Claude Code Full Course: Build AI Agents Without Coding

Clot code works really well on a day but then when you close the terminal and reopens the clot code again the next day

it would have forgot everything about your project what you did it on a previous day. So I built this whole

entire clot code course in [music] a way that each lesson builds on top of the mistakes that previous lesson have left

behind. And by the end of this full claude code course, you would have three agents working inside a claude plug-in

that you can install on anybody's system. And to do all this, you do not need to be a coder or a technical

person. Do all this by copy and pasting and talking to plot code in [music] plain English. Welcome to gen plugged.

My name is Dhiraj and this is my full claude code course free for you so that you can go from lesson number one to

lesson number eight and become an expert on clot code without knowing anything about clot code. So [music] let's take a

look at the full course road map now. So in this course we go through eight lessons. Each one [music] builds one

before it. In the lesson number one, you install cloud code and build your first working tool by just describing what you

[music] want in plain simple English. Then in lesson number two, you give that project a memory, a few skills and a

hook. So you stop reexplaining yourself every single time when [music] you open cloud code again on day two, three and

subsequent days. Lesson number three is where you connect it [music] with web and other tools and give it external

intelligence. That's what I call it. In lesson number four, you build a research agent that does its work on its own

window. And in lesson five, you build another AI agent that gets your writing cited by AI tools as well as helps you

with the search engine optimization of your content over the internet. Lesson six puts a team of three agents on one

topic. And lesson seven tags the whole team into a clot plug-in that [music] anybody can install on their system. And

finally, in lesson number eight, you run clot code on free models. You run clot code for free [music] even when you do

not have claude subscription. So let's get started with lesson number one. What you are seeing is a content

multiplication system I built with claw code. One article goes in, 10 plus platform specific content pieces comes

out of it. LinkedIn posts, tweets, substack notes, newsletter sections or YouTube scripts, all in my voice, all

ready to publish. And here is the thing, I built this entire system without writing a single line of code myself. I

described what I wanted in plain English and Claude Code built it for me. Well, I am a software engineer. However, I have

not coded for a very long time, maybe more than a decade. I have been doing content creation for almost like 17 18

years now. Started my journey in 2009. But when Claude Code came along last year, well, everything changed for me.

It completely changed the possibilities that are available to likes of us who are content creators, small business

owners or soloprunners who want to build real tools. However, do not want to go into coding or software development or

hire software developers in order to do these job. Well, that's where claude code comes into play. Today, I'm going

to start a cloud code course on this channel which will be a series of videos to come in the coming weeks. And by the

end of this series, you will be able to go from opening up a terminal or a blank terminal to all the way building the AI

agent systems to run or power your businesses. In this first video of the series, we are going to talk about the

basics. What is cloud code? How to install cloud code? How to set up cloud code properly and initiate your project

workspace. And finally, we are going to build a content multiplication from scratch so that you have a full

understanding of how to leverage cloud code and how to interact with it. And by the end of the lesson, you have a

working content multiplication system for running your business or for any needs of content repurposing. So you

will walk away with actually something to use by the end of the lesson. But what is this content multiplication

system? Well, we'll get to it in a bit. But let's start with the basics first. Before we install anything, let's first

understand what is claude code because it's not what most people think. Well, Clot Code is an agentic AI powered

coding assistant from Enthropic. It lives in your terminal and it can read your files, write code, run commands,

create entire projects all from natural language instructions or communication that you do with it. Think of it having

a senior developer sitting next to you who can access your system, access your files, understand you, understand the

problem that you are stating and build systems for you in a much faster and very iterative manner. The keyword over

here is agentic. It just does not answer your question, but it acts on it. It takes action. It brainstorms with you.

It writes content, create files, edit files, run test, commit to git repositories, manage your servers,

manage your entire cloud environment and whatn not. You describe what problem you are facing, what you want to solve in

plain English and cloud code is going to do or solve that problem for you. And how to use this cloud code? Well, there

are many ways of using cloud code. Some people use it directly in the terminal. Some uses visual studio code and an

integrated terminal or visual studio code plug-in. Then some people use with cursor or conductor. Some uses even

desktop app of claude or even the web version of claude code. But today we are going to use the terminal version

because that's where the most power lies and I'm going to mostly use integrated terminal with visual studio code. Now

let's quickly go to the screen share and start installing clot code and set up your environment. All right. As you can

see, I have opened up terminal on my Mac. If you are using Windows, you can open up the command line utility on your

machine and follow the steps that I'm going to follow. First, I'm going to head to the code.clude.com.

Over here, you can see that the documentation exists so that you can easily follow it. I'll also attach a

file that is having all the commands that we'll be running and all the prompts that we'll be using throughout

this tutorial. But this is a good reference which you can also refer at a later stage as well. So what I'm going

to do is it's first step is installing plot code. You can see that there is native install which is the recommended

one. Then there is homebrew and then there is vinget. What I'm going to do I'm going to use this native install

over here. Right now I'm using Mac OS which is supported also by Linux and WSL. So for that you are going to copy

this command and come back over here and just paste it and run it. Since I have already installed cloud code on my

machine, it is going to check whether everything is installed or not. But if you are installing it for the first

time, it is going to do the entire installation for you. Let's just wait for it. It takes a little time in order

to fetch it and then install the clot code. As you can see that it has started installing but it quickly said that it

has successfully installed the clot code on my machine. But for you it might take some time if you have not installed the

clot code. Now next thing how we can test it out? We can just write claude d-help. So if I write this then it is

going to provide some sort of help commands. As you can see the help commands are over here. That means

claude is successfully installed on our machine. If we go back to the documentation over here you would see

that there is this Windows PowerShell command as well. So you can use this Windows PowerShell command or you can

use the Windows cmd command as well in order to install clot code on your system. But because I am on Mac OS, I

use this particular command in order to install on my machine. So now the installation is complete. Next thing is

authentication of your claude code with your claude account. So the prerequisite of installing claude code is that you

must have if I go up you must have a terminal and a command prompt open. You have a code project to work with. So for

now we use this my home directory as a base and we install cloud code. That is completely fine because we are going to

set the home directory or project directory project workspace later for the project that we are creating but you

can install it from anywhere. So that's what it means that a code project to work with. And finally and the most

important one is the claude subscription you must have pro max or teams or enterprise plan in order to use claude

code. You can also use claude console account by using API as well. But I would recommend that you start with pro

version of the subscription of claude and then start using claude because it becomes pretty cheap or I would say like

you would get much more usage limits while using the claude subscription instead of using the API credits through

claude console. That is going to cost much more and API credits burn much faster in my experience. And once you

have set up that account, what you have to do is follow the second step. If we go down, you will see that second step

is to log into your account. So over here, we are going to start the clot. So I'll come over here and I'll say clot.

So this is going to kickstart the clot. So first time you'll type this command and press enter for any working

directory or any folder. It is going to give this particular warning or a safeguard measure that hey are you sure

that this is your own work? This is a folder that you trust. This is the folder in which your work is there

because you are giving it access to claude code and claude will be able to read, edit and execute files in this

particular directory. So that means you need to be sure that whether this particular directory is owned by you or

not and then only allow the access. So we are going to trust this because this is my system. So I'm going to trust it.

I'm going to say yes, I trust this folder. And you can see cloud code has started. And since there is so much of

space, let me just minimize a bit and drag it over here. So the first line tells us which version of clot code you

are using. So currently we are on 2.150. The model that it is going to use is Opus 4.6 and I am currently on my Claude

Max plan. If you are on cloud pro plan, you are going to see Claude Pro. So these are a few things. This is the

working directory you are in. So you can see it over here where you are working right now. So I'm right now working in

my home directory. We are not going to continuously work over here. It is just to show you how the login works and what

basics are there of cloud code in a terminal. Because when we start the project, I'm going to switch on the

project directory. So this is where you are going to start typing commands or start speaking or start telling claude

what to do. So this is your that chat interface where you are going to tell what to be done. First thing like we

said if I go back to the portal of documentation you can see we have to do / login in order to authenticate our

claude code with our main account or subscription account that we have taken from claude. So I'm going to come back

over here and I'm going to say slash and as you can see there are so many commands that has opened up. You might

not be able to see all these commands because these are special skills and commands that are created on my

particular system. And we are going to talk about how to create all this in rest of the lesson so that you have full

understanding of what I have created and what you can create. First thing is / login. So you can see sign in with your

anthropic account. That is what the meaning of this command is. And I'm going to press enter. What it is going

to do? Select the login method. You can use claude account with subscription pro max team enterprise entropy console

account or third party platform like Amazon bedrock foundry or vortex AAI. So this is the third way of authenticating

claude code on your machine. Whichever you have set up, you can use that. I'm going to use that claude account with

subscription. So I'm going to press enter. It is going to open up claude website for me. Clot.ai. And it is going

to open up this particular screen where it says claude code would like to connect your claude chat account. So

your account will be used to all these things. read them, review them and then basically you have to click on authorize

so that the system start getting used by claude code on your local machine. So your local machine is now authenticated.

You can see that it says build something great. You're all set up for cloud code. Now we are going to come back over here

and over here you can see as well login successful. Press enter to continue. So I'm going to press enter over here and

now cloud code is set up connected and working for you. But before starting up any kind of project or build anything

with Lord code, we have to first create a project workspace. I don't want to pollute the main home directory which is

there which is my own user home directory. I don't want to pollute that. I want a special project directory where

all my projects reside or sub projects reside. So we'll exit out. I'm going to just clear it and I'm also going to

close the terminal because what I want to show you is how I use cloud code where I use cloud code. I don't use it

in the terminal of the MacBook rather I use it in the integrated terminal of Visual Studio Code. So let's head over

there. If you have not installed Visual Studio Code, you can install it from this particular link code.studio.com.

If you are using Windows, it will show download for Windows. If you are using Mac, it is going to show download for

Mac OS. download it, install it and how it will look like is something like this. So this is the Visual Studio code

over here and the main reason I prefer using integrated terminal is that it gives access to me for the project files

as well as the integrated terminal all in one screen. Now Visual Studio is installed and open on your machine. Next

thing what we are going to do is set up the project directory in which we are going to create project which is content

multiplication system and for that either you can follow the GUI version like you can go to the explorer and

create those directory and come back over here or you can use the command line. So for that I'll go back to the

command line interface. So on the command line I'm going to change the directory from my home directory to this

claude code project masterclass directory that I have created. This folder or directory is already created

on my machine. That is why I was able to change the directory from my home directory to this directory. In this

directory, we are going to create our project. For that, you are going to use either MD for make directory in Windows.

On Mac, it is mktir. And I'm going to say content. So now the directory is created for our project.

Let's get into that directory. So cd. So you can see that the directory that we created exists in this particular cloud

code-masterclass. So now we are going to come back to the visual studio and over here it says

start. We are going to open up our own directory. So that's what we just created. So I'm going to head over here.

I'm going to head over here and this is the directory that we created. I'm going to open this. So now you can see that on

the left hand side the explorer is showing the empty directory. There is no file in it. That's why it is empty. And

that's what I was telling you that I can access all the project files over here. I can doubleclick it. I can open it, I

can see it, I can edit it, I can save it. So that's the power or that's the advantage of using Visual Studio Code

and a integrated terminal with cloud code inside it. Over here there is a chat session as well. I have also

installed Visual Studio CL code plug-in as well. You can use that if you want, but I would prefer that you use an

integrated terminal because it has all the power. So I'm going to click on this. I'm going to cross this and we are

going to go into the menu bar and say new terminal. In the menu bar when you start the new terminal you will see it

like this. [snorts] Here you can expand it. You can create more terminals from here. New terminal new terminal window.

You can split the terminal. So you can manage all the terminals and run multiple instances of cloud code in

different terminals integrated inside the visual studio. all giving you access to the project directory on the left

hand side so that you can view the files and I'll come to it and it will get more and more clear to you as we are going to

work and create this system. So you can see the terminal has opened inside our project directory over here that is

where we are going to start the clot code and for that you already know we just have to write clot and I'm going to

click on this and you can see welcome back right now you are in this particular directory you are on this

particular model you are on this particular plan so now the clot code is up and running into our project

workspace and I'm going to click on this square brackets over here as you can see maximize the size panel so if I click on

it. Now it is maximized. Now like I said here you are going to type anything or you can speak anything. Hey Claude, what

can you do for me? I generally don't use much keyboard nowadays. I just talk to my system via dictation. So this is

where you are going to use the natural language in order to communicate with cloud code or discuss or brainstorm or

create or build things. And the last thing is the status. So right now you can see that it's saying plan mode on

shift tab to cycle. So if I say shift tab, so now it is saying question mark for shortcuts. Now one more time I

pressed it, it says accept edits on shift tab to cycle again plan mode. So that's how you recycle between different

modes of claude code. So the two most important modes are plan mode and the second one is accept edits on. If you

have this particular mode on using shift tab you can recycle between these modes but if you have this mode on cloud code

is going to start doing the work. If you have plan mode on, it is going to save you a lot of hassle. And that's the

first important tip of this entire course that when you are starting up any kind of project with cloud code, always

start with plan mode. It is getting smarter and smarter every day as the new releases are coming. You can see right

now we are at 2.1.50. It is getting smarter and smarter and at times it automatically jumps from like

working mode in plan mode. If it thinks that the it requires considerable amount of planning, considerable amount of

change or it's a major change request that you are telling it to do or any type of change request, right? Not just

with code but even for content creation as well. If it thinks that it requires a lot of uh work to be done, it

automatically jumps into plan mode. However, I'll caution it to you that don't rely on that automatic jump.

Rather deliberately think that if there is a significant amount of thing that you want to build or create better start

with plan mode because this is where you can interact and brainstorm with cloud code first and finalize the plan. Okay

with the plan that hey this is what the blueprint looks like the entire plan how it is going to build a particular

feature particular system or create a particular type of content or any type of piece or any type of problem that you

are trying to solve with it. It is going to lay out the entire plan of the solution to your problem. First review

that and then approve that plan so that Claude can start working on it. This is the most important best practice. You

should never neglect the plan mode. So now enough of basics and setup of cloud code. You have cloud code up and

running. Let's build something and let's build a content multiplication system. But what is this content multiplication

system I'm talking about? Well, as a content creator and a YouTuber, I write articles on my Substack. If you have not

followed or subscribed to my Substack, go over there, subscribe over there. A lot of detailed tutorials comes out.

Then I publish videos as well. So for the videos, I have transcript of the videos. For the articles, I have full

articles. But I also have to post tweets, I have to post LinkedIn post, I have to create substack notes. All this

takes time. So what that means is I have to read these articles or watch these videos or read transcripts and create

these LinkedIn posts, tweets, substack notes, newsletters out of these already created content and that becomes very

tedious. I know chart GBD is there, Claude is there. I can just copy paste and ask it to produce it. However, that

is also more like a manual work because now I have started copy pasting the same content. Now I'm reviewing as well

whether my brand voice which how I speak, how I write is there or not. I have to tweak everything in a more

iterative manner using clot projects or cloud code or using charg projects. But there has to be a better solution to it

that understand what my brand voice is that understand how I write that understand how I speak and how at the

end I will write a LinkedIn post or a tweet or a substack note. So the system should be intelligent always enough to

understand that in take a particular piece of input as a YouTube transcript or an article and produces all these

facts out without me doing anything. I'll use NAN in order to distribute the content that this content multiplication

system creates it for me. That's the need of creating a content multiplication system which always

generates this repetitive type of work. The grunt work is being all done by cloud code for you in a more automated

manner instead of iterative manner. That's why we need this system and we are building it. So let me clear this

out. So let me come back over here. Press escape to cancel. Now we have cleared it out. This is where I'm in

plan mode right now. Like I said, we'll always start with plan mode. So, what I'm going to do, I'm going to paste a

big prompt over here. So, as you can see, it says pasted text one of 17 lines. And what prompt I have typed in?

Let me open the text editor over here. And let me show you. Let me bring this over here. So, this is the prompt that I

have pasted over here. I need a content multiplication that works like this. First one, I have a source content. It

could be blog post, article or transcripts in markdown files. I want to generate multiple content pieces from

each source, five LinkedIn post, 10 tweets, three newsletter section and two YouTube video transcripts or scripts

structured with the intro, main and outro. If I am basically using an article, I need this. If I'm using a

transcript, probably I'll suggest the system not to create that. The third point is very important because now over

here you are saying that maintain my brand voice casual practical no jargon. I'm going to share how it is going to

create the brand voice for us as well. While during this whole system setup you are going to watch that as well. So bear

with me. You can see pull actual quotes and examples from source material. All these content created should be saved in

an organized folder by content type. And finally include a content calendar as well showing what to publish and when.

And the last step is I want to run this with a simple command like node generate.js source article md. We are

going to further simplify these commands when I talk about commands and skills later in the lessons. But for now this

is the first lesson basic lesson. We just going to tell it that I'm going to just run this command and everything

else should be done by it. So that's the overall system that we have defined. You can see that I have been very very

explicit about my requirements over here because garbage in means garbage out. So you need to be very very particular

about and clear about what you need to do. The better the input that you are going to put into clot code, the better

the output will come out. So that's why it is important to think about your problem statement and then share the

task with cloud code. Lastly, I have couple of more things. So think step by step about it and build this system for

me. Create all necessary files, scripts and folder structure. So it is going to create it. We are not going to do

anything. We are going to start with an empty folder as you can see. Finally, make sure to ask questions or push back

if required at any stage. I always welcome other suggestions and alternative as well. That will make the

content multiplication system much more successful. This is where you are partnering it with it. It is not that

you are just giving it a command, it is going to execute it for you. You are asking it, opening up that channel that

if code is having some sort of like doubt or concern, it can push back, it can ask question. And that's where it

becomes more like a partnership. So that you are building a system together. First creating a plan and then allowing

it to build the system. This is the prompt that basically we have pasted in cloud code and I'm going to press enter

over here and it is working in the plan mode. So no harm is being done. Nothing is being created right now. Can see that

it has started working. It is going to explore the current project but current project structure is empty. It is not

going to find much over here. And as you can see it has started consuming the tokens and tool usage is already

consumed 37,000 tokens over here. Then another one is 38,000. So that is where the different plans of claude code will

come into play or claude subscription will come into play. That is why I use claude max because I'm a very heavy

user. I build a lot of things. I've already released five or six products and services through the help of cloud

code. So it has already paid for me. It is a $100 plan. I know but it is more like having a developer or senior

developer rather in fact sometimes tech lead play it as a technical architect as well at the cost of $100 a month. If you

are into building a lot of things if you have a system to create then I would recommend that you use claude max plan

because otherwise you'll keep on hitting the limits but for this kind of a simple project prop is enough so that you can

get a taste of how claude code helps you. Now you can see that it has started asking us question. You can see it is

saying before I design the plan I have some important question a fully built version already exist in a back of

folder. So it's a system that I already created when I publish the article. So you can see that now it also knows that

in my system a fully created system is already there whether I should copy and improve it or whether I should build it

from scratch. But you can see how smart it is that it already figured out that there was a fully built system already

there. Now what we have to do because we have to work through this and I have to show you the entire process. So I'm

going to press the down arrow key which helps me select these options. So you can see that either you can select all

these options or you can even chat about it as well. So if I select this it will open up the window. I can chat or enter

whatever thing I would like to say about this particular question. For this question we are going to just say second

option which is like build from scratch. I'm going to press enter. Now it is saying which AI provider you want.

whether you want open AI, whether you want cloud API, which API provider you want. So I'm going to say like you know

open AAI API keep as is. So this is the one that is going to be used when repurposing the entire article or

transcript into different content pieces. So I'm going to say keep open AAI as is. Now it say enhancements. So

all these questions it is asking me because it already found a system. It might ask you certain different

questions because you may not have created all these things, right? So for now I'm just going to say just get it

working and then like I'm going to say submit. Now all the questions have been answered. Like I said you may have some

different question and it might ask us different question again while working on the plan. So let's submit that. Now

you can see that it says got it build from scratch open AI focus on working system with better prompt. So now it's

going to again work again deliberate think about it and maybe ask certain question again or going to create the

entire plan for our review. So let's just skip to that part. All right. So you can see that the plan has been

created. Let's go and scroll up a bit and see what it is telling us in the plan. So this is the context that it has

written for us that build a content multiplication system that transforms a single markdown source, blog post,

article, transcript into 20 plus platform specific content. And this plan could be a little different for you

because like I said, I had already created this system for my Substack article. You are building it from

scratch. But you can also follow along and review your plan side by side. But it is definitely going to build the

system from scratch as I approve it. Over here it is saying that key fixes over the backup. So this is what it is

going to approve like due to max tokens it is going to increase from 3,000 to 4096 update the pricing for GPT4.1 mini

much better prompts and a major improvement of passing source content to all the generators backup only passes

analyze summaries. So the previous version that I had created only used to pass the analysis summaries. Here it is

telling us how the whole structure of the project would look like. The folder structure right at the top it's content

multiplier. If you don't understand code, you don't need to fear. Like I said, it is going to write the code for

you. All you have to see over here is these comments that are there. So for example, it is telling us that the brand

voice will reside in config.json. The example article for the input would reside in example- article md file. This

is where you will load your input example or input voice. Not just one example, you can upload multiple

examples over here. The prompt is going to live in this particular JSON so that you can improve or fine-tune the prompt

that it has created. Then rest of the stuff I don't much care about it. At the bottom you can see there is an output

structure. So LinkedIn post will come over here, tweets will come over here, newsletter will come over here, YouTube

scripts will come over here and calendar will be published over here. So this is basically more like a information to you

that how it is going to structure the entire system the AI system that you are building for this content multiplication

how it is going to structure it for you. So you just need to review that forget about or ignore the coding part of it.

Here it is telling us what is going to be the build order phase 1 phase 2 phase 4 5 6 and orchestrator and all. Here it

is going to tell you what exactly is going to happen inside the code and what is old new and finally at the end it is

telling us how it is going to verify it. So all these verification steps as well it has written at the bottom you can see

claude has written up a plan and is ready to execute. Now we have reviewed the plan. What we are going to say is

yes clear context and auto accept edits. You can reject the plan. Let's suppose if I don't like something over here I

can just type it over here or narrate it over here. So this is like your input field where you're rejecting the plan

and asking the clot to take these inputs first and revise the plan based on it. So that's how you can basically do a

conversation or build the plan together with claude. For now I am happy with the plan. So I'll go ahead and just say

clear the context which is going to save 42% of it and then auto accept edits that means that all the edits that it is

going to do I am not in the loop it is going to just run and create everything for us until there is absolutely some

permission is required then only it will stop and ask us to provide that permission but for us the first option

works so I'm going to press enter now it is going to start building the application files building this entire

system and overall on the left hand side in this tree area of the project you would start seeing sooner the files will

start populating the folders will start populating as per the hierarchy of the folder or folder structure we reviewed

in the plan that we submitted it so it is going to take its own sweet time so I'll skip to that part directly where it

has built the system for us so you can see that the files have started appearing in our project workspace

directory and claude has started outputting code. You are not writing code. I am not writing code. Code is

being written by a plot code. We just gave it in plain simple English or natural language what we want to build.

It figured out the entire plan. It figured out the entire technical architecture. It is figuring it out what

code needs to be written in order to implement this whole content multiplication system. So that's what I

mean that you don't need to be afraid of the word code in it. just wear your builder hat and start working with Lord

Code in order to build some coolest AI agentic systems or AI automations or regular system without the use of AI as

well. So in the to-do list you can see that two phases have been completed and now it is building the third phase. Now

it is going to create the fourth phase which is more like building prompts and then generators and finally the

orchestrator. Then it is going to verify if the entire build works or not works. You can see that entire thing has been

generated. And if I look at it, build is complete. Here is the summary of everything that has been created. You

can see 20 files built from scratch in content multiplier. All this code over here that you can see if I open up these

JavaScript files, all these codes nothing has been written by us. It has been written by cloud code. Whether it

works or not, we don't know. We are going to just use the system and then tell us what is broken, what is working,

what is not working. It is going to fix all this code for you. So, we can just ignore all this code part over here.

Let's go back. Let me expand this or maximize this. Now, let's see what it is telling us. Let's scroll down at the

bottom. So, this is what it is telling. This is the phases that it has completed. So, it has implemented these

fixes over my system that was there. But for you it might have created an entire new system. So that is also okay. Now it

says that validation tests which are passing or not. So it has also verified some of the use cases whether things are

working or not. And you can see it is telling us to run a full generation create the env file with your API key

and run this command. And remember that this is the command that was there in our initial prompt that we want to run

the entire system using a simplified command over here. And in the next lesson I'm going to tell you how to

create slash command. So you would be able to just write slash multiply content and the input path of the file

and it is going to work for you. So you don't even need to further write these with the help of slash command. So we'll

come to them in the next lesson. But for now it is telling us before running this I have to create an environment file.

And for that it has also created an example file over here as well. The env.ample and it is telling us that we

have to create env file. So what I'm going to do is I'm going to right click over here and create a new file and name

it env. All right. So that's what it is. Eenv is there. So from the example I'm going to copy this and I'm going to

paste it over here. So now over here I have to replace this particular thing with the open AI key. You have to create

open AI key or in the plan if you told claude to use claude or Gemini then those key pair values would be present

over here. But during the time of the creation I said that we will be using open AI key and in order to create the

open AAI key what you can do you can go to the platform open AAI. So I'll just say platform open AAAI. When you go over

there you can just register your account. I have account registered. So I'm going to login using that account.

And then you have to click on this API keys over here and create a new secret key. So if I create a new secret key

over here it is just say plot- test content multiply. So it is going to get into the default project. I'm going to

click on create secret key. So now the key has been created. I'm just going to copy it. Done. And I'm going to come

back over here and paste it over here. So my key has been generated. I have pasted over here. I'm going to save this

environment file. And I'm going to come back over here. Now I'm going to expand this and I'm going to tell it, hey, I

have created the env file and loaded the application open API key in it. Can you check and tell me everything is working

as per the expectation and then we can run the script. So you can see that I have just written it in plain

conversational English that I have created the environment file and then I'm just asking it to do its magic. So

I'm going to press enter. So you can see it's saying let me verify your environment file. So that means it has

found the environment file. It has found the open API key and now it is going to run the full validation pass source file

config promps API key all together. You can see that content multiplier system is running. Here is the status.

Everything validates and analyze correctly. The cost estimate shows about 1 cent with GPD1 mini. So now it is

telling us yes go ahead and generate. I'm going to say yes and then let's run it. Now it is going to run the system.

It is going to input the article that we have given it. So you can see that input example article over here. If I come

over here example article. So this 15minute content system that changed how I work. So this is the example that I

created earlier like this is one of the lessons from my substack. So it has used that as an example put it over here and

then basically it is running the system and creating multiple pieces of content over here. So you can see generation is

complete 20 pieces created for 15 cents. Now let me run the quality check for verification plan as well. So it is

going to run the quality test for us as well. If something is wrong, it is going to fix the system, selfheal it, self fix

and produce the output as per the desired quality or as per the quality standards that we have defined.

Everything is working based on the brand voice that has been input in this config.json file that I showed you. So

this is the output summary it has and the output was present in the output folder. So let's go over here. As you

can see there is a calendar, there is LinkedIn post, there is newsletter, there is tweets and there is a YouTube

script as well. So if I open this LinkedIn you can see this is the first one and key theme is overcoming decision

fatigue and context switching in content work. So this is like one variant. This is the second variant where theme has

been changed. This is the third variant, fourth, fifth. Similarly for the newsletter as well. You can see that it

has created the outline of it over here in the tweets. This is the first tweet, the second tweet, third tweet, fourth

tweet and all the way till 10 tweets. Now for the YouTube script as well, it has also created the YouTube script. Now

let's suppose we want to add a system to it. Let's suppose I want to add Substack nodes. So what I'm going to say is hey

all these outputs are fine. I'm happy with it. Can you create an other output layer? I want to generate substack nodes

as well. And those substack nodes should be scannable. For each input, I would like to create at least three to five

substack nodes or I'll say exactly three substack nodes should be created. Now it is telling us that it has again entered

into plan mode. The old plan is from initial build. Now I need to plan the new substack nodes feature. So that

means this is a new feature. Now it is going to build into the existing system that it has already created. So that's

how I don't know code. I don't even know like how it structured all these code files. Similarly for you as well, you

might not even know code. You just told it that hey I like the output that you have created for all these platforms. I

would like to add Substack notes in pure natural language. And as soon as it receives the input, it has started

working again and it is going to build the feature for us. So you can see that it has created the entire plan for

building this new feature of adding substract nodes generator to our existing system and I'm going to skip

all this coding part over here and I'm just going to say yes clear the context and start building. Now it is building

the substack feature for us. And this is the part that really got me hooked to this whole clot code thing because now

it feels so powerful that if I have to update a feature, I have to remove a feature, I have to add a feature, I just

need to converse it and explain it. Hey, this is how it is working, this is how I want it to be. The more clear you are,

like I said, the better output it is going to generate. I am just conversing as I'm like talking to you and creating

this video over here. And meanwhile, cloud code is doing the work. So this is just a practical example of what we can

do. For now, I'm going to skip to the part where it has completed the substract node generator feature for us.

All right. So verification checks have passed. The new feature is built. Now we can go and do something little different

in order to test the new feature. What I'm going to do, I'm going to go to my substack over here. So what we are going

to do, I'm going to just copy this URL from here. And then we are going to go back. So far how our system works is

that there should be an input markdown file which should be present in this input folder or anywhere on your system

there should be a markdown file that should be inputed to this system in order to create the multiplication of

the content for these platforms. What I'm going to do is I'm going to tell it that hey I have this URL go fetch this

URL get the markdown from it and input it as an input or contents for our newly developed system in order to create the

outputs from it. So I'm just going to narrate it. Let's see what it does. And I'm not going to run the command that is

there. I'm just still talking in plain conversational English. So that's what you need to remember that even if the

command has been created, you don't need to run the command over there. If you don't like it, you don't run it. You

just tell it to run it on behalf of you. It is going to do that job for you. I have narrated it. Here is the URL. I'm

going to paste it and I'm going to press enter. Now let's see what it does. So you can see it is telling us let me

fetch the article content and then run through the content multiplier system. We did not run the command that we were

supposed to run. It is going to run it based on the English or conversation or chat message that I have submitted it.

So now it has got the full article. It is going to save it as an input file and run the content multiplier. So it is

doing the work for us. We are not doing it. So you can see that a new file has been added over here. It has fetched the

markdown from it and now it is going to input this particular file that it has downloaded or created from a downloaded

article for over the internet and then running the entire system for us and now it has started running the entire

system. As you can see article saved now running the content multiplier on it. So all 23 pieces have been successfully

generated. It is telling that it is going to show the substack notes as well. So this is note one build real

apps by describing problems not specs and then like you know the entire note is there. If I open the abstract folder

over here you can see these notes right over here. One thing now you are noticing is that we generated three or

four types of variations including the verification that it was doing for us. So we generated the content first from

the example article then I gave it the link over here. However, this whole output folder is getting replaced again

and again. So if you have to build this feature of that this should not get replaced again and again. I'll just

leave it as a homework to you so that you can just keep on iterating over it. Tell it in plain simple conversational

English or natural language whatever you speak. Just tell it this is the problem that you are facing that hey all these

content is getting replaced for each and every folder for each and every input article. What can we do? How can we

preserve it for each article? It is going to brainstorm with you and implement the solution in no time at

all. Now let's take a step back and look at what you have actually learned over the past 30 minutes or so because I'm

sure that you have picked up much more lessons and learning that you might even realize. So the first thing that you

learn is what is cloud code, how it is installed and how it is set up from scratch on your system. Second thing you

get to know how to work in a terminal without being a terminal expert. Third thing we reviewed is what is plan mode

in claw code and how it actually works, how it actually supports you. Fourth thing was the most important one. How to

describe in pure natural language a particular problem to claude code and how it can build it for you. Fifth

important thing was how to embed your own brand voice, your own voice that you use in order to write or create content

into a repurposing system like this. And finally, I also shared with you how you can iterate, change, update, make new

features within an entire system that you have already built with the help of clot code without writing any kind of

code. And at the end, this whole content multiplication system that we built and you will iterate over it is for you for

the taking and you can use it, apply it right away for your own needs. If you are generating content in any ways, this

is a system that you can start using right away from this very day. And all this was very much the foundation of

cloud code. How to get used to it, how to start working on it. So by now you have a working tool, one article that

goes in [music] and LinkedIn posts, Substack notes, YouTube scripts and all that content

repurposing comes [music] out of it. You built all of it by describing the entire system in plain simple English. But here

is the problem. You close this terminal and open this project or content repurposer again tomorrow. Claude code

would have forgot [music] everything about it. So you end up explaining everything about it again, including

your brand voice, including your preferences, including how the output should be located in the output folder.

That is exactly what this lesson number two of my claude code full course solves as you give your project a memory called

claude.md file and you turn all the prompts that you keep on repeating into claude skills [music] and finally you

write the claude code rule that claude code does not break for you in this project.

Claude code forgot everything. That was my experience for the first week that I was using claude code. Every session

started the same way. I have to reexlain my brand voice. I have to reexlain the entire project at times. Even redescribe

what goes into the coding guidelines. This was more like talking to a new employer, new hire almost every day.

Then I discovered claw code has something called extension layers that many people do not know about. CLMD file

that acts as a permanent memory of the project. Now they have memory MD file as well and we are going to talk about it.

Secondly, cloud code has skills or slash commands that can activate or execute longunning workflows for you. They can

be invoked by either user or they can be invoked automatically by cloud code. Then third thing they have hooks. Those

are some of the rules that claude code cannot break. And finally, clude code has sub agents that can work

autonomously for you in their own context windows. In the last episode of this claude code full course series, I

built a entire content multiplier system live with you. That is the project that we are going to put base of this entire

full course and keep on building more and more use cases more and more powers using cloud code ecosystem into it so

that you have an end toend system. So last time we just have a project it does not remember what we did. So today in

this lesson I'm going to talk about all these extension layer concepts so that you have the right foundation of claude

code in place before we move into other important aspects or other advanced topics that claude code offers us. By

the end of this lesson you would have a working project which is again a content multiplier system but the concepts you

can apply in your projects as well. And this content multiplier system is not going to forget anything. It's going to

execute some of the tasks that are more like workflows in this content generation ecosystem and it will have

some safeguards in place that cloud code never can override. For example, never able to delete the output files. Now

before we build anything, let's talk about some sort of a mental model that we should be following when talking

about all these four layers of extensions in cloud code. For your mental model, what you need to

understand is the first layer that is there which is cloud.md file. This is more like a house rules that are there.

They are always active. They always load into the context of clot code execution of any of your prompts. They do not need

any sort of a trigger. They are there as a set of instruction that Lord code almost try to follow, almost try to

adhere always. This is where the entire project memory goes. So that you do not need to tell cloud code what this

particular project is. At the bottom I have kept it into the deterministic category because we always know that

claw.md file will get loaded by cloud code into the context memory. However the instructions which are there in

claw.md file may not always all the instruction get loaded or followed and we are going to talk about it in a while

later in this video. Then we come to the second layer which is skills. These are more like longunning workflows or the

tasks that you want to execute or the business domain knowledge that you have you have put into skills that can work

along with you when you are working on a project. These can be manually invoked by using slash commands and I'll come to

it what is slash commands and how you execute them or they can be automatic in nature as well that claude code

automatically figures them out and execute them or run them when needed. Then comes the hooks. You should treat

them as sensors. They are guaranteed to execute no matter what. So for example, you ask Lordco to delete something. If

you have a hook in place, it is going to block that delete operation. And finally, you have agents which are more

like specialist in nature. They own their own context window. That's the beauty of agents or you can call them

claw code sub aents as well. The main thing or main difference is that they run in their own context window. They do

not pollute the window or context of the running operation or running conversation that you have. They run in

background in their own context window. So you get to do more in the same session. And I have tagged all of them

into deterministic, probabilistic, deterministic, probabilistic. Like claw.md file hooks are categorized as

deterministic while agents and skills are probabilistic. What that means is probabilistic is the nature where claude

code or LLM or AI which is the LLM will decide whether I should invoke a skill or invoke an agent to do a particular

task or not. You do not have much control unless you explicitly specify and increase the probability to 100%.

For example, you tell in your conversation clot that use this brand voice audit skill. That means you are

increasing the probability to almost 100% and telling claude what to do. That means it is going to follow your

instruction and load that skill that you said. Similarly, for an agent, you can call an agent by name and instruct

claude code in your main session to execute an agent. That's what it means by probabilistic and deterministic. Now

let's get to the screen share or cloud code session and see some of these things in action before we deep dive

into each one of them one by one. So I'm going to switch to my visual studio window where I have opened up an

integrated terminal and you do not need to have the visual studio code. You can open the terminal that you have and

start writing claude in it. So as soon as I write claude in a terminal it is going to open the cloud code. So this is

how the clot code looks like if you are watching it for the first time. Over here is the project that we built in the

last lesson which was the content multiplier system and all the code base that you are seeing over here. None of

them was generated by us. Everything was generated by claude code. Today we are going to build on top of it. Before we

jump into details, let's quickly see some of the helpful commands or help that you can get in cloud code session.

So in order to get help something called slash commands that I said are merged into skills. Now that is what you are

going to refer. So if I write forward slash then you are going to get a list of commands that are available in your

system that you can invoke by just typing it out and pressing enter. For example I'm going to press /help. So you

can see that it is going to show all the help and available commands in your clot code. So I'm going to press this and

then I'm going to press enter. All right. So you can see that help center has been opened by using the help/

command and at the top there is this general where it shows all the shortcuts that are there. You can use exclamation

mark for bash command so that you can run the bash commands inside directly in cloud code interface. This one is

forward slash that we just used to in order to invoke the help command. The add symbol you can use to tag your file

paths or you know invoke any files, put any files in the prompt that you are using or interacting. Similarly, if I

press on the arrow key of the right, you would see all the commands that are there by default. And you can see there

is this branch command, agent command, add directory command, buddy command, which is pretty new these days. This

command I use most often because it clears the conversation history and free up the context. This is a tip that you

should be following specifically if you are on cloud code pro plan which is basically $20 plan that gives you less

usage limits. There are many commands that are there and another one that is very important one is the init command

and this is what we are going to use next because now we'll deep dive into the extension layers and we are going to

start with clot.md file. So how do you create a clot md file? You use the init command to initialize a new clot.md file

which will insert everything about your project into the clot.md file. So let's go ahead and start the demo of that

part. So I'm going to like click on escape escape quite a few time and now we have exit out. At the bottom you can

see I'm in plan mode and I talked about what is plan mode in the last lesson as well. Plan mode is when you talk to cla

code, it creates a plan what you need to do and then you can act on the plan. If you press shift tab once you are going

to toggle into different mode. So right now it's shortcuts mode. Now I am into accept edits on which is where cloud

code whatever I'll type it into this particular section or prompt or input field it is going to execute that. This

is where the execution happens. If the plan is already created or even if the plan is not already created, you can

execute simple prompt, simple instructions right away in this particular mode. So this is what we are

going to use right now. I'm going to press forward slash and I'm going to say init and maybe let's before I init

knows about our project. What do you know about this project? So you can see that it is trying to figure it out what

it knows about this project. Right now it has started reading the files, listing the directories but it does not

know. Right? So it is basically trying to identify what this project is. It immediately did not know. I have

cancelled it so that I can show you what is the power of clot.md file. So how do you create clot md file is using the

init command. So we'll press init. And now I'm going to press enter. And this is what is going to happen that it is

going to go and understand what it was trying to do earlier. It is going to understand the entire project and create

a claw.md file which will act as a project memory like I said before as well. That means next time when I type

this command what do you know about this project? it will already know because the claw.md file gets loaded into the

session as soon as you start the session of cloud code or basically try to interact with it. So it knows in the

context what this project is and it is going to immediately give the answer. So it is exploring the entire codebase. Now

it is going to find out each and everything about this project and then create the claw.md file which act as the

permanent persistent memory for this project. And you can see now the claw.md file has been created. You can see it

has purpose, the commands, the architecture, configuration, output structure and key patterns. Let's see

where it is. So it has outputed the clot.md file over here. And this is how the claw.md file looks like. At the top

it says this file provides the guidance to cloud code when working with code in the repository. Content multiplier. This

what this project does. It's a content multiplier. transforms a single source document into 20 plus platform specific

content pieces using OpenAI's API. This is what the commands are like node these are some of the code files it has

identified how to use that and how to set up the environment environment variables and then we have architecture

over here then we have configuration then we have output structure and where the output structure should go in and

then it has also identified some of the key patterns that we use in our project so this is how the claw.md file gets

created but generally what I do is I come over here I right click and I create a folder called clot And I'll

move this clot.m MD file inside it. So clot.m MD file whenever you are going to reinitiate the session it is going to

get loaded from the dot code folder. So that's where basically you should keep it that everything can be organized into

dot cloud folder about the claude or about the claude code session itself because this is where we'll put the

skills the agents the hooks everything will go inside the dotcloud folder. So I like to keep the claw.md file inside the

dotcloud folder but you can put it into the parent project directory as well. Both the locations are correct. Both the

location clot code picks it. So now we know how to create clot.md file. Let's look at what should exactly go into the

clot.md file. The session context window is very very important and you should not pollute clot.md file with junk like

it should not be like hundreds of line. This was a very small project but there could be large projects of huge code

bases where this claw.md file could have been bloated a lot. In that case what should be the guidance? What should go

into the claw.md file? Well in the claw.md file I say three things should go. First one is the map. What this

project is what it teaches and where things live. This is something kind of a map or blueprint of your project that

you are providing in clot.md file so that clot code can quickly identify things quickly know what the project is

about and quickly know what the text tag that it uses. The second thing comes the intent where we talk about why the data

flows this way. What should never be touched upon without any kind of permission. All that things that you

have inside your head more like guidelines more like guidelines that your team is following something that it

should not be doing all that intent type of pieces should go into cloud.mmd file and finally the playbook like what kind

of patterns you would like to use to be followed in your project for example if I'm talking to perplexity which is an

MCP server do I need to make all the calls like all the models are exposed do I need to tell cloud code that this is

the pattern that I follow. I follow like two calls to perplexity not more than two calls and using these models. This

is how I club my queries. All this is more like a flow how your test will run. How the overall flow of smoke test would

be that is all you can also tell as a playbook in the cloud MD file. So this is very important to keep in mind and

I'll show you a template or best practices followed claw.md file in a while as well. But then let's talk about

the three level of memory that exists in cloud code. So first one we have already talked about that inside your project

you can have a claw.md file in the dotcloud folder or in the parent directory it will. So this only loads in

this project right. So this is like the level one kind of memory which we call project memory. So the second level is

the user level memory. So this is basically stored in your user home directory where the dotplot folder is

present and then you have a claw.md file over here. This is in your home directory and it applies to all the

projects. So something that you want to apply to all the projects you can create a claw.md file inside the home directory

of do.clot folder then all your projects will share it. Finally we have automemory which is like a recent

introduction to all this memory layer of cloud code which is called memory.md file and mostly you would see that cloud

code itself is managing this memory MD file. If you want to override some of the instruction or put some of the

instructions you can do that. However, keep it very very concise because treat it as like a much more mini version of

the claw.md file and this always gets loaded some patterns that you always want to be followed that can go into

memory md file or you can ask claw code to write into memory so that it remembers over a period of time if it is

making the same mistake again and again. So now moving on this is how I try to structure my claw.md file. First one is

the critical rules that like always use this kind of pattern for specific task never modify. So these kinds of critical

rules should go. These are like must follows. Then I talk about the project viewer view. There we talk about text

stack key directories data flow design decisions that I have taken. Then how to work here? What is like build test

deploy commands look like? What is the pattern to follow? What lot gets wrong like this is more like teaching or

learnings over a period of time that you will gather. So recurring mistake number one the fix add the list every time

claude repeats an error. So these are some of the instruction that keep on getting wrong and claude keep on doing

it wrong for me I'll keep on adding over here and finally I have this deeper docs where I would say for deployment refer

this point cloud to separate files instead of cramming everything over here. This is something called

progressive disclosure where you can have architecture.md testing.md and deployment.mmd being referred as files

as pointer to these files in your claw.md file. If you have a very large clot md file chances are some of the

instructions that are there are going to get missed and then you will feel that hey claw.md file is supposed to be

deterministic but why it is not following a rule that was already there. That is the reason because your claw.md

file could have been polluted or got too long too verbose for the whole context to be loaded and some of the context are

being overlooked by claw.md file and that's what I meant when I said it is almost deterministic not always and this

is the reason why it is not always deterministic because majority of the time as the project codebase grows the

claw.md file if you keep on asking it to write it on itself it is going to grow and it is going to bloat into a very

massive file and claude will also start giving you warning that it has gone too big for me to understand, comprehend and

act upon. So you need to optimize it. Then you can ask claude to optimize it as well. But you can follow this mental

model of what five essential things can go along with the principle that I use of progressive disclosure. So that you

are just providing the file pointers. So now it understand there is a file pointer to follow a testing pattern or a

workflow or a playbook. It will basically go and fetch that file at the time it needs it. It does not need to

know the testing pattern when it is not doing testing. So that's the reason you should only have pointers. Now if we go

back to the project over here, let me just delete everything in this cloud. MD file and paste something that I had

created. So as you can see that earlier if we go over here there are like 56 lines and again you know if I paste mine

there are these 40 lines. So that means 16 lines of instruction you have saved and it covers much more. It covers all

the critical rules that I wanted to follow clearly at the top. These are must follow ones. It has all the project

overview. This is all about how to work over here. All the references and then this is about what clot gets wrong for

me. Don't change import required. Don't call OpenAI directly. Always use OpenAI client wrapper. And then deeper doc

references where I'm talking about this is where my brand voice exists. This is where platform specific branch exist.

This is where full brand voice guideline exists. And why we are telling it is because if I start loading all these

into claw.md file, it is going to get bloated. So whenever it needs a brand voice or platform specific prompt, it is

going to know that okay, you know platform specific prompts are present in prompts.json. I'm going to load now and

read the prompt.json file in order to execute the task for the user. If there is no task of using a platform specific

thing while you are interacting with cloud code, there is no need to go or load into the context prompts do.json

file because at the end you have these 150 instructions in total and 50 used by system prompt. So you have like 100 to

150 for you and you have to make best use of it. So this was all about claw.md file. Let's move on to the second layer

which is skills. So in the skills I'm going to cover the commands as well and like I said in the newer version of claw

code commands have been merged into skills but the older hierarchy is still backward compatible. So in the dotcloud

folder over here you can basically have the commands as a folder and you can have skills as a folder both are going

to continue to work to support the existing projects or the projects that have been created before. So everything

in the legacy command folders are also being treated as skill. So that's like one thing that you should know before we

deep dive into what skills are. But what exactly is a skill? Well, skill is a modular set of instruction that you give

cloud code which are repeatable workflows that can be executed over and over again. It can be as simple as a

simple prompt that use on a daily basis. For example, you might have created a prompt to check your brand voice, right?

So how your brand voices you have configured into pram and then you specify a link or you specify some text

as an input and you say that hey run a brand voice audit on this particular text or this particular link of mine

then claude code is going to do that. What you might be doing is you might be like saving that prompt somewhere in

your prompt pack or prompt library but claude code allows you to create a skill of it which could be like brand voice

audit skill. So that's where you extend the claude codes functionality or core powers into the skill because now you

are providing your own domain expertise or your knowledge to claude that this is how you do a thing or someone in your

team do a thing or you want to do everything consistently throughout the team. So you can define that as a skill

and everybody in your team can use that skill. So the same task can be performed in the exact same way that you have

defined. So this is more like your workflow routine or any type of task that you keep on repeating with the

domain knowledge with the expertise that you have with the flavors that you have you can put together into a skill which

is nothing but a YAML formatted piece of text that I'm going to show you. And as you can see the skills have two

invocation modes. One is the user invocable skill where you type in forward slash and the skill name appears

and then you run it because you have the control you want to execute a particular skill at the moment that you want. So

the control fires when you want to fire it manually. And the second method is auto invocable. This is where Claude

already understands that hey I have this kind of a skill with me which is more like brand voice audit. Why not I

execute this skill as Dhiraj is asking me to write a draft on LinkedIn. One workflow is that it drafts a LinkedIn

post and then I invoke a skill of brand voice audit. Right? So that's like one way of doing it that you are involved in

it. So that could be an example if you have configured your skill in user invocable fashion. But if you have

configured a skill in auto invocable fashion, then claude knows already that skill because it is going to load the

context which is there right at the top like the description and all into its memory and it knows that I have these

skills available. I'm going to basically auto invoke it because I know that at the end of it I'm going to as a user the

is going to ask me to do the brand voice audit. So it is acting on behalf of you already. So that is what the power lies

when you start systemizing things. These skills are very very important and you should able to understand what you

should give powers to clot code to auto invoke and what you should keep it to yourself. Now let's go ahead and build

some skills for this particular project. So what I'm going to do is come over here and click on this and create a new

folder called skills. Then in the skills folder I'm going to create another folder called quick audit. And over here

I'm going to create a new file called skill.md. I have already created a skill. So I'm

going to paste it over here. So this is where I paste it. So you can see that now I have description quickly audit the

draft article against brand voice structure and rules. Use forward/quick audit path to check any draft. The mode

is user invocable. So basically I think I have it written underscore. This should be like that. So user-ash

invocable this is how you structure or tell it that this is a skill that user will invoke this is not a skill that

claude will invoke. Now in the quick audit I have defined the prompt and the output looks like. So this is more like

a simple skill more like a prompt that I'm going to use continuously. So this skill goes over here and what we are

going to do is quickly create another skill. While we are at it, I'm going to create another folder brand voice check.

And again inside that we'll create a file called skill.md file and I'm going to paste it over here. So now you can

see I have description automatically checks brand voice compliance when claude writes or edits markdown files in

drafts or output directory. So we have draft directory or output directory. It is going to automatically flag the

forbidden phrases, long paragraphs and weak openings. And this is where the entire prompt is. You can see that I

have also provided the full brand guide at. So this is present in some other project of mine where brand knowledge is

there and this is where the entire brand voice guidelines are. So these are some of the core rules that should be

applied. These are brand voice config which is there. So this is like in the config and this is like the brand voice

key. So if violation found this is what happens. So these are the two skills that we manually created or you may be

able to download someone else or maybe you are able to use this skill. This has the right structure. However, what you

can also do is you can write something called plug-in and then you can go to the marketplace. So, this is where you

can add more powers to cla code by installing some of the plugins and the skills which are there. Over here, I'm

on the discover tab of plug-in. So, I can explore what all exist or browse through it. These are the ones that I

have already installed. This is a marketplace where these plugins or skills or agents are present. So, you

can like see there is this Claude plugins official marketplace. So I'm going to enter that and here you would

see that there is something called skill creator. So this is the skill creator command or plug-in that you are going to

use in order to basically create the skill for yourself. So what I'm going to do is press escape couple of times to

get out of it and I can say skill creator. So you can see that I have like couple of uh skill creators over here. I

think like it's overriding some of them. So I can like just mention skill creator document skills. So now it has invoked

the skill creator right. So it is basically going to ask what skill I have seen that you have loaded skill creator

what skill would you like to create or improve. So you can improve the existing skills as per the claude code guidelines

or you can create a new skill itself. So if you are going to define what you want the skill for you can paste your prompt

over here. It is going to convert the skill for you and add it either at the user level or at the project level. You

have to just guide it. So that's how you create the skills. Now let's whatever skill we have created, let's see them in

action. And before I do that, we have polluted this context or main context window with multiple types of things. So

what I'm going to do is show you the power of clear command. So as soon as I enter, it has cleared the entire

context. Now it is fresh and clean. So now the skills are there. Let's reload the claude code for this particular

session. So I'm going to exit out. So now I'm going to type again. So now I'm in another instance of clot. So now if I

type quick audit. So you can see that quick audit is now visible into the commands or I can invoke this skill. So

I can just say tab. Now I can execute it in the acceptedit mode. Earlier the whole quick audit used to take like four

or five prompts where I was in between copy pasting from one window to another. Now it is executed as a workflow with

just one single command because we enabled this command as user invocable. So I am the one who is going to execute

it whenever I need to do a quick audit of any of the draft that I have created. So this saves a lot of manual work. This

is where you step into the world of creating systems for you systems for your workflow. If I see the brand voice

check, you can see user invocable is false because I want claude to invoke it at any moment claude is creating any

type of draft for me or any kind of output it is editing in these directories that I have defined. So this

is more like automated way of enforcing the quality in place. So more power to your cloud code ecosystem that you are

building. So let's come back and quickly run a quick audit. So what I'll do is I'll just paste it here. So now the

skill is involved. It is doing a quick audit on our sample article and you can see that quick audit this is what the

critical issues it has found. So has an overall score of 4x 10. These are some of the major issues. These are some of

the critical issues. I did not write any kind of prompt. Basically we had the skill that got invoked and it did the

entire audit for us. You can see critical major minor issues all have been found. So that you can basically

have another skill that can fix it as well. So this was user invocable skill in action. Let's see what is like the

auto invocable skill or claw invocable skill looks like. So for that I'll just going to write a prompt. Write a

LinkedIn post about why consistency beats perfection for content creator. Save it to drafts folder.

LinkedIn-consistency.md file. So now we are writing to the draft folder and that's what we have in our

skill file that anything that is written into the draft folder must be flagged or must be run a brand voice check. So now

claude is writing this LinkedIn post for us. Let's see what it does. So right now it you can see that it has gone directly

to config.json because in the claw.md file we have mentioned that if you want a

specifically prompt kind of a thing then go over there. So you can see that it has written the draft and then it has

also done a quick selfch check against the brand rules and it has found out all these issues against the brand rules. So

that is how the skill got automatically invoked and you can see the power behind now the system that you are creating.

Now let's move on to the third extension layer of cloud code which is hooks. These are the rules or guardrails that

are going to always get invoked and claude code cannot escape it. How a hook is structured? You can have a hook with

event plus a matcher and a script. So event is when does it fire? For example, pre-tool use, post tool use, stop and

there are 15 more events which are there in the documentation. Matcher is which tools trigger it. For example, using

bash or edit write or any other tool name that you can think of. Then finally we have script. what it does like what

the hook does whether it floors whether it blocks whether it validates. So this is what a hook structure looks like and

the exit codes are more like everything is fine so you can go ahead and proceed. So at any moment for example this event

of pre-tool use or post tool use the script or the hook is going to get evaluated and if it exits everything is

fine then only clawed code further action will happen otherwise it would get blocked. If the exit code is two,

other is more like warning but it does not get blocked. So it will just note as a warning. And then there are four types

of hooks available. First one is the command. It runs a shell script which is the most common form of the hook and

that's what probably we are going to create. Second one is the HTTP. It calls a URL endpoint for you. Finally, it is a

prompt. It ask to evaluate something and then agent delegates to an agent. These are four types of things that you can

invoke. I'm going to cover the topics of hook in the next lesson. We are going to deep dive into all the 18 events that

are there, the entire life cycle of it and everything. But in this video, I'm going to keep it brief. So, let's get

into the visual studio and create our first hook. So, let me go ahead and so this is like our project over here. So,

again like for the hook as well, what you have to do is in the dotload folder itself, we are going to have the hooks

too. Again, you right click, you create a new folder called hooks. create a new file block dangerous actions sh. So this

is a script file that we are making and I'm going to paste this that I have already created. So what this script is

going to do is block some of the dangerous patterns that cloud code can execute without asking you while

executing a particular set of instructions. For example, remove RF like removing everything from particular

folder. Get push force, get reset, get clean, drop table, drop database. So you can add all the commands which are

dangerous in nature that you never want cloud code to execute. You want full control over them. So we have created

the hook file. But where does this hook file get configured? How does clude knows about it? That's where

settings.json will come into play for this particular project. And we are going to create a file called

settings.json. I'm going to like save this. And in the dotcloud folder itself, we are going to create a file called

settings dot. So in this file, I'm going to paste the hook configuration. So you can see that in the settings.json which

is in the dotcloud folder. We are saying hooks. This is where the hooks will start. All the hooks will get defined in

the settings folder. And I am creating this in the project folder of content multiplier. But I can easily create this

setting do.json or edit the setting do.json of my home directories do.tcloud folder where the settings do.json is

also present. I can add this configuration into that so that all my projects follow. As of now this project

is going to only execute this hook not all my project but for now this is just an example and you can extend it. So I

have pre-tools use as the event. So at the time of any type of tool usage it is going to execute the matcher is bash. So

I want like see if any kind of bash command is being invoked just before using this bash tool. Then the hook is

type is command. So I'm going to execute a command which is like a shell script. Where it is located? The command is

located in the project directory dotplot hooks block dangerous. I think I have some different name. So I'm going to

like say rename and then come back over here and update the name over here. Now I'm going to save it. So now our hook is

ready. Our cloud code has been configured. Let's restart the session so that it gets loaded like the settings

get loaded. So I'm going to like say exit. So it has exited. Let's clear. So I have resetted the clot just to double

ensure that settings have been loaded. So what I'm going to do is I'm going to say let's clean up the folder. Delete

everything in output folder. So we can start a fresh and let's see if our hooks fire or not. All right. Right. So it

says that output folder is already empty. So what I'm going to do is I'm going to restore this folder and let's

try again. Right? So you can see that it is like picking it from the context itself. So

let me exit this and then let me just go back and start again and let's exit the plan mode. Let's say it again. So now it

is trying to figure out how many files are there. So you can see that it has 24 files, six directories. So you can see

that the hook got called over here. It has blocked the action attempted to delete files in protected

directory/output. The output and draft folders are protected. User targeted delete for explicit approval. So that's

how you can see that hooks can protect you from some dangerous action or it can perform some automated task as well.

This is just more like a dangerous action but you can automate certain task at the time something is happening in

your workflow or as claude is executing in those 18 events that we are going to discuss in the next lesson. Now let's

move on to the fourth and final extension of cloud code that is agents. They are also called cloud code sub

aents. They have their own context window. So that the context window that you are using in your main session does

not need to share any kind of context length with these agents. They can work in parallel for you perform the

specialist job. If there is an output to collect, collect the output and return it back to the main session that is

there. So that's the reason they are also called sub agents as well that they are being orchestrated by your main

session as well. However, you can create an orchestrator agent that can orchestrate the other agents as well and

you can control everything from your main window. The main thing that you need to remember is that all these sub

agents because they use the context windows of their own isolated from each other. You can achieve much more things

in parallel while these agents work for you. So this is how the agent definition looks like. Again a plain markdown YAML

type of a file where you have a name which is more like researcher description. You can have a description

tools that you want to expose to this agent and the model. Model could be sonnet or it could be opus. Whatever

model you will define, agent will use that model of entropy and execute the task for you. But I'll keep it very

brief because in the coming lessons of this full course of cloud code, I'm going to build three live agents in

front of you in a lesson itself so that you can follow along and we can do a deep dive into the agents. how you can

create them, how you can structure them, why you should create them, when you should create them, all that we are

going to have a deep dive much like we'll have a deep dive of hooks in the next lesson itself. So now let's talk

about a decision framework that you should use in order to pick what, when, and how. So this is what I try to

follow. Does it need to run every time? No exceptions. Right? If the answer is yes, it is about the project context,

then I'm going to use clot.md file. things goes over there and if this is not a project context then I'm going to

use a hook over here again if the answer to the parent question is no then does it own its own context window then I'm

going to use yes as an agent if you want to perform within the main context window some sort of an task that could

be auto invoked as well then I'm going to have a skill in place so that's a decision model that I follow and if I

continue further you can see that again you know as recapping plot md file deterministic more or less like I said

it is going to always load but depending upon the lines or the heavier file or what you have in claw.md file some

instructions can get skipped so trigger is like always loaded project auto and user scope user skill probabilistic you

can have /name so these are user invocable then you have claw auto skills as well again probabilistic context

matching needs to happen project or globally it can be created when it says globally it is at a user level that goes

into the claw folder that is present in user home directory. Then you have hook which is always deterministic. You have

an event plus a matcher and then the hook gets triggered as we just saw. The scope can be project or global. And

finally there is agent. Again this agent is also probabilistic unless until you define or specify in your instructions

to claim otherwise it will always be probabilistic in nature. The trigger is

more like delegation. they own their own context window and run in their own context window. So this is how you need

to start thinking in terms when you are designing your AI systems that hey which component or which extension of claude

code I have to use in what type of work that I want to do in a particular project. So this is how a project

structure with all the extension stack looks like. The claude skills agents will also come over here as we come to

that lesson. These are two skills that we created. We created a hooks file. We updated the settings do.json over here.

Finally, this is more like a cheat sheet for you that I'm going to share in the description of this video as well and

the entire slide deck. You can also refer to the entire guide that I have created on my Substack. So this whole

guide can you can read as well. You can go ahead and subscribe to my Substack and read it over there if you are more

like a reader. So the project remembers who you are now, what you want to do in this project. It has skills that you can

call with slash command or with the intent. It has a hook in place that protects you from writing wrong things

or with the wrong brand voice. However, it is still locked to your system or the laptop or the MacBook that you are

using. It still cannot access the internet, the LinkedIn, the Substack drafts that are there. It cannot still

extract them and [music] repurpose the content that you have already published over the internet. It cannot see the

notion calendar that you may have. what kind of content you want to repurpose in time to come. So in this lesson number

three, [music] you solve this problem and you connect the AI to the outer world. I call it external intelligence