Video: Adding context

1. Video: Adding context

I've got my code editor open inside that small project, and I'm going to start up the development server with a npm run dev. And when I run that, I'm going to be able to navigate to localhost 3000 inside my browser and see the application running. So here it is right here. We are going to use Claude Code to do a little bit of work on this project, but first, there's something really critical I want you to understand around using Claude Code. Specifically, I want you to walk away from this course with a strong understanding of context management. You see, inside of a typical project, there might be dozens or hundreds of files, each with a tremendous amount of information. Whenever we ask Claude Code a question or give it a task, there is some ideal amount of information that Claude needs, just enough to help understand how to answer your question or complete your task. As soon as we start adding in additional information that's not relevant, Claude Code's effectiveness will start to decrease. So it is really important for us to help guide Claude towards relevant files or documentation inside of our project. Claude Code can certainly work without any handholding, but you'll get the best results if you provide just a little bit of guidance. So for the remainder of this video, I'm going to give you a bunch of different tips on how to give Claude the best context possible. To get started, inside my editor, I've opened up my terminal, and I'm going to start Claude Code up by running the Claude command. Whenever you run Claude Code in a project for the first time, I highly recommend running the slash init command. This gets Claude to take a deep look at your entire codebase. It'll figure out the purpose of the project, the general architecture, relevant commands, critical files, and so on. After the search, it'll summarize its findings and place them into a file called Claude.md. When Claude tries to create this file, it will ask for permission. You can either hit enter to accept, or if you don't want to have to grant permission to every file write request, you can also press shift tab, which allows Claude Code to freely write files in your project. I would encourage you to open up the Claude.md file that was generated and take a look at its contents. As I mentioned, the contents of this file are included in every request we make off to Claude. This file really has two different purposes. First, it helps Claude better understand your codebase, so it can find relevant code more quickly, and second, it serves as a location where you can give Claude some general guidance. Just so you know, there are multiple Claude.md files that Claude Code will make use of. There's a project level, a local level, and a machine level. The project level is what we just generated by running the slash init command. We are generally going to commit this file to source control, like git. We're going to share this file with other engineers, and it's going to have some project specific directions that we want to hand off to Claude. Optionally, we can also create a Claude local.md file. This file is not going to be committed, and you're generally not going to share with any other engineers. Inside this file, you might put in some personal instructions that you want Claude to follow just for you. Finally, you can have a global Claude.md file on your machine. This file will contain instructions that will be applied to all projects that you run locally. Now, I keep on mentioning giving Claude special or custom instructions, so let me show you an example of that. Let's imagine that Claude is using comments way too often in the code that it writes. We can address this by updating our Claude.md file. We can either manually modify the file, or a little bit of a shortcut is inside of Claude Code we can put in a pound sign. This puts us in memory mode. This allows us to edit one of our Claude.md files intelligently. So we can put in a request, like don't write comments so often. I'll then specify that I want to add this instruction to the project Claude.md file, and Claude is then going to merge this instruction into that file intelligently. If I then open the file up and do a search, I'll see that in fact, yes, it did add in that new instruction. Now that we've created our Claude.md file, I want to give you a better understanding of how to pull in specific context into a conversation. Let's imagine that we want to better understand how the authentication system in this project works. We could just ask Claude to tell us about it, in which case it would search over our codebase and find files relevant to the authentication system. That would definitely work, but it would just take some amount of time. Alternatively, if we already know some files that are relevant for the authentication system, we could mention them using the at character. When we mention a file, it will be automatically included inside of our request off to Claude. This is an excellent technique for pointing Claude in a specific direction. You can use the same syntax to also mention files inside of Claude.md. Let me show you an example of why that is really useful. Inside of the Prisma folder of this project, there's a file called schema.prisma. This file contains a complete definition of all the different tables and types of records that exist inside the SQLite database that is used to store information inside this project. Because this information is so important and relevant to so many aspects of this project, I might decide to mention this file inside of my Claude.md file. Let me show you how I'd do that. First I'll enter a pound to enter memory mode. I'll then mention that schema file, and specifically tell Claude to reference that file anytime it needs to better understand the structure of data inside the database. Once the update is complete, I'm going to take a look at the Claude.md file and just verify that the note was added. When you mention a file like this, its contents are automatically included inside of your request. So if I ask what attributes the user has, Claude can immediately answer without reading the schema file.

2. Let's practice!

Create Your Free Account

By continuing, you accept our Terms of Use, our Privacy Policy and that your data is stored in the USA.