Vollständiges Transkript anzeigen (8.098 Wörter)
This video will cover everything you need to know about Cursor, specifically the new version, Cursor 3.0. I'm going to assume no prior knowledge and do this completely for beginners. That means I'm going to go over every core feature, show you how to set up Cursor, how to create a new project, how to write code, or use AI to generate it, and then I'm going to get into all of the advanced features like sub-agents, side chats, working in the cloud, MCP servers, all of the important concepts that you need to make this a legitimate development environment. Now, I've been using Cursor ever since it came out. They've actually been a long-term partner of mine and loaded me up with tens of thousands of dollars in Cursor credits, and I'm glad to say that they sponsored this video, so the curriculum you see here is officially approved by them. That said, you can use Cursor for free. You don't need to pay for anything, and with that said, let's get into the video and do a full Cursor 3.0 course.
So let's start by quickly covering what Cursor is and how this differs from tools like Claude Code or Codex. Now, Cursor is a full integrated development environment. It's actually a fork of Visual Studio Code, so it's going to look very similar to that if you've ever used it before, and it essentially has a ton of AI features. This means that you can use this as an AI IDE or AI integrated development environment, where you can view the code, you can edit the code, you can add tools, you can see everything that is going on, and at the same time, fully utilize all of the AI features that you would use inside of something like Claude Code or Codex. The main difference I like to point out here is that when you're using a CLI-based tool or something like Codex or Claude Code, you're typically not actually viewing the files. You're not reading through any of the code. You don't really know everything that's going on, and you're kind of just blindly trusting the AI model. With Cursor, you get a lot more control, so sure, you can still do that. You can vibe code all you want, but you also have all of the powerful development tools that real engineers would use when they're actually shipping applications. So people that are serious software engineers would lean towards using something like Cursor as opposed to a Claude Code or a Codex, where you have a lot more control in this type of tool.
Okay, so in order to get started with Cursor, the first thing that we need to do is download it. Like I mentioned, it is free, but of course, when you're using AI models, if you use them a lot, you will need to upgrade and purchase a plan. So I'm going to go over all the pricing so you understand everything. But the first step is to download it. I'm going to leave a link to it in the description. Now, Cursor has recently come out with a bunch of new features, which means you can use this from your phone. You can use this directly in the web, if you don't have even like a decent computer, but for most of us, we're going to want the desktop application. So I'll leave a link to it in the description. I've already got the Windows version, but you can download it for Mac or for Linux, on pretty much any machine you have, okay? So download Cursor.
Now once you download Cursor, it should prompt you to create or sign into an account. You can also do that directly from the Cursor web page. So if you don't already have one, sign in or create a new account here with Cursor, and then you can quickly have a look at the pricing. Now, if you've never used this before, definitely just stick with the free plan, and as you start using it, if you like it, you can upgrade to the Pro, Pro Plus, or Ultra plan, or go up to Teams or Enterprise, depending on what you're doing. Now, the pricing works really similarly to any tools you may have used before, where you have a certain amount of usage across different models. You have a lot more usage on Cursor's models themselves, which are Grok and Composer, which we'll go over in a minute, and then less on the frontier models, like Fable, for example, or GPT-5.6, but you can always pay for additional usage if you want to. Okay, so at this point, I'm going to assume that you've downloaded and installed the desktop application and that you've signed into your account or created a new one if you didn't already have one. Once you do that, you should be brought to a page that looks something, let me go here, like this. This is kind of the default view that Cursor should bring you inside of. Now, if it looks a little bit different, don't worry, because they're constantly updating the user interface, but you should see this kind of blank agent window that looks something like this.
Okay, and I'm going to go through the entire user interface, so it will make complete sense, and you'll know where to go, what buttons to press. But first, I want to quickly discuss some of the major changes that have come in Cursor 3.0, because some of you may have used previous versions of the tool or seen my other tutorials on it. Now, currently, Cursor is on version 3.0. They keep making minor changes, so when you watch this video, it might be slightly different, but there was some major changes that were made as opposed to Cursor 1.0 or 2.0. And we have kind of three waves of changes that I want to quickly go through. Now, the first wave happened about six months ago, where effectively what Cursor did is start added skills, sub-agents, lots of plugins. They have their own Cursor marketplace, and they added modes like Debug mode and Design mode, as well as side chats, which are super cool, which we'll look at in a second. Then in about April of this year, so 2026, when I'm filming this video, they switched to an agent-first interface. This completely changed the look of Cursor. So you no longer had the default IDE view, which you can still go back to, and it showed you a bunch of agents, and the idea was to be running multiple agents at the same time in parallel and using their cloud. And then recently, they added a bunch of other features, in the past two, three months here, like new models. Composer 2.5, which is really good. Automations, Bugbot, cloud sub-agents, the iOS application, and then, of course, very recently, they added the Grok 4.5 model. You can see July 16th was this post. It's literally yesterday from when I'm recording this video, and this is in partnership with SpaceX. So you may have heard Grok before on Twitter, X, whatever. Well, they actually trained the model using compute directly from SpaceX, and Grok 4.5 is now extremely good, and you can see quickly how it kind of lines up in the benchmarks at a much cheaper price per token than you would pay for comparable models like Fable, for example, or GPT-5.6 on X-High mode. Anyways, point is that they have a bunch of cool features. Let's go through them and let me show you them directly in the user interface.
Okay. So when you open up the user interface, you're going to be brought to a page that looks like this. Now, I promise I'm going to show you actually writing code, but it's really important that you understand where you're working and how to move around in here because there are a lot of features. So the view that should open by default is the Agent view. This is where you're just going to see a prompt window, similar to something like ChatGPT, right? And from here, you have a bunch of different options. So first, you have the model selector, right? You can see that you can change the model. There's a ton of models here. And if you have no idea what to use, you can flick it into this Auto mode. This is typically going to prefer using Cursor's model, so the Composer 2.5 model and Grok 4.5, but you also can switch between pretty much any other model that you want, but keep in mind if you're using models like Fable 5 or ones that don't come directly from Cursor, it's going to eat up your usage faster. So for example, if I switch to Fable 5, well, I can use this, but it's just going to go through the usage quicker than if I were to use their standard models. Okay. So this is kind of the first view, right? And you'll notice that there's a button in the top right-hand corner called IDE. It might move when you're watching this video, but if you click this, it's going to switch you into more of the development mode where you're going to be able to see what looks like a much more traditional editor. This is the IDE view, this is the Agents tab view. The Agents tab view is mostly meant for people that are not technical, where you're just going to be prompting and getting some kind of output. You can still view the code and see what's going on, but you'll be able to view multiple agents at the same time running here. You can see it's showing me all my different repositories that I had, and the idea is to use this more for multitasking when you have multiple different agents running at the same time. You also can expand all of the windows here, so on the right-hand side, you can see changes that have been made, you can see a browser, okay? On the left-hand side, if you move this, you can expand this kind of Agent view where you can search through stuff, automations, settings, right? There's a lot of stuff. I'm going to go through it one by one, but I'm just trying to give you a lay of the land. So again, important thing, Agent view, kind of the de facto view, especially if you're not technical. You're going to prefer using this, and then the IDE view here where you'll actually be able to view all of the code and use traditional IDE features, which I will show you in a minute. So for right now, before we go into the IDE view, I would prefer for us to stay in the Agent view because this is going to be simpler for most of you watching this, and we're going to start running a few prompts, setting up a project, and just getting kind of the base ready to go before we move into anything too complex.
All right, so we're in the Agent view. I'm just going to make this full screen, and we're going to start creating a fresh project. So let's say you're brand new, you're coming into Cursor, you've signed into your account from down here, right? And now you want to actually start working on a project. Well, the first thing you should do is change the folder that you're working in. So you're going to notice there's kind of a little button right here that says the folder that you're working in. Now, it's confusing because I have a folder called Google Gemini Pro, but this is the folder or the directory that you're going to work in. So what we want to do here is we want to scroll all the way down, and we want to press on new folder, okay? So if you're starting completely from scratch, you press new folder, from new folder, we're just going to make a folder to work in. So I suggest going to your desktop, making a new folder. I'm just going to call this Cursor Demo or something, and then just select this folder, okay? And create it. All right, so I have it open here, which is bugging a little bit, but you can see I've selected this folder. Now you also notice there's a bunch of other stuff popping up. You'll see cloud, PC, no repo, use existing, whatever. Don't worry too much about that. The two you're probably going to go with is use existing, which means you're going to connect to a folder that already exists, or you're just going to make a new one. All right, so this is our first project, and now what we want to start doing is prompting. Now, before we prompt, we want to understand how to do that, how to do that the most efficiently, and what a good prompt is. So if you're coming to Cursor, the idea is that you want to build something, right? You want to build some type of coding-related project. So we first need to understand what it is that we want to build, and this is really up to you. Now, for this demo, I'm just going to build a game in the browser because I find that's the most visual and easiest to demonstrate to you, but you can build pretty much anything that you want. However, when you are building, there's a lot of decisions that you need to make up front, which I'm going to help you do. So what we want to start doing right away is first of all, put this into Plan mode. Now there's a bunch of different modes, whenever you're starting out, I recommend that you put this in the Plan mode, we'll look at the other modes in a minute, and for the model, I suggest that you go with Cursor Grok 4.5. Okay, we're going to talk about model selection and modes and all of that in a minute, but for now I just want to get us started. Now, what I want to do is I actually just want to begin by kind of dumping my thought process on what it is that I want to build into Cursor. Now, in order to do that, I recommend that you use a voice dictation or a voice-to-text tool. So on my computer, you'll see that I'm using something called Wispr Flow. Let me quickly demo to you. Hey, I'm just coming up with a plan. Can you help me come up with the plan for the Super Mario Bros. game that I want to create? Whatever, I'm just doing some random stuff, and you can see that I immediately get all of my voice, translate it into text with all of the proper punctuations, spelling, all of that, and it will even format things and give me bullet points. Now, I'm using a tool called Wispr Flow. I'm just going to quickly open it for you so you can see what it looks like here. It's free, you can download on Mac, Windows, and it allows you to just directly talk into your computer. Now, Wispr Flow has been a long-term partner of the channel, they've also kindly sponsored this video, and they work natively inside of Cursor. What that means is that it can actually tag files with the Cursor tool, so that the dictation or the voice-to-text that you get is better. The reason I recommend using this, and again you don't need to pay for the Pro version like I am, is because you can speak much faster than you can type, and when you speak, you typically give much deeper, richer prompts that are going to give more context to the model so it can give you a better output. So what I like to do, right, once I've got this downloaded, is I like to start by just dumping my full thought process into the model and telling it, "Hey, this is what I want to make, this is why I want to make it, and this is who it's designed for." If you can answer those three questions at the beginning, you'll kind of be off to a good start and you'll see what I mean. So, in our case, I want to make a platformer game. So what I'm going to say is, "Hey, I'm making a YouTube video right now. I want to do a demo to showcase Cursor's capabilities, and I would like to create a platformer game. Something like Super Mario Bros. I want it to be a little unique. We don't need to go with the same color palette or the exact same design as Super Mario. I don't know exactly what I want, but I know that I want a platformer. I want to be able to move. I want to have some animations, and I want this to work in the browser so that I could share it with other people and that they could use it by going to a simple URL. I don't want anything too complex. I want to focus mostly on the visuals, the gameplay, and making a fun game in the browser." Whatever, okay? This is what I'm doing. And then lastly, kind of as I kind of forgot, I'm going to say, "I would like pretty much anybody to be able to play this, let's say between ages 10 and 60. It should be pretty user-friendly and use standard keyboard commands that most people are familiar with." So you can see I'm adding a lot of detail, I'm explaining what I want, and then the last part of the prompt that I always recommend that you put especially when you're just starting to build this out is the following: "Please ask me any clarifying questions or any design decisions that you need answered before we proceed with the build." So what I'm doing is I'm prompting the model and say, "Hey, ask me any questions, because I know I might not have given you all of the details that you need," and then you'll see when I press enter here in Cursor that it actually now has a feature where it can directly prompt you with questions in the user interface. So hopefully, in a second, it will pop those up for us, uh, and allow us to just directly press the answers. Okay, so you can see that it's asked us some questions. Sometimes it will actually give you like a pop-up and you can just press the, uh, the buttons directly, but in this case, we're just going to type them in. So it's asking us, "Hey, what visual do we want?" I kind of want to go with like this forest one, so I'm just going to paste that as option one, and then for the demo scope, um, I want to go with, hmm, maybe B. Slightly richer so we have double jump, power-ups, and a bunch of stuff. Also useful if you already know: controls, art style, um, let's go with, keep the controls arrow keys or WASD, okay? And for the art style, I'm going to go, let's go with, uh, kind of chunky pixel art. For sharing, we'll deploy it later, but don't worry about that for right now. Okay, so I'm answering all the questions. I'm going to press enter, and ideally, what's going to happen now because we're in this Plan mode is that Cursor's going to start creating a detailed plan for us. We can review the plan, and then only once we're happy with that, we can start executing it and building this out. Now, I promise I'm going to go into a lot more features, but I'm trying to keep this really beginner friendly. Again, the setup is make a new folder, select the model, put it in Plan mode, use Wispr Flow or your voice to give a really detailed starting prompt, ask the model to ask you clarifying questions as we did here, answer them, and review the plan that it comes up with, and then once we've got this plan and we're happy with it, we can move on to start building which is the phase that we're at now. Okay, so you can see very quickly, it's come up with the plan here. I want to quickly review and see what it does. Now, notice this plan is actually created using something called a markdown file. A markdown file is a file that ends in .md, and this is a very common format, kind of similar to like a text file or a code file, that's used when working with AI agents. Now the reason I'm mentioning that to you is that the formatting you see here is in what's called a markdown format, and recently they actually added the ability to have, um, what is it, like, you know, visual charts and stuff in markdown and render directly in Cursor. So we can see that we can move this around if we wanted to make it a little bit larger, we can make it full screen if it's kind of hard for us to see, and then we can open and close it on the right-hand side and view all of the stuff like the forest platformer demo here which will then trigger the plan. Okay. So decisions, we've made that, the tech approach, it's going to do this in HTML5, okay? The gameplay loop is going to look like this. Of course, if I was building a real app, I would review this a little bit more carefully, but now what we want to do is proceed to build. Now, when we're building, it's going to directly give us an option to either build in parallel or to build locally. We can also build in the cloud. I'm going to stay away from the cloud right now because I don't want to confuse you, we'll look at that later. For now, I suggest you build everything locally on your own computer. Okay, so to build locally, we can literally just press this button, and then there is this build in parallel mode. If you do that, what will happen is Cursor will actually create multiple agents for you that will start tackling the plan in parallel so that ideally it can get done faster. There's no reason not to use this, so I'm just going to use the build in parallel mode, which is automatically going to parallelize multiple agents if it's possible to do that, and then start building it. So again, to trigger that, I've just pressed this little down arrow and I'm going build in parallel. Okay? So now what's going to happen is we've got our plan, we've got some to-do list, and it's going to start setting up multiple agents in the background that will start building this out for us. And you can see that it's created a sub-agent. We'll look at those more in detail later, that's going to start building the app. So while this agent is building the app, let's talk about a few other features that we have. So first, let's just move this to the side so we can view the window in full. So you'll notice that right now we're in multitasking mode. The reason we're in multitasking mode is because we did run in parallel, and what we can actually do right now is that if I wanted to do something else, so I wanted to have another agent running while this one is building, I could just type a prompt right here. So I could start building another game. I could press new agent, and I could open up another project and start working on multiple at the same time, and that's kind of the benefit of this, this view, this agent view, where you can manage multiple agents in parallel. You also can expand all of the windows here, so on the right-hand side, you can see changes that have been made, you can see a browser, okay? On the left-hand side, if you move this, you can expand this kind of Agent view where you can search through stuff, automations, settings, right? There's a lot of stuff, I'm going to go through it one by one, but I'm just trying to give you a lay of the land. So again, important thing, Agent view, kind of the de facto view, especially if you're not technical. You're going to prefer using this, and then the IDE view here where you'll actually be able to view all of the code and use traditional IDE features, which I will show you in a minute. So for right now, before we go into the IDE view, I would prefer for us to stay in the Agent view because this is going to be simpler for most of you watching this, and we're going to start running a few prompts, setting up a project, and just getting kind of the base ready to go before we move into anything too complex. Now, what I can do is I can remove this multitasking mode, and when I'm just in this default mode here, all this means is that I'm just chatting with the agent normally. So I'm going to wait for one task to finish, then I'm going to run the next one, and we can queue up multiple tasks by typing, you know, "hello world" or whatever the next task is that we want, and it will kind of go in order where we wait for agent to finish, then it goes to the next task, next task, next task, sequentially. Now, there's some other modes. You're going to see we have the Debug mode. Debug is when we want to fix a particular bug or problem. So we don't want the agent to build something new, but we want it to find what's wrong. So for example, the game's not loading, we flick it into Debug mode. Hey, the game's not loading, can you figure out the issue? Plan mode, we already covered that. That's what we start with at the beginning to generate a plan for our application. Multitask mode, we know what that is, and then Ask mode is simply a mode that you can use if you want to ask questions about the codebase, but you don't want Cursor to make any changes directly, okay? Additionally here, you have images. You can actually upload images directly to Cursor, so if there was some inspiration for an app you wanted, you can just copy an image and paste it right inside of here and it can use that and view that. If the app's not working properly or you want to do a revision on something, you can take a photo of it, screenshot it, send it to Cursor, tell it to change that. And then we have skills and MCP servers, which we'll look at in a minute as well as models. So I want to quickly talk about models which you can also access from the model selector right here. So when it comes to selecting the models, you're going to notice there's a bunch of options. Now, first you have this Max mode. The Max mode refers to the size of the context window, and when you enable this, you will increase the context window size, which is going to use more usage and is only necessary when you have relatively large projects. So if you're building something simple like us, you don't need the Max mode, you don't need to enable it because it is going to cost you more in terms of usage credits. However, if you're working in a really big codebase, you definitely want to enable this, and it will expand how much context or information the model can reason on at once. Next you have the Auto mode, as we discussed. If you're not sure what to use, you can flick it into here and it will usually get the job done for you, and then other than that, you can pick the individual models that you want. Now, I'm not going to go through all of them. Again, my suggestion is that if you're doing something really simple that's typically web-based, use Composer 2.5. If you're doing something a little bit more complex, you can use this Grok model. This is the new model, it's very good, it's a little bit more expensive than Composer, but it's still very, very cheap relative to something like Fable or GPT-5.6 or all of the other models that you see here. Now, these other models like Fable 5, for example, or GPT-5.6, you can use them as well. Typically, I will only switch to those models if I'm not getting done what I need with the models built into Cursor. So maybe they've, you know, had 10 iterations, and they can't fix a bug. Okay, I'm going to switch over to Fable, and I'm just going to use it for something that's much more complex, get a fresh perspective and see if this model can maybe handle it. Fable's not working? Okay, I'm going to switch over to GPT-5.6. So that's how I would use the models. I would keep default Composer 2.5 and Grok 4.5. Again, Composer, if you want a little bit lighter usage, not as complex. Grok, if you want something a bit more challenging. And then you just switch into these higher-tier models when you need kind of a fresh perspective or something's broken. That's how I personally use them. Now, from these models, you can also edit them. Now when you edit them, you can change the effort or the reasoning level. The higher effort you go, the slower they will be, but typically the better result you will get. You then have the option to enable Fast mode. When you enable Fast mode, it's going to make these models much faster in terms of the number of tokens or code that they're outputting, but it will use more of your usage credits. So it's up to you. I typically leave Fast mode on, but again, if you're really constrained on cost or using the free version for example, you probably want to turn that off, okay? And same with all of these models, you'll see there's a bunch of options. You can change the reasoning, the context window, and in most cases, sticking with medium reasoning and the lower context window, especially if you're not building something complex, is totally fine. Okay, so while I explained all of that, the app has been built. Now, you'll notice with Cursor that it's actually able to automatically run the application for you and typically we can view that by looking in the Cursor browser. Now, right now it hasn't actually put it in the browser for us, but sometimes it will actually load it up directly here. So if I want to view this app, but I don't want to go to my own web browser, usually when I press this button here, it will load it directly in this browser window that's built into Cursor. So again, the way I activated this is I just closed this, I pressed browser, you can see that it's here. I pressed on where the application is hosted, it says serving it right here, and boom, we have the app. Then from here, you can see I kind of get this forest platformer game. Okay. Looks like I'm stuck and, uh, I can't get out. So, oh wait, there we go, okay. You know, we're going to have to make some changes or maybe I'm just really bad. There we go, okay. You know, we're going to have to adjust a few things because this is a bit too challenging for me to play, but you can see that we have the browser open. Now as well as the browser, on the right-hand side of our screen, you'll see changes, terminal, and then files. So if we open up terminal, we can run a terminal shell directly in here, so we don't need to leave this window, and then if we open up files, or, uh, changes here, we can see any changes that have been made. However, you're not going to see a lot of information here unless you have what's called a Git repository initialized. Git is a version control system, it's used by software developers to keep track of all the changes in their project. Highly recommend that you enable it, it is a piece of software that you need to install, but if you want, you can simply just tell, um, Cursor here and let me change back to the model to do it for you. So I'm going to say, "Hey, can you please enable Git and just create everything? I don't understand how Git works, but I just want you to make a repo for me so that we can save all of the changes." So it's exactly what you can say, if you have no idea what Git is, just say, "Hey, make it," and if it's not installed, it actually may be able to install it directly for you from the command line, and then these features on the side where you're viewing the files and the changes should start working a little bit better. Okay, now I made this Git repo, and also create a file called Git ignore, okay? I'm not going to explain exactly what that is if you're not familiar with it, but you're going to notice that it's not really showing me any changes right now. The reason why is that the changes that were just made actually were already what's called committed to Git, so it's a little bit more challenging for me to kind of review them right now, but don't worry, we'll make some more. However, I will show you that if we go to the Files view, we can press this little expand thing on the right-hand side to view all of our files, and we can now view the files that we have. So we can see we have these audio files, camera files, and I can directly modify the code right inside of here, if I want to do that. In most cases, you probably will not touch the code, and if you are, you likely won't do it from this agent view. However, just showing you that if you want to see what the structure of your project looks like, you can do that directly from here, okay? So you can see that we have all the code, I can even make a new file if I want to, I can make a new folder, I can refresh to see any new stuff that has popped up. Okay, so I know that was a lot of stuff. What I want to do now is just add a follow-up prompt, have it build this out a little bit more and make it a little bit more playable, then we'll look at some more features. So I'm just going to say, "Okay, this is pretty good, however, it's very fast, it's a little bit difficult to jump. I find that I'm not dying, but also, when I fall through the holes, can we tidy this up, clean it up a bit, make it a little bit easier to play, and make it not so fast so that I can move a little bit slower and it feels a little bit more controlled." Whatever, okay? And you're going to notice also, by the way, I just want to show you this, that I can directly tag files in Cursor using the @ symbol. When I do that, I can reference documentation, files, past chats, the browser, um, different files, and you'll notice that I actually did that using Wispr Flow. Because Wispr is directly integrated in Cursor, I can just say "at levels.js." And when I say that, it actually knows how to tag it, and it's kind of cool that it directly tags it inside of here. There's a few other ones like that. But anyways, if I do this, I just want to show you kind of the change flow here in the IDE view because it's a little bit different. You'll notice that now it's going to go directly in here, it's going to make a change, and this time when it makes a change, it's going to be a lot easier for us to review it and revert it back. So you can see that it's showing us the changes, I can kind of scroll through them here, and there's actually this keep file and this undo file. Now, keep file effectively doesn't really do anything because the files already kept by default, like the changes have already been made, but if you want to revert the change now, and you can go between them, you can press undo file or you can undo all of the changes by pressing this little button right here. So for me, when I'm making larger changes, or when I am doing something that's supposed to be a little more precise, or with code that I really care about, I work in the IDE view again, because this for me is a little bit easier in terms of actually reviewing the code and making sure that everything is good, and then you can add this to chat, you can comment on it, or you can just undo everything and it will get rid of all of that very easily so that those changes will be reverted. Additionally, if you do actually want to write code, I don't know who's doing this anymore, but let's say, I want to do something, there is a tab auto-complete, so if I start typing, so again, just kind of showing you here, you can see that I can press tab, and then it will auto-fill if you want to do that. I don't know if you guys are writing code in the editor, which I think many people are not doing anymore.
Okay, so I just made a fresh agent and I now want to talk about things like skills, um, commands, rules, and some more advanced features when you're creating a more serious project. Now, Cursor and many other agents have the ability to create something called project rules. Now, these are essentially rules, right, or text, that's going to be injected in the agent anytime you start working with it in this particular project. So for example, you may want to have a rule that says, "Hey, I only want you to use JavaScript, I don't want you to use TypeScript. I want you to write everything, or have every file be less than 200 lines of code, or I want you to never do XYZ, whatever." Things that you don't want to have to keep repeating to the agent and that you want it to remember. If that's the case, you can put it in a rule. So to create a rule, it's actually easier just to ask the agent to do it. Just say, "Hey, can you please create a simple rule for me? I want this rule to state that we should always use JavaScript as opposed to using TypeScript or any advanced frameworks. I want you to write everything or have every file be less than 300 lines of code, and everything is modular and factored very well." Okay, so there we go. We're going to go ahead and press on enter, and you're going to see in a second that it will create a new folder called .cursor, inside of that it will create these rules, and these rules are things that will be automatically injected into the prompt anytime you write code. You can see we have these rules that get created in this markdown file with this description and kind of always apply thing, and there is other ways to make rules, but it's always easier just to ask the agent, and it will put it in here. Now as well as rules, you have skills. Now, skills are essentially reusable workflows you've probably seen in other tools before, that allow you to just to create a doc or a spec for how something should be done. Now, there's skills that are built into Cursor. You can see we have skills like automate, right? Right? If we keep going, banner design, brand, browser automation, canvas. I don't know how many of these I installed versus which ones were automatically here, like create rule, right? Which it used just to make a rule. Create sub-agent, design, whatever, but you can use these, and when you use them, they will reference the skill file which looks something like this. I just pressed on it so you can see this is what the skill file is. Now if you want to make your own skills, you can do that and then that will give you a command that you can run to run them repeatedly. So for example, I'm going to say, "Can you make a project-level skill that just does a quick review and ensures that the game is playable at the correct speed? I just want this for a quick demo, so let's call it like Speed Skill, where I just want to make sure the player has the correct speed." I don't know, this makes no sense for a skill, but I'm just showing you if you want to make a new skill, you can just say, "Hey, make a new one," and I just want to show you kind of what it looks like here, because it will use a skill, funny enough, to make the skill. It should put it in a new folder for us here inside of .cursor, and then we're going to be able to execute it. So you can see if we go here, we now have a skill, speed-skill, it's created this for us, I can keep it, and then if I want to use this now, I can type /speed-skill, you can see that's my custom skill, and I can run it and then it's going to go and execute what we have in here for checking the speed of the player and it says, boom, everything is good, we can continue. Now, all of this stuff, you can also view and manage from the agent's window. So if you go to the agent's window, uh, not that, where's the agent's window, and you go to customize, you can see plugins, MCPs, skills, sub-agents, rules, commands, hooks, right? So all the skills that we have for user ones, um, sub-agents, rules, commands, hooks, and all of this other stuff that you can manage and this is now where I want to start talking about plugins and MCPs. So, what are plugins and MCPs? These are essentially additional functionality that you can add directly inside of Cursor. Now, you can see that I've actually added a GitHub MCP server, I have a browser MCP server, and then in terms of plugins, I have this browser base one as well as Zapier. Now, if there's a particular plugin you want, you can browse the plugin marketplace here, for example, maybe you want to connect to Slack, or you want to connect to Figma, you can just do that, and then what will happen is it will give you a bunch of skills that you can use in Cursor. I'm not really going to go through them because it's really dependent on your workflow, and then same thing with MCPs. If you've seen an MCP server that you want to add, you can just add it and it should give you the config or the instructions from where you're adding it. If you want a good one to add, I would recommend adding the GitHub MCP server so that you can just tell it to directly do commands on GitHub and it will do that for you. Just nice now that's kind of put this all in one place, and again, if you want to manage this as well from the IDE, because I'm just showing you both of the views, you can go to the settings in the IDE, and from here, let's make this a little bit larger, you have all the same options. You have like Gits, plugins, rules, skills, tools, and MCP, so it's available in many, many different places that you can check out. All right, so sorry for the cut here, but there's a feature that I wanted to bring in here that I forgot to mention to you, which is called side chats. Now, side chats is a new feature inside of Cursor which is super cool, which allows you to select something in the current conversation and add it to a side chat. Now when you add it to a side chat, this will allow you to have multiple agents running at the same time, but to reference something that was in the original chat and maybe ask a question about it or clarify it without interrupting the current agent. So for example, if in this agent I said, "Can you just make some minor changes to the game? Let's change the background color. Let's have a better main menu that also does character selection." I'm just doing a random change, right? I'm going to ask it to do that, and then while this is kind of chatting, let's say it pops something up that I want to ask a question about, or I want to go back and be wait, what is that make what is this? You know, I want to make a change here, whatever, right? I can just select something, so I'm going to select this and add it to a side chat, and now I can say, "What is the best size for the character? What do you think?" And I can chat on the side, right, while the main agent is running, and it's kind of the same thing as sub-agents, but I'm able to reference parts of the original conversation, so super cool feature, relatively new, just wanted to jump here and show it to you. Now let's go back to the rest of the flow. Okay, now I think the last major thing to go over quickly is automations. Again, something that I'm not really going to dive into too much, but that you can do using tools like Slack, for example, GitHub, um, Linear, whatever, the all these different tools that you can connect effectively, and what you can do is make an automation so that agents will be automatically run either based on a schedule, so for example, once a day, or once an hour, or based on the trigger of a particular event. So for example, I can go, new automation, I can say, "The trigger for this automation is a Slack message, or a GitHub message, or a Sentry issue or whatever," and then I can say, "Hey, I want you to do this whenever this gets triggered." You can set up memories, you can add tools. Again, this could be a whole video on its own, so much to go over here, but I'm showing you this exists, pretty new feature, and it's something useful again if you're working on real legitimate code bases and you want something to run quite frequently. Okay. So with that said guys, that's going to wrap up this video. I promise you a full course, I promise you a lot of features, I know there was a ton of stuff that we went over here, but I wanted to try to be as detailed as possible. If you guys enjoyed, make sure to leave a like, subscribe to the channel, and I will see you in the next one.
