How to use Claude Code for beginners

Key Takeaways

Oh, nice. Claude Code has an update. I just got to restart to apply it. I wonder what's in it. Wait a minute. What if I built a change log tracker for Claude Code that shows the best practices and tips on how to use it for beginners? Genius. I wanted to make a very quick beginners guide to Claude Code for everyone that's starting to use this for the first time with some tips and best practices on how to use it. And I thought what better way than to do a claude code change log tracker. I did this with cursor before and I thought it would be a good idea to do it again because when claude code updates and I restart and I'm back in with the latest version, it would be good for me to know what those changes are and if there are any new features, commands, anything that I can use to make my experience better. And that's what we're going to do. So this is assuming that you have Claude code set up already. We're just going to type claude inside this empty project to get cracking. Now, it's going to ask us if we trust the files in this folder because this is the first time using it. I'm going to go with yes. And we are here. Now, you might have noticed I'm using the terminal in cursor here. But really, I don't like this. I don't want to be here. I want to use my more native like right hand sidebar in curs. So, we have this chat window over here, which I really like. I want to use that but with clawed code. So what we can do is select commandshiftp and then go to shell command install cursor. If you're using VS Code, you could run this command instead, but we're in cursor. So we're going to install this. And when we next load up, you'll see over there it flashed up with the IDE connected. We are now connected to cursor, which means we no longer need to use the terminal. So I can close this down completely. Delete that. And we can press this new shortcut command escape which will open it right where we want it on the right hand side. We're connected to the IDE. And now we can do some interesting things inside of cursor. It will know which files we're in. We can send it text from the file or line references from the file as well. And to check this, you can also do / IDE and it will tell you and you can choose which IDE you want to connect to. We're already connected to cursor. We can see that right there. So, I'm just going to quickly close this because I want to show you one more thing. If you like cursor's notification when it plays the little sound when it's complete, you can also do that in Claude Code with this command here. It's just setting a preferred notification with a terminal bell. And that should now when Claude Code has finished doing something, it will play a sound, which is great because sometimes I'm away on social media not paying attention to what Claude is doing. But if it needs my input now, it can tell me. So we can go back into Claude to continue building this project. A relatively new feature is the plan mode. So we're going to use that for this by doing shift tab tab. And you can see now we are in plan mode. This is where you want to be to just chat with Claude Code to figure out a strategy of how you're going to build either an entire app or a feature in your app and you can just chat back and forwards to really nail it down before continuing. This is very important. You need to give it a lot of context and information because you are still the pilot. Claude is just the co-pilot helping you to build things much quicker, but you need to do most of the like thinking really yourself. Now, what I'm going to do is jump over to repo prompt and go into uh a worker that I built in claw code before which tracks cursor change logs. And I'm just going to copy the source folder and a bunch of these files here. Now I can click copy over here which will copy all of these files to the clipboard ready to go. I'm going to head back into claude code. One thing, if you're copying and pasting a huge amount of text, you don't want to do it in here in the in the in the window. If we paste it here, it can cause clawed code to be very slow and sometimes crash. It doesn't work. So, what we really need to do is create a file here. So, we're going to go new file. We're going to call this, we're just going to call it context.md. And I'm going to paste that information in from repo prompt, which is everything that we built in another project, which you can check my channel for when I covered some of the information on how I did this. So what we can do now in plan mode is at that context file which is a much better way of giving it the information that you want through a file. And now we can just ask it to oh you might notice I did a multi-line here. You can either do shift line line or you can actually do option line line with no backslashes. So typically if you want to do a new line inside of claude code you would do backslash and then enter and then it can do it back slash enter and you can keep going. However, there is a better way of doing it with a slash command. It's called terminal setup and it will add the bindings to do this for you. This one has already found it because I have it set up in cursor. This will just add the ability to do option enter and we can add new lines as we want so that we can keep our messaging clear. So, back in plan mode, we can now add this context along with a prompt and telling Claude how we want this to be built. So, I'm basically giving it the the URL for the change log and some instructions. We're just going to set this up as a very simple dashboard for now and then see how it does it and then we can add Telegram and other notifications and things later. So, the first thing it wants to do is read this change log and we can go to number two here and allow it to do that so it can understand what it needs to track. It's already gone and checked the existing cursor change log that we passed in as context. And now it's going to propose the plan. So here's the plan. It's proposing to create this Cloudflare worker. It's given us the project structure. It's giving us the differences between this and the cursor example that I gave it. A list of the basic features of how we going to build this thing and also a database schema as well. Now you can either continue planning by choosing no here and giving it some more information or you can then say yes I want to proceed. In this case we're going to ask it to proceed and build the worker for us. So it's now built the the worker for us. So we can go to our terminal really quickly and just do npm rundev to see if it actually works. So now we have a dashboard here. If we click check for updates, it says no no new updates found. Latest version is 1.08 which is incorrect. So, what I'm going to do is go command, control, shift, and 4, which on the Mac allows me to take a screenshot. And I will do that. Now, to paste an image, it's not command V as you might think. It's actually controlV. And then you'll see this image here. Latest version is actually 1.0.25. So, it can analyze the image, see that it's got it wrong, and hopefully go back and double check this. If you already have an image, by the way, you can drag and drop that in. My workflow is to create a screenshot on the fly and then paste things in. That way, I haven't got files lying around everywhere in my system. So, it found one new update and then it did update it here. So, this is pretty good. One thing you might want to check as well is /model, which will allow you to choose the default model that it uses. Now, the recommend is Opus 4 for up to 50% of usage limits and then Sonnet 4. I personally would recommend just going for Opus. Okay, it will reach the usage limits much faster, but if you're on the $200 max plan, then you might as well make the most of it and then it will switch back to Sonnet after. So, definitely switch to Opus for the absolute best coding experience you can get. I made a mistake and closed Claude code halfway through a session, which is actually a teaching moment because if we go back in with command escape, it's going to open up Claude as normal. But if we just do control CC C to close it and then do Claude C, it will continue with the last session where we left off. And you can see here we'd set the model and then I got cut off. So what I've done is added a dev.valls vase file with my telegram keys and we're going to ask Claude to basically add that functionality as well. So if we go to shift tab tab, we're going to go into plan mode and now we'll ask it to do that. A very simple prompt on basically running this schedule every 1 minute to check for updates. We need to be able to test locally as well. And then the last thing that I'm going to do here just to show you how this works in clawed code is you have a few options of how you can instruct this. So level one is to say think about this and it will allocate a little bit more thinking budget before it makes any changes or before it actually comes up with the plan. The second level after think is think hard about this. Okay. And that's another level of thinking. The third level, can you guess where this is going? Think harder about this is the third tier of how much thinking time Claude will give before it actually proceeds with the plan or or coding if you are not in plan mode. There is one more as well. Can you guess where this is going? If you thought think hardest, you would be incorrect. It's actually ultra think about this and that will give it the highest level of thinking before it proceeds. So we'll say ultra think about this before you plan. Okay, it did a lot of deep thinking there for over a minute to plan out what it wants to do. And now it's going to add the telegram notification along with a scheduled handler and update everything that we need inside of the Cloudflare worker to do this. And so we're going to proceed. And that's it. It worked first time. It added a check and notify endpoint that I can just add to the URL to test this. And you can see here the change in my Telegram. I can click copy. I could paste this to social media and instantly go viral, I imagine. But this is great. I could take this a step further now and maybe I could do some sort of AI summary on this. I could maybe even get AI to write me a social media post talking about the changes. But that's pretty good. So now at this point we are good to go. We have a decent little project but we don't have over here we don't have a claw.md file. Claw.md is our rules if you like very similar to cursor rules or or memory of the project. We don't have that. What we can do is type slashinit and that will actually check our entire project and write it for us based on what we have built so far. So look at this a full claude MD file which outlines the project completely how to run it all the different endpoints how to test it key features schema of the database how to run the scheduleuler common tasks and anything else we might need. Now what if we don't like this introduction to the project here you can see that as I've done this here over on the right hand side it says it's got one line selected. So I can just say, can you rephrase this? And it will know exactly what I mean when I say, can you rephrase it? Because it's the line that I've selected. This is all part of the Claude code integration inside cursor and it's redone it. Now I could also if I wanted to highlight that and I could do command option K and I can send in the reference to that line here. But with the newest version of Claude Code and the ID editor, you don't need to do that. You can literally just highlight the line over here and it will know which line you're talking about inside Claude Code. We can also type /config and go to configuration here, which will determine what we use inside of Claude Code. Maybe if I open that a little bit, we can see it more. So, we have autocompact, which I'll talk about in a sec. Using to-do list, yes, we want to use that. notifications. We've got the terminal bell, which actually I noticed it didn't run, so maybe it's just not working properly. I I'll look into that. Now, the diff tool, we can do it in terminal. We can do it auto. Terminal will only show it inside of clawed code or auto will show it over here in cursor as well. So, let's say that we think something is missing from here. So, for example, I know that Nex.js is very difficult to be deployed on Cloudflare. So I might want to add something in here to say that we only want to use really basic vanilla HTML and JavaScript for the front end of this project. I could do that by just doing the hash and it's going to ask us where we want it. We'll do project memory. There you go. Development guidance never use Nex.js. So you can add things very quickly with a hash. Now even though we've used this a lot in this project, we haven't had any issues with context window yet. Nothing's come up over here. The first option is /clear, which will get rid of the entire conversation history and free up context. But don't do this until you have saved some information about what you've done so far. Shun shun sh and basically write a recap of what it did and it will then use that as the starting context for the next chat. You can give it an optional custom summarization instructions there, but generally you can just use /compact. Now, one thing I've noticed with this, it's not actually that good. It tends to I don't know when that when I continue the conversation after having a compact or autocompact, which is when you just run out of context window and claude does it for you anyway. It it tends to lose the context. The summary doesn't seem to work for me basically. So, I actually built my own system for this, which I'll do another video on at some point. Um, but I built custom slash commands that will store memory for me. But let's just try it anyway. We'll compact the conversation. So, we can see it's done a compact summary here. We can do ctrlr to view that summary, which is pretty comprehensive to be fair. So, it's talking about the initial request, investigation phase, all the information about this particular session. Now, one thing I have noticed here actually is that because I actually closed cursor by mistake, I have to come back into claude code. It doesn't have the context of the original conversation when we first started the project. So, that's just something to remember. One last thing here is that there is a slash command for cost. But if we run it, we'll see that I am on a max plan. So, I can't actually see what this costs because I just paid $200 a month and that is it. Let's exit out here. C. And then we'll command W to close that. back into cursor terminal with command J and I'll open a new tab here you can actually run npx cc usage very simple command install the package and it will tell you how much you've spent based on the tokens so you can see that this actually only cost us $644 if you have any tips and tricks that I haven't covered in this video let me know and if you have ideas for videos that I can do that you might be interested in. Let me know that as well.