Claude代码详解（及使用方法）

所以你想使用 Claude Code。

So you wanna use Claude Code.

你想充分利用它，但又不清楚具体该怎么做。

You wanna get the most of it, but you don't know exactly how.

这是一门速成课，教你如何精通 Claude Code，我们会用最简单的方式讲解。

This is a crash course, how to master Claude Code, and we explain it in the most simple way.

互联网上有成千上万，真的是成千上万的 Claude Code 教程，但没有一个像这个这么简单。

There are thousands, literally thousands of other Claude Code tutorials on the Internet, but there are none as simple as this.

我请来了拉斯·米克教授。

I brought on professor Ras Mic.

他登场后，以最简单的方式分享，让任何人都能使用 Claude Code 创建令人惊叹的初创公司和软件。

He comes on, and he shares it in the simplest way so that anyone could create jaw dropping startups and software using Claude Code.

我们将为你提供详细的步骤，从初学者的角度出发，教你如何使用终端，如何进行提示词设计。

We're gonna give you the exact steps for how you can set it up, thinking about the beginner, how to think about the terminal, how to think about prompting.

但如果你坚持看到本集的结尾，会有一个技巧与诀窍环节，我认为非常有价值，我迫不及待想看看你能创造出什么。

But if you stick around to the end of this episode, there's a tips and tricks section, which I think is super valuable, and I can't wait to see what you build.

我们请来了拉斯·米克做客播客。

We got Ras Mic on the pod.

到本集结束时，人们会学到什么？

By the end of this episode, what are people gonna learn?

希望你不会对Claude Code感到不知所措。

Hopefully you're going to not feel overwhelmed with Claude Code.

我知道终端很吓人，像个大怪物，但我将给你一份使用它的完整指南。

I know the terminal is scary and it's a big boogeyman, but I'm gonna give you the blueprint, how to use it.

我还会分享一下，这可以说是使用Claude Code或任何智能代理的终极速成课。

I'm also gonna share, consider this the ultimate crash course on how to use Claude Code or any agent effectively.

好吧，我们开始吧。

Okay, let's get into it.

所以，我认为这些节目的最佳开场方式就是共享屏幕。

So, I mean, the best way to start these episodes is with sharing our screen.

当我们思考使用AI构建应用程序时，无论是使用Claude Code、OpenCode还是Codex之类的代理，都有几个关键点你必须时刻牢记。

So when we think of building applications using AI, using some sort of agent like Claude Code or OpenCode or Codex, whatever it is, there's a couple of things that you always have to keep in mind.

你知道，这些原则其实从未真正改变过。

You know, the principles never really change.

我们必须明白的一点是，你的输入有多好，决定了你的输出有多好，对吧？

One thing that it's important for us to understand is however good your inputs are will dictate how good your output is, right?

我们现在正处在一个模型强大到离谱的阶段，如果你产出的全是垃圾，那是因为你给的本来就是垃圾，对吧？

We're getting to a point where the models are so freakishly good that if you are producing quote unquote slop, it's because you've given it slop, right?

曾经有一段时间，模型还不够好。

There was a time where the models weren't good enough.

曾经有一段时间，我们对模型生成的代码质量有着严重的疑虑和问题。

There was a time where, you know, we had serious qualms and issues with the quality of code the models gave us.

但现在，我们开始进入这样一个阶段，就连我自己，现在审阅的代码都比写的还多。

But now we're starting to get to a point where even myself, like I'm reviewing a lot more code than I write.

我从来没想过，能在2026年初的几个月里说出这样的话。

And I never thought I'd be able to say that in the early, months of 2026.

因此，我们必须深刻理解我们的输入——它们有多好、多精确、多清晰，这和输出同样重要，并直接决定我们输出的质量。

So very important for us to understand our inputs, how good they are, how precise they are, how articulate they are, are just as good as our outputs and will dictate just how good our outputs will be.

我希望人们这样理解这个问题：想象一下，你是在向一位人类工程师传达这些信息，对吧？

And the way I want people to think about this is Greg is like, imagine you were communicating this to a human, to a human engineer, right?

如果你给出的指令很简略，任何从事客户项目的人都会意识到，大多数客户只会告诉你表面的一件事，但你必须从中挖掘出他们真正想要的深层需求。

If you give them sparse instructions and if anyone is in like client work, you realize that most clients, they tell you one thing, but you have to sort of extract the deeper thoughts of what it is they want.

当我们与这些智能体、与Claude Code协作时，同样需要非常非常精确地构建我们的输入。

It's the same way when we work with these agents, when we work with Claude Code, we need to be really, really precise with how we build our inputs.

那么，我所说的输入指的是什么？

Now, what do I mean by inputs?

我的意思是我们的产品需求文档、待办事项列表或计划，对吧？

What I mean is our PRDs or our to do list or our plans, right?

比如，人们会给你不同的名称，你知道的。

Like there's, you know, people are giving you different names.

这其实并不重要。

It doesn't really matter.

本质上都是同一件事，对吧？

It's all the same thing, right?

当我们想到PRD、待办事项清单或计划时，我希望我们能这样思考：假设我正在开发这个产品，对吧？

And when we think of a PR deal, when we think of a to do list, or when we think of a plan, I want us to think in such a way as this, let's say I'm trying to build this product, right?

比如说，我不知道，Greg，你有什么产品点子吗？

Let's say, I don't know, Greg, any product ideas that Me

你有产品点子？

have product ideas?

是的。

Yeah.

实际上，问他是最好的人选。

That's actually the burst pest best person to ask.

对吧？

Right?

假设我访问idealbrowser.com，然后

Let's say I go on idealbrowser.com and

我正打算去呢。

I was just going to it.

我正要访问它。

I was just going to it.

是的。

Yeah.

从Idea Browser中挑选当天的创意。

Pick pick the idea of the day from Idea Browser.

它说这是一个用于诊断设备文本流失大量重复访问的工具。

It says it's a diagnostic tool for appliance text losing hundreds of repeat visits.

你看，我不知道这是什么意思，但假设我知道它的意思。

See, I have no idea what that means, but let's say I know what that means.

本质上，当思考这个创意并打算将其打造成一个完整产品时，你通常会这样想：如果产品X能实现Y、Z、A、B和C，那么我要如何达成它呢？我会从功能角度出发。假设格雷格刚才提到的这个应用有四个核心功能。

Essentially, when thinking of this idea and looking to build this into a full fledged product, generally the way you're going to think is, if product X does Y, Z, A, B and C, How I would reach that is I'm going to think of features, So let's say there's four core features to this application that Greg just mentioned.

如果我实现了这四个功能，我们就可以合理地认为已经拥有了这个产品，对吧？

And if I have these four features built out, we can safely assume that we have said product, right?

我们设计PRD、待办事项和计划的方式，应该是让代理或模型去构建出所有这些功能，对吧？

The way we are to design our PRDs to do lists and plans is such that we want the agent, the model to build out all these features, right?

因为所有这些功能组合在一起就是我们的产品。

Because all these features put together is our product.

你经常看到，人们会描述一个产品，而不是描述具体功能，然后对AI感到沮丧。

You see a lot of times people will describe a product, not describe features and will be frustrated with AI.

比如，AI应该能神奇地知道你在想什么。

Like AI is supposed to magically know what you're thinking about.

顺便问一下，Greg，到目前为止我讲得清楚吗？

By the way, Greg, am I making sense so far?

完全清楚，我完全同意你的观点。

A 100%, I'm with you.

是的，所以我们真的需要从功能的角度来思考，但这里有个精彩的部分：在开发功能时，模型常常面临一个问题——你开发了一个功能，或者让模型开发了一个功能，但我们不知道它是否有效。

Yeah, so we really need to think in features, but here's the cool part, when developing features, oftentimes the issue with models is like you'll develop a feature or like, let's say the model develops a feature We don't know if it works.

我们不知道它是否做对了。

We don't know if it did it the right way.

这时候，借助Ralph正在发展的这些强大功能，我们就可以引入测试了，对吧？

That's where with all the cool Ralph stuff that's happening, we can introduce tests, right?

假设模型构建了第一个功能。

So let's say, the model, the agent builds feature one.

在进入第二个功能之前，我会让模型编写一个测试。

Before moving up, moving on to feature two, what I'm going to do is I'm going to get the model to write a test.

如果这个测试通过了，我们就继续开发第二个功能。

If that test passes, then we'll work on the second feature.

如果这个测试通过了，我们就开发第三个功能，对吧？

If that test passes, we work on the third feature, right?

所以我们终于进入了一个可以用这些模型真正构建重要项目的时代。

So we're finally entering an era where you can really build something serious with these models.

所以与其只是跟你们讲规划，不如我们一起来实际做一次规划？

So instead of telling you about just planning, why don't we do actual planning together?

我要打开我的终端。

So I'm gonna pop up my terminal.

我知道大家都害怕终端，但说实话，如果你不会用终端，就去问AI，这简直是最简单的事情。

So I know everyone's afraid of the terminal, but in all honesty, if you don't know how to use a terminal, ask AI, like it's the like simplest thing.

如果不行，你甚至可以下载 Claude Code 应用，进入代码模块，指定一个你想操作的文件夹，然后使用这个应用。

And if not, you can even download the Claude Code app and go on code section, give it a specific folder you want to work on and use the app.

真的，完全没有理由不使用 Claude Code。

Like there's literally no excuse to not use Claude Code.

如果你害怕，那也太弱了，直接上手用 AI 就行，你手头有所有工具。

If you're afraid, boohoo, just jump into use AI, you have all the tools.

话虽如此，我现在就输入 Claude，然后打开 Claude Code。

That being said, I'm just going to type in Claude and we're going to have Claude Code open.

通常人们做计划时，会按 Shift+Tab，对吧？

And usually how people plan is they'll click shift tab, right?

然后进入计划模式，你可以这样说：我想为我的营销公司开发一个 TikTok UGC 生成应用。

And then you have plan mode on and you can say, let's say I want to build TikTok UGC generating app for my marketing agency.

我到处都能看到这些 UGC 应用。

I see like these UGC apps everywhere.

请帮我制定一个计划，并把这个写成产品需求文档（PRD）。

Please help me create a plan, write this in the, in PRD.

MD 文件。

MD file.

所以，大多数人就是这样设置规划的，对吧？

So this is how most people have planning setup, right?

你会让 Claude Code、Cursor 或其他任何代理帮你做规划，让你把它保存在某个文件里，然后它会说很乐意帮你规划，并向你提出一些问题等等。

You'll tell Claude Code or Cursor or whatever agent to do the plan for you, and you ask it to be in some file, and like it says, it'd be happy to help you plan this out and it'll ask you some questions, etcetera, etcetera.

但我发现有一种更好的方法可以获得更简洁的规划，这种方式能让你更深入地思考权衡、顾虑、UI/UX 决策，因为大多数时候你让 AI 对某些决策完全自由发挥，这可能会导致你最终得到一个自己并不兴奋的产品。

But I found that there's a better way to get an even more concise plan and this way it actually gets you to think a lot more about trade offs, concerns, UI, UX decisions, because most of the time you're sort of allowing the AI to have free reign over certain decisions, which I think will lead you with a finished product that you're not excited about.

而这就是调用一个特殊工具。

And that's invoking a special tool.

我本来想给你们看看那条推文，但不幸的是，现在 Twitter 正在宕机。

I was gonna show you guys the tweet, but unfortunately Twitter's down right now.

但 Claude Code 有一个名为‘询问用户问题’的专用工具。

But Claude Code has a specific tool called Ask User Question Tool.

这个工具的作用本质上是围绕你的规划细节进行提问，对吧？

And essentially what this tool does, it starts to interview about the specifics of your plan, right?

所以我要输入这个提示：阅读这个计划文件，使用‘询问用户问题’工具就任何细节对我进行深入访谈，包括技术实现、UI/UX方面的顾虑和权衡。

So I'm gonna drop this prompt where it says, read this plan file, interview me in detail using the ask user question tool about literally anything, technical implementation, UI, UX concerns and trade offs.

我把‘implementation’拼错了，别笑话我。

I spelled implementation wrong, do not judge me.

这样做的效果是，它会超越我们现有的计划，开始向我询问更多细节。

And what this is going to do is it's going to go past the plan that we have and start to ask us about my new details.

所以让我们先完成这个计划。

So let's finish off this plan first.

我就默认这是内部使用，技术栈用 React。

I'm just going to accept this is internal use, tech stack will use react.

我只需要核心功能，我会提交答案。

I just want core features, will submit answers.

然后你会看到，Claude Code 可能还会问我们几个问题，但总体上这就是计划了。

And then Claude Code, you'll see, might ask us a few more questions, but this will generally be the plan.

对。

Right.

所以这不仅仅是计划的问题。

So it's it's not just it's not just the plan.

而是正确的计划。

It's the right plan.

对吧？

Right?

就像你之前说的，往回滚动看看，那些功能和是的。

Like, to what you were saying, like, go back scroll back up here, the features and yeah.

功能和测试。

The features and tests.

我是这么想的，不知道你是否同意，如果你让Claude Code给你造一辆车，它其实并不知道车是什么。

Like, the way I think about this, and I don't know if you agree, is, like, if you ask Claude Code to build you a car, it doesn't really know what a car is.

它不理解，你需要方向盘、收音机，还有轮子。

It doesn't understand, like, you need a steering wheel and a, you know, a radio and you need wheels.

所以最难的部分是，要以简洁清晰的方式解释清楚这些是什么，而这正是这次访谈所要做的。

So the the the hard part is trying to figure out is basically explaining what those things are in a really succinct and clear way, and that's what this interview is basically doing.

它在逐一解释每一个功能，然后我们将对这些功能进行测试。

It's it's explaining each of them, and then we're gonna test each of those features.

没错。

Exactly.

这么说吧，举个简单的例子。

Like, of it this way, like a simple example.

假设你让AI代理为你构建某个特定功能，对吧？

Let's say you ask the AI agent to build you a specific feature, right?

它会如何呈现这个特定功能呢？

How is it going to present that specific feature?

你希望它出现在仪表板上吗？

Did you want it in a dashboard?

你希望它是一个弹窗吗？

Did you want it to be a modal?

它必须是一个独立的页面吗？

Did it have to be a separate page?

当你没有指定这些细节时，它会为你做出假设。

Like when you don't specify these minute details, it will make the assumption for you.

而且有了Ralph循环之类的东西，你可能会构建出一个完整的应用程序，但它并不完全符合你的期望，对吧？

And with Ralph loops and all these type of things, like you might have a whole application built out and it's not exactly to the liking or the expectations you had, right?

所以让我继续。

So let me continue.

我会在这里做一些选择，以便我们能继续下去，然后点击提交。

I'll just make some selections here just so we can move on and then hit submit.

然后我会暂时暂停这个规划，接着粘贴这段内容。

And then I'm gonna pause this planning here and then I'm going to paste this.

我会说：读取这个计划文件，我会给这个计划文件打上标签，它叫PRD。

I'm gonna say read this plan file and I'm gonna tag the plan file it's called PRD.

MD。

MD.

我们这里就有这个文件。

We have that right here.

我会说，向我提问关于这个问题的细节，或者我甚至不需要标记它，因为上下文里已经有了，但我只是想让你看看这有多烦人——它会不断问我关于这个计划或这个应用想法的问题。

And I'm gonna say interview me the details about this question or I don't even need to tag it because it has it in its context but I just want to show you how annoyingly annoying this is going to get meaning it's going to keep asking me questions about said plan or said app idea.

注意看，它写着‘第一轮核心工作流和技术基础’，对吧？

So notice how it says round one core workflow and technical foundation, right?

它可能会问你的一些问题，是你可能根本不知道的，因为你并不懂技术。

And some of the questions it might even ask you are things that you might not know about because you're not technical.

那当我遇到不懂的东西时，我该怎么办，格雷格？

So what do I do when I don't know something, Greg?

我会复制这段内容，然后去我选择的聊天机器人那里，不管是Claude、Chad、GPT，随便哪个，我都去问它问题。

I'm gonna copy this and I'm gonna go to the chatbot of my choice, whether it's Claude, Chad, GBT, whatever, I'm and gonna ask it questions.

所以如果你还记得，之前它问过我一些关于这个应用的通用问题。

So if you remember earlier, it asked me generic questions about the app.

现在它问的是：‘从头到尾生成用户生成内容视频的理想工作流程是什么？’

Now it's saying, what's your ideal workflow for generating UGC video from start to finish?

注意看，现在的问题变得更加具体了。

Like notice how the questions are even more specific now.

所以它说的是线性、一步步、基于模板、批量处理、迭代式、对话式。

So it says linear step by step, template based, batch processing, iterative, conversational.

假设我选择了这个，它会问：应用程序该如何处理代理API的成本和使用量？

So let's say I select that and it says, how should the app handle agent API costs and usage?

现在它开始谈成本了，对吧？

So now it's talking about costs, right?

同样，当你只有基本计划时，大多数情况下这些内容并不包含在计划内，对吧？

Again, most of the times when you just have a basic plan, this is not included in the plan, right?

假设我们想设定一个硬性预算。

Let's say we wanna have a hard budget.

你希望使用什么样的数据库和托管方案？

What database and hosting approach do you wanna use?

大多数观看这个视频的人可能根本不知道。

Most of you probably watching this have no idea.

所以我可以把这个问题复制下来，去聊天机器人那里问，什么样的决定是最好的。

So I can copy this over, go to chat GBT and ask what's the best decision.

这是我的现状。

This is my current situation.

然后你继续下去，不断提交答案。

And then you keep going, you keep going and you submit answers.

所以当你使用这个‘询问用户问题’工具时，问题会变得更加细致。

So when you use this Ask User Question Tool, the questions become more granular.

它问我关于核心工作流程和技术基础的问题。

So it asked me about core workflow workflow and technical foundation.

现在它要问我关于用户界面、用户体验和脚本生成的问题。

Now it's going to ask me about UI, UX and script generation.

如果你注意到它最初提出的计划，即Claude Code的默认计划，那相当基础。

If you notice the first plan that it came up with, the default plan for Claude Code, it was pretty basic.

现在它在问我：好吧，我想要用什么AI来进行脚本生成？

Now it's asking me, okay, what AI do I wanna use for the script generation?

我会用Claude。

I'll use Claude.

你想要什么样的UI风格和美学？

What UI style aesthetic are you going for?

简约干净、仪表盘为主、创意工具感、聊天优先。

Minimal clean, dashboard heavy, creative tool feel, chat first.

对吧？

Right?

所以，希望格雷格，我能说清楚，当我使用这个‘询问用户问题’工具时，被问到的问题会变得多得多。

So hopefully, Greg, I'm making sense with, like, how much more questions I'm being asked when I'm invoking this Ask User Question Tool.

是的。

Yeah.

这完全说得通。

It makes complete sense.

而且最终你会使用更少的令牌。

You're also you're also gonna use less tokens in the end.

对吧？

Right?

因为你是对的？

Because you're right?

要

Gonna

是的。

Yeah.

关键是，你的计划越完善，输入越精准，提供给模型的初始文档质量越高，最终的结果就越好。

Because the thing is the better your plan, the better your input, the better the initial set of documents that you give the model, the the better the outcome.

如果结果更好，就不用来回反复了，对吧？

And if the better the outcome, there's no back and forth, right?

大多数人会陷入一个循环，只做一个基本计划，然后按你说的去做，但你并没有说清楚。

Most people will have a Ralph loop running, it'll be a basic plan and it'll do what you told it to do, but you weren't specific.

所以现在你又得回头，可能再运行一次循环，或者反复做各种修改。

So now you're going back and then maybe you're running another loop or you're going back and doing all these changes.

但如果你一开始就做对了，花时间在规划阶段，我百分百相信你会节省大量成本，这也能帮你理清很多思路。

But if you get it done right, if you invest the time in the planning stage, I 100% believe you'll save a lot more money and this will help you clear up a lot of ideas.

比如，我们刚才提到的这个TikTok用户生成内容农场，我们希望它怎么搭建？

Like, for example, this idea that we just had, this TikTok UGC farm, how do we want it set up?

我们希望它是基于搜索的扁平结构吗？

Do we want it to be flat with search?

我们希望它是客户活动素材的形式吗？

Do we want it to be client campaign assets?

有很多这类细微的细节，你根本没考虑到。

There's a lot of like these minute details that you're not thinking about.

因为你没考虑到，所以你让Claude Code替你做了这些假设，对吧？

And because you're not thinking about it, you're allowing Claude Code to make those assumptions for you, right?

结果呢，等它消耗了大量的token之后，你又得回头去修改，对吧？

Which at the end, after it's burned through a ton of tokens, now you're going back to change, right?

如果我们从一开始就做好充分的规划，就能省下大量麻烦，希望人们能认识到这个‘询问用户问题’工具的价值。

We can save so much headache if we do the proper planning from the beginning and hopefully people see value in this ask user question tool.

确保你在提示词中明确说明这一点，希望Greg，这样解释你能明白。

Make sure you specify it in your prompt and hopefully Greg that that made sense.

确实如此。

It does.

所以，对于这个Claude速成课程，我认为第一步是学会好好规划。

So I would say step number one for this Claude crash course is I would get good at planning.

我要非常、非常擅长规划。

I would get really, really good at planning.

我要学会很好地生成这些内容，对吧？

I would get good at generating these, right?

你看，它一直在问我问题。

Like, look, it keeps on asking me questions.

如果你注意到我们用Claude生成的第一个计划，它只有两组问题，就已经准备好可以开发了。

If you notice the very first plan that we generated with Claude, it was two sets of questions and it was ready to build.

但这次，它在问我：我有没有基础头像、自定义头像、多场景视频？

But with this, it's asking me, do I have basic avatars, custom avatars, multi scene videos?

我该怎么处理存储？

How do I wanna handle storage?

我想立即下载视频吗？

Do I wanna download the videos instantly?

云存储、外部存储，软件工程中有太多东西要考虑了。

Cloud storage, external storage, like there's so much to software engineering.

在我们上一个视频中，我在Twitter上分享了这个观点，我不知道是你还是别人，但开发个人软件很容易，而开发别人会使用的软件却非常非常困难。

And I think in our last video, shared this on Twitter, I don't know if it was you or someone else, like software, building personal software is easy, but building software others are going to use is very, very difficult.

如果你没有胆量或责任心留出一点额外时间来做规划，那我敢保证，你生成的任何东西都将是AI垃圾。

And if you don't have the audacity or the decency to set up a little time, a little extra time to plan, then I guarantee whatever you generate is going to be AI slop.

你可能会怪罪模型，但真正的问题在于你自己。

And you might blame the model, but really the problem is you.

所以，投资于你的规划，花时间使用规划，不要使用Cursor或Claude Code提供的通用规划模式。

So invest in your plans, spend time using planning, don't use the generic plan mode that cursor or Claude Code has.

我会使用Claude Code，然后指定‘询问用户问题’工具。

I would use Claude Code, and then I would specify the Ask User Question Tool.

它会不断用问题打扰你，就像它一直在问一样，对吧？

It's going to continue to annoy you with questions, like it keeps asking, right?

因为直到它确切知道你想要什么，它才开始构建。

Because until it knows exactly what it is you want, it won't start building.

所以我认为这是使用Claude Code进行构建的第一步。

So I would say that's step number one to building with Claude Code.

第二步，大家都在谈论Ralph，这很令人兴奋，但我不会用它。

Step number two, and everyone's talking about Ralph and it's exciting, but I wouldn't use it.

我不会用Ralph。

I wouldn't use Ralph.

我不用Ralph的原因是，如果你刚开始，格雷格，你怎么能想象自己完全不会开车，却直接买了一辆特斯拉，指望它的自动驾驶功能？理论上很酷，但或许先学会开车、掌握转向、过弯，甚至学会在别人插队时骂几句，再使用全自动化版本会更好。

And the reason I wouldn't use Ralph, if I was just starting out Greg, is because how are you going to, like imagine this, like imagine not knowing how to drive, but then buying a Tesla for like the self driving stuff, like cool in theory, but maybe it's a great idea to know how to drive, how to steer, how to hit the corners, how to maybe yell at someone when they cut you off before you get the full automated version.

我想说的是，当你熟练制定计划，并与AI协作开发每个功能、测试每个功能时，你会逐渐培养出对产品构建的直觉——就连有人把Vibe称为QA测试，你也能通过亲自一步步实践获得这种感觉。

I say this to say, because when you get good at developing plans and then working with the AI to build each feature and testing each feature, you start to develop the sense on product building, like, even I heard someone call Vibe QA testing, you get this sense by going one on one yourself.

这就是为什么过去几个月一直在和Claude Code较劲的人，现在能非常熟练地使用它了，因为他们花时间在没有这些疯狂自动化循环的情况下亲自构建。

This is why a lot of people who were fighting with Claude Code all these months are really, really good at using it now because they spent the time building without using these crazy automation loops.

所以，如果你是第一次使用Claude Code，或者刚接触它，那么第一步就是制定一个好计划。

So if you're using Claude Code for the first time or you're just getting into it, good plan, number one.

第二点，不要用Ralph，多加练习。

And number two, get your reps in by not using Ralph.

所以一个一个地开发功能。

So develop the features one by one.

既然你已经有了计划，就可以直接告诉Claude Code：嘿，好了，我们来开发第一个功能吧。

Now that you have your plan, you can literally tell Claude Code, hey, okay, let's build the first feature.

你尽管去做好了。

You know, go ahead and do it.

等功能完成后，你可以测试一下，问它：我该怎么测试这个功能？

And then once the feature is done, you can test it out, ask it, how can I test this?

我该怎么运行这个应用？

How can I run this app?

我不会一开始就跳去用Ralph。

I wouldn't jump into using Ralph right away.

先不用Ralph，自己动手开发。

Build without Ralph.

但假设你现在已经完成了这些练习，并且对Claude Code感到得心应手了。

But let's say you've built these reps now and you're you're comfortable with Claude Code.

现在你听说了各种技能，比如MCP、提示词。

Now you hear about all these things skills, MCP, prompt.

代理。

Md, agent.

代理。

Md.

还有别的吗？

What else is there?

某种东西。

Something.

你听说了各种规范、插件，还有Ralph，所有这些东西。

Md, you hear all these conventions, plugins, you have Ralph, all these things.

那么，我要如何才能完美地使用Claude Code或任何代理来构建东西呢？

So what do I need to perfectly build something using Claude Code or Any Agent?

我跟你说实话，这些东西其实都差不多。

I'll be honest with you, most of these things are all the same.

PromptMD 和 AgentMD 都只是 Markdown 文件。

PromptMD and AgentMD are just markdown files.

插件就是多了一点额外功能的技能。

Plugins are skills with a little bit extra.

要成功使用这些代理构建东西，首先你需要一个好计划，对吧？

What you need to build successfully using these agents is first of all, you need a good plan, right?

那些文档，也就是产品需求文档（PRD）。

Which are documents, which is the PRD.

我们刚刚生成的 MD 文件。

MD we just generated.

然后你需要记录下进展。

And then you need to document the progress that's being made.

任何熟悉 Ralph 的人都知道我在说什么，对于不熟悉的人，Ralph 循环的妙处在于以下几点。

For anyone who's familiar with Ralph, you know what I'm talking about, for those who aren't, what's cool about a Ralph loop is as follows.

Ralph循环本质上就是，你有一份需要完成的任务列表，也就是所谓的PRD。

A Ralph loop is basically, you have a list of things that need to get done, whatchamacallit, the PRD.

MD文件或计划，你把它交给AI模型，模型先处理第一个任务，完成后会将进展记录到另一个文件中，然后继续下一个任务，直到完成整个列表为止。

MD or the plan, you give it to the AI model, the model works on the first task, it finishes, it then documents it in another file and then it goes again and it stops until it's completed the whole list.

这本身并没有什么特别的，但如今它变得极其强大，是因为模型的能力变得越来越强。

Now this isn't anything special, but the reason why it's now super powerful is because the models are getting so, so good.

但这里有个问题：如果你的计划很糟糕，PRD写得很差，那这一切都无济于事。

But here is the issue, If you have a terrible plan, if you have a terrible PRD, this doesn't matter.

你只是在把钱白白送给Anthropic。

You're just donating money to anthropic.

如果你真想这么做，那我祝你好运。

And I wish you the best of luck if that's what you want to do.

但如果你想确保你的token不被浪费，你就得投入精力去打造一个优秀的PRD。

But if you want to make sure that your tokens are not wasted, you're going to invest in a good PRD.

无论是MD文件，还是优秀的计划文件。

MD file or a good plan file.

格雷格，到目前为止我讲得清楚吗？

Greg, am I making sense so far?

完全清楚。

100%.

好的。

Okay.

你把重点讲得很到位。

You're driving the point home.

是的。

Yes.

接下来我简单谈谈拉尔夫。

So I'll talk a little bit about Ralph now.

所以对于拉尔夫，

So with Mr.

我们该怎么使用它？

Ralph Wiggum, how do we use this?

现在这个方法衍生出了很多不同版本，大家都在加入自己的风格。

Now there's a lot of different iterations like people are coming with their own style.

我马上会给大家展示我配置拉尔夫系统的方式。

I'm gonna share with you my Ralph setup in a second.

格雷格，还有一点我得说清楚：Claude Code有一个插件，叫做拉尔夫·威金插件，我不建议用这个。

Greg, one thing I will say is Claude Code has a plugin, a Ralph Wiggin plugin, I wouldn't use that.

我不推荐用它的原因是，连发明整套拉尔夫系统的人都反对用这个插件。

And the reason I wouldn't use that is even the person who invented the whole Ralph system is against it.

这并不是拉尔夫最适合的用途，但我只是想分享一下它的运作原理。

It's not the best use of Ralph, but I just want to share this concept of how Ralph works.

它会从头到尾执行我们的计划，一步步搭建出所有功能，不完成就不会停止。

It's essentially going to go through our plan and it's going to build out each feature step by step, and it's not going to stop until it's done.

如果你的计划可行，那这个工具就非常好用。

This is cool when your plan rocks.

但如果你的计划一塌糊涂，那它就会变得很糟糕，完全起不到作用。

If your plan sucks, then it's terrible, it doesn't matter.

现在来说说如何设置Ralph Wiggum，我有自己的设置，我不希望任何人以为我出于任何原因在推销自己的设置，但我之所以建立自己的设置，是因为我的Ralph循环能完成几件事。

Now, in terms of how to set up Ralph Wiggum, I have my own setup and I don't want anyone to think I'm shilling my own setup for any reason, but the reason why I built my own setup is there's a couple things my Ralph loop does.

第一件事是确保有一个计划和PRD。

The first thing is it makes sure that there's a plan, PRD.

MD文件，以及一个进度文件。

MD file, and there's a progress.

TXT文件，但除此之外，它为每个构建的功能都会编写测试并进行代码检查。

Txt file, but it also, every feature it builds, it then writes a test and it then lints.

基本上，这确保了每个构建的功能确实能正常工作，对吧？

And basically what this does is it makes sure that every feature that's built actually works, right?

因为如果第一个功能没做好，继续做第二个功能是没有意义的。

Because there's no point on working on feature two if feature one doesn't work.

如果第一个功能没做好，测试失败了，你猜AI模型会怎么做？

If feature one doesn't work, if the test fails, guess what the AI model is going to do?

它会回去继续处理第一个功能。

It's going to go back to working on feature one.

一旦测试通过，我们就开始处理功能二。

And once the test passes, we work on feature two.

然后当功能二的测试通过后，我们就处理功能三，对吧？

And then once feature two test passes, we work on features three, right?

所有这些都很棒，但我还是要回到同一个观点：如果你的计划很糟糕，那么Ralph循环就无关紧要了。

All this is awesome, but I'm gonna go back to the same point, if your plan sucks, then the Ralph loop won't matter.

要设置这个循环，你可以在这里找到GitHub链接，如何设置？说实话，我都不打算解释了，格雷格，人们可以直接复制链接，发给Claude，然后说：我想运行这个Ralph循环。

Now, in order to set up this loop, you can find the Git up here, how to set it up, honestly, I'm not even going to explain it, Greg, people can literally copy the link, pass it to Claude and then be like, I want to run this Ralph loop.

它会明确告诉你该怎么做。

And it will tell you exactly what to do.

这就是如今模型的强大之处。

That's how good the models have become.

但我来给你演示一下它实际运行的效果。

But I'll show you an example of this running.

所以我有一个简单的PRD。

So I have a simple PRD.

MD 文件。

MD file.

没什么特别的。

It's nothing crazy.

只是为了让你明白这个观点。

It's just to show you the point.

但这里有几个任务。

But basically there are a couple of tasks here.

我想构建一个带有若干基础端点的简单服务器。

I want to build a basic server that has some basic endpoints.

我只是想让你看看我的 Ralph 循环是如何工作的。

And I just want to show you how my Ralph loop works.

所以当我运行这个 Ralph 循环时，再次说明，如果你不知道怎么运行，就把 GitHub 链接粘贴到 Claude Code 的代理中并询问它，它会告诉你怎么做。

So when I run this Ralph loop, and again, if you don't know how to run this, the you paste the GitHub URL in Claude Code in your agent and ask it, and it will tell you how to do it.

我有几种不同的配置。

I have a few different configurations.

如果我想用开源代码，我可以使用开源代码；如果我想用Codex，我也可以用Codex，但我现在就用Claude Code，直接运行这个脚本。

I can use open code if I want, I can use Codex if I want, but I'm just gonna use Claude Code and I'm just gonna run the script.

基本上，它会开始逐项执行每个任务，正如你所见，它会更新产品需求文档，并持续工作下去。

And basically what it's going to start doing is it's going to start running through each task, as you can see, and it's going to update the PRD and it's just gonna continue to work.

现在我可以走了，对吧？

Now I can go and leave, right?

我可以去忙我的事，Greg也待着，这个循环会继续运行。

I can go about my day, Greg hang and this loop will continue to work.

我会看到，无论五分钟后、三分钟后、十分钟后，不管多久，它都会完成所有任务，构建出一个可用的产品，这很酷，但如果我回头去看原始文档，而计划本身不好，那也没用。

And I'm going to see that at some point, whether it's five minutes, three minutes, ten minutes, however long this is, this is going to finish all the tasks, I'm going to have a working product built and all this is cool, but it doesn't matter if I'm gonna go back to the original document, if the plan isn't good.

技能当然很棒。

Now, skills are great.

MCP也很棒。

MCP's are great.

所有这些不同的Markdown文件都很棒。

All these different markdown files are great.

如果你的计划很好，那你就会为自己带来巨大的好处。

You would do yourself a serious service if your, if your plan is good.

所以，成功使用Claude Code的关键在于你有一个极其出色的计划。

So the key to successfully building with Claude Code is you have an absolutely great plan.

如果你使用‘询问用户问题’工具，你会在计划上花费大量时间，甚至开始让人感到厌烦。

And if you use the Ask User Question Tool, you will spend so much time on the plan where it starts to get annoying.

这并不有趣。

It doesn't get fun.

但那些专注于这一点的人最终会获得更好的输出结果。

But those of us who focus on this will end up having better outputs.

我们继续吧。

Let's continue.

如果你注意这里，我的Ralph循环仍在继续，并且已经完成了第一个任务。

If you notice here, my Ralph loop is continuing to go and it took care of the first task.

我已经能看到一些生成的文件了。

I can see some files already generated.

如果我查看进度的话。

If I go to the progress.

文本文件里你可以看到，格雷格，它已经开始取得一些进展了。

Txt file, you can see, Greg, it started to make some progress.

它正在记录这些内容，而且这种情况会继续下去。

It's documenting that, and this is just going to continue to work.

它会持续运行下去。

This is just going to continue to run.

所以人们有不同的迭代方式。

So people have different iterations.

我知道AMP代码团队有自己的迭代方式，不同的人也有自己的迭代方式。

I know the amp code people have their own iteration and different people have their own iteration.

这其实并不重要，对吧？

It doesn't really matter, right?

有人的Ralph可能更好。

Someone's Ralph is, could be better.

有人的情况可能会更糟。

Someone's can be worse.

有人可能会做所有这些事，这没问题，但别陷在细节里。

Someone's gonna be all of that is cool, but don't get stuck in the weeds.

关键在于你如何在一场精彩的演示中清晰地表达，创造出完美的输入，因为如果你创造了完美的输入，我们就达到了一个模型会给出完美输出的阶段。

The main sauce is how you can articulate perfectly in a beautiful presentation, create the perfect input because if you create the perfect input, we have reached a point where the models will give you perfect outputs.

所以这是我给大家的主要建议和速成课。

So that's my main, tip crash course for people.

使用‘询问用户问题’工具，先不要用Ralph；如果你打算用Ralph，就要明白：如果你的计划很糟糕，你只是在把钱捐给Anthropic，而我觉得Anthropic已经足够有钱了，不需要你的捐款。

Use the Ask User Question Tool, build without using Ralph, and if you are going to use Ralph, understand if your plan sucks, you're just donating money to Anthropic, and I think Anthropic has enough money that they don't need your money being donated to them.

阿门。

Amen.

阿门。

Amen.

还有没有什么其他人们需要知道的小贴士或技巧？

Is there anything else people need to know, like little tips and tricks?

我注意到，你没有用 Mac 终端，而是用了 Ghosty。

I noticed, you know, you're not using the Mac terminal, you're using Ghosty.

是的。

Yes.

是的。

Yes.

说实话，这全都是个人偏好。

So honestly, it's all preference.

对吧？

Right?

所以，你用哪个终端之类的，全都是个人偏好。

So, like, the terminal you use and all this stuff is all preference.

我来谈谈我的看法。

Here's what I would say.

咱们来列个技巧清单吧，技巧和窍门。

Like, let's have a tips and tricks list, tips and tricks.

首先我想说的是，现在的拼写真是让人惊叹。

So first I would say is, my goodness, spelling today.

首先我想说的是，使用‘提问’工具，那个具体叫什么来着？

First I would say is use the ask, what was the specific tool?

我只想确保自己不会忘记，那个‘询问用户问题’工具，我之前忽略了，不知道为什么没人谈论它。

I just wanna make sure I don't forget, ask user questions tool, slept on, I don't know why no one's not talking about it.

真的，我看到Anthropic团队发的推文，100%我会在规划时用这个工具。

It literally, I saw the tweet from the Anthropic team, 100% I would use that when planning.

第二，别过度沉迷于MCP技能之类的。

Number two, don't over obsess, obsess on, MCP skills, etcetera, etcetera.

我不是说不要去接触这些。

I'm not saying don't get into these.

我不是说不要去阅读它们。

I'm not saying don't read about them.

我不是说不要用它们，但我几乎可以肯定，这些并不是你的产品不成功的原因，对吧？

I'm not saying don't use them, but I can almost guarantee you, these things are not the reason why your product isn't working, right?

大多数时候，是因为你的计划太烂了，对吧？

Most of the time it's your plan sucks, right?

这是第二点。

That's number two.

第三点，我会在没有使用Ralph的情况下先做出一些东西之后再用Ralph。

Number three, I would use Ralph after I've built something without.

原因在于，听好了，如果你是个有钱有势的大佬，资金充裕、不在乎花销，想把钱捐给Anthropic，那尽管去用Ralph吧。

And the reason being is, again, listen, if you are a baller shot caller and you have all the money to blow and you don't care and you wanna donate money to Anthropic, go ahead and use Ralph.

但如果我们面对面坐着，而你连一样东西都没做过、没部署过，连我和Greg都点不开你建的任何网址，那你根本没资格用Ralph。

But if we were to sit here eye to eye and you haven't built anything, deployed anything, there isn't a URL that I myself or Greg can click on that you've built, you have no business using Ralph.

你真的完全没资格用Ralph。

You literally have no business using Ralph.

我建议你先好好掌握提示词技巧，用某种计划（不管是什么，AG1、Claude Code、Open Code都行）先做出点东西来。

I would first get good at prompting and building something using a plan, whether it's whatever, AG1, Claude Code, Open Code, whatever.

等你把东西部署到Vercel上，有了一个可访问的网址，我们都能用上了，那时候你再用Ralph也不迟。

Once you have something deployed to Vercel or like there's a URL and we can use it, then you can use Ralph.

第四点，这有点细节，但上下文比以往任何时候都更重要。

Number four, this is a little in the weeds, but context is more important than ever.

很多时候，Claude Code 或甚至 Cursor 都会告诉你已使用的上下文百分比。

And a lot of times Claude Code or even Cursor will tell you what percent of context has been used.

我一般不会超过50%，比如 Anthropic 的 Opus 4.5 模型上下文限制是20万标记。

I generally wouldn't go over 50%, meaning like the Anthropic model Opus 4.5 has a 200,000 token context limit.

在我看来，一旦你使用的标记超过10万，也就是在同一会话中，模型的表现就开始下降。

The moment in my opinion, you've got over a 100,000 tokens, meaning you're using the same session, it starts to sort of deteriorate.

这就是为什么有些人，比如 Greg，会说：‘我一开始做得不错，但后来越来越差。’

That's when you have people, Greg, who say, oh, like I started off good, but it started going bad.

这是因为你塞进了太多的上下文。

That's because you've filled it with so much context.

最好的理解方式是把自己代入进去，对吧？

And the best way to think about this is like yourself, right?

比如说，我们去上一节英语课，或者任何别的课，教授一直不停地灌输信息、信息、信息。

Like, let's say we went to some English class and or some, you know, whatever class and the professor just kept dumping information, information, information.