走一遍。 下面四步，请按顺序做，每一步都别跳过「记录心得」——这是整个方法的灵魂。In Session 2 you saw: building a knowledge base from scratch, then using it to serve clients and generate a PPT. But understanding ≠ being able to do it. Real skill comes from taking this workflow into a domain where you have your own knowledge, and doing it yourself. Follow these four steps in order. Don't skip the "write reflections" part — it's the soul of this method.

第 0 步Step 0 先搭好你的学习工作区Set Up Your Learning Workspace

打开 Claude Code，新建一个 空目录 作为你的课程学习项目（比如 我的AI课程-S2）。 空的就对了——接下来的内容由你和 AI 一起长出来。Open Claude Code and create an empty directory as your course project (e.g. MyAICourse-S2). Starting empty is correct — you and the AI will grow the content together.

建议的目录结构（边做边建，不用一次到位）：Suggested directory structure (build gradually, not all at once):

我的AI课程-S2/
├── 课程心得/
│   ├── ppt/                  # 把课程 PPT 放这里
│   ├── 跟我做-提示词.md       # 下载包里的演示提示词清单，每步带「这步在干嘛」（很有用！）
│   ├── 我的心得.md            # 随手记，想到啥写啥
│   ├── 不懂的名词.md          # 听到没懂的术语/概念
│   └── 课程总结.md            # 让 AI 帮你生成
└── 知识库项目/                # 你自己领域的知识库（作业主体，见第 3 步）
    ├── raw_material/
    ├── wiki/
    └── output/
💡 怕 .md？用 .txt 一样行。Afraid of .md? .txt works just as well. 上面那些 .md 文件，你嫌麻烦就全建成 .txt（系统自带记事本就能写）， AI 照样读得懂、照样帮你总结。.md 只是排版能好看一点，不影响用——别让格式挡住你动手。You can create all the .md files above as .txt (Notepad works fine), and AI can still read and summarize them. .md just looks a bit prettier — don't let formatting stop you from getting started.

第 1 步Step 1 记心得 + 让 AI 帮你总结Write Reflections + Let AI Summarize

先你写，再让 AI 补。Write first, then let AI fill in. 顺序很重要——AI 是来帮你深化的，不是替你回忆的。Order matters — AI is here to deepen your thinking, not recall it for you.

  1. 在 我的心得.md 里，尽量回忆课上听到的内容，记下你的体会。不用写得完善，碎片、半句话都行。In 我的心得.md, recall what you heard in class and write down your impressions. No need to be thorough — fragments and half-sentences are fine.
  2. 在 不懂的名词.md 里，记下你没太懂的术语和概念（比如 raw_material / wiki / Skill / Graph View）。In 不懂的名词.md, write down terms and concepts you didn't fully understand (e.g. raw_material / wiki / Skill / Graph View).
  3. 把 PPT 放进 ppt/ 目录；再把下载包里的 跟我做-提示词.md 也放进 课程心得/——它是课堂演示用到的 提示词清单，每步还标了「这步在干嘛」，AI 翻它能准确还原每步在做什么。Put the PPT in the ppt/ directory; also place 跟我做-提示词.md from the download package into 课程心得/ — it's the prompt list used in the class demo, each step annotated with "what this step does," so AI can accurately reconstruct each step.
  4. 然后让 AI 结合你的心得、PPT 和 跟我做-提示词.md 帮你总结。多问几次，每次问得更细，最后要求它 输出一个 课程总结.md 文件，方便你以后回看。Then have AI summarize based on your reflections, PPT, and 跟我做-提示词.md. Ask multiple times, going deeper each round, and finally ask it to output a 课程总结.md file for future reference.
📎 跟我做-提示词.md 为什么重要Why 跟我做-提示词.md matters：PPT 是「讲了什么」，跟我做-提示词.md 是「实际发了哪些提示词」。 你对某一步没看懂、或想不起演示细节时，让 AI 翻 跟我做-提示词.md 回答你，比凭空解释准得多。The PPT covers "what was taught," while 跟我做-提示词.md contains "what prompts were actually used." When you don't understand a step or can't remember demo details, having AI reference 跟我做-提示词.md is far more accurate than explaining from scratch.
Prompt 示例Prompt Examples（直接抄，再改成你自己的）： (copy directly, then adapt to your own):

第 2 步Step 2 亲手走通课堂演示（Part A）Walk Through the Class Demo Yourself (Part A)

只有自己做，才有真体会。You only truly learn by doing. 对着下载包里的 demo/跟我做-提示词.md，从一个空目录开始， 把今天的财务顾问演示自己跑一遍：建三层库 → /wiki-extract 写第一条 wiki → /wealth-client-qa 收集客户 → /ppt-master 出 PPT。约 30-40 分钟。Follow the download package's demo/跟我做-提示词.md, start from an empty directory, and run the financial advisor demo yourself: build the 3-tier structure → /wiki-extract write the first wiki → /wealth-client-qa collect clients → /ppt-master generate PPT. ~30-40 minutes.

Part A 只是找手感、验证环境，不用提交。真正要交的是 Part B（第 3 步）。Part A is just for getting the feel and verifying your setup — no submission needed. The real deliverable is Part B (Step 3).

第 3 步Step 3 换成你自己的领域做作业（Part B）Redo It in Your Own Domain (Part B)

作业不是交差，是 把这套流程搬到一个你有真实知识的领域——你能判断 AI 输出对不对的领域。 保险 / 律师 / 医疗 / 教育 / 房产都行；实在没有，就跟着做理财库。Homework isn't just a formality — it's about taking this workflow into a domain where you have real knowledge — a domain where you can judge whether the AI's output is correct. Insurance / law / healthcare / education / real estate all work; if you don't have one, just follow along with the wealth KB.

完整六步在 homework-guide.md 里The full 6 steps are in homework-guide.md，这里是骨架： — here's the skeleton:

步骤Step做什么What to Do要交吗Submit?
Step 1-2选定领域 + 建三层目录（raw_material/ wiki/ output/ + CLAUDE.md）Choose your domain + build the 3-tier structure (raw_material/ wiki/ output/ + CLAUDE.md)—
Step 3先建 /wiki-extract Skill，用它写 3 篇 wiki，每篇写完自己 Review 补行业判断Create /wiki-extract Skill, write 3 wiki entries with it, review each and add your domain expertise是Yes
Step 4用 superpowers（/brainstorming）针对一个具体客户/场景做定制方案Use superpowers (/brainstorming) to create a custom plan for a specific client/scenario—
Step 5用 /ppt-master 出一份 PPT，截图留好Use /ppt-master to generate a PPT, save screenshots是Yes
Step 6 ★（进阶可选）封装你领域的客户咨询 Skill(Advanced, optional) Package a client consultation Skill for your domain否No

关键是迭代Key is iteration——质量不够好，先改 output/ 里的方案/画像，再跑一次。知识库质量 > 提示词技巧。 最重要的一步是 Step 3d：每篇 wiki 写完，把 AI 没提到的行业经验补进去——这一步让它从「AI 的系统」变成「你的系统」。if the quality isn't good enough, first improve the plans/profiles in output/, then run it again. KB quality > prompt engineering. The most important step is Step 3d: after writing each wiki, add the domain experience that AI missed — this turns it from "an AI system" into "your system."

建三层目录的 PromptPrompt for Initializing the 3-Tier Structure（在新空目录里打开 Claude Code）： (open Claude Code in a new empty directory):
「请帮我初始化一个 [你的领域] 知识库项目，按三层结构组织：raw_material/（原始素材，留来源链接）、wiki/（结构化知识条目）、output/（客户档案，隐私区不进 Git）。创建这三个目录 + 一份 CLAUDE.md 说明结构和隐私规则，并把 output/ 加进 .gitignore。Please initialize a [your domain] knowledge base project, organized in a 3-tier structure: raw_material/ (raw materials, include source links), wiki/ (structured knowledge entries), output/ (client profiles, private — keep out of Git). Create these three directories + a CLAUDE.md explaining the structure and privacy rules, and add output/ to .gitignore.」

第 4 步Step 4 让 AI 帮你给我写一封邮件 = 提交作业Let AI Help You Write a Submission Email

最后，结合你这一路的 心得体会 和 作业产出，让 AI 帮你写一封邮件发给我。Finally, let AI help you write a submission email based on your reflections and homework output.

📌 关键说明Important note：这封邮件 就是你的作业提交——下面三样 一样都不能少。 让 AI 帮你写，不等于让 AI 帮你省。所以 prompt 里要把三项都点给 AI，让它逐项填满。This email IS your homework submission — all three items below are required. Having AI help you write doesn't mean having AI do it for you. Make sure your prompt specifies all three items so AI fills each one in full.

提交三样（邮件正文必须全部覆盖）：Three Required Items (all must appear in the email body):

  1. 3 篇 wiki 片段3 wiki excerpts——把 Step 3 写的 3 篇 wiki 内容贴进邮件（或截图），最好带上你自己补的行业判断。Paste the 3 wiki entries from Step 3 into the email (or screenshot), ideally with your own domain expertise annotations.
  2. /ppt-master 生成的 PPT 截图PPT screenshots generated by /ppt-master——Step 5 PPT 打开后的 1-2 张关键页截图。1-2 key slide screenshots from Step 5's PPT.
  3. 200 字心得200-character reflection——这次和 Session 1 的作业有什么不一样的感受？可参考：「工具」和「系统」有什么不同？哪里 AI 的答案让你想修改或补充？如果你有真实客户，会怎么用这套流程？（200 字以上，不设上限）What felt different from Session 1 homework? Consider: what's the difference between a "tool" and a "system"? Where did AI's output make you want to revise or add more? If you have real clients, how would you use this workflow? (200+ characters, no upper limit)
Prompt 示例Prompt Example（直接抄，把括号里的路径换成你自己的）： (copy directly, replace paths in brackets with your own):
「请结合我的知识库项目（知识库项目/ 目录）和课程心得（课程心得/ 目录），帮我写一封发给老师的作业提交邮件。下面三项必须完整出现在邮件正文里，一项都不能省：Please combine my knowledge base project (知识库项目/ directory) and course reflections (课程心得/ directory) to help me write a homework submission email to the instructor. All three items below must appear in full in the email body — none can be skipped:
  1. 我 Step 3 写的 3 篇 wiki：原样贴出内容，标出我自己补充的行业判断；My 3 wiki entries from Step 3: paste the full content, mark the domain expertise I added;
  2. Step 5 的 PPT：说明在哪个领域、为哪个客户/场景做的（截图我会另附）；Step 5 PPT: describe the domain and the client/scenario it was made for (I'll attach screenshots separately);
  3. 学习心得：≥200 字，写清楚『工具 vs 系统』的体会、以及哪里我修改/补充了 AI 的输出。Learning reflection: ≥200 characters, clearly describing the "tool vs system" insight and where I revised/supplemented AI's output.
末尾再补一段我还想请教的困惑。语气真诚、简洁。写完先列一个清单核对上面三项有没有遗漏。End with a paragraph about questions I still have. Keep the tone sincere and concise. After writing, list a checklist to verify all three items are covered.」

发送方式How to Submit：邮件发到 austin.aicourse@gmail.com，主题必须是 【Session 2 作业】+ 你的名字。Email to austin.aicourse@gmail.com, subject must be 【Session 2 Homework】+ your name.

我想看到的不是完美，而是 你真的动手了、真的在想。困惑和卡点尤其欢迎——那正是下一节课我能帮你的地方。I don't need perfection — I need to see that you actually did it and genuinely thought about it. Confusion and sticking points are especially welcome — those are exactly what I can help you with in the next class.

一句话总结One-sentence summary
写心得 → 让 AI 总结 → 走通演示 → 换成自己领域建库出 PPT → 反馈给我。
每一步都亲手做，每一步都留下记录。这门课的能力，是「做」出来的，不是「听」出来的。Write reflections → Let AI summarize → Walk through demo → Build your domain KB and PPT → Send it back.
Do every step yourself, document every step. The skill in this course is built by doing, not listening.

—— 看不懂任何一步，截图发学习群，老师和同学帮你。没有笨问题。If anything is unclear, screenshot it and post in the student group — the instructor and classmates will help. No question is too basic.