0:00 Claude Code can now do in a matter of hours and minutes were used to take developers entire weeks, but only if you actually know how to use the tool to get the outputs you're looking for. In this video, we're going to run through 30 key concepts to help get you the most out of Clawed Code and how to actually use it day-to-day. But first, what is Clawed Code and how does it differ from the normal chat interfaces that we tend to use with Gemini, OpenAI, and Claude itself? Well, think of the chat interfaces as literally just inputs and outputs for conversational threads. Yes, they can take some action, but it's a bit more limited. Maybe creating a word doc or updating some, for example, notion templates, anything like this, but it can't really build anything or go beyond that. That's where cloud code comes into play. We're able to go
0:46 through and create our very own applications, websites, and much much more because Claw Code will literally help us write the code that sits behind it. And the great piece about that is you don't need to be super technical to use it. You just describe what you want. It can build the website, set up databases, install anything on your machine. So, you need to watch the permission section later on on that to secure it a bit more. But essentially, you now can get outputs by just talking to it. And yes, you can actually get it to generate outputs without you having to code anything. But a little bit of coding knowledge always helps to really understand what it's doing and the outputs it's giving you. Now, Claude Code was built by developers for developers. But because it's so powerful, we can mold it and adjust it to different use cases such as using it
1:32 to build business operating systems, generate marketing material on autopilot, or build entire apps from scratch. If we come down here, we've got three main different ways for us to be able to use it. We've got the desktop version that we're able to use within the app. Next up, we've got the web app. And then last but not least, I'm going to tie the two together with our terminal and IDE. The IDE stands for integrated developer environment that can help build out all of our application requirements, files, everything that we need to make it run and so we can see it and edit it nice and easily. You're going to have two main options for IDE. You can come across and use VS Code, which is totally free to use, and it's backed by platforms like, of course, Microsoft, or
2:18 another really popular one is platforms like Cursor, where we're able to go through, and they have their very own AI agent, or we can just turn that all off and use it. Nice and simple. It doesn't matter which one you choose. If you're brand new, I recommend starting with VS Code just cuz it's a bit easier to start off with. Now, when you start your screen up, you should see something that looks like this with the terminal on the right and the empty space on the left. So, now we're going to be able to actually talk to our clawed instance and actually get the outputs that we want. Now, this is called the art of prompt engineering. And there's two different ways that you can get started. The first way is literally just typing hide there and whatever you would like into the specific chat window on the right. We're also now able to use forward slash and voice to make sure that we've got our
3:03 voice mode turned on so that we can talk to it nice and neatly just by holding the space bar. But check out my other video on the channel that covers that in more details. Now, say we want to go through and actually give it lots of information or we've got some reference documents like a PRD for product requirements document or MRD for marketing requirements document. We can come to the left hand side plus for a new file. We can do prompt and then for example start.md. This is going to create a very simple text file for us to be able to use and reference within our project. So now we'll be able to go through we can hold down for example with whisperflow. Hey there. So we're going to be going through and generating a brand new application which is all about helping a
3:49 beginner's guide to claude code. We want it to look amazing. generate lots of example outputs, the different prompts they should use, the way that they can interact with it, all the different commands, and much much more. Now, that was a very, very short prompt, but as you can see there, we would be able to make this much much longer, and then all we would do is come across to Claude, say, "Hey, please check out the and then add symbol for that reference document on the left hand side." So now if I press tab, we're able to reference that specific document in this conversation thread as we go through. And then once you've pasted your prompt in there, another thing I recommend doing is we've actually got something called planning mode. You can do shift and tab to make
4:36 sure you come across to this specific mode. But what it will enable you to do is work with claw code on that more planning element first before it goes through and actually builds anything. This way we can get that best output hopefully quicker because you're doing that back and forth to really clarify any of the requests how it will complete the actual task anything in the space get it to think step by step before you actually get it to run and generate the output. Another great thing about claw code is it isn't just a text entry that we can do. We can also do something called multimodal and this is where we're able to give it different inputs. So you saw for example we can do text. We can also add files. We can also add images. And hopefully in the future, we'll be able to do video directly as
5:21 well. Say that you really liked this image. You could rightclick, copy that image. Come back across to your chat interface. Now, I'm using a MacBook, so I can do command and v or you can do control and v to paste if you're on Windows. Now, as you can see there, it's referencing image one. Now, we can send that off. And what it's going to go through and do is try and replicate that exact dashboard because it's going to analyze it, really see how it works, then generate the single HTML file. So that's going to be basically the web code to see it on our browser for us to be actually replicating it nice and easily. So if you have screenshots of process flows, documents, PDFs, Excelss, CSVs, text files, anything like this,
6:06 we're able to share it with clawed code as well. If you don't want to paste it straight into here with the certain permissions that you've got within Claude as well, you'd be able to go through and tell it the file location on your machine. So, for example, a transcript that I've got for a video, we' be able to say, "Please access the text file here." And then we can paste the command in there, which shows that in this use case, it's stored on a 2 TBTE SSD. We're then looking in the specific transcript folder for a text file. txt to really understand what it is. And there we go. In plain simple English, it's gone through and made those changes. So, if we come back across, as we can see here, we've now got that dashboard ready to go. Now, we can talk through some of the permissions. So, you can start it up
6:52 with claude. You can do clawed enable auto mode or you can come through and do claw dangerously skip permissions which is where it's going to actually just run all of the commands that it needs to versus waiting for you to approve specific commands. That has it pros and its cons cuz it can build a lot quicker but it will literally run any command on your desktop. So you got to be very careful if you're using this mode. That's why the auto enable mode is become really popular because it's the best of both between asking questions and executing safe commands. And now if I press enter, what we should see is at the bottom here, we've now got bypass permission on. So it's going to enable it to keep going through. Of course, we don't want to be using bypass permissions for everything where it's
7:38 literally got the ability to run any commands that it wants. So what we're able to do is create a settings.json JSON file, which means that we can say what it is allowed to accept and complete, but also what it's not allowed to access. So this can be it's allowed to read all of the files that it's got access to, but it can't write to any of them. Or for example here, maybe we've got an environment file which if I come across has lots of different API keys, lots of different details which we don't want to be sharing directly with the LLM. That could be the same for maybe a credentials file that we've got as well. So, what we're going to say in our settings is it's not able to read any of this information and it can't go through and write for it. As well, we've also
8:25 got some commands down here where it's not, for example, using the curl request able to go through and download any specific curl files, i.e. download repositories if it was told to use that command. One thing I highly recommend though is making sure that you always deny access to any environment variables, any of your secrets or credentials because of course we don't want that to be going through to the AI model at the end. Now, as we run through this demo, what you're going to see is me doing forward slash a lot. These are called slash commands which are built-in commands and we can create our very own what they call skills as well which we'll touch on later on to execute a specific task process or instruction that we can get clause code to do. So
9:10 some of the built-in ones include for example help. This will show you all of the different commands that you're able to use. As you can see here we've got a bunch of the different inbuilt ones that we can use as well. What this will do is have a specific instruction that Claude can run. Now, as you start chatting through with Claude code, what you'll find is there's something called the context window. When you first start talking to it, it's going to have a really low amount of information that it's stored readily available for that conversation. But as you chat more and more and more, it's going to be storing more of that information to use and reference in the future. Now, we're going to have two commands to help deal with this. We're going to be able to do forward slash and then compact, which is going to keep the conversation history,
9:58 but it's going to summarize all of this information so we can use some of it in the future conversations. The other way, if you really want to open it up, is just to come through and do clear. And this will literally remove any of that conversation that you've had up to date and free up that entire context window for you to continue on. And that will stop that context window getting full and leading to what we call context rot. This is where the outputs start getting worse and worse and worse because it becomes overwhelmed with all the information. It starts losing sight of what you're trying to achieve, the outputs you're trying to get. And to check the amount of context, you can go through and just type context in here and you'll be able to see the total context usage for this current conversation at the moment. But what if
10:44 you're having a previous conversation with Claude and you wanted to bring it back? Well, you're able to do forward slash resume. This is then going to resume a previous conversation that you had so that you can actually continue it nice and easily. So, if you do forward slash and then resume and if you just press enter now, you'll be able to see the different conversations that you had in the past. So, you can press enter and bring it back into that conversation fold like you've never left it. Next up is cost. And that all comes down to two main different ways that we're able to use it. There's going to be the use of our account. So, for example, here we can pay for a pro account or a max account using Claude. But if you're starting off, I recommend just going through using the pro version. But if you're starting to want to get more out
11:31 of Claude Cove, you're running into those limits constantly, then I would actually look to upgrade to the max plan. The other way you're able to be build is actually through API pricing. Now, what you'll see is you've got the million input tokens and you've got the million output tokens. What this means is it's going to cost you $5 per $750,000 words that go into the model. Then once it's done all of its basically analysis, it's generating the output for you. You would then be charged $25 per 750 words that get put out the back of it. So, think of it as around one token equals 75 of a word. It is worth noting that when you look at the market because the clawed models are actually some of the most expensive for you to be able to
12:17 use, but they are some of the highest quality that you will get for both that personality as well as the output that you'll be able to achieve. To check how much you're spending, you can do forward slash and then cost and you'll be able to see the exact cost. Of course, in this example, we're using our orth login details, so we're not being charged anything at all. We can also come through and do stats as well. And this is going to show our current activity, how much we've used, of course, the favorite model that we want to use, the total tokens, the longest sessions, and much much more. And if you want to tab across, you can also see it by the model as well. The difference between the models that you're able to use with Claude is you've got Haiku for that fast and cheapest option. You've got Sonic,
13:02 which is honestly the best allrounder for many people to use day-to-day, but you've also got that most intelligent model with Opus, which is incredibly good and the best I think on the market at the moment for helping you to get those big picture thinking done, as well as the execution of the code as well. And when you want to choose which model you want, just do forward slash and then model. And then, as you can see here, you got the different options for which one you want to use. By default, it's going to use Opus 4.6. six, but now Sonic 4.6 is actually incredibly good for lots of those day-to-day tasks or if you got a quick light question, then definitely look to use HiQ instead. When we come up to the larger models, like for example using Opus, we're also able to go through and change the effort.
13:48 You'll see at the bottom there for low, medium, and high and max effort. Or if you come through forward slasheffort, say you're working through a long piece of work. Now you'll be able to go through and do effort tab and then choose the one that you wanted to use. So now if I do low, we're going to change the reasoning model the effort that it's going to put in now down to low instead. Now we're going to run through claude.md. And this is a really important file because we got two different variations we're able to use within our project. This is going to help us basically every time we wake up a new claude code instance is going to understand the outcome we're trying to achieve any goals or rules that we put in place. So for example, we could say only use
14:34 British spelling. We're able to do forward slashinit for initialize if you haven't already got one created and it will read your entire codebase and project that you've got set up and create a brand new version. We'll run this in the background. Of course, we got nothing set up at the moment. So it should go through and just create an empty one on the left hand side here. Or you are also to go through and generate it manually. And we just do the plus button here. And then we just type claude.md which will create that markdown document for us to go through and add all of the details in. And there you go. As you can see we've now got that set up. So what this means is every time we start up claude within our terminal, it's going to make sure to read this command before
15:19 it goes through and execute any specific processes. So if you got some hard and fast rules, definitely make to add them into claw.md to help get those consistent outputs in the future. Now that's how we able to use it within of course that folder or potentially project level. But if you wanted to impact every single conversation that you have, then we've got the global version. This will impact every single project with those specific instructions and commands. So if you find that you have a best way of working with Claude code and you want that to be repeated every single time you engage with it, just make sure to go through ask Claude to update the master and global MD file and it will go through and do it nice and easily. The reason you want the difference between the two is because
16:05 sometimes you may have more project specific information such as that is a claude agent for lead generation and it needs to be the best it possibly can be for that versus the global project is just consistently helping you get the best outputs across lots of different use cases. Now Claude has also got its very own memory. You're able to do forward slashn memory to be able to bring it up and edit it. You'll be able to see it's got it automatic memory that it can turn on. So every time you chat with it, it learns your preferences, what you want for good, what you want for bad, of course, what you're looking to achieve, everything like this. As we go through, we can also turn a brand new feature called all to dream on. And what this means is every now and again, it's going to go through and store some more information so that you can reference it
16:51 in future projects as you need. Now, if you're serious about AI, come join us in the AI edon school. It's completely free to get started. Hundreds of members all over the world. There's a full masterclass section. Templates you can grab and use straight away. No fluff, no algorithms, no ads. Link and QR code are on screen now or just simply check the description and you'll see it there. Flags. These are additional commands that we want to give Claude at the start to tell it how we should operate. So for example, we can do claude dash dash and then model. And then we'll be able to say exactly the model we wanted to use. So for example, I would type Claude-Opus-4-6. This is going to start it up using that specific model if we hadn't already said
17:37 it. Another one is we can come through and do PL-P and you're able to use these in the commands as well. So for example here, if you say hi there and then do -p, all of these flags will work in there as well. And then after this, all you need to do is type the command in there you want it to execute. So for example, we need the Chrome marks to tell it what we want. We could say build me a landing page for a consulting business. So now what it will do is if I execute this, it's going to start up our clawed code instance and then it's just going to run through in the background to help get us the output without necessarily seeing everything as it executes the code. Another popular one is for example claude dash verbose. And this will basically ask Claude to go into lots of details for the output that
18:23 you're trying to get. Next up, we've got skills. These are pre-written instructions that we're able to give to Claude to execute a specific process or retrieve an output that we're looking for. We're able to store it in what's called a skill.md file. And an MD file is just pure text that we're able to use to store all of this information. I'll show you an example in a second, but just to run through the entire structure, we're going to give our skill a specific name. We also need to give it a description of what it's doing. We can also give it some instructions in what's called a markdown format. And I'll show you what that means in the example. We can then go through and add extra resources. So explaining scripts, i.e. maybe it's got a Python command that it
19:09 wants to run to execute a specific calculation. We can also add any reference documents that you want it to load such as maybe your marketing strategy. And then last but not least, we can add any assets as well. So say perhaps you've got a specific actual icon that you wanted to use your logo's business, then you'll be able to add it in there nice and easily. To create your very own ones, you just need to do forward slash and then skills creator. So now if I put this into here and send it off, Claude is going to help me build this into an instruction, a recipe. And there we go. We've just got everything back. So what we're able to do is now do forward slash and then linked in post generator. We can do tab to bring it
19:54 into the conversation. And then we could just go through and say and we'll send it off. Now what is go through and do and as we can see it's going to write that LinkedIn post for us straight away. So as we go through and improve and enhance our skill, it can sound more like us. We can use the best hooks, we can generate it on autopilot and much much more, which is the power of skills. We can also share these with other people because if we come to the leftand side here and we just drop down see for example the skill dock, we can now see all of that information. So, we could share this with someone else. They can make their own version or if we've got different team members in a business, we can make sure they all use the same one in the future. Now, we're going to run through plugins. These are extra abilities that we're able to give our AI
20:40 agents, our clawed instance to help achieve specific outputs that we want to. So, we've got different ones that you can see here in the official marketplace repository. You can add any of these that you want nice and easily. If you do tab, it will take you across to see all of the different ones that you've already got installed so that you can see if any of them need to be updated, you want to deactivate them. We can also see the different marketplaces we've got installed and if there's any errors for the plugins that we've used. So for example here if we wanted to go through and improve the claude.md management we would go through for many of you starting off just installed it at the user scope if you're comfortable exactly what you're going to be using it for. What I also recommend if you're
21:26 ever trying to install external ones that don't come from the official marketplace and just come across the tools like Virus Total which in this use case are totally free to get started with and it will give you security analysis of is there any potentially malicious elements to that specific file you want to use. But if we come back across and say install for you now it's going to have installed it as we can see up the top here. And then we just need to do forward slashreload plugins. And it's going to make sure to reload and update all of the ones that we've currently got installed for our project instance or for our global plugins. Next up, we've got MCP servers. Now, Claude will automatically work with all of the files on your local machine. But what if we want to access different tools like for example notion, air
22:12 table, slack or GitHub? What we're able to do is give it the power of model context protocol servers which basically enable us to access all of the different commands for executing specific tasks i.e. do this task in your application, give me this data back, everything like that that through the MCP server because it knows all of the different possible abilities it's able to do. And that's because for your business and many other businesses, you don't just live in one system. They all need to be able to talk to each other. You need to be able to access data from different ones and pull it through to complete processes. If we come across to the official documentation for MCP servers with clawed code, you'll see here that we've got lots of different popular ones and
22:59 just the simple command to install them. I'll include all of the links down below so you can find this nice and easily. And if you wanted to install a brand new one, for example, Bitly, just come through and copy this command here, open up a new terminal on your machine, paste in there, and now it's going to go through and add that specific MCP ability to our Claude Code instance across all of our projects. Now, everything so far we've been running within one single conversation, but there's a feature that's really important to understand called sub agent. What this enables us to do is in our main conversation is going to be our main AI agent. But say we need to complete multiple different deliverables, multiple different outputs for one specific task. Well, then what
23:45 we're able to do is a create our very own agents, which I'll show you in a second, but we can get it to go through and complete specific outputs. So maybe you need one to be a researcher to investigate a specific market awareness around a tool. Then maybe we're developing a new app, a new feature, we can go through and use the tester and the developer both for different lenses to see if we're actually getting the best outputs. And then last but not least, say we're doing some LinkedIn copywriting. Again, we could have our very own subwriting agent to complete that specific task. Each one is going to run with its own context window. It's going to help get us better outputs and we're able to run them in parallel. So it speeds up that development versus having just one conversation running
24:31 with our main AI agent at any one time. Think of it as having sub agents specialist where we've got our main agent as the coordinator and generalist to get your outputs. But Anthropic have also taken this one step further because we've now got what's called agent teams. Now, previously with sub agents, they're not actually able to talk to each other throughout the process. You send off one piece of information and it sends it back once it's done. What we can do with agent teams, which do use a lot more tokens as a result, is they're able to have context around what each other's doing. So, they understand, for example, if we're changing the API, how would we then need to test them? but also they can work with our main agent in a coordinated way to say how we need to
25:18 update any of the tasks we're completing adding any relevant of course context to help get the output and much much more but it does come at that token cost which will mean that you run into your limits a lot quicker and I would only recommend it for really complex build or lots of complex changes where they all interconnect with each other if you're getting it to go through and just write a newsletter then trying to get it to write a blog blog, then trying to do maybe an image for a post on social media. They can be three different sub aents that don't have to talk to each other. To set up your very own agents, you can do forward slash agent. As you can see here, if we press enter, so we could do create a new agent. If we say we're going to do project, we want to generate it with clause cuz we can
26:04 explain it what we want to do. So now if we send this off, it's going to go through and try and generate a description for us for this specific agent. And we can also say if we want to have access to any of the tools. So again, you'll be able to go through turn any of this on or off. We can also come down to the advanced options, which as you can see here, there's tons and tons of different ones to choose from. We're now going to have our very own agent start to be created. We also need to make sure that we define the model that we want to use because otherwise it's going to go through and by default use the exact same model as the main conversation. So we can come through say we want this one to be sonet. We can also say if we want it to have a particular color. So we could do blue
26:51 the same color as LinkedIn and press enter. We're going to say user scope for this. We're going to do press enter. And now we've got it all ready to go straight away. If we want to use this agent rather than doing forward slash for linked in as a skill, what we're able to do is an at symbol, we can then do linked in come down to our content strategist. So now if I press enter on that, it's going to start up our agent to go through and get the outputs that we're looking for. And there we go. It's now asking us the questions to help get it output ready to go. Of course, if you're ever invoking one of these agents and you want to change the model, just make sure to go through and tell it, hey, I want to use Opus or I want to use Sonnet to make sure it doesn't default to whatever your main one is using. Now,
27:38 what if we've gone through and tried to generate an output with our clawed agent, but it's actually gone down the wrong train of thought. And that's where we're able to use checkpoints, where we're constantly saving different versions in the background. So we can then use for example rewind to then restore it to a previous part of the conversation. So now if I restore conversation here, we can go back to that main part that we were talking about beforehand and we wouldn't have the maybe error workflow or error process. It was starting to go down. So now we're back at the start and we could actually change that entire conversation thread in the future if we weren't happy with how it was going so far or we made a mistake and we needed to come back up to a previous version. So now that we're generating all of these files, where do
28:23 we actually store them? Well, that's where we're able to use a platform called GitHub. When you log in, you should see something that looks like this. On the left hand side, you'll have something called repositories. This is just think of it as your folder of file structure for where you're storing all the information. So, for example, this trade flow AI that I did for a previous video. As we can see in here, it's now got what's called a readme file. So, all of the information about this specific project. And if we come back up to the top, we can also see that it's got all of the files in here. And this is great because it can help us track the different changes or the latest version for us to access anywhere in the world. So, in case there's any issues on our local machine, we don't need to worry about it getting wiped or any file corruptions. You'll see that we've got
29:10 the plugin and the MCP that we're able to use. We also need to make sure that we've got it downloaded on our local machine to authenticate us in to be able to use it. And to do that, just come across to this site here. We need to make sure that we're installing Git locally, which is going to track all of the changes for us and connect us to our GitHub account. And then you just need to come down and run this gh or loging command within your terminal to set everything up. Next up, we got something called work trees. Now imagine you have to go through and ship multiple different updates for specific features at a time. This is where we can help run different tasks in parallel. So for example, in one chat, we could ask it to go through and create an actual new
29:55 dashboard for us. In another one, we could ask it to go through and actually try and fix any of the bugs that we've got with our checkout page at the moment. Or last but not least, we could also test maybe updating that landing page for a client at the same time in parallel. For these to work though, make sure that you've got your project on GitHub. And work trees aren't just for developers. Think of them as productivity multipliers. you're able to have three different agents completing maybe three different tasks at the same time to help speed up your workflow and get the output. Next up, we've got hooks. Now, what these are are specific tools that have been developed by Claude Code. They run automatically in the background. Don't use any of your tokens, but what they enable us to do is
30:41 every time Claude saves a file, it will format it automatically. Every time Claude maybe runs a command, it would log it for you so you can see it in the future. Or every time you end a specific session, it's able to go through and back up everything. As you can see here, we can notice all of the different ones they've got set up. So you can go through and change these to exactly what you want. So as you can see here, lots and lots of different ones for us to be able to use, change and adjust. If you want to edit any of these, it's nice and easily. We can chat to our Claude agent or we can go through and edit the settings.json. JSON in our files so that we can actually tell you what it can do, can't do, and the hooks we need to run our project. And there you have it. That's the top key claude code concepts that you need to be aware of whether you are starting off with your projects or
31:27 you want to be running it overnight. Don't worry, you don't have to memorize all of these. Just make sure to bookmark and save the video so you can come back as you need it. And if you want to learn the latest and greatest about Claude Code, make sure to check out the channel for the latest updates because we're constantly showing what you're able to do with it as well as many other platforms as well. And as always, stay tuned for more around AI automation and have a great day.
