CLAUDE CODE Full Course For Beginners (DATA DOMAIN Edition)

Everyone wants to learn plot code in 226 because of its growing demand. But where

is the structure guide? What are the resources I need to refer? What about

prerequisites and how to start? I know I know this is a situation right now. That

is why I have created this complete beginnerfriendly 7 hours long claude code masterclass for you where you will learn claude versus claude code claude code architecture cla code agentic tools

cla.md5 file building customized scales

cla.md5 file building customized scales in clot code plugins and marketplace building sub aents in clot code attaching hooks for automation connecting MCP servers with clot code

local MCP servers and hosted MCP servers and and and much more. I literally

covered everything that you need to know as a developer with respect to clot code. The beauty of this course is cool

code. The beauty of this course is cool handdrawn doodle visuals which help you understand complex concepts with ease and not just this all of the notes all

of the code files all of the resources as well that we are going to follow in this particular course will be provided to you for free. In short, you will start this master class as a total beginner who doesn't even know the

meaning of claude and you will end up becoming a confident industry ready claude code developer. And I'm serious.

I have literally put my heart while creating this amazing claude code masterass. So if you want to support my

masterass. So if you want to support my hard work, simply hit the subscribe button right now and just write I love claude code in the comment section and let's get started. So what's up? What's

up? What's up fam? Happy Sunday and welcome to this amazing Claude code master class. I know I know I know you

master class. I know I know I know you are very very excited to learn claude code because claude code is non-negotiable in 26. Let me just tell

you it was good to have in 2025. In 204

it was like very nice to have but nowadays it is like non-negotiable. If

you do not know about clot code, people will treat like this person doesn't know clot code. Can you imagine? Do not

clot code. Can you imagine? Do not

worry, do not worry. An Lamba is here and I will just tell you everything in cloud code that I know personally. Okay,

because claude code has so many things, so many things. So many things and as I just mentioned, um I have also added the recent upgrades in the claw code that just happened last week. So all of these

upgrades will be covered in this video as well. And before wasting your time, I

as well. And before wasting your time, I just want to tell you the prerequisites because prerequisites prerequisites are very important for this video. What should you know? um

this video. What should you know? um

very basic Python language and that's it very basic very basic how to print how to do this thing and that's it because claw code is totally a different thing we work with LLMs we work with tools but

when you know some basic programming language basics very basics then then you can just understand what's going on behind the scenes because this course is not just showing you how to make LLM

calls how to get the response not at all we going to level up really really quick but yes we will start from scratch that means Means you do not need any prior knowledge in claude. You do not need to know what is entropic. You do not need

to know what is claude. But you do not need to know what are cloud lms. You do not need to know what are claude um code. What are the tools available? Um

code. What are the tools available? Um

what is claude desktop application?

Anything. You do not need to know anything related to claude. Just basic

Python language is more than enough.

More than enough. Perfect. Now let's

quickly see the agenda of this video.

Agenda of this video is very very simple. You are a total beginner. Even

simple. You are a total beginner. Even

if you know some things I would say bro this video has covered so much of things. So do not feel like I already

things. So do not feel like I already know MCPS I already know about claude code. So this is this video for me

code. So this is this video for me obviously. Yes. So if you are a beginner

obviously. Yes. So if you are a beginner yes I am assuming that you are a beginner. So this video will make you a

beginner. So this video will make you a confident claude code developer. You

will be able to integrate claude code in your day-to-day activities. If you are a developer in any domain, data domain, software domain, any domain you can integrate clot code. And I will just show you how you can use clot code

because once you understand clot code how does it behave you can just use it anywhere in your day-to-day activities in your anything system. Okay, simple,

very good. So, this is the entire agenda of this particular video. And yes, as I just mentioned, I will be using amazing cool cool graphics for this particular illustration so that you can understand

the concepts and see not just understanding it's also about retaining the information in your memory and when you just learn from these graphics from these live illustrations not like

pre-built illustrations and not just the diagrams which are available in the official documentation, you will actually understand the flow. So that is why I thought like why to why why not to just create our own uh illustration instead of just referring to the

documentation. Yes, we'll be referring

documentation. Yes, we'll be referring to the documentation whenever it is needed for syntax for definitions. But

yeah, for understanding point of view, we will be just using these cool graphics. Okay, and let me just give you

graphics. Okay, and let me just give you a quick sneak peek about our um working directory that we have built in this course. So we will be just building

course. So we will be just building entire thing from scratch. You can see agents, you can see agent memory, hooks, skills claw.md settings.json whatever

skills claw.md settings.json whatever

we have built here will be built from scratch and everything is available in my GitHub repo as well. You can just clone it. But you would not require to

clone it. But you would not require to clone it because I will explain you everything. So you can just build it on

everything. So you can just build it on your own as well. Simple, simple,

simple, simple. Okay, very good. Now

next agenda is clear. Um, everything is clear. So now what? Now before

clear. So now what? Now before

proceeding I will just take a few minutes just to congratulate the people my fam my my people who have just cracked their offer letters from their dream job and they're really really

happy. So I want to just highlight the

happy. So I want to just highlight the success of my fam. So congratulations to all of these people.

Wow wow wow wow. list is very long and I want this list to be very very very very very long, right? Okay. So yes, if you

also want to just succeed in data engineering domain especially and I would say AI data engineering domain, just hit the subscribe button right now because this particular channel will

help you a lot to grow in this domain because this channel um publishes all of the in demand tools and technologies, tutorials, projects, everything related to AI and data engineering. Simple. So,

simply hit the subscribe button and drop a lovely comment in the comment video.

Comment video in the comment section.

Okay. And at the end, this is me. This

is me, this is me. And that's it.

Simple. Let's get started with the master class. So, let's get started with

master class. So, let's get started with our claude code master class. First of

all, anal is claude code and what is claude? What is claude models? Because

claude? What is claude models? Because

everywhere we go we see claude and there's like new name for claude and something claude and something claude and something and you know what you need to understand the difference between all

of these major components and then only you'll be understand claude code in detail common sense right so I will assume that you have never ever heard

this word called claude code okay can I just make this assumption even if you have heard it's good but let's say you even haven't heard about Claude. Okay.

Ever. So everything was started everything was started um just few years ago. So everything was started with

ago. So everything was started with something called as claude. Now what is claude? Claude is nothing but you can

claude? Claude is nothing but you can say a component a product by a company called entropic.

By called entropic. Okay. So everything

was actually started with entropic.

Okay, simple. So, entropic

and if you have not heard about entropic, it's fine. It's totally fine and I will just tell you what is entropic as well. So anthropic is basically you can say another AI company

that is like booming right now similar to open AI which has like LLM models large models but those models are very compatible for the coding very

compatible coding and day-to-day activities other than just doing the research work those models can do a lot of um research as well but the major and primary goal is like coding side more

okay so entropic has some models entropic has some what models models.

Very good. What kind of models? Let's

forget about claude as well for now. So,

first of all, they launched some models.

So, let's say they have a model called um they have basically three types of models. So, first is haiku. Where is my

models. So, first is haiku. Where is my font? Where's my font? Okay. So, first

font? Where's my font? Okay. So, first

is haiku.

So, this is first type of LLM model that they have. Okay. This is the smallest

they have. Okay. This is the smallest one. Smallest means like not very very

one. Smallest means like not very very small as compared to other two categories but haiko is also very powerful because entropic is entropic man and you will get to know a lot of

things and I think you already know because I see your love already on the you can say promotions for entropic lord or whatever. So, haiku is the first type

or whatever. So, haiku is the first type of um category you can say and the second category is sonnet which is my personal favorite

because sonnet is the balanced approach not very small not very parametric high perfect for all the work. So my personal favorite is sonnet okay and then we have

one more type which is like very very very optimized models which are actually very very good. Okay. And these are called

very good. Okay. And these are called opus.

Perfect. So this is the paradigum of entropic. And they came up with these

entropic. And they came up with these three major models or basically these three categories basically. Okay. These

three categories 1 2 and three. Now anal

where is claude in between this? Okay.

So entropic came up with this particular paradigm. Haiku sonnet opus. And even if

paradigm. Haiku sonnet opus. And even if I just go to any browser and if I just search let's say um anthropic

models you will see all of the models will be based on these three names um reject all cookies um where is the models model model so claude 4.6 6.

Where are the models, bro? We need

models. Oh, yeah. So, currently they have Opus 46, Sonnet, Haiku, Opus 46, Sonnet, HiQ, Haiku 45 and whatever. So,

they have all of these different one variants, but all of the models are based on these three categories. Haiku,

Sonnet, Opus. Haiku is the smallest one.

Sonnet is the balanced one. Opus is the biggest one. Simple. Now Claude came up

biggest one. Simple. Now Claude came up into the picture because all of the model used the name called Claude Claude

Haiku Claude Sonnet Claude Opus similar to Open AI. Open AI is the parent

company but all of the models start with GPT GPT4 GPT 5.2 Cordex GPT5 like this.

So similarly entropic is the parent company and all of the models start with claude claude haiku claude sonnet claude opus. So this was the beginning point of

opus. So this was the beginning point of claude make sense. So I can put claude in

make sense. So I can put claude in between this because claude is nothing but just the term that he that they use simple.

Okay.

Okay. So is this a claw code bro? Hold

on. This is not claw code. This is just the claude claude paradigm. Okay. We

need to learn step by step everything.

Okay. So

perfect. Let me just remove this.

Yeah. So

I hope you understood this. Simple. So

this is your cloth. Now when they made these models obviously we were able we who developers we were able to use

clawed models through API calls we were able to integrate the clawed models in our chat bots all of these things were happening where we were just making API calls we were getting the response back

perfect now they thought that openai has a chatbot obviously it's called chat GPD so they thought we should also create a you can say um chat chatbot where people

can directly access our models and they can just get the answers back in return like day-to-day Q&A, day-to-day research, day-to-day questions, anything right so they came up with this concept

called claude desktop.

Claude desktop so this is the application that they created or basically a wrapper they created on top of claw on on top of their claw models.

So they are saying bring claw to your desktop. Simple very good. So you can

desktop. Simple very good. So you can just download in Windows, Mac, Windows ARM. So this is the Q&A version. Q&A

ARM. So this is the Q&A version. Q&A

version like basically you can simply ask the question similar to chat GPD. If

I open chat GPT so this is what this is just a wrapper. This is just a front- end application built on top of what the models which models GPD 4, GPD 5, GPD

4.1 whatever. Simple. So clawed desktop

4.1 whatever. Simple. So clawed desktop came up with this solution called clawed desktop application. There's a web

desktop application. There's a web version available for it. But people

love using it on their local desktop.

Why? Because it is very very powerful and very easy to use. So they just think like we can directly download the application because when you download the application now you will know the claw code as well and so on. So that is

very powerful. So people love using it

very powerful. So people love using it on their desktop. I have already downloaded it on my desktop. So I also love using it on my desktop. Simple. So

don't worry I will just ask you to download it but not not now I want to explain a lot of things before that.

Perfect. So now this is our claw desktop. So this one this was the first

desktop. So this one this was the first part of it like claw desktop. After

these models okay they launched something called as clawed let's say desktop app.

Simple. This was created.

This was created. Okay. Perfect.

Simple. So now we have app. Now we can do what with this app. We can let's say you are a user. If you have any question

you can say um hey what is the syntax of this particular code? Hey what is the uh you can say I am facing this error. Can

you please solve it for me? Whatever you

want to ask. So you can literally go to any cloud model. Let's say you picked um let's say this model. This is one of the cloud model and

you simply picked let's say haiku haiku 45 simple so you pick this model and what is this model just you can say the llm

and then you will get the response back simple this is the flow this is the flow simple cloud desktop application this is

the sole reason behind building the application What? Then something great

application What? Then something great happened. What? Now something happened

happened. What? Now something happened and you already know the answer. Claude

code.

Now you already understood what is claude like what is the beginning point of claude and what is claw desktop. Now

you can understand the claude desktop.

Now you were able to access this claude application. Okay. You were asking hey

application. Okay. You were asking hey um what is let let's say I'm asking questions like this. Let me just open my cloud code. Okay, I will I will just

cloud code. Okay, I will I will just tell you how you can also download it.

Don't worry at all. So this is my clot code or not clot code just clot desktop and two free version. It is not a paid or upgraded version. Okay. And you do not need that like you can get that but

we will just talk about that in the video as well. So now this is my claw desktop application. Okay. So I can

desktop application. Okay. So I can literally ask anything. I will simply say um bro I am

facing an error while writing my pispar code. So I will just give this prompt prompt and I will get

the response. Simple. Very good. Now I

the response. Simple. Very good. Now I

was also asking some questions related to my desktop application or you can say desktop work. Let's say um I want to

desktop work. Let's say um I want to copy all of my just ignore this MCP server

because I was building on MCP.

So I want to copy all of my uh let's say files to a destination folder

but only the let's say CSV files I ask this thing okay and I can pick the model so I can pick let's say haiku hiku is a smaller version by the way I'm using free version so let's use 46 why to save

the tokens right we are not paying for it for now so let's say CSV files so I just literally So I have got the response. What is the response? I would be happy to help you

response? I would be happy to help you copy only CSV file to a destination folder. Could you provide a few details?

folder. Could you provide a few details?

What are the source destination folder paths? So I can just give them as well.

paths? So I can just give them as well.

Like source path is this, destination path is this. And you already know the answer like what will be the answer. It

will give me the step-by-step implementation. Anala the same thing we

implementation. Anala the same thing we can do with chat gpt as well. Exactly.

Exactly. That's why there's no difference between cloud desktop and chat GPD for now. Okay. for now. Okay.

Now what happened? Now what happened and let's see let's let's try to first of all explore this thing cloud desktop. So

I want you to download cloud desktop application right now and just create an account in that because it will be required for the next discussion. So how

you can just download the cloud desktop application. Simply go to cloud download

application. Simply go to cloud download let's say cloud desktop application because there are two things claude code and claw cloud download. You need to download claw desktop application. So

simply say claw desktop application and then download.

Okay. Then you will land on this thing and make sure you are downloading the application bring claw to your desktop.

Simple and then you can click on Windows or Mac whatever your system is. So click

on this thing and I have already downloaded it. So it will again download

downloaded it. So it will again download it but it's fine. So you can install this application and then you just need to sign in like normally you do and just

make sure that you are able to use cloud desktop application simple just try to do that because I want to show you some small small things because these are required for your learning journey for

this clot code so here when you'll be creating your clot account you will see this thing which is you can say your free version use clot for free pro version and max version. So now the

thing is for now for now you just need to pick the use claude for free when we'll be just working and learning about claude code then it will be your choice if you want to go for pro version or for

the API version. A what is API version?

API version means like you just create the API and use it like the other models. So it's not very very special

models. So it's not very very special thing that you are hearing for the first time. But for now you do not need to

time. But for now you do not need to worry at all. Simply go for clot for free. Simple because here you can simply

free. Simple because here you can simply say create create edit content create code visualize data and whatever the chatbot can do. Okay. So let me just

click on claude for free. Okay. And you

can just ignore this thing. Why? Because

I was just building some MCPS. So it is saying ask me anything and I understand.

Okay. I understand. So this is my very first chat. So I can say enter your name

first chat. So I can say enter your name Lamba.

Do you want bro?

Let's go. All set.

Oh man, I just created this new account just for you so that you can see this thing as a new thing. And let me just delete this MCP for now. And yeah, so

this is the very first ad that you it will just generate for you. Okay. And I

just clicked on randomly because it was just keep on asking something. So I can simply click on delete. Simple. So this

is the homepage where you can see the claw code. Simple. So here this is the chat

code. Simple. So here this is the chat version and these are all of the models that you can use. In the free version you can use haiku and sonnet. Okay good.

Now the thing is now the thing is first of all you need to click on this particular icon. Okay. And here you will

particular icon. Okay. And here you will see all of the details related to your account. Simple. Here if you want to

account. Simple. Here if you want to upgrade plan anytime you can click on this button. You can upgrade it. It's

this button. You can upgrade it. It's

your choice. And first of all I will go to settings because I want to show you something. So if I go to settings this

something. So if I go to settings this is my settings page where you can just change so many things in cloud desktop application as for your convenience. So

obviously I'm not going to give you this particular tutorial. I will directly

particular tutorial. I will directly take you to the developer option because we are the developers. So here here what is this thing developer. So the this is the area that we will be using when we

will be working with claude quad and we'll be building MCPS. And if you do not know about MCPS, do not worry. But

I'm so sure that you would have heard about this word MCP. MCP MCP. So I'm

just giving you the um hint that we will be just building MCPS as well. So do not need to worry. So for now I can simply click on delete. And why this is failing? This is failing because I have

failing? This is failing because I have not turned on my MCP server because I'm running MCP locally. So that is why it is just like giving this error. See it

is running on my local host and it is like turned off right now. But do not worry at all. I will just create a new one from scratch. So it is fine. So this

is the thing. Okay. So now just tell me An Lamba this claw desktop application is done. You can just go to chats. I can

is done. You can just go to chats. I can

just talk to this particular chat and everything. Where does claw code come

everything. Where does claw code come into the picture? Because everyone is telling that we can use claude code in cloud desktop application. And what is

cloud code? Yes. Now your groundbased

cloud code? Yes. Now your groundbased fundamental or you can say highlevel overview around cloud is completed. Now

you know what is clawed and what is clawed desktop application and what are those clawed terminologies that we use.

Okay. Now you are all set to understand the technical areas. Okay. But let me just tell you one more thing when you be just talking about sign in completed.

Yes we know sign in completed. Okay let

me just close this. Okay perfect. So

when we just working about working along claw code the learning graph for for claw code is not like this let's say like this no do you know what is the

learning graph for claw code because I'm preparing you beforehand so learning graph is like this it starts from here it goes like this it goes like this why because you'll be you'll be feeling like

hey where I'm just progressing to you are yes progressing but there are so many things that you'll be just circling back and you'll be just exploring it again and again and again because this

is the thing. It is what it is. Claude

code is not like very simple thing or very easy thing that goes in the straight line. But because you are

straight line. But because you are landed on this video, you should say that this is the most easiest thing.

Why? Because we are on this channel, we are on this video. Simple. Okay. But do

not worry. Even if you feel sometime that hey what is this? It's fine. when

we were learning about these we were also saying hey what is this so do not worry at all just re-watch some segments of the video practice more and that's it simple good so now let's do one thing

now let's directly jump onto the claude code now what is claude code now after developing the claw desktop application and let me just say this thing claude

code is amazing amazing but claude code is not the only thing or you can say the only application or any you can say um

component in the AI which is the sole development or you can say sole developer of this particular thing. Let

me just tell you what do I mean. So

let's copy this thing cla code. Now what

is claude code? This is very interesting. Okay. Because now you

interesting. Okay. Because now you understood the hierarchy. Now what? Now

what is the hierarchy? So now after cloud desktop we got claude code. Okay.

We got claw code. Now what happened in claude code? Claw team said we are

claude code? Claw team said we are providing our models in the form of claude like claude haiku claude sonet

blah blah blah. Any user can ask anything from this. Yes. And as we know that it is the fundamental behavior of

any LLM that it generates tokens, it generates text, it responses us back. We

all know that. Very good. Now this

person says hey let's say this person says what is um let's say pi spark

okay so claude will say um pi spark is and very long definition simple so this

person will get the response back oh simple this is very simple okay perfect now if this person says um claude

Just or you can say please create a pispark script

for me. What? What? What did he say?

for me. What? What? What did he say?

Create a pispark script for me. Who are

you? Who are you? If you want to ask me something, if you want to ask me for any code, feel free to do that. I will

provide it provide you to the I will provide you the code.

But now you are asking me to create a piscript pi pispark script for you. This

is one step ahead because claude can generate the code.

Simple claude can generate the code.

Claude cannot generate a file. Claude

cannot generate a script for you. It can

generate the code only. But now you are saying create a pi pispark script for me. On top of it let's add some more

me. On top of it let's add some more challenges. Please create a pispark

challenges. Please create a pispark script for me and and let's say and copy or let's say put the

script in in the let's say scripts folder. Oh, so

now this person is saying create a script and also put the script in the scripts folder.

Oh, okay. And let's say if the scripts folder is not there, Claude needs to create that folder for that person as well. Is it possible? No, not at all.

well. Is it possible? No, not at all.

LLM is just a model which responds us.

That's it. Can Chad GPD do this thing?

Obviously not. Gemini, no. No one can do this thing. But now, let me just say

this thing. But now, let me just say that it can do this thing. What? Yes.

That's what claude code is. So now what will happen? This claud is LM. Yes.

will happen? This claud is LM. Yes.

Claude is very smart is like one of the best coding agents that we have coding LLM sonnet 45 or 46 the latest one 4.6

the best one obviously opus 4.6 is the best one but I'm just talking from a you can say developer point of view when you need to just like use that model personally okay because opus 4.6 is like

way too expensive. Sonnet is very good.

Simple. So now Claude has the best LLM.

That means Claude can write code for anything efficiently.

Now what Claw team did? Claw team

created some tools. What tools? Okay.

Tool one.

Okay. Tool like this.

Tool like this. Tool like this. Tool

like this. Okay.

Now what will happen? So these tools are attached to the clot. Simple.

Okay. Now there are some built-in tools like so many built-in tools not just four. But for example let's say we have

four. But for example let's say we have these four tools. This tool,

this tool can perform um file system level tasks.

Simple. This tool can do that. How? If

you have some fundamental knowledge about Python or any programming language. So, we know that we always get

language. So, we know that we always get some OS level modules. Let's say in the Python we have OS module. With the help of OS module, we can create the folder.

We can create the file. We can put the content into that file. Simple. This

shows your fundamentals of Python. If

you have fundamental knowledge, it's fine. Okay. Now, this is fine. Okay.

fine. Okay. Now, this is fine. Okay.

This is fine. This tool can perform that task. Similarly, there are some tools

task. Similarly, there are some tools which can let's say um access your website or all of these these things.

But let's keep it very simple. So, let's

say this tool is there which can perform those things. So when claude needs to

those things. So when claude needs to perform any task related to file system related to that person's uh you can say file system folder files whatever claude

lm will generate the code and it will hand over that code to tool basically it will make a tool call that tool call

will perform the file level tasks how how is it possible because claw code

claw code lives in your terminal. Clawed

code lives in your system.

Okay, that means it can run bash commands. That means it can run all the

commands. That means it can run all the subprocess commands. Subprocess commands

subprocess commands. Subprocess commands means let's say it wants to create directory. It will simply say MKDS and

directory. It will simply say MKDS and create the directory whatever. Okay, so

it can perform those things.

That is why that is why clawed code cannot be accessed on the web. It can

only be accessed on the terminal.

Wherever we have the terminal, wherever we can execute the commands.

Make sense? Make sense? So that is why Claude has created these amazing tools like next level tools. Next level tools.

Okay. And these tools are actually performing those tasks and they are leveraging the LLM model under the hood.

This is your claw code. When they

combine this clawed model with these tools, it becomes clawed code. This is

the most simplest explanation that you would hear for claw code.

Now ana how do you know that it is connected to tool? How do you know this thing? Because see if you have developed

thing? Because see if you have developed some or you if you have some basic knowledge about agentic AI how agentic behave if you have worked with langraph lang lang chain is not like very very

agentic but let's say lang graph if you have developed those agents you would know that it is not magic you can do the same thing by yourself as and don't worry I will also show you my light

version of cloud code I have just created just for this tutorial how you can even create your own cl code you do not need to depend on cloud cl code but obviously cloud code is very efficient.

Cloud code is like very very you can say um well written by the team dedicated team and I have just written the code in just like few minutes just to demonstrate that this is something that

we can also build. It is not out of the box. It is not like a black box. We do

box. It is not like a black box. We do

not know what is going inside and we do not know what is coming outside. No,

everything is predicted. Everything is

like very simple. Very simple. Just a

tool call. Simple.

So this is clawed code under the hood.

Simple. Now if we want to work with claw code we need to install these tools right install these tools where are these tools because obviously we do not

need to download the model lm is like hosted on their servers right and we just use that server using the API or subscription but we need to download

these tools because these tools will live on our system will live on our system. Okay, now comes the point that we need to download the

clot code. Simple, very good. I hope you

clot code. Simple, very good. I hope you understood the overall concept of cloud code like what is claude code. So that

is why I say if I need to just define it in the statement as well. So clude code is nothing but let's say claude code

is your or let's say a genic graph. Why

graph? because we are just building a graph LLM in the middle and they have the tools and obviously multi-step graph. So claw code is agentic graph

graph. So claw code is agentic graph which uses claw lms under the hood. This is the oneline statement for claw code and I think this is the most simplest one that

you can just use. And obviously if you want to memorize those documented definitions go on Google learn it memorize it. If you want to understand

memorize it. If you want to understand it this is a definition. Clawed code is nothing but the agentic graph which uses clawed llms under the hood. Simple,

simple, very good. Okay. And you know what? If you understood this concept,

what? If you understood this concept, now you'll be able to learn claude very easily because now do you know what what you need to do? Let me be honest. Now

you just need to explore its capabilities and that's it because now you understand the architecture and if you have understood the architecture let's say 60% 70% it's fine because rest of the 20 30% you will understand when I

will just show my code as well that I have personally created with respect to clot code just to demonstrate this thing okay do not feel like on you build your own claw code it's not like that it's like a very light very light version of

clot code very basic version of aentic graph simple okay because I want to give you the best knowledge So I just thought like hey let's create this as well so that you can also learn.

Okay. So now what we need to do we need to get started with claude code download or you can say installation. So how you can just get started with it. Now yes

claude code is not for free. First of

all let me just tell you claude models are not for free. If you want to just use claude for free you can only use cloud desktop application and do the Q&A work. But if you are learning, if you're

work. But if you are learning, if you're data engineer, if you are developing something, it is a very common sense that you need to make API calls with the models. And when you need to make the

models. And when you need to make the API calls with the models, you need to pay to the servers, right? It is not providing everything for free. There was

a time in the beginning when this was free for some period like way back like 2 years back not two, I would say 1.5 years back. But now it is not for free

years back. But now it is not for free and they cannot just provide this thing for free. Their models are expensive. So

for free. Their models are expensive. So

if you want to learn claude code yes you need to pay for some APIs right and what will be the bill it depends like how much APIs you are using simple so now there are two options now if you want to

use cloud code there are two options let me just tell you so now you want to get started with clot code so there are two options the first option is API which I also recommend you as a developer

because you are smart enough you should not go with subscription directly it totally depends upon your use case subscriptions are very very good. I'm

not denying that. But you should be convinced before going for the subscription. It's not like everyone is

subscription. It's not like everyone is taking subscription. I will also take

taking subscription. I will also take the subscription. No, you do not need

the subscription. No, you do not need that for now at least. So first option is API.

This is the best one to get started because you simply pay for whatever you are using. Okay. Second one is

are using. Okay. Second one is subscription that you can just take.

It's up to you.

Okay. Now if you if you want to just use a subscription what you need to do simply go to your cloud code or basically cloud desktop and click on your account and click on upgrade. When

you click on upgrade plan you will see this window. You can just pay like this

this window. You can just pay like this particular for me it's like 24 CAD. Um

so I can just pay this and it is like for yearly for monthly it is like 28 CAD. I know it is like expensive so

CAD. I know it is like expensive so obviously I'm not going for it. So like

it's not expensive expensive if you compare the usage when you are just calling the APIs it is actually meaningful but if you're learning if you're just trying to learn cloud code you do not need to directly pay $28 in

the beginning once you are well versed with it once you know how to work with cloud code then it will make sense a lot of sense because the work it can do for you next level bro next level like you

cannot even imagine like the work it can do because the moment we'll be learning about MCPS you will be blown up I'm just telling MCP is like next level thing. Okay. And

with pro plan you get a lot of well-connected MCPS as well. So that is the game changing thing but yeah don't worry you will just learn about all the things. So we have two options okay APIs

things. So we have two options okay APIs and subscription. So for now if you want

and subscription. So for now if you want to take subscription that is like one step just go pay done you have the subscription and once you have the subscription you can just use a claw code. Okay. So if you have taken that

code. Okay. So if you have taken that subscription woohoo congratulations and just wait now. Now if you just want to go with APIs so simply go on

clawed code console simple and here what you need to do you simply need to go to sign in developer platform or whatever you just see so there's like console dedicated console

and you can just see that thing so let's try to see this okay so it will simply ask you to login you can just login with any account but I will recommend you to login with the account that you have

used in the cloud desktop application it will be like better to sync Okay, simple and yeah, simple, very simple. So, let me just open in Google

simple. So, let me just open in Google because Microsoft Edge is like next level browser. You can see the web pages

level browser. You can see the web pages like this cloud console.

So, this is the page. Very good. Cloud

console. So, this will show my personal um obviously plat um not platform basically dashboard. So now here you

basically dashboard. So now here you first will click on get API key obviously but not actually you then you you cannot like directly click on this get API key because even if you create

the API key it will not work because you first need to put some amount. Okay. So

click on API keys and here these are some of my API keys. I just refreshed all of my API keys because there were like so many API keys. So I just deleted all of the API keys more recently I

think. Yeah. And see 9th of March. So

think. Yeah. And see 9th of March. So

now what you can do now you can go to first of all billing okay and here you can buy the credits simple so just put whatever amount you want to put okay and

then it will be there for at least one year you can just use that amount throughout the year it is not like monthly amount that you have to have to use it for like within 1 month no put the amount use it for the entire year I

think it is like there for one one year at least simple simple it's up to how much you want to add. I will say just start with 10 bucks and just see whatever your balance is and yeah and

even like if you're learning something if you're learning a technology then obviously um it depends like how much APIs you're calling so I will say if you're learning do not just keep on

hitting the APIs randomly so just behave smartly because this particular thing is not something that you you can just play around it is not like free so this is the billing just add some bucks when you

have some amount here then you and just go to the API keys. Then what you can do simply go to API keys and then click on create key once you have the amount.

Okay, click on create key and then you can pick the workspace and name your key. You can just put any name. Let's

key. You can just put any name. Let's

say LLM API or your beautiful name API whatever. Okay. And then that's it.

whatever. Okay. And then that's it.

Simple. By the way, you do not need to create an API key because when you install cloud code, it will automatically create the cloud code key as you can see in my case. Why there are two cloud code keys? Because there are

two areas where I use claw code. One is

with VS code and one is through my terminal. I love using in both the

terminal. I love using in both the areas. So that's why you are seeing that

areas. So that's why you are seeing that I use this and I used it never. Okay.

Simple. So just create claude code key.

Um you can just name it anything. It's

up to you. So now what we need to do we have the key ready. That means we can get started with the claude code installation. Simple. So now you need to

installation. Simple. So now you need to download claude code on your desktop. It

is a kind of you can say package that you download. It is a kind of you can

you download. It is a kind of you can say um package or you can say graph that you are downloading on your system so that you can just use those tools.

Basically you're downloading the tools.

Simple. So how to download cloud code?

Let's say you have the API ready. You

are all set. Okay. Now go to cloud code and then you can say cloud code download. simple uh or you can say cla

download. simple uh or you can say cla code uh documentation because meanwhile you will also explore some documentation. So claude code overview

documentation. So claude code overview and here in the getting started you have overview. So in the get started you will

overview. So in the get started you will see terminal if you want to download through terminal you can just use this command or let me just find that command for you. So if you are on windows okay

for you. So if you are on windows okay let's talk about windows because I personally use windows. So if you are on Windows, you just need to open your PowerShell.

PowerShell, okay? Not cmd, not command prompt, powershell. And you need to run

prompt, powershell. And you need to run this command. IRM https claude.ai and

this command. IRM https claude.ai and blah blah blah. And if you open cmd by any chance you do not have access to powershell or whatever, you can run this

command to download the claw code. Now

you are downloading those tools.

Remember claude code. Very good. Then if

you are on Mac OS or let's say Linux, okay, or even if you have WSL installed in your Windows, okay, then you can also use this command simple cloud command and that's it. Okay, so this is like

native installation. Then if you have

native installation. Then if you have homebrew, you can just use this thing.

If you have vinget for obviously windows, but I will recommend you download through this particular link.

So this is the best place to download the thing. And how did I access this

the thing. And how did I access this thing? Simply search claude code

thing? Simply search claude code documentation and go to overview. Then

you will see this thing as the first thing. Get started.

thing. Get started.

Simple. Very good. And do not go to VS Code desktop application. No, these are something else. You are downloading the

something else. You are downloading the cloud code. Okay. Okay. Makes sense.

cloud code. Okay. Okay. Makes sense.

Very good. So I have opened one AWS EC2 instance just for you because I have already downloaded cloud code on my system. So I do not want you to miss any

system. So I do not want you to miss any step. So I thought like let's create a

step. So I thought like let's create a fresh EC2 instance for Windows and let's try to download cloud code from scratch.

So that you will also see all of the steps from scratch. Simple. See how much an Lamba love his data community. Yes, a

lot. So let me just grab the command. So

this is a command that I have. So I will simply run it on cmd let's say. Okay.

And let me just open cmd here. So this

is a very fresh EC2 instance like not very very fresh but yeah it's fine. So

let me just run this command and let's hit enter. And now let's wait. So it

hit enter. And now let's wait. So it

will download or install that clot code.

This machine is not like very very quick. So I can expect some of the delay

quick. So I can expect some of the delay because I think it has very few RAM. Not

very few. It has like 4 GB of RAM. So

it's fine. So now you can see like setting up clot code and installing clot code native build latest. So it will install clot code. It can take some time because it is like big file but not

sure.

Oh native installation exists. Okay. So

I maybe I would have downloaded it. No

worry at all. But yeah setup native.

When did I install it? I just downloaded this like I just created this instance like few days back. Uh

okay maybe I don't know. So okay this is installed. In your case it would be

installed. In your case it would be installed as well. So let me just close this terminal and let me just open a new terminal. So we need to confirm as well

terminal. So we need to confirm as well right that our claude is installed. So

you can now simply um run the command called claude and that's it and hit enter. So it will say claude is not

enter. So it will say claude is not recognized as an internal or external command. That means our claude is not

command. That means our claude is not actually downloaded. Makes sense. So now

actually downloaded. Makes sense. So now

what we can do, we can simply confirm like what is happening. Let me just go and open a new

happening. Let me just go and open a new terminal. So I just ran the command

terminal. So I just ran the command again because I didn't read this step which is like very very important. So it

is saying that we have installed clawed code on your system and it exists there.

See now whenever we'll be run the cloud command obviously it will look for the executable like where I need to execute the command so that path is not added in

my user system or you can say environment variables now I need to add that particular path in my system as well because once I have that path then what will happen whenever I will write

claude it will go to that executable and see that's why I just created a new VM so that you can also see these errors you can just troubleshoot these error Make sense? Let's try to troubleshoot

Make sense? Let's try to troubleshoot it. So here now we will be just adding

it. So here now we will be just adding the environment variable. So basically

what happens I I don't know like if if it will happen in your case because you are already using your system and you already have some executable commands but when you just get started with a

very fresh VM you can see these issues like when the um path is not added and when the path is not added it cannot execute anything. So let me just create

execute anything. So let me just create that particular thing. Hit the window button. So then we can simply say system

button. So then we can simply say system variables. System variables. What was

variables. System variables. What was

written here? I have just taken a screenshot. Where is that? Oh yeah,

screenshot. Where is that? Oh yeah,

environment variables.

Environment variables. Okay. Edit the system. Yeah.

variables. Okay. Edit the system. Yeah.

System environment. This is like very very not risky like very careful step because if you mess up with any particular variable your other applications will not work. So be very very cautious. Okay. Now we need to open

very cautious. Okay. Now we need to open the system variable. We have opened it.

Click environment variables. Okay, where

are involvement variables? Where are

environment variables? Okay.

Oh, here involvement variables. So now,

oh, here's a path. And you know what?

This is the same thing that we install when we install any application. Let's

say we are installing Python. But Python

is very nice because we are downloading Python using the UI. So it already prompts us, hey, do you want to add this path? We say, yeah, yeah, yeah. So this

path? We say, yeah, yeah, yeah. So this

is exact that exactly that thing. But we

didn't add anything. That's why we do not have anything. Okay. Okay. So now

what we need to do um under user variables. Okay. Here user variables.

variables. Okay. Here user variables.

Okay. Click edit. Okay. Edit. And then

click new. Okay. Click new. And then I just need to paste the path which is my claude administration.

Oh. Okay. That's it. Click okay. And

then simply say okay. Okay. So now we again just open a new terminal. And if

it doesn't work then I would need to just restart my machine. And I would ask you to restart your machine as well because sometimes it needs to process that information and sometime it just caches the results.

Now if I search CL now it is waiting that means it is working. Fingers

crossed. Let's see. Claude claude

claude. Yeah. Boom. Very good. And you

know what? See, I also used you can say your claw desktop to get the steps step by step because I know the steps were written on the screen. But whenever this is my pro tip, if you do not want to

mess up with your machine, with your system, if you see something related to addit the environment variables, be very cautious and just make sure that you are following all the steps very carefully.

Okay. So now we are getting started with the cloud code. So let's focus on this thing. So now what is it saying? This

thing. So now what is it saying? This

particular uh icon says that your claude is okay started. Okay, running. So now

it is saying dark mode, light mode, whatever. So I will simply say dark mode

whatever. So I will simply say dark mode and you just need to use your arrow keys to get started with it. So I will simply hit enter dark mode. Now it is saying how you want to use your claw code. Do

you want to go with subscription? Um not

really. Do you want to go with entropic console account that means API? Yes. So

I will simply say second option and then hit enter.

Now it will open the browser for me.

Why? It will ask me to just simply login. But in your case, you can also

login. But in your case, you can also login.

Make sense? Now I do not need to just login through this machine obviously because this is my just VM. But but I want to show you the moment you log into your account. Okay. Do you know what it

your account. Okay. Do you know what it will do? It will create one clawed code

will do? It will create one clawed code key that you would have seen in my screen as well. Maybe I would have blurred it. But yeah in the API keys

blurred it. But yeah in the API keys even if if you have created one API key for you it will create a new API key called claw code key because this will

be used by the claude simple. So then in your terminal not like terminal terminal in your you can say portal dashboard in the claude dashboard you will see one

more API key automatically created by claude simple claude code key okay nothing fancy very simple very good the

moment you have all of these things okay then I can just minimize this thing because now I know I can just switch back to my machine okay because see I wanted to show you all the things all the things from scratch so that you will

not be stuck at any stage because I know you are really really excited to master cla code. So I want to enjoy your

cla code. So I want to enjoy your excitement as well. So let me just open my cmd quickly.

Here is my cmd. Okay. So now I will simply search clo. So in my case it is already you can say verified. It is

already um it has already created the API key. Simple. So the moment you will

API key. Simple. So the moment you will create the API key not you basically claude code once you have the claw code key in your account simply close the

terminal and open a new terminal or maybe restart the machine. Okay then

write claude and then you will see this thing. So what is this thing now it is

thing. So what is this thing now it is asking that I want to perform some things in your terminal. Do you trust me? Um I will say yeah I trust this

me? Um I will say yeah I trust this folder. Okay. Hit enter.

folder. Okay. Hit enter.

uh hit enter perfect so this is your code claude claude okay now you can see that every information is written here I am using haiku 4.5 model I am using API

usage billing and I am using my you can say um account name or workspace name that I have created simple and this is the terminal location that means you can open cloud code in any terminal and

don't worry we'll be just opening this thing now um in VS code because that is the best thing we can do as a developer efforts. So now an Lamba you are using

efforts. So now an Lamba you are using haiku 4.5 and you are asking us to or you can say you have assigned us sonnet 4.5. See I have not assigned by default

4.5. See I have not assigned by default it picks the sonnet 4.5 but you can just change the model. How you can run a command. Let me just increase the size.

command. Let me just increase the size.

So there's a command this is called like slash commands. So you can simply say

slash commands. So you can simply say slash and then model. Okay. And then hit enter. Then you can just pick any model.

enter. Then you can just pick any model.

So I want to pick let's say default.

Default is like recommended. That means

it will simply pick 4.6 and there's a like cost return and this cost is like for 1 million tokens. So, set is like $6 for 1 million token for input and $22

for 1 million tokens. 1 million tokens is like a big deal. Okay. Then we have Opus model which is the most most most performant model that we have. And then

we have Opus 1 million context. This is

like a little bit like more expensive.

Then Hiko is the most cheapest one. $1

for like 1 million token and $5 for just like 1 million output tokens. So I have picked haiku. You can also pick haiku.

picked haiku. You can also pick haiku.

Happy happy. Good. Now llamba let's say we are using API and in the middle I want to let's say um jump onto subscription because I love subscription

because I want to use a lot of things.

Okay. Then you can simply say login hit enter and then you will see the same thing again. Subscription console and

thing again. Subscription console and just do whatever you want to do. Simple.

And if you just want to escape, simply say escape to cancel.

Uh, and now if you just want to exit from this particular clot code, simply say backslash exit. It will say bye. Simple. One mc

exit. It will say bye. Simple. One mc

file. It's fine. So, oh, I'm on cmd, not on powershell. So, okay, this is your

on powershell. So, okay, this is your command prompt. Simple. And now, now if

command prompt. Simple. And now, now if we just open let's say another terminal just to confirm like everything is fine.

can simply say cmd and here simply search claude and it will open everything and every time it will ask do you trust me if yes fine if not exit simple so this is your claude code

installation and I hope that you are able to download cloud code and if not I would say just follow the steps that I have just mentioned from scratch okay I have literally created a fresh VM just

for you so that you can get started with the clot code because I know some of you can face those path level errors so that's why I created those VMs and I Hope your claude code is installed.

Simple. Very good. So now let's talk about the developer toolkit. So now

obviously we'll be just writing a lot of code and we'll be just implementing claude code in so many examples. So yes,

we can access everything through terminal but it is not a very good practice because being a developer we can access everything through terminal but even when we want to access terminal

we should not directly open command prompt or like let's say powershell we should open the terminal through a kind of code editor okay so here you would need one code editor of your choice I

will recommend using VS code if you do not have VS code then you can use anti-gravity you can use your PyCharm whatever but VS code is the recommended

one because we have some native support we have claude code extension within VS code so it'll be like better to install it there okay so VS code and then you should have Python installed on your

system because you'll be just using some Python related task as well in the clot code so just make sure that you have VS code editor installed and I would recommend you to go with it if not it's

fine you can just use any code editor second one is Python and I Assume that you know how to download this thing. Simply go on Google, search

this thing. Simply go on Google, search VSO download, download it. Go on Google, search Python, download it. And make

sure that Python is like Python's path is all um already added in the environment variables. And you do not

environment variables. And you do not need to do it manually. You just need to install it. And in the beginning, you

install it. And in the beginning, you will see that checkbox. That's it. So we

just need these two things and that's it. It's more than enough for this

it. It's more than enough for this particular course, this particular tutorial. So okay. So once you have all

tutorial. So okay. So once you have all of those things, let me just open my terminal as well. Let's say my VS code as well. So let me just open it. So this

as well. So let me just open it. So this

is my VS code and in your case, it can be a little bit different because you would have your own themes applied in your system. So yeah, this is it. And

your system. So yeah, this is it. And

this is our VS code editor and I have opened one empty folder as you can see called clot code. You can also create one empty folder, but just make sure it doesn't have anything else because we

are starting everything from scratch.

and just create an empty folder and just open it. And if you do not know how to

open it. And if you do not know how to do it, it's very simple. Go to file and go to open folder. Then just pick any empty folder that you want to pick for this particular project because this

will be our working directory. Right?

Very good. Now first thing first, you need to open the terminal. So I will simply open the terminal. You can press control shift and back tick or you can

go to three dots and go to I think view not view terminal. Yeah, new terminal.

Yeah, I always use like shortcut. You

can also use shortcut. It's fine. So,

this will just open one PowerShell for me. And first thing that I need to just

me. And first thing that I need to just do is pip install UV. Why? Because nowadays we do not use pip as a package manager. We

always use uv and I can simply use like download uv. I already have UV but in

download uv. I already have UV but in your case it will be downloaded. So this

is a kind of Python package manager which is very very fast and you will actually feel it because it is I think built on top of rust. So it is actually very quick. So yeah it saves a lot of

very quick. So yeah it saves a lot of time and you will feel this thing when you are building um you can say your MCP server because in MCP servers we need to run the you can say Python packages

which are hosted on Pi library. So in

those particular scenarios when you use UV to download and run those packages it works very very quick quickly. So yeah

this is everything. So you have now UV very good just make sure that it is downloaded you can check the version of it UV version so in my case it is like 0.9 in your case it will be like

something else maybe it's fine okay so now we have the UV very good now what I will do I will initiate the project that I want to do so simply run uv init uv

init is a command which will initiate git and virtual norm and everything for your project you do not need git don't worry at all at all but in my case I already have get in my system so it will

just show you but you do not need get at all. So simply hit UV in it and all

all. So simply hit UV in it and all these files will be created automatically. What are these files?

automatically. What are these files?

These files are basically the package files that are very helpful when you are creating the package. Okay. So just

ignore it. The main thing is now we need to say u sync. So now what u will do?

Usync will create a virtual environment and it will sync all of the requirements or you can say dependencies that I would have defined in this particular tumal file. Currently we have nothing but

file. Currently we have nothing but still we need a virtual environment. Hit

enter and you will see virtual environment. Simple very good now

environment. Simple very good now let's try to turn on this particular virtual environment. There are two ways

virtual environment. There are two ways you can close this terminal open a new one. VS Creator will automatically open

one. VS Creator will automatically open the terminal with that virtual environment or you can say

um dot when scripts activate. So now you can see my virtual

activate. So now you can see my virtual norm is activated in I think Mac it is like slightly different. You can just check on Google you will get the command. I think it is called bin and

command. I think it is called bin and then bin and then something. Okay. So it

is exactly same. So now what we need to do first thing first we need to create our environment file okay environment

file and I will create a new file and it will be called as env simple in this env

we need to create a variable which is called entropic API key and here you

need to put your entropic API key. Now

you do not need to copy the clawed code key because you cannot see that API because it it is already created. You

need a fresh API key. Simple. So just go to your um platform for entropic for claude where you created the API key.

Just go there and paste the API key here in the file. Okay. So here you can simply see your enthropic API key here.

Simple. Okay. But I will just replace this file with my original API key. Let

me just do it right now. So this is my API key and you can see that I have pasted it here and I now I don't need to worry about this API key at all. Why?

Because I'll be I will be directly importing this environment variable into my scripts and that is the you can say recommended approach whenever we deal with the API keys.

Okay, very good. Instead of hard coding it in in our files. So first of all I want to show you the normal LLM call that you make with any entropic LLM

model and how claw code makes the difference. Simple. Okay. So let me just

difference. Simple. Okay. So let me just create one file and let me just say lm call. Okay. And I will simply add

lm call. Okay. And I will simply add ipynb. This is a kind of Jupyter

ipynb. This is a kind of Jupyter notebook or you can say IPython kernel format. Here it will also prompt us to

format. Here it will also prompt us to download the IPI kernel. If you get that prompt, simply say okay. Okay. So, let

me pick the kernel. First of all, I will simply say Python informance. Yes. And

it is saying why it is not showing me the virtual environment. Why? Uh okay.

Let's try to run any command and then it will maybe recommend us. No worries at all. Uh because it should show us uh

all. Uh because it should show us uh existing Jupyter server. know by the norance h something is let me just restart it let me open a new

terminal or let's create something okay so now what we need to do we want to make an LLM call within okay so in order to run anything with any LLM or you if

you want to make any LLM call the best way or you can say easiest way is using lang for it obviously we are not learning langen but I'm simply using the library to make a quick LLM call because

yes you can create um one client with entropic which will take some time and which is not a recommended approach as well because if you want to use multiple LLMs then you should always use a

library agentic library let's say lang chain lang graph okay so first of all I will go to my terminal I will say uv

lang chain and then entropic okay and then it will simply add all of the things that we have and if you just go to dovnv you should see all of the things here in the not in the scripts in

the library. So entropic entropic

the library. So entropic entropic entropic. So now it has all of the

entropic. So now it has all of the things that it needs. So now what we need to do simply say let me just turn on the code. Okay. So now you can say

let me just try switching the kernel by the environments. Oh it displays our

the environments. Oh it displays our claw code environment and then it instantly hides that. Why?

Why? Let me just click here. Um

not actually let me just refresh this VS terminal. Let me just see what's wrong.

terminal. Let me just see what's wrong.

So I have simply refresh my VS C editor and sometime it behaves like this. It's

fine. So now let me just try one more time. Python moments. See what is what

time. Python moments. See what is what what's wrong with you bro? And let me just hit refresh button from here. Oh

yeah finally. Oh man. So now our you can say clot code is here. and cloud code 3.12. Yeah, perfect. Now our version one

3.12. Yeah, perfect. Now our version one is selected. So now what we need to do,

is selected. So now what we need to do, we simply need to write from lang chain entropic import chat entropic. Simple. So this is one

chat entropic. Simple. So this is one kind of you can say class that we use to make any LLM call with the anthropic models any model. Simple. So now what I

will do? I will simply load my

will do? I will simply load my environment file which is this one right because I have the API key here and if I want to make any LLM call to the entropic model I have to use my

environment file to fetch the environment variable. So for that I

environment variable. So for that I would need to add one more package and the package name is env from Python. So

I will simply say uv env. Perfect. And here I can write from

env. Perfect. And here I can write from env load env. And then I simply need to run the function load env. And done. And

now it will say install kernel. Yes,

install the kernel please. Please

please.

And it will install all of the packages for IPI kernel. And that's it. Good. So

now this cell is running.

And now what we need to do, we simply need to confirm that our API key is working fine. So I will simply use OS

working fine. So I will simply use OS module for that.

And I will say if uh OS dot get env or let's say yeah os do.get

environment function get env and I want to get enthropic API key.

Okay if os.get env is none then simply say raise value error else you can say is set simple. So yeah, we have our

entropic API key. So that means it is set. Simple. This is a very common

set. Simple. This is a very common practice that we follow in the beginning of the code so that everything is working fine. Okay. So our API key is

working fine. Okay. So our API key is set. Now I can call the function. So

set. Now I can call the function. So

I'll simply say llm equals chat and cropic. Now I need to pass the model

cropic. Now I need to pass the model that I want to call. I can pick any model basically. Um I can pick let's say

model basically. Um I can pick let's say claude haiku and then let's say four five. Let me

just bring the models cla models.

Claude models and models overview.

So I have these choices. Okay. I can

simply let's say copy this one. Copied.

Very good. Let me just put it here. So

this is the model that I want to use to make the LLM call. So this is my LLM.

Okay, this is the variable that has created one LLM for me. LLM object. Now

I can use this LLM to get the response back from the clawed model. Simple. So I

can simply ask anything. Let's say I want to ask um give me the

code to create a pandas data frame. Simple.

Okay. With three columns and five rows.

Simple. So I will simply run this thing and I will get the response back from the LLM. Simple. So here I can see AI

the LLM. Simple. So here I can see AI message content equals to blah blah blah. Let me just make it more

blah. Let me just make it more structured. So I'll simply say response

structured. So I'll simply say response equals to this response dot content.

Simple.

Okay. So this is the response basic data frame with three columns five rows and it is it has written the code for me.

This is the code. See when you see three back text Python. So now from here this is a pure Python code. Wow. So it has written the code for me for creating the

pandas data frame. I already mentioned that cloud models are very very efficient to build the code to write the code. We already know that. But now do

code. We already know that. But now do you know what? I do not want to just create this code. Okay. Now what do you

want to do? I want to I want claw to create a Python file for me and write this code in that particular file. Wow.

Yes. I want this thing because that is called agentic behavior. That is the core reason I want to use clawed code right and so far clawed code has not come into the picture. It will come into the picture like it is already there but

you need to understand the paradigm first of all make sense do not feel like hey it is still not there what are we learning you need to understand what we we exactly try to solve with clot code.

So I have the code. Now I have two options. I can simply copy the entire

options. I can simply copy the entire code. Okay. And I can create a file by

code. Okay. And I can create a file by myself and I can paste the code within that file. That's exactly what are we

that file. That's exactly what are we doing so far with chat GPD with any kind of Gemini copilot whatever. Right now

let me just show you the magic. Okay.

Are you ready? Okay. And see now you will say Anish Lamba do you know what you are not able to generate the file um from this claude model because you are

dumb you are not asking claude model to generate the file you're simply saying give me the code just ask it to just give you the file it will give you the file really okay let me trust you okay

let me just ask it um give me the code to create a pandas data frame okay with five rows and let's use a multi-line um command

Okay, with five rows and because you are saying let me just try this as well because you think that it it will work.

Please also generate a Python file to store the code.

Okay, simple. Please let me just add one more please. Simple. Let me just run

more please. Simple. Let me just run this now. Let's wait if it if it it can

this now. Let's wait if it if it it can create the file. So here's a code to create the panda data frame. So it has again written the code exact code see

the entire code is here simple it has not generated the file even if I have requested it to create the file it cannot and you know the

reason because your fundamentals are already done that's why I just showed you and just talked about all of the fundamentals in the beginning because now you can relate it better now you do

not need to um understand it again right so now let's see what will happen. So

now we know that these are the normal API calls that we make okay to get our work done. And the same thing is

work done. And the same thing is happening in charge GBD and whatever.

Now you need to go to extensions here this square button. In my case it is at the top. In your case it will be on the

the top. In your case it will be on the left hand side. So just click on extensions and now you need to use the extension called claude. Why? See, you

can use claude in the terminal for sure, but this extension is a better way to use claude. Why? Because you can just

use claude. Why? Because you can just work with it in a chat friendly manner because you are already familiar with chat based or basically chatbot style um

agentic behavior. So it will like feel

agentic behavior. So it will like feel very much familiar simple. So let's

install that extension. And don't worry, I will just show you the terminal as well. It's very simple. So simply search

well. It's very simple. So simply search Claude and this is the official with 6.4 million downloads. So simply click on

million downloads. So simply click on install and this claude is installed and where it is that where is the extension.

So you will see this new icon called claude code. This is the claude code

claude code. This is the claude code extension for VS code. Now you need to configure it obviously. So simply click on this. And now in my case it is

on this. And now in my case it is already configured because I just re uh uninstalled it just to show you how to download it because it just used the cached credentials and it's done. In

your case it will ask for the login which lo which which login exactly the same login that you have already done in the terminal subscription based API console or the third one is like totally

different. So simply pick the console

different. So simply pick the console based if you are using API simple simply pick um console based if you're using API it will take you to the account simple and it will create another claude

code key anula in that case you'll be having two cloud code keys it's fine so what will happen whenever you will use this extension it will use that that clot code key if you will use a directly

the terminal it will use the different clot code key it's fine it's totally fine I also have two clot code keys it's fine okay so let's say you are also reached at this particular stage. See

this is exactly the same chat application that you use right so that's why I just I'm just recommending it to you so that you do not need to feel like hey why do we need to work in terminal

some people do not like it simple so here what is the thing very simple very similar to charg all of your past conversation will be here if you want to delete them you can delete it so first

of all here is the thing here you will get the option in clot code that how I need to behave who I clot So I have enabled one feature called ask

before edits. That means claude cannot

before edits. That means claude cannot edit my code base automatically. It will

take my approval which is a very good thing. Whenever you are working on a

thing. Whenever you are working on a critical project, you should enable this feature. Ask before edits. Simple. What

feature. Ask before edits. Simple. What

are the other options available? Let me

show you. Edit automatically. That means

it will automatically edit everything.

Okay. And plan mode. Plan mode is for the tasks which you know that it should plan. Why? If you are just performing a

plan. Why? If you are just performing a multi-step operation and when you know that each step is very very important.

So you should always go with the plan mode. So what it will do? Claude will

mode. So what it will do? Claude will

explore the code first of all and then present a plan entire plan in front of you before doing anything. It is a very important feature when you are asking

claude code to build a very huge code so that you do not waste the tokens. For

example, let's say you randomly asked it to do something, right? Generate the

code. You got the code and you didn't like it. So obviously now your tokens

like it. So obviously now your tokens are gone, right? It has generated the code and you are paying for it. So

whenever you want to work like this, you should always plan. You should always ask it to first give me the plan. If I

want any changes in the plan, I can make the changes in the plan. Then it will edit the code and then it will generate the code. Simple. But just for the

the code. Simple. But just for the simple example, we can go with ask before edits. These are like for small

before edits. These are like for small tasks. When you are simply giving one to

tasks. When you are simply giving one to two tasks and you know like there's like just one to two tasks. So you do not need a plan for it, right? But plan one is very important. Good, simple, very

good. Now what is this thing? Code. This

good. Now what is this thing? Code. This

is just the you can say context. Now

claude the benefit of clot code is it uses the entire code base as a context.

But if you want to provide more context or if you want to ask cloud code that I am asking my questions related to the current file put more focus on the

current file then you can just put here the file and I didn't put it manually.

If the file is opened and the same file is open on the left hand side and then I open the clot code it will automatically put that file in the code. Simple so far

so good. Very good. Now behind this face

so good. Very good. Now behind this face behind this face this this anchamas face there's a button to attach the files.

Okay this button is called attach something. Let's say upload from

something. Let's say upload from computer add context. Let's say you want to add some screenshots. Maybe you are seeing some errors. Yes claude is very smart. It can extract the text from the

smart. It can extract the text from the screenshots and it can process that information as well. Wow. Yes. So you

can even add some context from your own computer as well which is not the part of this codebase but yes it can be the part of the other folders or anything.

Very good. Now at the end you will see the command menu which is the slash. So

you do not need to worry about that because all the commands can be written here. So now this area this is your

here. So now this area this is your area where you write everything. If you

want to write technical commands of claw code let's say back slash this back slash that back slan. Remember that we just ran the command let's say back/model to select the model. So you

can run all of the commands using slash.

It will run the technical commands as well. See if I just click on model it

well. See if I just click on model it will give me the options. I will simply say switch model and I will simply pick haiku. So that is fine. So set model to

haiku. So that is fine. So set model to haiku. Very good. Now the beauty of this

haiku. Very good. Now the beauty of this box is you can even use your natural language to ask claude to do the work.

And whenever you want to ask anything from claude to do not use backslash or just slash because slash is for technical commands right and these are

also called as like slash commands. So

now what we are going to do I will simply say using my natural language the same prompt.

same prompt.

So I will put the same prompt and you will see the magic. I have put the command. Okay, let me just remove the

command. Okay, let me just remove the unnecessary space.

I know like it has some optimization techniques which replaces the space with null but why to take risk right because our API is attached to it. Okay. So

let's hit enter. Boom. Now you will see thinking stewing whatever blah blah blah blah boom something happened something

happened in our codebase. So now it has literally written the code for me using my codebase.

It has not saved the file. See I have also got the option allow write to my C drive cloud code which is the folder and then the name of the file. If I say yes

only then this file will be saved otherwise it won't save it. This is the power of ask before edits. Simple. If I

say yes allow all edit this session that means you can edit all of the file and you do not need to ask for it but which is not a good thing. So I will simply say yes. Boom. Now it has literally

say yes. Boom. Now it has literally created it and it has said done. I have

created this code. Wow.

Okay. Same model. Haiku. Haiku this is also haiku see same model I have literally used the same model see haiku 452510

01 okay claude hiku 4525 1 01 ditto same model but here it didn't do the work here it generated the file let me just check the file if I just go to my

codebase on the left hand side this is the file see create dataf frame py wow wow same llm model same brain

this time how it created the file. You

know the answer because I made sure that you know the fundamentals. The answer is here tool calls.

This same claude haiku model made a tool call and passed on the code. It said hey just take my code okay run this file and

then just do the work for me. Then just

do the work for me. Wow.

Literally wow.

It's not magic. It's not magic at all.

It's just a tool call. Right? If I ask you a very simple question, okay, just think about it. Forget about claude code. Forget about age. Just imagine

code. Forget about age. Just imagine

that we are living in 2021.

Yeah, because in 2022 we got the chat GPD in the month of November. I still

remember like it was my semester and examinations and my friend was like, "Hey, hey, hey, let me show you something. Just ask anything and you

something. Just ask anything and you will get the answer." I was like wow. So

yeah let's say we are living in 2021. If

I ask you that I will provide you the code can you create a file in my repository. You will say what's the big

repository. You will say what's the big deal? What's the big deal? I can do

deal? What's the big deal? I can do that. I can literally use OS module. I

that. I can literally use OS module. I

can create a file. I can write my code in that file and that's it. That's

exactly it is doing. But how? In an

agentic way.

In the version that I have just explained, I need to hardcode that logic which cannot be dynamically implemented.

It can be dynamically implemented but there'll be like a lot of work we need to do. Clot code is saying you do not

to do. Clot code is saying you do not need to do that work again and again. We

have already done the work for you. We

have best data or not just data like developers. Okay, we have the best

developers. Okay, we have the best developers and we literally wrote one entire agentic graph for you which has the built-in tools. You can use the

tools and use our LLM models to do the work. Simple and let me just show you

work. Simple and let me just show you what are the tools available to us through um your claude code. So I will

simply say uh claude code built-in tools. I can also go here in the

tools. I can also go here in the built-in tools. I can ask because see

built-in tools. I can ask because see this is our personal developer. I can

ask anything. I can simply say hey what are the built-in tools do you have? But

obviously whatever you are asking you are using tokens right. So that is the thing. So now I simply take you to the

thing. So now I simply take you to the claude code tools claude code overview cloud code documentations uh

where is built-in features built-in tools. So let me just search. So these

tools. So let me just search. So these

are the built-in tools that I have literally found out for you. So see

read write edit bash globe gp web search, web fetch, ask user questions.

Wow. So what it did actually behind the scenes, let me show you. It read my codebase. Okay. And I would say it

codebase. Okay. And I would say it didn't actually read my code because I didn't ask any question related to my codebase. But I can ask it, right? Then

codebase. But I can ask it, right? Then

it use write function to create new files. Wow. It would have used bash as

files. Wow. It would have used bash as well to maybe let's say create the new file using bash command using subprocess or it can use os module as well. It's

it's it's its choice but it can also use a git commands. Wow. It can also use globe if it wants to find the files based on the patterns. GP just to fetch

the file. Web search if it wants to just

the file. Web search if it wants to just use any information available on web. So

these are the built-in tools that we get with clot code. Wow. These are all the tools and these tools are made by clawed clawed team and they are providing it to

us so that we can also perform our task.

So model is LLM with these tools become claude code.

Wow. So everything is being done by LLM.

Yes, that's not being changed but the agentic behavior is being added through clawed code basically through these functions. And these functions are

these functions. And these functions are such like nicely written because if you write the function you will hardly write some edge cases but it is actually handling all of the edge cases and the

way it fetches the context from your codebase is outstanding is literally outstanding with the speed it is also important. So this is the framework. Let

important. So this is the framework. Let

me just show you step by step so that you can just understand. So let's say you are this person okay and we just used plot code. We just said, "Hey, we

want to do this." So what happened actually? So we made a request to Haiku.

actually? So we made a request to Haiku.

Simple.

We made a request to Haiku. Okay.

Now when we made a request to Haiku, this request generated some code obviously. Okay. Generated some code. So

obviously. Okay. Generated some code. So

let's say we have some code already.

uh let's say some code but this is a textual code this is what this is a textual code so let not use a file because you'll be get confused

so let's say this is the response that we have got so let's say some code is written blah blah blah blah blah and and

okay so this is the code now this code this code will go to a tool Let's say this is the tool.

This is the tool and this tool is called let's say write and this write tool is used to create the files for us. And this tool takes

the parameter as you can say code or basically anything that we want to write in the file. So parameter is very simple any

file. So parameter is very simple any code anything any text anything simple.

So here we called this particular write function using this as the input here simply like this simple. So this

write function did what? Created one

file.

Okay. This is the file. Okay. If you are coming from Python background, you can imagine that it created a file using with operator like with open file as w

and written the code and it wrote this file inside a folder simple within our codebase because that

is our repository. Simple. This is the architecture.

This is the claw code overall architecture that it uses under the hood. Simple.

the hood. Simple.

This is your claw code. Wow. Yes. This

is your claw code. Obviously there are like so many things but this is like very basic very basic operation that it performs. Simple. So this is your claude

performs. Simple. So this is your claude code. Now when you combine all of these

code. Now when you combine all of these things all of these things it becomes claw code. But without that it is just

code. But without that it is just clawed.

But now it is claude code. Wow. This is

the entire architecture of claude code.

Is it helpful? Obviously because all of the tools, all of the functions are coming from the claw team.

Wow. Now let me show you. I have also created a very similar light version of claw code where I will also ask the same thing and it will create the file for me. Let me just show you that code as

me. Let me just show you that code as well. So this is my my very light

well. So this is my my very light version of clot code. do not do not make a fun okay I have just literally created in just few minutes because this is like very light version I just want to show you that it is just like tool calling

and that's it okay so I have just obviously using lang lang graph because that is one of the best libraries or you can say packages or you can say framework that you can just use for agentic solutions and now you'll say we

als lang lang graph I have literally created the most detailed courses on both lang and langraph if you want to learn learn it but after this video just focus on cloud code right now but

if you want to learn I have created the courses on YouTube it's available for free go and learn okay very good so we are not diving deep deeper into this code I'm simply showing you this is the

very simple very small agent that I have created for my claude code simulation okay so what I have done I have literally created this tools and I have also added one more tool which is like creating the directory and I have like

create file similar to the cloud code and then I have also instal like created one tool to install the packages If it doesn't have those packages, it can install that tool as well. So I have like added some more.

Just kidding. Claude can perform all of the things. I love cloud code and cloud

the things. I love cloud code and cloud code has all of the tools available. It

can literally perform all of the things using subprocess and all. But yeah, I have literally used very very very small subset of it.

Okay. So now what I will do, I will perform complex task. Why? Because I

will not simply say give me the code to create pandas data frame and this just please also generate the file. This is

the same um you can say by the way prompt but I will also add some complexity to my um agent. I will also say hey create a directory for me called

my project. Then inside it uh create a

my project. Then inside it uh create a python file. So let's see if it can

python file. So let's see if it can create a folder on its own and then it will just put the file into it. Okay. So

let's see. Let me just click on run all because I didn't test it. I I tested it here. So I I I think it was working

here. So I I I think it was working fine. Let's see this time as well. So it

fine. Let's see this time as well. So it

has done. No, it is still running. Okay,

response equals to agent.invoke. Okay,

I think it is done. It is green. 3.4

seconds.

Yeah, I think it is done. So, oh yeah, it's done. So, I can see the project

it's done. So, I can see the project called my project. It this folder was not there before. And let me just check the what we have inside the folder.

Okay, so I have the file called dataf frame creation.p. Let me just open this

frame creation.p. Let me just open this file. Wow.

file. Wow.

Ah wow. So this file is created by my agent and like my own clot code and this file which is I think this one this file is created by clot code. I think my code

is better. Okay. So just kidding. So

is better. Okay. So just kidding. So

this is my file dataf frame creation.

Okay. So this is the file that it has created. Wow. Literally wow. And my claw

created. Wow. Literally wow. And my claw code can even install the packages.

Okay. Because I have the tool for that as well. But yeah that is out of the

as well. But yeah that is out of the box. So this is a kind of very quick

box. So this is a kind of very quick agent that I have built. You do not need to worry about like how I build this code because this is like totally different area. This is like lang chain

different area. This is like lang chain lag graph how we build our agents.

Currently we are learning how we can use cloud code in our development and how we can leverage cloud codes because see cloud code has built amazing tools. So

we do not need to rewrite those tools because it is actually taking care of all the edge cases. It is able to run a loop between the LLM or you can say our natural language. my tools would not be

natural language. my tools would not be like very very efficient but yeah cloud code is very good okay so do not feel like hey we also want to build this agent it's not a thing that we are just trying to learn here that is like

totally different stuff lang lang how we build the agentic agent and all make sense so here we are learning the claw code simple so let me just close all the tabs that are not necessary for this

thing okay and let me just save this as well and that's it so this is also done as you can see that we have created this project and it is working absolutely fine Fine. Very good. So now we need to

fine Fine. Very good. So now we need to look at the claude code components because we have seen an amazing amazing amazing feature like tool calls. But

initially when claude code was like a very new thing obviously now also it's a new thing but earlier when it was new in

the market really really new it only had those tool calls and that's it. Then

obviously like me and other a lot of people said hey we can also build our own cloud code and we can also make tool calls what's what's the big deal why should we use clot code makes sense

because see if I build oh oh sorry so if I if I if I build my own clot code what is the advantage advantage is I can use

any lm GPTs claude gemini I don't have any bender lock in no but when I use cloud code I have to use the llms by

clude Right? So then claude code might

clude Right? So then claude code might have also conceded this thing. Okay. Now

the thing is they are actually or they have actually improved claude code a lot. It's not just now a kind of agentic graph which just makes the tool

calls. Now you can literally use it in

calls. Now you can literally use it in your organization. That's a very big

your organization. That's a very big statement. How we can just use it in our

statement. How we can just use it in our organization. Earlier cla code cord was

organization. Earlier cla code cord was just for like independent developers where everyone have their own cloud code. Everyone is just generating the

code. Everyone is just generating the code on its own. For example, let's say let's say because the very first component that I'm just going to discuss is very very very

um you can say inclined towards this problem. So let's say there's one

problem. So let's say there's one developer. Let's say let's bring one

developer. Let's say let's bring one developer. Let's say this developer's

developer. Let's say this developer's name is let's say Rahul. Okay. or let's

use a different name because every time I use Rahul and if there's like real Rahul give me a random name or let's say two random

names simple simple uh what is this

let me just take names Rahul and and Raj and Siman No Rahul and one more let's say this one this is

Rohan Rahul and Rohan okay RR so let's say Rahul and Rohan so these are the two developers and both of these developers know how to use clot code and they are they both are very happy they are just

generating code they are just making their life more efficient as a developer and they are actually becoming more and more productive in their day-to-day activities and they are generating the

code which is actually like meeting the standards of us Senior data engineer obviously. So now this person is using

obviously. So now this person is using clot code. Wow. This person is using

clot code. Wow. This person is using clot code because both are watching Anala's videos.

You can watch any video bro. So let's

say using clot code and in order to obviously work in an organization let's say this is organization where both are working right now. Let's say this is

their office and both work in this office. Okay. Now all of that code is

office. Okay. Now all of that code is being stored here obviously like for this organization. I hope so you got my

this organization. I hope so you got my point like whatever you're building you're building for an organization right so all of the code is being stored here let's say they have their code

repository okay and all of the code is here now the thing is these are the two developers cloud code cloud code and they are actually writing a very very very like highle code very

high level and very detailed and meeting all the edge cases and wow such a code is like looking like a wow so now the

thing is the person Rahul Rahul is building the code which is actually very different from the code that Rohan is

building what does it mean for example you have just seen the problem when I used clot code okay I developed something else let's say in my case if I

just open my VS code editor if I just show you the file that it has created for me so it has created a file called create dataf frame py one thing okay to note let's say instead

of just asking to create this data frame I'm asking it to create a function so it will pick any function name let's say um

dataf frame pandas and then this is the code right now claw does not know the standards of your organization

claw does not know the naming convention that you use for your code there are so many standards because writing a code is one thing. Writing an

efficient code which gets aligned with your organization is a different thing.

Are you getting my point? Everyone can

write code. Okay, everyone can write code. But why organization takes time to

code. But why organization takes time to on board even a prodeveloper? Because

that prodeveloper needs to understand the standards. What type of case you use

the standards. What type of case you use for your coding? It can be camel case.

It can be snake case. If you do not know about these cases, basically whenever you use underscore, it becomes snake case. When you use small with capital

case. When you use small with capital combination, it becomes camel case. So

there are so many standards. This is one thing. Second thing is whenever you

thing. Second thing is whenever you create the utilities, you try to keep the naming convention like that. Maybe

you use some prefix for development environment, QA environment, production environment. There are so many

environment. There are so many standards.

Yes, Anala, we can feel you really. So

now now the thing is you will say what's the big deal do you know what I am a smart developer h I know I know now you will say I'm a smart developer and I can

provide all of this information in my context let's say in this my VS code so anal do you want what do you want first of all let me just open the past conversation let me just continue the chat you will say what do you want tell

me do you want camel case no worries just tell me I will write it here um let's say this is the code wow H let's say please create whatever we said okay

and I will say please use the camel case this is the camel case camel case for the coding and blah blah and anla now

you are happy now you are happy now see I have put everything here very good that is expected from you very

good so now the thing is every time you are making a request to clot code every time you need to pass this information.

H okay will you pass this this information every time let's say you want to fix a bug immediately and you are spending 15 minutes to type uh please use camel case please use this

are you serious are you serious you will not use it right so now what you will do now do not say that I will copy the entire prompt and I will paste it again and again bro see claude code has

already solved this problem so this problem is solved by something called as claude md file yes you heard it claude MD file. Okay. So now what we need to do

MD file. Okay. So now what we need to do instead of doing it like this instead of doing it like this we can include one

middleman or you can say initiator.

For example, first of all we are talking about claw.md. Let me just create a

about claw.md. Let me just create a subheading just for your notes because I also need to provide these notes otherwise you'll be shouting lamba where are nodes? So here are nodes. So let's

are nodes? So here are nodes. So let's

say claw dot MD file simple okay claw MD file make

sense so now what this claw MD file will do claw MD file let me just use this one perfect claw MD So now this claw MD file

basically MD means markdown file so this is just a markdown file this is not like very difficult programming language with extension called MD. Not really. This is

just a markdown file. Markdown file is just like adding comments. Readme like

readme.md. If you have open any GitHub, so let's say if you open any readme like any GitHub, you will see readme.md,

right? Common sense. So this is exactly the same. This is a readme for claude.

the same. This is a readme for claude.

So every time every time it will execute any command, it will who it who it means claude. Claude will always look for this

claude. Claude will always look for this particular file in your working directory. Why? Because now claude code

directory. Why? Because now claude code has made this change into their you can say platform.

Now every time you are asking something let's say create a function create this create that it will use this file

every time. Let's say like this claw

every time. Let's say like this claw empty file. Let's say this is a kind of

empty file. Let's say this is a kind of text file. Okay. So let me just bring

text file. Okay. So let me just bring the text file txt. Wow.

txt. Wow.

So let's say this is a kind of MD file claw dot MD. Simple. This is the file that you create at your root directory.

Remember root directory. Now what will happen? Whenever this per person is just

happen? Whenever this per person is just asking anything by default it will always read everything that is written in the file. Okay, everything by default you do not need to mention hey please

read the cloud.mmd please because I do not have time to write everything again and again. No it it will automatically

and again. No it it will automatically read that file every time. So if you say create a function clot code will say yes I will create the function but I will first of all read the clot md. Now in

the claw.md you will write everything about all the standards. Use camel case always um let's say create a commit for every change you make because some people like creating more and more

commits right in the working directory.

Um use this particular file notation. Um

always add comments like this. Always do

this like all of the standards that you want to create for your organization and it is totally subjective to the organization.

Simple. You cannot just like keep on using different different cloud files.

Cloud.md no not at all. Simple. So now

how we can just work with that thing.

Let me show you. So now claw.md

should reside in your uh let me just close it should reside uh yeah claw.md

file should reside in your root directory here. So let me just click on

directory here. So let me just click on this blank canvas and let me say claw md simple claw.md but claude has

literally created a special directory called dotclaude dot claude here you can literally put everything which is related to the claude so that you do not need to mix

the files okay so what I can do I can literally create a new folder called dot clot and I and move this file into this

directory. So every time my claude will

directory. So every time my claude will go to this folder dotcloud and it will read everything which is there not everything everything for like especially claw emptyd file. Now in this

cloud empty file which is the empty file for now I can literally put anything for example I can say coding um you can say

guideline and I can put all the points always use camel case for variable

function and class names. Okay, use

descriptive name that clearly indicate the purpose of variable because some coding um tools will randomly generate ABC D. What is this ABC D? Use

ABC D. What is this ABC D? Use

productive names. Use names which are like easy to understand like what exactly you're trying to do. Coding is

not just like coding coding. Now coding

is like the one can understand it even if the person doesn't know the coding.

keep functions small and focus on single task because some coding tools randomly generate like very very very edge case specific function and they literally put every all the focus on the edge cases.

Okay, sometimes you do not need that right. So this way you can provide all

right. So this way you can provide all of the guidelines that you want in your clot in your clot.

Simple very good now obviously let me just show you the professional one of the clot md file in the demo. I can also go on Google just to search what are the good claw.mmd files. I can also open the

good claw.mmd files. I can also open the official documentation that claw.md file

like how it should like look like. Um I

can literally go here clot code and let's go to let's say extension extension extend clot code. Yeah,

perfect. So here we have clotmpd file.

So let's read some of the best practice because there's nothing wrong and right here. There's nothing wrong and right.

here. There's nothing wrong and right.

You can literally write anything. It's

your description. It's your demand. It's

it's your standards. There is nothing wrong and right here. But we can still like read what are what are the best practices by claude because claude will be reading that thing. So let's see what

they are saying. So they are saying the page covers how to write and organize cla.md files and blah blah blah blah

cla.md files and blah blah blah blah blah. So here they are just talking

blah. So here they are just talking about cloud.md versus auto memory.

about cloud.md versus auto memory.

Automemory is something like when it automatically fetches the context. But

now you know it is not auto memory. It

is the it is it is a standard for your entire codebase. Simple and as you can

entire codebase. Simple and as you can see it is loaded into every session.

Session means every chat every chat session. So in every chat session it

session. So in every chat session it will load that particular claw emptyd every time. Simple. Very good. Scope.

every time. Simple. Very good. Scope.

What is the scope? Project for sure but organization as well. Wow. Yes. Okay.

And here are some of the claw MD files.

And here are the directory like where you should just put it manage policy project instructions. So you can also

project instructions. So you can also put it in the root directory called cloud.md or inside dotcloud folder. I

cloud.md or inside dotcloud folder. I

like putting the claw.md file inside dotcloud. It is like much more cleaner

dotcloud. It is like much more cleaner and nice. Okay. And they are just simply

and nice. Okay. And they are just simply putting everything like size target under 200 lines. That means you should not write more than 200 lines. Why?

Because when you provide too much context to LLM then also it is not a good thing. Okay, keep it like short,

good thing. Okay, keep it like short, brief and meaningful to the point information simple. Okay, very good. And

information simple. Okay, very good. And

then you can just read some of the things that how you should just create this empty file and blah blah blah blah blah blah. Okay, and they're also

blah blah. Okay, and they're also showing like how you can just do that.

And now I will use I will use claw desktop application to generate one claw

emptyd file. I will simply say I

emptyd file. I will simply say I want to generate a nice uh formatted

claw dot MD file for my Python project under two lines 200 lines.

Simple. It will give me the code. And

let me just create a new chat first of all.

Perfect. So now I will get this thing from the sonnet simple because it will already do the hard work for me and I can just make the modification. I don't

need to write everything from scratch.

That is like you working with AI from AI. Wow. What type of file? So it is so

AI. Wow. What type of file? So it is so nice. See cloud desktop is so nice. It

nice. See cloud desktop is so nice. It

is also asking me what kind of project I'm building. Let's say I'm building on

I'm building. Let's say I'm building on um data engineering.

Data engineering bro.

What should claw.md cover? um coding

style first of all and project overview and uh dev setup and tools. Perfect. So

now it will just ask those questions.

Now it will create the file for me. This

is not claw code. This is cloud desktop application. And now it is saying what

application. And now it is saying what tools frameworks. Um let's say for now

tools frameworks. Um let's say for now pandas and I can skip it if I want to.

If I do not want to answer it um now let's skip. Yeah, now it's too much. Now

let's skip. Yeah, now it's too much. Now

it's too much.

So now let's wait and it will write a very nice detailed cloud.mmd file for me under 200 lines obviously. So now let's wait and let's see what it has created

for me. So it has started creating all

for me. So it has started creating all of the claw.md file for me. I know this is not something that I have to just use

it but this is a very highle generic idea that you can use in your claw.md

file. Simple. So it will just simply write all of the things that it should follow and it should be a very detailed one not very very detailed but at least it should cover all of the things for your project and again it is totally

dependent on your particular use case.

So see it has actually write every it written everything. So first of all it

written everything. So first of all it has written project overview. So what is this project about? Then we have models utils even we do not have this thing because we didn't share everything with that but it just took the dbt as the

reference because in dbt we have all of these folders available. Then data flow is like inest validate transform environments get clone and all of these things has written. So this is like very

demo claw.mmd file and it should look

demo claw.mmd file and it should look like this because I wanted to show you one of the demo files as well. Okay. And

obviously rest is up to you like how you want to create your um you can say your claw.md file. But now let me just write

claw.md file. But now let me just write my own very small very small context cloud.mmd file because I don't want to

cloud.mmd file because I don't want to waste tokens on every session. So let's

create a very minimized version of claw.md. But yeah this is a kind of

claw.md. But yeah this is a kind of demo. You can ask for any chat gbt or

demo. You can ask for any chat gbt or claw desktop to give you the detailed one as well so that you can just feel the real world experience as well.

Simple. So this is my claw.md which has like very very minimum information but very useful. So I have literally written

very useful. So I have literally written the overview first of all like what are we doing in this project and then project structure organize code to logical modules packages use descriptive folder file names and whatever coding

standards these are the coding standards. So we are using camel case

standards. So we are using camel case here. Yeah use camel case for functions

here. Yeah use camel case for functions and methods and use exceptions thoughtfully catch only with like what you can handle. So this is for like uh you can say your catch and try method or

you can also say try except method that we have in Python. Then dependency

management as well for all of the requirements txt or pip project.l

documentation as well if you want to just put anything in readme just put it right away so that we do not need to dependent at the end. Then some version control. Commit often with meaningful

control. Commit often with meaningful messages. Use branches for new features.

messages. Use branches for new features.

Bug fixes. Keep the main branch stable.

Notebook contribution guidelines and some of the documentation it can read if it wants more information regarding anything. Simple. So this is a very

anything. Simple. So this is a very common very generic claw. MD but very meaningful. So now if you will ask claw

meaningful. So now if you will ask claw to do anything. So let me just open a new session. You can even open the

new session. You can even open the previous chat but let's open a new session. And sometimes you even need to

session. And sometimes you even need to start new session manually because whenever you do something in clot code you should start a new session. Okay. So

if you want to delete the previous chat you can click on close this thing so that it will not just take some context from there as well because by default it takes a context from the previous chat as well. So let me just remove it and

as well. So let me just remove it and let me just open a new chat. So I'll

simply ask the same thing that I asked before and let's say let let me just ask like I want also want to create the function this time. Okay. So let me just ask the

this time. Okay. So let me just ask the same question. Let me just close the

same question. Let me just close the terminal and I will say create

a pandas create a function.

So if it follows my guidelines, it will use camel case. And you already know by default all of the LLM models use snake case. Everyone knows that. Okay. So

case. Everyone knows that. Okay. So

let's see if it is working fine or not.

So simply initiate the chat.

So let's see what it will do and you will see. Okay. Perfect. So now I can

will see. Okay. Perfect. So now I can simply say yes, I am happy with your code and let me just open generate data frame. So here you can see that it has

frame. So here you can see that it has used camel case. Wow. create sample data frames. This is called camel case where

frames. This is called camel case where the first word starts with small letter and the next word starts with capital letter and so on. So this is the thing

and because I told it to add meaningful dock strings. So you can see that it has

dock strings. So you can see that it has also added meaningful dock strings and whenever you work with agentic behavior of your data engineering stuff more context you will add in your dock strings better your LLM can understand

the context. So you should always add

the context. So you should always add rich context to your dock strings. These

are called dock strings. Okay, good. So

this is the function. Very good. And

this is a code that it has simply created for me. But this time using claw md. Why? Because it will read that

md. Why? Because it will read that information. And because I mentioned

information. And because I mentioned this thing, let's say camel case. So

what if I just replace it with snake case. Okay, what will happen? What will

case. Okay, what will happen? What will

happen? Okay, let me show you what will happen. So if I just go to and go to

happen. So if I just go to and go to this prompt and copy this let's say perfect. So this is a new session. Let

perfect. So this is a new session. Let

me just create a new session. No history

at all. Let me just run it. This time it will create a file with snake case. Why

it is reading my claw.md simple see create data frame. Wow.

Wow. That's how it uses our claw.md.

claw.md is like the standard is just for the standards and that's it sector simple very good I hope you got an amazing understanding of cloudmd so make

sure that you provide to the point information in the cloudmd so that it follows some standards every time do not

over complicate it do not put a lot of information because it will use that file as a context and context is everything whenever you are working with

APIs, right? Good. And if you also want

APIs, right? Good. And if you also want to see the cost for that you I will also show you how you can just track the cost. Literally, you just need to go to

cost. Literally, you just need to go to clot code and simply type / cost. Okay,

if I hit enter, I will see the cost for it. Total cost. This is this because

it. Total cost. This is this because this is a new chart. So, let me just click here.

So, if I say cost, perfect. Because this

is like literally very less code. So it

didn't actually generated million token for this but yeah makes sense and you can also explore one more command which is very important it's called context because when you say context you will

actually see what are the context for your code if I hit enter it will generate a very detailed very detailed you can say

breakdown so this is the breakdown so context usage so system prompt is taking 5.9,000 wow 5.9,000 tokens like very less as

compared to 1 million token but yeah so 5.9,000 token is called system prompt and from where I'm getting this thing system prompt is like claw.md

really yeah claw.md here and claw md or any markdown file that you would have created in your root directory root directory means because see there are

two things one is project and one is a system root like your entire desktop entire system. Sometimes we add some

entire system. Sometimes we add some properties in the system as well so that entire system can use those properties.

In my case it is 5.9 to 5.9,000 tokens including all of the readme files, all of the claw.md files um um system. So

this is simple. Okay. So now the second thing is system tools. What do you mean by system tools and why you are seeing 19,000 tokens here? Let me just tell

you. So here if I just scroll down I

you. So here if I just scroll down I have MCP tools as well. What type of MCP tools and what is MCP? You do not need to worry about that thing. But I have one MCP installed at my root system. And

I don't know why I didn't delete that MCB before starting this. But I will just delete it right now. So this is MCB. You can just literally remove this

MCB. You can just literally remove this thing for now. I will explain MCB at the end. Okay. And you can say

end. Okay. And you can say apart from MCP I have some built-in tools. Really? What type of built-in

tools. Really? What type of built-in tools? Built-in tool means the tools

tools? Built-in tool means the tools that I just showed you in the beginning.

Okay. System tools means the tools that I just showed you in the beginning. Read

and then write, GP, all those tools.

Okay, makes sense. Then we have some memory files. Then we have some skills.

memory files. Then we have some skills.

We'll just talk about everything. But

this is a kind of context that you can just see within your cloud code. Simple.

And you are also seeing that I have one memory file. Memory file is just like

memory file. Memory file is just like claw.md stored at the root directory.

claw.md stored at the root directory.

Simple. Okay. So this is it. This is it.

Simple. And you just need to worry about this thing. Okay. You actually do not

this thing. Okay. You actually do not need to worry about that thing because this is a kind of context that it uses to perform all of those things. If the

context will not be there, it cannot do anything because it's just an LLM.

Simple. Very good. So here I will just take your few seconds and I just want to make a quick request. If you haven't subscribed the channel yet, simply hit the subscribe button because your one

subscribe hit will actually help me a lot and it takes a lot of efforts to create tutorials like this and providing it for free. So if you genuinely want to support my channel, support my hard

work, simply hit the subscribe button right now and let's get back to the master class right now. Now after this important component claude.m MD file

let's talk about the most hyped and most important component in today's modern claude code era. It's called skills. If

you just search clo I am so sure that you will see skills skills skills skills skills skills skills skills skills everywhere because it is very important.

Skills is the backbone for the modern claude code era because it is like recently added. Not a very very recently

recently added. Not a very very recently but yeah it was not there in the beginning and skills actually changed changed the game literally changed the

gra game game changed the game because it brought amazing superpowers to our normal claude code which is just an

agentic graph. Why? How and what

agentic graph. Why? How and what everything we will discuss. So skills.

So let me just try to explain this concept because this is a very critical concept and it is very important to understand. Okay, skills. So for

understand. Okay, skills. So for

example, let's say this is a developer called Rohan. So now Rohan knows how to

called Rohan. So now Rohan knows how to work with claw code.

Rohan also knows that there's a claw.md

file to provide the standards. But now

Rohan is let's say a kind of data engineer or let's say data analyst or let's say BI analyst or let's say data scientist or let's say blah blah blah whatever varieties are there in the market. Okay. So let's say Rohan is one

market. Okay. So let's say Rohan is one of those varieties.

Now Rohan performs some tasks on a daily basis. Right? Very good. Now Rohan wants

basis. Right? Very good. Now Rohan wants to delegate those tasks and he does not want to perform those tasks on a daily basis. First of all, he can make

basis. First of all, he can make mistakes. Okay. Second, he obviously

mistakes. Okay. Second, he obviously needs to spend some time to do all of the work again and again. Now Rohan

wants to delegate the task and he wants claude code to utilize it.

Okay. So now what Rohan will do? Rohan

will use claude code like normally but Rohan's claude code is much much much more powerful because now Rohan's claude code is connected to something called as

skills.

Okay, skills. This is lesser skill one.

Okay, this is skill one. This is lesser skill two.

This is less skill three. Simple. Okay,

perfect. These are the three skills that Rohan's clot code has. Perfect.

Now, what will happen and what are these skills? Skills are let's say the normal

skills? Skills are let's say the normal English word that we use for any kind of capability, right? Let's say if I say

capability, right? Let's say if I say what what are your skills? You will say by the way, what are your skills? Just

tell me in the comment section. So now

skill let's say someone someone will say um I can sing okay this is my skill if I just talk about the technical world someone will say um I can write APIs

better this is my skill anyone can say this right someone will say um I can transform the data better I know how to transform the data I know how to look at that lens someone will say I can

visualize the data better I can make the dashboards better obviously now yes So far we have dedicated people for these things but now that time is coming when

there will be just one person performing all of these things right this is the AI era and this is the high time to learn all of these skills so now

this person Rohan can build skills for each tasks let's say there's a skill for API calls okay there's a person who has a very rich understanding of API person

knows how to pass tokens how to pass beer tokens how to take care of credentials, how to do all of those stuff. Okay.

stuff. Okay.

Now, there's a person who knows how to transform the data and how to convert the data or how how to migrate the data.

Okay. Now, there's a person who is a very very good data visualizer. Okay.

Who builds like dashboards like crazy amazing dashboards. Okay.

amazing dashboards. Okay.

So now what I can do I can say Claude come here I will provide you these skills I will provide you these skills once you are trained on these skills

will you do the work for me claude will say recharge the API I will do the thing for you that's it so now let me ask you one question whenever you want to learn

a skill do you spend some time let's say you are learning clot code now are you spending some Yes. Once you learn the skill, do you

Yes. Once you learn the skill, do you need to spend the same time, the exact same time that you spend while learning?

No. Exactly. So this term skill is exactly the representation of that particular paradigm. So what will

particular paradigm. So what will happen? We developers spend a dedicated

happen? We developers spend a dedicated time to create very detailed skills. For

example, I am creating a skill for API calls. So I will create the skill. Okay.

calls. So I will create the skill. Okay.

in detail exact steps will be written like how to do this, how to do that, how to just go to next step, all of the things will be written like how to make the API calls. I will provide all the edge cases. I will provide all the

edge cases. I will provide all the things. So the I basically I will bring

things. So the I basically I will bring one API person. I will sit with him.

Okay, I will say hey just tell me whatever you do. Okay, just tell me. I

will keep on writing. I will just create all of the edge cases everything. I will

spend some time obviously but that is just like one time work. I will create a skill for all of these things. Now

whenever I want to perform any kind of task related to API calls, I can literally run the skill and that's it.

Clawed code will have will happen will you can say take care of everything.

Clawed code will have all the context that it needs. Wow. Not just context, how to run the scripts, everything. It

is not just a context.

It is much more than that. It is just a kind of orchestration. Yes, it is like context plus orchestration. Then

context.

Okay. Then there there's a person. Okay.

So I will say hey um how do you migrate the data? How do you just pick the file

the data? How do you just pick the file format? Hey, come here. Come here. Come

format? Hey, come here. Come here. Come

here. Okay. Sit with me. So just tell me um if you're dealing with CSV file formats h um do you just um try migrating data in the OC format or park format? If park performant like what

format? If park performant like what will be the um decision or you can say what will be the um points that you will keep in your mind maybe based on number of records or whatever. home just just tell me just tell me. Okay. So these are

the things I will put some time. I will

create a skill. Once the skill is ready I will say Claude code can you do this for me? Yes I can do that for you. Very

for me? Yes I can do that for you. Very

good. Visualize um come here come here come come here. So okay what you do?

Okay what KPIs do you build? What kind

of formulas do you use to build KPIs?

What are the important KPIs that we need to build? I will write down all of the

to build? I will write down all of the formulas. I will just build all of the

formulas. I will just build all of the things. Then I will say clot code. Can

things. Then I will say clot code. Can

you just run these things? Yes. Perfect.

This is insane.

This is insane. That is why skills are literally amazing thing. If you want to automate stuff and if you do not want to keep on spending some time that you like spend on daily basis, you literally

define what you want to do and that's it. That is it. Simple. How to create

it. That is it. Simple. How to create the skills? Don't worry. We'll just talk

the skills? Don't worry. We'll just talk about this thing in detail very much in detail because this is a backbone. Once

you understand this thing, all the other concepts are like so easy. This is the backbone because this is integrated with the future components that we are going to discuss. Skills is the backbone. So

to discuss. Skills is the backbone. So

we'll be putting extra extra time to create the skills. And don't worry, we'll be just creating the real world skills like literally real world skills like what you can actually do. Okay. So

this will be a detail section. Don't

worry. And just for now, let me just give you a quick overview very quick overview on the skills written in the official documentation. So we are saying

official documentation. So we are saying that we want to extend our clot code with skills. Okay. Yes. So now what are

with skills. Okay. Yes. So now what are skills?

Skills extend what Claude can do.

Create blah blah blah. We will just talk about it. This is just like giving us

about it. This is just like giving us things. So it is saying bundle skills.

things. So it is saying bundle skills.

It is also fine. Getting started. Okay.

Create your first skill. This example

creates a skill. Bro, put something so that we can read what is a skill. Um

yeah, they have just put the technical things. That's it. So now you I think so

things. That's it. So now you I think so have some overall understanding like what are skills and why do we create skills just to perform repetitive tasks.

Just to perform repetitive tasks simple and I can see that they have created some skills for front end. Let's say if you want to just handle front end there will be a skill for front end. There

will be a skill for deploy as well I guess. Yeah. Uh I saw deploy somewhere

guess. Yeah. Uh I saw deploy somewhere there was a skill. I don't know if they have removed that. Yeah, there will be a like um skill called deploy. So

basically we create skills for the tasks for the activities that you perform on a daily basis right and what we do we set we create the skill in detail so that

claude code can implement it easily so that you do not need to jump in the middle of every execution and help claude code to debug the things. That's

why provide a very rich context again short simple to the point simple okay and skills obviously it's it's not just a med file it's is it's

not just you can say context it is way more than that and you will actually see where we'll be just building skills okay so this is your skill under the hood like what it does and this way you can

make your claude code much more powerful much more powerful like it already has so many amazing built-in tools now it has some standards now you are providing

skills that's why claude code is like booming right now and claude code is like actually taking over the AI world especially in the coding right so now

let's see how we can create skills in cloud code and now let's actually try to build the real world skills okay let's see so now let's try to create our very first skill and being a data engineer or being any data professional not just

data engineer we work with APIs a lot we fetch data from APIs a lot a dot. So now

what we need to do let's try to create the simple skull to fetch the data. This

is our very first skull okay that we just trying to create. Now how we can create the skull. So we have just read the document obviously and we'll just refer the document

again okay whenever we need it but it's very simple. So let me just go to my VS

very simple. So let me just go to my VS code reader and perfect. So this is my VS code reader and this is my clot. So

now in order to work with skills okay first of all I need to create a special folder inside the dotload really yeah if I just show you the

documentation as well okay let me just say code because I don't want you to say hey you are just making things by yourself

not exactly everything is written in the documentation and yeah oh man again I just opened the one

the best um browsers and let me just remove the unnecessary URL. Oh wow. I think I also have some

URL. Oh wow. I think I also have some bookmarks, favorites bar, uh cloud code.

Yeah, because Microsoft Edge is like top level, you know, browser, you can easily find the web pages. So that's why I have to bookmark.

So now in the skills, let me just go to extend cloud code. Perfect. So now in the skills, so now it's up to you how you want to create the skills. Again you

have both the options system level or project level. But I will again ask you

project level. But I will again ask you to go with the project level because every project is unique and every project should have different skills.

Yes, I know some skills are common but we are not building skills for the entire you can say repository of the projects. No, we are simply focusing on

projects. No, we are simply focusing on our project only for now. But yes, it is very simple. If you know how to create

very simple. If you know how to create skill for your project, the difference is you just want you just need to deploy or basically you just need to add that particular folder in your root directory

as well. So for now let me show you

as well. So for now let me show you because you will learn it by actually doing it. So you need to go inside this

doing it. So you need to go inside this dotcloud folder simple where you have claw.md file and within that folder

claw.md file and within that folder within that folder you need to create a special folder called skills.

Okay. Now within the skills folder you need to create all of the skills. So

here within this folder you need to create one more folder called let's say I want to create a skill called fetch or let's say yeah fetch fetch API. Okay.

This is my name that I have just kept it for my skill fetch API simple. This is

my folder fetch API. Within this fetch API you will define your skill. Oh, that

means skills. Okay, folder skills. All

the skills within that which skill I'm talking about? Fetch API skill, right?

talking about? Fetch API skill, right?

So, fetch API skill will have the skill.

Now, within this folder, there is one the most important file that we need to create. It's called

create. It's called skill dot MD. Wow. Yes. This is the file. In this file we will write or

file. In this file we will write or basically we will create our skill.

Simple. If you have understood my previous conversation that I have just made regarding skills that we spend our dedicated time to create to refine to

test our skills because once the skills are ready we can use it on a daily basis. But yes this is a tutorial. So

basis. But yes this is a tutorial. So

here we will not spend much time in creating the skill but still we will learn how we can create a skill. Simple

got it. Very good. So now as you know that I want to create a skill for fetch API. Simple. Before doing anything

API. Simple. Before doing anything I need to register my skill with clot code so that claude can fetch the skill.

Okay. How we can just do that? We need

to use YAML. Okay. If you do not know about YAML, YAML is a kind of configuration file format similar to JSON. But why do we use YAML? Because

JSON. But why do we use YAML? Because

YAML um consumes less tokens because it does not need to worry about curly braces and all those square brackets and closing brackets, open brackets. So,

YAML is becoming more popular because it saves token. Okay, if anyone ask you why

saves token. Okay, if anyone ask you why do we need to use YAML if we have JSON, simply say we need to save tokens. We

are living in the AI era. So now what you need to do, let me show you the demo file. Yeah, see. So in this file we

file. Yeah, see. So in this file we first of all need to write this command.

So this is empty file. Obviously this is a kind of markdown. But within the markdown we can create the YAML like this. Okay. YAML like this. Now this is

this. Okay. YAML like this. Now this is the name of the skill that we are providing. So in my case it is fetch

providing. So in my case it is fetch API. Simple fetch API. And just keep the

API. Simple fetch API. And just keep the name same. Just keep the name same.

name same. Just keep the name same.

Fetch API and fetch API. It is not hard and set rule but it is recommended clawed code will find it easy to fetch the skills make sense because this is a

name that it will be used that it will use to register the skill and if we have the resonating folder with it okay so it will be like easier for it to fetch so always try to use the same name that's

why I'm not seeing now blue line earlier I was seeing blue line right so this was the issue so now after name we need to provide description yamba

Whenever we just provide description, it will be easier for claude code to identify that when it needs to call this skill.

Because whenever we work with claude code, ideally ideally we should ask it to hey run the skill, hey run that skill, hey run the skill. But sometimes

we are in hurry. Sometimes we forget to add the skill command skill hint in our prompt. Right? In that scenario, claude

prompt. Right? In that scenario, claude code is smart enough to automatically detect the resonating script uh skill for it.

Why? And how? Because first of all, claude code will look into the skills and it will try to fetch all of the skills that it has. If it will find any

relevant skill for that task, it will utilize that skill and it will understand by this description because power of LLM. It can process a natural

language really really faster and efficiently and it can perform that cosign similarity between the uh vector.

Again we are not going deeper inside the LLMs. If you want to learn about LLMs, I have made a dedicated course on LLM as well for free on YouTube. Just go and check if you want to. Okay. Where you

will learn how LMS actually work under the hood. Transformer model and decoder

the hood. Transformer model and decoder only transformer model encoder only transform model encoder decoder transfer model everything. Okay. So that is like

model everything. Okay. So that is like technical thing. But yeah here we are

technical thing. But yeah here we are learning plot code. Simple. Okay. So

this is my description. So this is first part of it. After that I just need to provide my usage. How I want to use the skill. Simple. And now let me just tell

skill. Simple. And now let me just tell you the requirement as well. Requirement

is very easy. So for now what we are trying to do we are trying to fetch the data from an API. And obviously here we

do not have like real API like Oracle or let's say any kind of database API. Here

I will be using my GitHub as a repository. And here you can see that I

repository. And here you can see that I have so many CSV files right? So many

CSV files. dim customer, dim date, dim product, dim store, fact returns, fact sales, all of these files are here.

Simple. Now I want to fetch this data but not like fetching fetching like I want to really fetch this data that we fetch in the production. In production,

we do not use synchronous behavior. We

use asynchronous behavior where we use event loop um core routines just to save the resources. If you do not know about

the resources. If you do not know about asynchronous behavior um you can just learn about it. It's like a very fundamental thing in Python like not very very fundamental it's like advanced thing but it's a part of Python

programming or any programming language.

So I want to fetch this data using that particular thing. H okay makes sense. So

particular thing. H okay makes sense. So

now what I will do I will create a skill for fetching this information and I will just ask my claude code to store this information into a directory and I will

just provide all of the steps and I will create a skill around that. Make sense?

Obviously we can test it. We will see the errors. We will fix it. And then it

the errors. We will fix it. And then it will like be a iterating step. Simple.

So okay perfect. So now what I can do I will simply say that I want to so let's say usage and I want to use this. So I will simply say

to use fetch API skills. You can call it with appropriate parameters. Make a

request in API as an example. Let me

just make it better to use fetch API skill.

um or let's say you need to fetch data from an API

from not an API from the following APIs. Okay, from the following APIs or I can also say step one, step two just to make it more

robust because it's up to me how I want to just provide the context. So let's

say step one f data from API. Simple. So you need to fetch the data from the following APIs and then I can just provide the API list and here I need to provide the URL. So

what is the URL? So let me just grab the URL. So I can click on this file. You

URL. So I can click on this file. You

can also see the URL. Okay, just click on it and then just click on raw and then just provide this URL. So this is the URL that I have. Okay, perfect.

Perfect. So here I will just provide the URL. So this is the first URL. Okay.

URL. So this is the first URL. Okay.

And then I have second URL and here I just need to change it to dim I think store.

And then I have third URL. And here I just want to

third URL. And here I just want to change it to I think dim date. Is it dim date or what? Yeah, dim date. And then

dim product. Okay.

Then let's say dim product and then let's say fact sales.

And then at the end fact returns simple. So these are all of the APIs that I want to use. Simple.

Okay. Perfect. So this is step one.

Okay. This is step one.

Perfect. So this is my step one. It will

fetch the data. Okay. So or you can say yeah fetch data from the APIs. You need

to fetch data from following APIs.

makes sense. Or let's say you need to make you need to make um Python

uh API or let's say yeah you need to make Python API call API calls to fetch data from the

following URLs. Perfect.

following URLs. Perfect.

using asynchronous functions or let's say for using asynchronous HTTPX because we do not use request library here in synchronous in asynchronous because

request library is synchronous Python is a part of synchronous Python httpx is the asynchronous version of the fetching the data so I'll simply say you need to just do this so this is my step one once

this step is done then I will say step two step two handle API responses after feting data you need to handle appropriately. Um, not really because

appropriately. Um, not really because this is just CSV data. After fetching

the data, you need to create a directory named um

having name with current time stamp, current data name in the format of this thing and save the fetch data CSV files in that directory. Perfect. Yes. So it

will create a folder and then it will just save the data or just to be specific I can also say like where it needs to create that folder because otherwise it will create the folder at the root location but do not worry whenever you are just creating a skill

you will see errors even if you do not see errors you will see unfamiliar behavior but you need to train it that's the concept of a skill you need to train

it right so I will simply say um the location of the directory should be uh let's say

skills. Yeah. Or let's say dot claude

skills. Yeah. Or let's say dot claude should be dotclaude skills fetch API data and that's it. Perfect. So this is my second

that's it. Perfect. So this is my second step. Perfect. Two steps are done. Very

step. Perfect. Two steps are done. Very

good.

Step three uh I will say logs because logs are very important whenever we have fetch the data because sometimes APIs do not behave the way it should. Okay. And we

should always log the responses.

For example, I have six to seven APIs.

Okay, one API is not working. So, I need to just log the information which API worked and which didn't. Simple. Um,

store the or let's say login.

Simple.

Um you need to create a log directory at logs. Okay. And log the status of

at logs. Okay. And log the status of each API not the ejp call. And so inside logs I can create create a log directory

add this with name current date and time in the format this and save a log file in that directory with the name fetch API.log log. Yes,

not log. Oh, yeah. Let's let's say it's fine. The log file should contain

fine. The log file should contain the information about the API calls made including what APIs

called, what were successful and what were not. Any errors

encountered during the process because obviously we do not need to just make the API calls and it didn't work, bro.

We need to see the errors as well. So

this is my skill. This is my one skill that I have created.

Simple. This is my one of the skills that I wanted to create just for the fetching the data. I can make it more robust. I can make it more and more

robust. I can make it more and more detailed. But we are simply learning how

detailed. But we are simply learning how we can just create the skill. Make

sense? And if I just drag the documentation and we can just also see when you can say one demo skill. Um so

they are also like using skill like very very specific to the point skill. do not

need to put garbage stuff just to be more more detailed because everything will be used in the context. You do not need to waste your tokens, right? Make

sense? Very good. So now I have just provided very like to the point thing.

Okay, very to the point thing. And here

I know like it can just perform this thing. Make sense? So now what I can do,

thing. Make sense? So now what I can do, I will ask it to just fetch the data. So

now what it should do, it will first of all go to each URL. It will run the code in my terminal obviously and then it will handle the responses and then it

will log everything.

I have not asked it to create any kind of code but it will like do it everything behind the scenes. Simple. So

this is my skill one skill only one skill that we have created very to the point. You can make it more detailed but

point. You can make it more detailed but just be to the point as much as you can.

You just need to make sure that you have provided enough information to the LLM so that it can just understand the flow and that's it. And when I just write step one, step two, step three, LLM understands that it cannot perform step

two before step one. This is common sense, right? And you need to provide

sense, right? And you need to provide this information. Simple, very simple

this information. Simple, very simple thing. Okay, so this is skill. Okay,

thing. Okay, so this is skill. Okay,

this is skill. Now, if I want to see how many skills do I have? Okay, so let's see. Let me just open one claw session

see. Let me just open one claw session and I'll simply say skills. So we do not have direct command

skills. So we do not have direct command to access the skills. Okay. So if you want to still see like how many skills do we have? But when I write the name of

my skill that means back like normal slash and then fetch API. Wow. So it has registered my

fetch API. Wow. So it has registered my registered my skill in cloud cloud code.

Why? Because of this architecture.

Because of this particular metadata that I have provided. Wow. Yes. When you

click here, it will trigger this particular exercise fetch API. Simple.

What? Literally yes. I have not even provided any kind of prompt. Any kind of prompt. Now it is saying you want me to

prompt. Now it is saying you want me to run this skill. Yes. I will simply allow it because obviously it needs to perform some function. So it will perform some

some function. So it will perform some batch command. See async io httpx and

batch command. See async io httpx and this is the code that it is trying to run in the terminal directly. I have no code. I literally do not have any code.

code. I literally do not have any code.

See I do not have any code. I have

literally created a skill and I triggered it and I will simply say yes.

Now what will happen? It will keep on performing all the things. And now I can see that it has also opened one window in my screen. It is saying let me try a simpler approach. So it will say yes

simpler approach. So it will say yes because it just popped a one machine. It

is called Python install manager. So I

think it is just trying to download Python. It it doesn't know like I have

Python. It it doesn't know like I have the Python as well. So it will just simply do that. Let me run the script with the correct Python. It will simply say yes and

perfect fetch API worked and wow man literally wow I can see data I can see logs literally wow do you know what do you know what what

actually happened a lot of things happened back in the scenes let me just show you first of all what happened if I go to data boom I have all the files dim

customer csv dim date dim store fact returns fat sales. Wow, logs. I have

logs and I can also read the log.

Wow. See total six successful six failed zero.

Wow. And this is like a daily log. See

date as well.

This is actually amazing. I have not touched any code. I have literally created this skill in front of you in just few minutes. just provided few

steps and I triggered it using slash skill and that's it and that's it and now do you know what let me just

explain you the flow what exactly happened I said /fetch API that is my skill name simple now

I am satisfied with the skill now if I want to perform this entire process I just need to come here and say fetch API and that is it. I do not even

need to provide a prompt. Why? I can

provide a prompt. No worries at all. I

will just show you that as well. I can

say um can you please fetch the data uh using the skill fetch API? I can say that. But I do not need to say that as

that. But I do not need to say that as well. Why? The moment I say fetch API,

well. Why? The moment I say fetch API, okay, it will directly go to my skill.

It will read everything. It will do all the things that I have asked it to do.

Wow.

And how it performed everything because it I didn't provide any code right it created the code first of all see this is the code it tried running it but it

got some error see exit code 49 it didn't even ask me hey this is the code just handle it on your own it fixed the code how it installed some packages

because one thing that I missed I didn't ask it to run my Python through this environment through this virtual environment obviously we need to install async IO HTTPX and all those things, right? It it would have I know it would

right? It it would have I know it would have installed all of these libraries in my global Python environment and it would have used that environment to run my code. Obviously, that is a skill.

my code. Obviously, that is a skill.

See, we run it, we see the mistakes even if those are not mistakes, we will see the areas of improvements and we can improve it, right? Obviously, that is

that is expected. But now what it did it ran that command with new you can say your um your executable python executable and then it keep on see it

was saying python was not found because it didn't actually know like where is my py python then it said python version and which python simple and it did

everything on its own did I did I ask it that hey if you are finding difficulty um just go to my this location you to find the Python. Did I

did I ask anything? No, it automatically solved the issues on its own using clawed code. Clawed code is actually

clawed code. Clawed code is actually doing all of the things behind the scenes, resolving your errors automatically. I am simply allowing it

automatically. I am simply allowing it by saying yes, yes, yes. And that can also be automated. That's not a big deal. And it is running everything

deal. And it is running everything through bash, through terminal. See, it

found my Python executable at this location. at this location. See, wow. At

location. at this location. See, wow. At

this location, I can expand it and see this is not my Python virtual moment.

This is my normal Python. And it is saying now let me run the script with the correct Python command. And then it ran the same command. Wow. And then what it did the same thing that I just asked

you to do save the location, save the result at the given location. And that's

it. See, it actually followed it. Where

is hallucination? Nowhere. Why? This is

agent. This is not just like LLM responses. This is a feedback loop that

responses. This is a feedback loop that is running behind the scenes in the cloud code.

Wow. And once you are satisfied with this thing, it's okay. You can just keep on running the skill. This is fetching.

Now just imagine like obviously we have taken a very basic example because I know I need to just be focused on clot code. But for example, let's say I am

code. But for example, let's say I am fetching my data from let's say Kafka, from let's say event hubs. Can I do that? Obviously.

that? Obviously.

Do I need to write any code? No. How?

Because it knows the code. I just need to tell I just need to create the orchestration and orchestration as well.

I do not need to write any code. Do I

need to write any um you can say airflow code? No.

code? No.

Because airflow is used for orchestration, right? This is my

orchestration, right? This is my orchestration in pure English. Not even

like pure English, like normal English, normal day-to-day English.

Forget about orchestration through coding. Forget about orchestration

coding. Forget about orchestration through drag and drop softwares. It is

pure natural language orchestration of the work. Can you imagine the power of

the work. Can you imagine the power of it?

Make sense? So now you can make the skill even more better by creating that hey um use this particular environment.

Let's say step this is step two and let's say this is step three. Okay, this

is step four and step one will be step one pick the Python environment.

You need to before you start make sure to pick the Python environment in which you want to run the code. You can do this by clicking on Python environment.

Not really. I will say you can run install dependencies

using my dot VNV environment. Okay,

environment. Okay, which is located at and then I can just provide the full

location. I'm just trying to help him

location. I'm just trying to help him because it is better, right?

Simple. Okay, which is this one dot VNV.

Yeah, it will just find the Python executable. So now it can pick that

executable. So now it can pick that thing. Make sense? Then I can just save

thing. Make sense? Then I can just save this thing. And before running that

this thing. And before running that thing, I can just make sure because I'm so sure it didn't pick my virtual moment because I didn't ask it to do it. So now

I can say UV pip list.

Um but if it is smart, it can pick that as well. But oh yeah, wow, it actually

well. But oh yeah, wow, it actually installed HTTPX in my virtual norm. That is amazing. Okay, so

virtual norm. That is amazing. Okay, so

that means it took care of virtual nom as well. That is amazing. Let me just

as well. That is amazing. Let me just confirm it by running the command in my cmd. Let me just check if it has

cmd. Let me just check if it has uh pist uh here. Let me just check

uh here. Let me just check uh httpx. No, I do not have httpx. Wow,

uh httpx. No, I do not have httpx. Wow,

this is my global global environment.

Okay.

Wow. This is actually smart. And you

know what? I am not even using sonet. I

am literally using haiku. IO is like child. Just think about sonnet. Think

child. Just think about sonnet. Think

about opus model. Think about it. Opus

model has the context window.

Context window of 1 million token. Can

you imagine 1 million token context?

Crazy. This is crazy. Oh,

should we just try it with NLP this time? Let's let's do it because

time? Let's let's do it because obviously it will create a new folder obviously because this is like recording the time as well, right? So let's do it.

Let's create a new session. Okay.

Okay. Let's do it. So I will go to new chat. And I have so many chat. Let me

chat. And I have so many chat. Let me

just delete all of these ones.

Perfect. Let me just click on plus new session. Okay. So let's start by

new session. Okay. So let's start by using NLP. Um I want to

using NLP. Um I want to fetch the data from the APIs.

Okay, I want to fetch the data from the APIs um using my skills. Simple. I am

not even specifying which skill. Let's

say it will automatically fetch it.

Okay, let's say processing read. So it

read all the skills. See read. I will

help you fetch. Boom. It picked the skill. Why? using description. Very

skill. Why? using description. Very

good. So now it will create the script.

See this is the script. So now it has actually created the file for me. File

for what? File for actually fetching the data.

Using what? NLP. Using what? NLP. But

now you can say do we need this thing?

Yes. For the next component I need this thing. So that is why I'm simply saying

thing. So that is why I'm simply saying yes. Okay. So now it will say I have

yes. Okay. So now it will say I have performed the script. Let me run the script. So now it will run the script as

script. So now it will run the script as well. You can say yes.

well. You can say yes.

Now it got the error. Okay. Now it will say allow this command. Yes. Do this. It

will automatically solve all of the errors. All of the errors. Perfect. All

errors. All of the errors. Perfect. All

the six CSV files are there. Yes.

Boom. Boom. But this time like something different happened. But that was

different happened. But that was required and that was expected. Okay.

Let me tell you what happened. First of

all, I said I want to fetch the data from the APIs using my skills. This time

or let's say um last time it generated the code. This time as well it generated

the code. This time as well it generated the code obviously but last time it ran the entire code in the terminal.

Yeah. Okay. It ran the entire code in the terminal.

This time it created the file and then it ran that file and then it fetched the data and performed the rest of the stuff. Yes, simple. This is the

stuff. Yes, simple. This is the difference. Now why it happened? Because

difference. Now why it happened? Because

first of all I didn't just say fetch API because when I say fetch API it will not do anything on its own. It will simply follow my steps. But when I use NLP it will actually do something else. it

will actually assume that we need to do this that this this that. So it will try to do some extra stuff. So this is a trade-off. Okay. Whenever you just want

trade-off. Okay. Whenever you just want to perform exact stuff just call the skill. But sometime we do not just want to perform exact stuff.

For example, let's say um you have created this skill, right? Very good.

You use it on a daily basis. Perfect.

Now let's say after 1 week you have got let's say two to three new APIs new urls. Okay. Now you can say could you

urls. Okay. Now you can say could you please run these APIs just for one time and you can use my skill for the

reference and that's it. It will run the new APIs and it will refer your previous previously built API just for the

reference. Simple simple so that is a

reference. Simple simple so that is a trade-off that is totally dependent upon the requirement if this is just a kind of single trigger because some of you

can say hey we are data engineers right and we can simply orchestrate it I know but what if let's say I want to just do something else using the same skill can

you run your pipeline just for new APIs or let's say just out of the box thing I want to ask anything using my NLP can you do that with that pipeline no but with clot code you can do that you can just orchestrate the stuff as well and

do not worry I'm just coming on your point where you are saying that we can just build the scripts to orchestrate everything I'm coming on that point do not worry at all but are you imagining

like what is happening here okay amazing so now let me just show you what actually happened we got another folder because this is a new time frame right all the six files are there 1 2 3

4 5 6 perfect and here another folder for logs see perfect amazing man amazing And it has created one file as well. Um

where is that file? Fetch API data. This

is the file. Now let me just tell you amazing stuff because now we got you will say an Lamad skills are amazing.

Are you are you satisfied with skills right now? Because I am going to share

right now? Because I am going to share amazing stuff with skills. Skills are

not over. It's just started. Okay. What?

Yes. Yes. Yes. So now

now what are we going to do? We are

going to enhance our skills. First of

all, our fetch API skill is amazing.

Okay. So, you have two options to call using NLP or directly call the skill using slash command. Slash command is more convenient. Okay. I personally love

more convenient. Okay. I personally love it. But if you want to just use NLP, you

it. But if you want to just use NLP, you can use it. Simple. Very good. Now, we

have the script. By the way, now do you know what? Do you know what? Let me just

know what? Do you know what? Let me just tell you amazing thing. First of all, this is like doing everything. Fetch and

save CSV. So it just fetches the data and save the CSV and then it just logs the information as well. Everything it

performs very good. So I don't want to touch it. So it will be performing as it

touch it. So it will be performing as it is now.

Now I have the skill. Okay. Fetch API. If

you remember my architecture and you know what what it has done initially remember our Python was like throwing some error. So it literally opened

some error. So it literally opened Microsoft store in my screen. Let me

just show you. See. So this is the window. It has already opened for me. I

window. It has already opened for me. I

was like who opened it bro? Then I was like okay clot code is running. So it

opened this thing Python install manager because it was not able to find Python.

So it literally opened my Microsoft store to download the Python for me so that it can run that code. Can you

imagine? Can you imagine that it can access anything that you have on your machine but obviously with security with safety because you allow it to do that

otherwise it won't do that.

Okay, perfect. Mhm. Okay, so now now this skill is done, man. Oh man, this is this was amazing by the way. I loved it

because writing the code on the fly using HTTPX is not a very easy thing.

Plus logging the information every time when because logs are very very important.

Logs are very important. And can you imagine how long it took? How long it took? 2 minutes less than 2 minutes just

took? 2 minutes less than 2 minutes just 1 minute and you can replace this particular scenario with let's say you are

download not downloading basically fetching the data from let's say um Oracle rest APIs you are fetching the data using Azure event hub you are

fetching the data from anything from anywhere from anywhere just provide the information step by step and that's it you do not even need

to just worry about the code. But if you are someone that like if you are someone who really wants to see the code, ask it to generate the file, confirm the code, test your skill

because skill is actually self-explanatory concept here. You need

to train it.

You cannot expect I will create a skill and it should work on the very first run. But you are creating a skill. Okay,

run. But you are creating a skill. Okay,

you are not just generating a response.

Okay, so when you create a skill and when you test it let's say 10 times, 20 times if you are satisfied now you are okay. Now you can rely because you have

okay. Now you can rely because you have trained your personal clawed code which is purely built on your project and that's it.

Simple. Very good. So this skill is done. Okay. Very good. Very good. Very

done. Okay. Very good. Very good. Very

very good. Now we are going to upscale our skill. What anala

what do you want to do? Let me show you something. We can even extend our

something. We can even extend our skills.

What? Extend our skills. What do you mean extend our skills? So just think about it. So let's say this is my skill.

about it. So let's say this is my skill.

Here I'm just putting everything right.

Here I'm putting every information right? Okay. Let me just answer your

right? Okay. Let me just answer your first question. I know you have one

first question. I know you have one question. Let me just answer that first.

question. Let me just answer that first.

Anala, I can also add this information in my clot. MD

as well here and I can just add that particular step called my skills and I can say whenever I type skills perform these actions. Obviously this is LM uh

these actions. Obviously this is LM uh this is LLM. We just need to provide the context. We can provide context

context. We can provide context anywhere.

That is a beautiful question because I also had that question. So then I got to know one thing which is very very important. So now what will happen? Let

important. So now what will happen? Let

me just explain this answer. By the way, if you didn't have this question, do not worry that that shows that you are not learning. You are learning. I'm just

learning. You are learning. I'm just

kidding. So but yeah, if this is your question, this is a very good question.

Beautiful question because this shows your attitude of how and why. Obviously

everyone talks about how but why? Why?

Why should I do this? Okay. So let me tell you why.

Is my mic working? Yes, it is working.

It is working. I can see the red light.

Do you know what? I just bought this. I

don't know why but now it is just blocking my face. But I can adjust.

Don't worry at all. So now what will happen? Let's say this is what was his

happen? Let's say this is what was his name?

Robin. No Rohan. Rohan. Rohan or Rahul.

Yeah. So Rohan. So now Rohan whenever Rohan calls claude code make sense. Now

let me just take this this entire diagram and let me just type the heading as well.

Um let me just say oo okay why um skills

when we have claw md simple this is your question if this is your not if this is not your question still you should have this question simple so let me just explain

why so now what will happen every time This person cla this uh Robin no Rahul Rohan Rohan Rohan Rohan.

So Rohan every time Rohan makes the request to claude code. You already know this concept that every time this clot

MD file will be injected automatically.

Automatically obviously right. Why? To

provide the context for the project.

Simple automatically.

So we have the tokens here. We have the tokens here. Wow. Do you know what?

tokens here. Wow. Do you know what?

If I put everything in the claw.md

even if I'm not calling a skill, even even if I'm not calling or using any skill, I am wasting my tokens because I'm just passing the entire

context to claude code. So you mean to say that we do not pass context of skill to claude code? No, not at all. We only

pass the context to claude code when claude code invokes the skill. That

means whenever I just use claude code, let's say this is my session. Okay,

let's say this is my session. So here if I ask something else, if I say bro, how are you? Whatever. Okay, do you know

are you? Whatever. Okay, do you know what will go to the clot code? Obviously

claw.md file in the beginning of the session in the beginning of the chat then only this part of the skill will go only

this metadata why because then only it will know that I have the skills otherwise how it would know but after this thing nothing will go to the

context after this thing only this body this is called body of the skill body of the skill will go to clawed code only when clawed code invokes our skill

otherwise it will stay in our skill folder that is why we just create dedicated skills and obviously we have step-by-step implementation which has like so much of thing to do with skill

that you will just learning right now so now you should have this clarity okay good by the way this was like a beautiful question beautiful I also had this question I was like why

I'm just using skill I can just put everything in glmd then I got to that it's all about cost, it's all about tokens, okay? It's all about getting the

tokens, okay? It's all about getting the work done. Because with cloudmd, you

work done. Because with cloudmd, you cannot provide so much of context. It

will be totally confused because when you have one to two skills, you can think like that. You have 20 30 skills, bro.

Bro, are you serious? So, I hope you got the answer like why do we use that?

Okay. Okay. Makes sense. Makes sense.

Simple. Now let's talk about so what we are just talking about okay skills is this one okay so now we were just talking about extending the skills okay now what is this extending the skills

this is very important and this actually makes skills so so so powerful by the way this is like recently added feature skills like extending the skills

now what we're trying to do we know that we create skills using our step-by-step guide using our markdown file using our skill.md file basically. Now I want to

skill.md file basically. Now I want to show you something else. Okay. So let me just bring the documentation bro. Here's

the documentation.

So here um by the way this documentation is amazing to be

honest. Very good documentation. Uh okay

honest. Very good documentation. Uh okay

perfect. Um share skills. No not shared skills.

Maybe I just need to take my own example because I was hoping I can just take an example from here. Restrict tool cause blah blah. Oh yes. Yes. Here is the

blah blah. Oh yes. Yes. Here is the thing and you need to understand this thing please please carefully. Okay. Now

now we can extend our skills. How and

what does it mean? Basically you provide everything in skill.md. Yes. But

sometimes you need to provide some references.

Yes, it can be some kind of references.

Let's say there are some references that I want to perform or let's say there are some references that I want to use whenever it is needed in your skill. For

example, you are fetching some data.

Okay. And you are doing some more operations as well. And within that you want to provide some references. For

example, you can create a kind of markdown file where you have shown some of the errors that your organization has

already faced and resolved.

Makes sense. Or you can say that you have some secrets and keys that you put in the references file. Make sense? Kind

of thing. Okay? I'm just trying to provide you the realtime scenario here. You can also provide you can say a

here. You can also provide you can say a kind of documentation for performing a specific task. For example, you are

specific task. For example, you are asking claude code to do a specific thing. It's a very good thing. Claude

thing. It's a very good thing. Claude

code is trained to generate the code.

But sometimes what hap what will happen is let's say you are asking claude code to um write a pispark code and performing

those things. Okay. But sometimes let's

those things. Okay. But sometimes let's say there's an upgrade in pispark in spark for example spark declarative pipelines which is a very big upgrade.

Simple. Now you are expecting clawed code to generate the code. Maybe

clawed code is not trained on that particular code. Maybe maybe it is a

particular code. Maybe maybe it is a latest upgrade. So what it will do? It can use

upgrade. So what it will do? It can use that information whenever it will be needed.

Whenever cla code will feel like I want to use this as a reference. So you can add this information whenever you need any information regarding generating code.

There is a dedicated folder for you called references.

Make sense? So you can even provide some more context. But whenever it is needed,

more context. But whenever it is needed, you are just trying to help claude code as much as you can. If you are coming from agentic background, what is agentic

background? Let's say you are coming

background? Let's say you are coming from a background who who is like aware of AI stuff. Okay. What is agentic background?

So now the thing is you are simply creating a personal repository on you can say personal directory for claude code just to help claude code.

Here you can also provide some information of your organization.

Maybe you are creating a skill where you want to perform some operations which are only available on your organization.

It is not openly available publicly available. You can provide that

available. You can provide that information in the references. Wow.

Okay.

First thing. Second thing that you can do is scripts. Now being a data engineer, I know I know you had this question. Okay. Some of you would have

question. Okay. Some of you would have this question because I also had this question. So I will be just covering all

question. So I will be just covering all of the questions. Now the thing is I want to

perform something. Okay. I have a skill

perform something. Okay. I have a skill but for that skill for that skill I am a senior data engineer okay in my

organization bro I'm just taking an example do not feel like so what what should we do I am just normal an Lamba okay I have no one

so let's say you are a senior data engineer in your organization now in your organization You already have built some scripts.

You already have built some files, some code files, right? Those code files are running exactly fine. But now you need to integrate those files, those scripts

into your skill.

So instead of asking clawed code to regenerate the code every time because you have written the code for it and you are you know that it is running fine. So

now you can create a dedicated folder called scripts and instead of asking claude code to generate the code and run it, you can simply say bro

go to this folder and run the script and that's it. Wow.

that's it. Wow.

What is the advantage of it? Double

advantage. Advantage number one is you do not need to rewrite the code or you can say get the code rewritten from

cloud code. Second advantage is it will

cloud code. Second advantage is it will not use the tokens. Wow. Because it is not generating the code. So those tokens are saved.

It is not running the code in the terminal. Tokens are saved. It just need

terminal. Tokens are saved. It just need to or let's say it just needs to run one line of code in the terminal. Let's say

Python your code file name py you are hardly spending 15 20 tokens and let's suppose that script is such a

huge script let's say 2,000 3,000 lines of code you have written maybe it's not a big deal it's not a big deal. So you

are literally saving at least at least let's say 3,000 lines of code and in each line you will write almost let's say 10 tokens 30,000 to 40,000 to 50,000

let's say lines like tokens. Wow

60,000 tokens versus 10 15 tokens. This

is called smartness. This is called optimizing your skills to the next level.

Plus performance. We know that our scripts are running fine. So this is called extending skills. And this is where skills become even more powerful.

Like amazing stuff.

Wow. Wow. Wow. Let's try to see it in real. So now what we did we learn? Let's

real. So now what we did we learn? Let's

summarize it. Okay. Because I want you to remember the things because now we'll be just implementing it and now we will not be just revising the concept in the VS code. VS code is for implementation.

VS code. VS code is for implementation.

Okay.

So here we have Rohan.

Perfect. And here we have a folder.

And this folder name is skills. Simple.

In the skills folder, we have skill.md

file. So let's say skill.md file.

Perfect.

Or let's write it like this. not a big um first of all skill name. So we have let's say fetch API. Let's use that same um skill. Obviously we'll be just

um skill. Obviously we'll be just creating a new one but just for this demo. So let's say skills and here we

demo. So let's say skills and here we have fetch API.

This is also a folder.

Okay. So this is first folder remember within that we have what scale.md very

good very good within that we have what remember references

perfect references very good spelling okay references and along with that you have scripts Okay, you have scripts and here you can also put some references just for the

example. So here in the references you

example. So here in the references you can say this is my references documentation

if you want to save okay and say document references and here you can have Python file because Python is a

like generic script that we use simple for data engineering especially. So this

is the hierarchy.

I hope it makes sense. Okay, I hope kill.md.

kill.md.

And here we have references. And here we have scripts. Simple. And here it is

have scripts. Simple. And here it is connected to this one. Perfect. And here

this is connected to this one. Perfect.

This is your hierarchy for skills.

Make sense? So this is your skills.

Now what exactly we need to do in the skills? So here what I will do I will

skills? So here what I will do I will create another skill called migrate called what? Migrate. So now in the

called what? Migrate. So now in the migrate skill what I need to do I need to migrate this data to a different location. Okay. And here obviously you

location. Okay. And here obviously you can imagine that you can migrate your data to Amazon S3 bucket here. Obviously

I'm not bringing S3 bucket here. It will

not take time okay to bring it but it will divert the focus because first we'll be creating AWS account then we just creating S3 bucket then we just creating IM role then we just creating keys so it will just divert the focus I

just want to keep it entirely for cloud code but being a data engineer I'm just telling you that you can do it as a homework you can create a migrate skill that I'll be just creating right now for

the local migration you can migrate the skill for the AWS S3 migration makes sense because being a teacher as well. I

just need to provide you that homework.

Okay, because see I want you to learn and you to master. That's it. That is

the entire agenda of this video and this channel. Simple. Okay, make sense? Okay,

channel. Simple. Okay, make sense? Okay,

so we are extending the skills right now. Okay, so now the requirement is

now. Okay, so now the requirement is very very simple. We need to migrate our data. So let me just open VS code

data. So let me just open VS code editor. Here we have fetch API, right?

editor. Here we have fetch API, right?

And here what I will do I will create another folder in the skills and this will be called as let's say migrate h this migrate skill is very important

here what we need to do we need to first of all let me just close this and close this terminal perfect here what we need to do in the migrate folder we will be

migrating the latest data every time from this fetch API/ data okay so this is the latest folder so I'll be just fetching all of the information

basically all of the data files from this particular folder simple and not just fetching the data being a data engineer we work with big data file

formats we prefer columner file formats so I want this data to be converted to be migrated in the form of park file

simple we can create the skill right but here I do not want to just run the code on the fly because here within the migrate folder I will create yes

skill.mmd which is non-negotiable

skill.mmd which is non-negotiable but along with that I will create scripts and along with that I can create references I do not have any kind of references by the way but I can still

create the folder just for the future use case I might add it whenever I need to but I can just keep it right for now good okay this is skill.md of it. So

here in the scripts now first of all let's try to create the script because script is very important we need to test it. So here I don't have any Python file

it. So here I don't have any Python file but let's imagine see we need to relate it to the real world right. So now what is the requirement? We need to migrate

this code to the new folder right and here I will be creating let's say a folder called data simple and here we'll be having the migrated data the latest folder so this

script what this script will do this script will fetch the latest information from this location from this one okay this folder data it will fetch this folder the latest one and then it will

automatically read all of the files using a for loop very simple and then it can let's say perform the data migration using any kind of data frame library here obviously we are not using pispark

but you can use pispark but in order to use pispark again it will just divert the focus and I want you to be focused on cloud code but you can use pandas for that okay so here we'll be just using

pandas just to create it because pandas do not need any kind of special installation simple common sense but when you can do it with pandas you can also do it with pispark I don't know

like what is the big deal between both it's just about so now I will be just using pandas to convert or migrate this data to the park format right in this

particular folder so let's try to create a script simple and how I will just create the script I can use claude obviously or I can just write the script

by myself let me just complete that so I have created this particular file okay with the help of obviously claude

convert to pocket Okay, so this is the code that I have got and just make sure that you need to download two dependencies pandas and pyro because these are required in order to work with

data frames and converting into park.

Simple. So these are some of the functions that we can just run and I have not tested the script. I can just test it in front of you. I just got the code and if something is wrong, we can just fix it together. Okay. So I can

simply test it. Python

convert to park. py simple. Let's see

what is the response and if it runs fine then we can just add it as a step in our skill. Okay. So let's wait.

skill. Okay. So let's wait.

Uh okay. So

it can take some time obviously because it is performing some things. Okay.

Converting files. Okay. Oh wow. It

worked. Wow. I can just check in the data. Oh man. Perfect. This is actually

data. Oh man. Perfect. This is actually amazing. These are all the park files.

amazing. These are all the park files.

Wow. Amazing man. Amazing. Okay,

perfect. These are the column file formats and with the same folder because the requirement was I wanted to first of all fetch the latest file like latest folder. Okay, this is the latest folder.

folder. Okay, this is the latest folder.

It it didn't read this one. Okay, so

this is like next level code. Okay. Oh,

so it just skipped this folder and then it only read this folder and I wanted Okay, I wanted to keep the folder name same. Why? Because if you have worked

same. Why? Because if you have worked with data migration things, if you keep the destination folder name same as your source folder name, you can debug it

better. If you see any problems, if you

better. If you see any problems, if you just keep it different, you cannot actually find the things in case of debugging. By the way, we are not

debugging. By the way, we are not talking about how to improve the governance pipelines and all. So it's

fine. Okay. So now this is good. We have

the script ready which is like actually very good. So now what I will do I will

very good. So now what I will do I will move this script and I will just keep this folder empty. So I don't need to worry about this thing. So I can simply say delete this folder. Okay? Because we

will just test it with the skill. Okay.

Perfect. So now what I will do I will move this file create convert to bucket control X and in the

migrate under scripts I will put this simple. Now let's try to create the

simple. Now let's try to create the script um skill. So in the skill first of all we will create name. So name will be let's say

uh let's keep it migrate. Okay

simple migrate description.

Um this will help to or let me just get the detailed skill from here because this was really really good. Fetches data handler. Yeah. Yeah

good. Fetches data handler. Yeah. Yeah

it's almost the same. migrates the

fetched data stored or let's say migrates the fetch data stored in the yeah directory to a new location.

Use when you want to move or back up your data or move your data after. Not

backup, I would say move also. Use when

you want to move your data to a different directory or when you want yeah to a different directory.

Simple simple. So this is my metadata. Now

simple. So this is my metadata. Now

let's say usage and if you do not know how this autocomp completion is working for me. So this is like another agentic thing I'm using. It

is called GitHub copilot. If you just click on this thing toggle chat um this is exactly that thing or if you just click on this icon this one. This is the

thing. So I'm just using um GitHub's

thing. So I'm just using um GitHub's agentic code you can say assistance. So

it helps in code completion and I think it is free for some daily limits but if you just want to use it further then yes you need to pay monthly subscription.

Yeah. So yeah. Yeah. If you are just wondering. So first of all you will

wondering. So first of all you will simply say pick the Python environment.

So before you start make sure you pick the Python environment and this is the environment. Okay. Perfect. Step two

environment. Okay. Perfect. Step two

migrate data.

Yes. So here comes the step. Here we do not need to here we do not need to ask it to create the code run the things

whatever migrate data. All I will say run the Python script to migrate the data.

Perfect.

You need to run the Python script stored

at migrate and after migrate we have references not reference scripts

okay scripts migrator py this is like first step so I can simply say migrator py Perfect. What else we need to do? We

py Perfect. What else we need to do? We

can just ask it to do some things. Um,

we can just ask one thing that after migrating the data, do we need to store logs? No. Let's keep it like very

logs? No. Let's keep it like very simple. Okay, because this is just like

simple. Okay, because this is just like data migration. So, we do not need to

data migration. So, we do not need to add useless context just to increase the token size. We just want to perform this

token size. We just want to perform this thing. Okay, simple. Good. So, this is

thing. Okay, simple. Good. So, this is very straightforward. So, now if we just

very straightforward. So, now if we just go to claude, you already know how to just fetch it. Simply use slash and then migrate. Okay. The moment I click here,

migrate. Okay. The moment I click here, it should create this data folder. Okay.

It should populate this. So it is simply saying should I run this? Yes, run it with my virtual environment.

And let me check the skill documentation first. Yes, read it. So see it is now

first. Yes, read it. So see it is now reading my skill.md.

So it is now reading the convert. py.

Now it is running the file. Perfect.

Exit code one. CSU to packet converter and I can see one file converted.

Okay, calculating. Let's wait. Let's wait. I

calculating. Let's wait. Let's wait. I

need to fix. Okay, there's one issue.

What issue? Run the CSV to bucket conversion script with UTF encoding. Oh,

so it found something. Okay, with UTF encoding, which is a very common issue with the CSV files, but I think it will just fix it on its own. Perfect. So, it

makes sense. It makes a lot of sense. So

now I have all of the park files.

Amazing. So it has also provided us a cute summary. Okay, that's good. That's

cute summary. Okay, that's good. That's

good. Oh wait, I can actually ask it to just provide me this thing. I think this is like actually good. Rows, column,

size, and whatever. Wow, because this is all of these things I'm just printing in the I think so in the print statement.

Let me just check because if I have this thing conversion completed successfully print this h okay it's fine we are simply storing

the logs but you can do whatever you want see sky is the limit okay so make sense so I have got all of the files in the form of pi very good and now we

literally saved a lot of tokens because all of the code which is written here in the script it didn't generate this code it didn't generate this code it and E1

pass this code in the terminal to run it. It simply ran one simple command

it. It simply ran one simple command which you can see it here as well. If

you see it here as you can see input is this one. See this is the input. Simple.

this one. See this is the input. Simple.

It just ran this bash command and that's it. Wow. Wow. See this one. Wow.

it. Wow. Wow. See this one. Wow.

But yes, it read the skill.md. Obviously

that is a non-negotiable. That's how it just runs the thing. So this is it. This

is it. This is it. And you know what?

When Nvidia's CEO says that very soon or I think he said in the next 5 years and

already 2 years are already passed so 3 years more. So he was saying that in the

years more. So he was saying that in the next 5 years we will replace the programming with the natural language and I can see that happening easily.

See just tell me one thing what I have put it here. Did I write any code? Did I

write any code here or here? Did I write any code?

Did I write any code?

That doesn't mean that programming is not important. Programming is important

not important. Programming is important and you need to understand the programming because you will be the one validating it, fixing it, evaluating it,

expanding it, all of the stuff. But

the time it used to take to write the code will be gone. Will be totally gone. You

are just a validator. You are just the one who is taking care. Okay, this is running fine. Okay, this can be handled

running fine. Okay, this can be handled like this way because 3 years are left, right? Like still three years are left.

right? Like still three years are left.

It's just like almost almost two years and you can see the big shift because when he made the statement, everyone literally laughed. And you are saying

literally laughed. And you are saying bro this is not possible. We know that we have charge GPT but this is not possible now.

So this is a big shift in that I and I don't know like why people are going crazy all of the all of the people are going crazy who were not there during the invention or you can say during the boom

of computers. Okay but what happened

of computers. Okay but what happened after that? What happened after that?

after that? What happened after that?

Everything changed, right? But it

changed for good cause. Now AI is here everything will be changed right but for good cause

learn grow and that's it that's it that's it so now let's complete our skill number three as well okay and this skill number

three will be like critical one because here we just need to perform much more things what so after making the API call so let's say I'm just repeating my skill to fetch

the data in the morning. I am repeating my skill to migrate the data in the afternoon. And then after migrating the

afternoon. And then after migrating the data as well, I now want to run the visualize skill. Okay. So now in the

visualize skill. Okay. So now in the visualize, I want to create some KPIs which is very simple thing.

Okay, which is very simple thing like major major KPIs, number of orders, number of sales and whatever. So now in this skill I can

tell that what KPIs do I need what are the orders what are the things I need like let's say I want total number of orders by date I want total

number of returns by date um I want total number of orders by customer all of these KPIs I can just ask it to create a dashboard and display these

KPIs how will it do under the hood very simple very simple very simple and Now I want you to create this skill on your

own. Make mistakes.

own. Make mistakes.

Iterate it then make mistakes then iterate it and then you will see that it will work and at the end just to confirm that what you have built and what I'm

going to build just come and confirm and that's it. This is learning right? So

that's it. This is learning right? So

just create the third skill on your own and I know you can build it and I know this is the most critical one among these three but now you are in the shape that you can build it.

Make sense? Okay. So now let's do it.

You build your own skill. Okay. Okay.

Very good. So now what are we just trying to do? We know that we have the data. If I just open the data we have

data. If I just open the data we have dim customers, dim date, dim product, dim store, dim return, dim sales. Right.

So what I will do? I will create my another skill and this will be called as let's say visualize simple and inside the visualize I will

create skill.md

create skill.md do I need scripts um I don't want to write any formulas for aggregations and those DAXs and all not at all I do not

want to provide any references so I will simply create skill.md because I wanted to give you some reference how to you can how you can just extend the skills so now you have the idea simple and one more thing. You do not need to keep the

more thing. You do not need to keep the folder names as it is. You can just put or keep any folder name. This is just for like a standard references, scripts and all of these things. Simple. But you

can just put any folder name because it doesn't care what is your folder name.

You just need to put the folder name in the skill.md. That's it. Here you will

the skill.md. That's it. Here you will say scripts. It will go to scripts. Here

say scripts. It will go to scripts. Here

you will say code file. It will go to code files. Simple. Okay. Good. So

code files. Simple. Okay. Good. So

skill.mmd. Let's try to create the skill.md. So I will simply say name and

skill.md. So I will simply say name and here name is visualize um yes description visualize data using libraries and tools and you know what we

can actually create a dashboard using HTML okay just to make it like really really beautiful and I can also maybe use mattplot lib just to visualize

data like how does it look like whatever we can do let's use HTML and let's see like what does it create I know like it will simply create one HTML with CSS and

styling and I can actually ask it to create one script first of all and I can run that script at the end because every time it will just running so much of

HTML code and you know whenever we just run HTML with CSS we need to just write so much of code so instead of just spending those tokens I can create a file create a script and then I can just

provide it as a script okay so let's try to create one folder for scripts as well that will be a better choice. So what

will be the flow here in the skill.md?

We will simply say first of all get the variables.

Okay, get the variables like all the KPIs then feed those KPIs in the HTML in the dashboard.

Simple. Or let's say just create a dashboard for us. It let let's see what what it will create. Okay, then you can just iterate it. You can simply say um create our um KPIs and then you can just

um build our dashboard based on that.

Okay. So visualize data using various libraries and tools. Um yeah it is fine.

You can just provide some description long enough as well but it's fine. So

let's say usage and here I will simply say make sure to pick the environment and the Python environment is this one because it will be using pandas to

aggregate the data.

So I will say use pandas to um build KPIs.

First of all you need to use pandas library to read c not cs file to read the park files stored in the yes data directory. But

here I want to build the dashboard on top of my latest data. So I'll simply provide the uh command here that I want to read just

the latest data because here I have migrate and then this folder okay read the park file stored in the

latest folder at this time simple then then yes so this is the thing you need to use the pandas library to read the bucket files

you need to use the pandas library to first of all you need to read do not use pandas library you need to read the pocket file stored

in the pocket files after reading data you need to use pandas to build the following kas yes total sales total returns net sales average sales per store per product very

good very good very good average sales per customer very good so these are the KPIs that I want to build for my dashboard what will be the formula we do not need to tell because everyone knows how to find the total sales, sum of

total sales, returns, total returns, net sales, gross minus whatever, then average sales per store, whatever. So,

we do not need to worry about those KPIs. Claude will handle all of the code

KPIs. Claude will handle all of the code for that. So, these are the KPIs. Okay,

for that. So, these are the KPIs. Okay,

this is the step two. So, once we have the step two ready, once we have all of this information ready, then I can go to step three. Yes, use mattplot lib to

step three. Yes, use mattplot lib to build visualization. After building the

build visualization. After building the KPIs, you need to use mattplot lips eon to build. Uh but how I would just

to build. Uh but how I would just visualize it. Mhm.

visualize it. Mhm.

So here we can use map by the way. So

let's use it because it will be like pure Pythonic instead of using HTML CSS because some of you will say hey what is this written HTML P tag BR tag

CSS class whatever. Okay. So let's see.

So after building the KPIs you need to use ML to build the following visualization. sales trend over time,

visualization. sales trend over time, sales per store, returns by store, returns by customer and all of these things. Okay. And then you need to save

things. Okay. And then you need to save the v visualization as PNG files. This

is very good. You need to save the visualization as PNG files. Oh, wait.

PNG files are expensive. No, bro. No.

Yeah, PNG files are expensive in cloud code. You need to um display the

code. You need to um display the visualation in Jupyter notebook. Um

PNG. Why? No, bro.

You can just save it in PNG files for the first time like just for fun. But

why should I do it? You need to display the visualizations um in a Jupyter notebook. Yeah, we can just say Jupyter notebook. If I run the Jupyter notebook,

I want to see it whenever I say visualize. I want to see it on my

visualize. I want to see it on my screen. I want to do it something. I I

screen. I want to do it something. I I

can do it with HTML CSS, but I want to do it with Matt plot lab. Um

okay let me just find one way. So I did a little research on Google and I got to know there's a specific function in plt

basically mcl or plotly we can save the images directly using code but we cannot rely on like cloud code to generate the image or save it. So I can say you need

to display or let's say you need to save the visualization in the as PNG files in the directory visualizations. Okay.

visualizations. Okay.

And I will say visualizations but in the visualizations I should also create the folder by folder because I'm just creating the pipeline. So host

folder and then the latest folder. Yeah.

So in the directory let's in the folder in the folder with the same name as source folder.

Okay with the same name.

Okay. You need to save the folder.

Okay.

Date time. Okay. And I can say using plt dot save f. Yes, because it can just run that function and then python will just save it instead of claw code will just

create the visualization on its own. So

this is better and after that you can run the uh visualization code

file.

Yeah. Scripts. uh you can run the because it is not creating the file actually. So it cannot run it actually.

actually. So it cannot run it actually.

So yeah let's skip it because we have the um functions for converting it so it's fine. So okay let's test it because

it's fine. So okay let's test it because this is still let's see the errors not a big deal. So I will simply say uh

big deal. So I will simply say uh visualize.

Okay let's hit it and let's see what do we get imagining. Wow.

So let's wait and let's see what do we get.

So now it is just creating the script and yes now you know that you can even create a script for so it has created the script that's even better. We can

just reuse the script. Okay, that's the benefit of just creating the script.

That's it. So, it has also created the script visual visualize data. Very good.

And it has just saved here in the scripts. That is amazing. And I need to

scripts. That is amazing. And I need to install. Oh, yeah. It needs to install

install. Oh, yeah. It needs to install some libraries like matt plotly uh all these things.

Okay. Allow this batch command. Panda is

okay. Yeah, matt. Okay. Perfect.

UV not found. Wow. UV note found.

Okay, so it added those Python libraries. Perfect. UV sync. Yes,

Python libraries. Perfect. UV sync. Yes,

you need to just perform UV sync.

And let's wait. Let's wait. And perfect.

Loading data. And okay, it made some changes in the file. Makes sense. Makes

sense. And it made the changes in the script directly. Wow. So that is

script directly. Wow. So that is amazing. So now data loaded

amazing. So now data loaded successfully. Wow. Oh yeah. I can also

successfully. Wow. Oh yeah. I can also see the visualization folder. Perfect.

But it is saying something. So I can simply say yes. Maybe something is left.

So in the visualizations, oh it is empty. Okay. So it is just like maybe

empty. Okay. So it is just like maybe trying to create some visualizations and storing it in the form of PNGs.

Very good. And if the visualize data py is fine, I can also run this file at the end just to pop up the visualizations that we can also do. Let me just click

on the file and see in the meantime. pl

Okay, so it is not doing anything. So it

is again making some changes. It's fine.

So plt.clo Okay, makes sense. So it is not simply keeping the connection on. So

it will simply create the visualization.

Oh, it has created the visualization.

Wow. Wow. Wow. Wow man. Amazing.

Amazing. Now, do you know what? I can

make some changes in my skill and I want to make some changes in front of you so that you will also learn how to iterate the skills, how to perform those things because once we have the script, we do

not need to keep on spending the tokens to perform the things again and again.

Make this edit. So, it is like keep it is like keep keeping the what what is

it? Okay. So now it keeps on editing the

it? Okay. So now it keeps on editing the file based on the KPIs because KPIs are something that you need to

first of all build the logic you need to write the formulas and then you need to perform the things make sense so that's why you are seeing so many edits because

I didn't provide any logic I didn't provide any kind of you can say formula but now it has everything and now it is on now it is a kind of reiterated skill.

Claude code iterated this skill multiple times based on our demand.

If you talk to any BI person, if you will simply say um build this KPI, build this KPI, build this KPI, the person will say bro just

give me the exact formula as well. Even

the person builds the KPIs 24/7. 24/7

means 24/7 of 9 to5.

But still they need formulas. Why?

Because they cannot build formulas on their own. But here it is building the

their own. But here it is building the formulas. Why? because you do not need

formulas. Why? because you do not need to provide the context. It is using its own knowledge to build the formulas.

Formulas are not something like which cannot be built which cannot be predicted. If I if I say you need to

predicted. If I if I say you need to um return the data of returns by store what is the big deal? You can just provide the aggregation but we didn't provide any kind of formula right so

let's check the pictures. So this is wow man amazing amazing amazing man amazing all of these data

analysis stuff is being done by claude code automatically just think that I can even let me just iterate the script okay so I can simply

say um instead of performing step two and step three I can also say simply run my visualize data py because this is very detailed Python script which has

done everything.

Okay, so we can even do that as well if you want to if you do not want to keep on changing our code. Simple.

And I can put this visualize data script here in the scripts folder.

Where is that? Okay, here.

Oh man. Visualize.

Let me just use control minus. Perfect.

So here I can put the scripts here.

Perfect. I can move it. Now I will reiterate my scale because I want to save my tokens. Okay. And I want to uh Okay, I want to simply run the file

simple because I have everything written here. So I will also save some tokens

here. So I will also save some tokens from the skill and I will do direct execution of the skill. So I'll simply say um run the Python script to

visualize data and generate visualizations. So you need to run

visualizations. So you need to run Python script stored at skills visualize scripts. Um

scripts. Um then visualize py. Perfect. This script

will generate this. Perfect. So I saved so much of tokens. Perfect. Now even if I delete this particular folder, let's say delete.

Okay. Now it should perform everything.

Let me just call this scale for one more time. Visualize and this time it should

time. Visualize and this time it should be very very smooth because we have itated our Python code multiple times.

And in your project I can see how self Python file project any specific details file. What

file. What API do you want to visualize? What data

do you want to visualize? Okay. Do you

have data in the CSV, JSON? Okay. What I

did? Let me check what I did. So in the Python file, what do we have? Visualize

data py. Okay.

In the visualize, I said go to claude skills and then visualize. Oh,

scripts. I didn't provide the right name. Okay, makes sense. Scripts and

name. Okay, makes sense. Scripts and

then this thing. Let me just open a new session and let me just say visualize.

I will help you visualize data. What

data do you want to visualize? CSV

existing data frame. Why it is asking me this question? Once you provide this

this question? Once you provide this data, I will prepare the data if needed.

Okay.

Visualizes. Oh, because in the description I have not provided the enough detail description. Visualizes

data stored in this thing. Integrate

visualization on the data. Use this

skill when you want to create visualization on your data. Perfect.

Makes sense. So it's question was valid.

Let's see. Let's try this time.

Visualize. Let's see. So now this time it is saying run the file. So good boy.

Yes.

Oh, I see what was happening. So, I

think it was just treating visualize as the internal command. So, I think so because it didn't even read the skill before. H it can happen because

before. H it can happen because visualize is a very common thing. Oh,

perfect. Perfect. We saved a lot of tokens and our script is working fine.

Wow. Wow. Wow. Wow. Wow. Wow. So, see

now this is the power of creating script because see we are technical people, right? We are developers. So we can

right? We are developers. So we can integrate the coding into our automation into our clot code. But cloud code is very popular when you want to perform

non-coding task as well. Let's say you want to also perform login you can write the code but it will require some steps.

You can totally rely on clot code for that. If let's say non-technical person

that. If let's say non-technical person let's say a manager wants to perform any kind of thing person can do that because clot code can handle all of the orchestration based on the prompt. Then

if you also want to perform some non-technical tasks after that, let's say you want to uh create a draft email, you can just do that. If you want to perform so many other steps, you can do

that. Simple. And in the visualiz data,

that. Simple. And in the visualiz data, you have amazing visualizations ready.

Wow.

Wow, man. Literally. Wow.

Now just think that what if I connect PowerBI with my claw code and if I say just build the dashboard for me

obviously it is so easy for that yes we need to talk about how to connect those external tools okay using MCP and we do not even need to connect to this thing okay because

PowerBI is smart enough to integrate their own agentic graph within the PowerBI itself so that's Why? You see um you write a

that's Why? You see um you write a prompt and you get the dashboard ready.

Boom.

So this is the thing. This is the thing and this is all about the skills that we have learned in clot code and I hope that you have gained very strong

knowledge about skills and how to create skills with respect to technical knowledge as well. Okay. So if you just go here read the documentation you will find that okay we are in good shape.

Yes, you should read this documentation because you will find some new things, some useful things. But I would say most of the things are already covered. So if

you just want to save time, move ahead.

You can just go ahead because rest of the stuff you can just apply or you can learn while applying it in the real world. Make sense? Because this

world. Make sense? Because this

documentation will be keep on changing, keep keep on growing. So it's not something that you have to have to learn everything in one go. Go with the flow.

Simple. So this is all about skills.

Was it helpful? I know it was very very helpful. Just let me know in the comment

helpful. Just let me know in the comment section because I tried my best to explain this critical concept called skills in detail. By the way, there's another concept called skills 2.0. Okay.

And now we'll be just talking about that as well. What is skills 2.0? Now, not

as well. What is skills 2.0? Now, not

exactly skills 2.0. A kind of skills 2.0 or you can say a helper function for skills. Helper function for skills like

skills. Helper function for skills like how we can just create better skills.

Make sense? So, let's talk about that as well. So now we are just going to talk

well. So now we are just going to talk about the skills 2.0 kind of 2.0 know but when you will see it you will actually say this is actually very important and you know what it is a

recent upgrade just I would say last week not even last week I would say this week today is what today is what today is what today is I don't know today's

what but I know it came I think this week on Monday or I would say Saturday Sunday so less than seven days let's say forget about this week last week less than seven days so I want to provide

that knowledge as well This obviously obviously bro obviously so now in order to work with that thing first of all let me just tell you a little bit about plug-in okay we will just talk about plug-in in detail later in the video so

plug-in is basically the package that we install some plugins basically packages are built by anthropic to support our work okay so they have created an

amazing package it's called skill create wow yes so they have created one skill basically what is a package a skill. So

they have created one skill called skill create. That means there's a dedicated

create. That means there's a dedicated skill which helps you build skills.

Why it is useful?

How how do we just work with it?

Obviously we will just look at it. But

before we have some prerequis are creating a skill and you do not know how to define a skill in detail. you

tried chat gpt cloud desktop and so many other things. Claude code knows what it

other things. Claude code knows what it needs to use. So this skill has a lot of questions based on your skill demand.

It asks you those questions in the real time and create the skill.md file for you in detail. And not just skill.mmd

file all of the resources as well. Wow.

And not just this even if you have skills ready the skills that we have created if you want to evaluate those skills with the time let's say you have built one skill today okay after 2 to 3

months our LLMs are more powerful so where you are using some manual work now LLM can handle that work you can evaluate your skills and you can make changes in your skills wow so this is

amazing but if you want to use this thing we have to install some plugins so if I just go to claude now where is my claude Where's my claude? Okay, here

claude code past conversations. Why it

is like this? Let me just restart it or let's say refresh it. Reload window.

H. Okay, so this is also reloading.

Okay, makes sense. Okay, so here we have our clot code. Simple. So here what we need to type, we need to simply type plug-in. Plug-in. Okay, so you will see

plug-in. Plug-in. Okay, so you will see the option called manage plugins. If I

click here, forget about these things.

These are like already installed plugins at my root location. But in this project, I do not have any plugin. Do

not worry at all. So in order to enable plug-in, you need to go to the marketplace.

You add the marketplace and then you can only see the plugins. In your case, marketplace is all will also be empty.

But do not worry. Let me just show you how you can just access those things. So

here first of all you need to simply bring the documentation. So there are two major plugins offered by enthropic and you need to install both. Okay. So

if you just go to cloud documentation, claude code documentation. Okay. And

here if you just go to pre-built plugins, right? So in the pre-built

plugins, right? So in the pre-built plugins, you will scroll down and you will find this link called plug-in installed built name Claude plugins official. And this is the same plug-in

official. And this is the same plug-in that I have installed here, Claude plugins official. What will happen after

plugins official. What will happen after adding this marketplace? All of the plugins that are built inside this marketplace will be available to me. All

of the packages, all of the skills that are already created by Entropic, it will be available available to me. Okay. So, you can also

get this script um you can also get this uh you can say your link and I can just provide you this link as well. Not a big

deal. I can simply write this thing in

deal. I can simply write this thing in the readme.md. Okay. So this is the

the readme.md. Okay. So this is the first link. Okay.

first link. Okay.

And this is the first link that you need to just get and command is this one.

So you can also copy this direct command. I will just provide this thing

command. I will just provide this thing in the readme. You can just check it.

Okay. Read me or I will create a dedicated file. Let me just create one

dedicated file. Let me just create one file. Um let's say

file. Um let's say URLs MD simple because in the readme I will just put the readme stuff not the URLs

because so let's say URLs so all of the documentation URLs whatever we are referring to I will just put it here simple so this is the command so now in order to run this command let's say this

one you cannot run it in the in in this chat no you cannot for that you need to initiate the terminal Okay, how? Let me

show you. So there's like one more command. Let me show you. So this is

command. Let me show you. So this is like one marketplace. There's one more marketplace and this marketplace is actually a kind of GitHub repository that is created by Enthropic. Okay. So

if you just go to skills, you will find this particular repository. If you click on skills, you will see that they have created so many skills. Wow. And this is a skill that we are just trying to cover

which is called skill create which is here skill creator.

Okay. And you can see that the last commit is like last week. So they have literally pushed this particular script last week.

Okay, less than in less than 7 days.

Skill creator. Okay, so this is a skill it will be available to us when we will enable this plug-in. So now in order to enable this GitHub plug-in, you can go to this readme file and here they have

provided this command which is called this one. This one. You can just run

this one. This one. You can just run this command and I will just put it here as well.

So let's say command and this one. So okay. So now I do not have this particular plug-in because I deleted it so that I can just show you the demo. So this is the command that

the demo. So this is the command that will enable the skills and skills means that skill creator skill as well.

Simple. So I will just copy this command here and I will not run it here in the clot code from here. So what I will do I will open the clot terminal. So open the

terminal write clot. it will ask you for the API key for the first time. If it is not asking that means it is already verified. But in your case if you are

verified. But in your case if you are running it for the first time in the cloud code it will ask you. Okay. Very

good. Then I can just paste this command here. Okay. And hit enter. Now it will

here. Okay. And hit enter. Now it will automatically fetch all of the skills and it will put it here. And now if I just go to clot code which is this

normal conversation. If I search plugins

normal conversation. If I search plugins manage plugins and if I just go to marketplace I will see that marketplace here. Wow. And here in the plugins now

here. Wow. And here in the plugins now you need to search the skill creator let's say skill

creator. So this is the skill skill

creator. So this is the skill skill creator at the rate creator plugins official because this is the part of creator plugins official this one this github repo. Okay this is the command

github repo. Okay this is the command that I just already showed to you. So

this is the command which is available.

So I can simply enable this enable plug-in. Wow. So it is saying now you

plug-in. Wow. So it is saying now you have enabled this plug-in. Okay. Now I

simply need to restart. I can restart it. So now if I just say /ash skill.

it. So now if I just say /ash skill.

Okay. This is a kind of skill but we have installed it as a kind of plug-in right? So it will not be like available

right? So it will not be like available as a like skill creator whatever like this but you can still use it. Don't

worry because this is available as a plug-in. Simple.

plug-in. Simple.

I know this is like a little bit confusing but it is very easy. You just

need to install like you you just need to treat plug-in as a kind of package that is provided by entropic or basically clot code. Okay. And we have a dedicated section to discuss about um

your plugins and marketplace. But

currently because we just wanted to use this special feature that's why I showed this thing to you. Okay. And you know how to install all of the things. All of

the things are here. You just need to run these two commands. Okay. in the

terminal not in the chat. Simple. Very

very very good. So now let me just show you the skill creator that we are just talking about and we have just installed plug-in and we have also enabled the skill creator skill. This is a skill

that is directly created by the entropic. So if I just go to their repo

entropic. So if I just go to their repo and if I just go to the scripts um not scripts uh there should be plot code. Let me just go to skills. So

plot code. Let me just go to skills. So

they will have like dedicated repository. Yeah. Skills here. So if I

repository. Yeah. Skills here. So if I just go to skills folder, I should see that skill called skill creator. If I

open this folder, what do we have inside this? We have these folders.

this? We have these folders.

Now here the most important thing is obviously the skill.md file because that defines how the skill should behave.

Second thing is evalu. So this is the evaluator for our skill. So let's say I want to utilize this skill and I want it to evaluate my skills that I have

defined all the skills or let's say a specific skill. So I can utilize this

specific skill. So I can utilize this skill. This is more of you can say

skill. This is more of you can say monitoring and governance skill. It is

not directly related to development but when you work with claude code a lot I think this is the feature which is like recently added just two weeks like I think last week it was updated

it will actually help you a lot because you do not need to govern your skills.

Now Claude code has created a dedicated skill just to govern your skills. How is

it behaving? Do we need to make some changes? Um can we just enhance it? Can

changes? Um can we just enhance it? Can

we just enrich it? So it will automatically give you the suggestions.

So you do not even need to take care of your skills because obviously when when people were creating so many skills in clot code. So the operations cost was

clot code. So the operations cost was obviously heavy because there there would be a dedicated person who would be just going through the skills maybe on a daily basis or at least on a weekly basis like everything is running fine.

Um no bugs nothing. So now it can take care of these things. Why? Because

obviously eval viewer what it will do it will read all of the logs as well of each skill that you would have created.

So it would read the errors if they would see any errors any improvements any deprecations any deprecations because in the logs you also get the deprecation messages. So it will take

deprecation messages. So it will take care of all the packages as well which is amazing and if you are working with lang lang graph I would say you will see those deprecation messages very

frequently because they just keep on changing their mod modules and utilities very frequently very frequently okay so how we can just take care of this thing let's first of all try to read the

skill.md file and don't worry I have

skill.md file and don't worry I have also provided the link to this particular repo if not I can just add it uh this is GitHub skills let and just

add it. So let me just open it. Where is

add it. So let me just open it. Where is

skill? Skill creator. Yeah, perfect.

Here. So in the skill.md file, we will see all of the instructions written. And

this will be a very contextrich skill obviously. Why? Because see it will help

obviously. Why? Because see it will help us to create the skill from scratch. So

it has to cover all of the relevant questions which are required to create your skill because this skill can be used by anyone. data engineers, data

analyst, data scientist, nontechnical people, managers because they can create any skill using the skill. So obviously

this will be very contextrich skill.

Okay. And whenever you will be invoking it then yes you would need to spend high tokens because see this is such a detailed version of it. But obviously

whenever you want to create such a skill you have to provide context otherwise the quality will not be that much high right and it will just keep on writing okay you can just take care of this thing you can just take care of that

thing and you can just read it if you want but the intent is it helps you to create your skill simple if you do not know how to create it or if you just want to create a skill from scratch and

everything okay so it has just simply put the description run loop.py run

eval.py. So they have also created some scripts that they can just run to evaluate your skills. So if you just go to skill creator you will see some scripts. So there are so many Python

scripts. So there are so many Python files they have created just for us just to validate our skills. So if I just go to run eval.py it will just take care and it will run all of the skills and it

will see like if there is any chance of improvement or anything. See everything

is written for you. But yes this is for your governance. So this is a kind of

your governance. So this is a kind of governance layer they have added in claude code right. Okay. So now if you just go to CL um VS code by the way this is optional thing if you want to use it

you can because it will require some heavy tokens but I can just show you. So

what it will do like I will just show you and yeah so what you can do if you want to test it because I don't want to test it I don't want to spend so much of tokens but what it will do as we know that we have the skill ready. So what

you can do you can simp so now we do not have the skill like this let's say skill create now we have skill creator okay so I just refreshed it you can just do it like this but I would not recommend you

to just directly use skill creator let me tell you why because I know that you can just use skill creator directly but skill creator is like very very very

very very generic very generic so what you should do you should say instead of using skill creator because it will automatically invoke that skill. You do

not need to mention skill creator here.

And it's not fruitful to mention it here. You simply need to say I want to

here. You simply need to say I want to create a new skill

to let's say um send the daily report to my manager regarding

the data fetch, cleaning, reporting whatever. Then it

will understand the context of your query and it will directly automatically call the skill create and it will keep on asking you the question.

But what is the benefit between writing this prompt or directly calling the skill create because now it has automatically some context. So the

questions that it will ask will be more inclined towards requirements. Simple.

So it will keep on asking so many questions like what you should do. Okay,

what you need to do, okay, what will be this thing? What will be that thing?

this thing? What will be that thing?

Because obviously it need it it needs to create the skill for you. So it will ask so many questions. You can try it if you want to but I don't want to try it. I

know like how I can just create my skill. Okay. But this is for those who

skill. Okay. But this is for those who do not even need to just write the skill on their own and they just want to build everything from scratch. Simple. So this

is a recent version or you can say recent addition of in the skills version. So that's why some people are

version. So that's why some people are promoting it like skills 2.0 skills 2.0.

It's not actually skills 2.0. It's like

adding a governance layer on top of your skills because it's not skills 2.0 at all. Simple. Okay. So this part is also

all. Simple. Okay. So this part is also crystal clear and we have got amazing understanding of skills. So now let's talk about this amazing concept in

claude code sub aents and this can be a little bit um you can say overlapping um scripts or not scripts like those skills

but it is totally different. Let me just tell you how and why. So basically sub aents means so first of all you have the parent agent what is the parent agent

cla code obviously so let's bring our claude code as our parent agent so whatever we ask we directly talk to claude agent basically claude code this

is our parent agent right very good parent agent perfect so now this parent agent we just talk to this parent agent simple and let's say

Rohan Rohan talks to this particular parent agent.

Simple. Okay. And all of the tools tools means skills. Okay. And obviously

means skills. Okay. And obviously

built-in tools. But built-in tools are like provided by claude code automatically. So that is totally

automatically. So that is totally different. So whatever skills we have

different. So whatever skills we have created, let's say these skills are available to this claude code parent agent. Simple. Yes. So we just talk to

agent. Simple. Yes. So we just talk to this parent agent and it does the work and it is simple. Now we have a concept

of sub agent. That means we can create one sub agent. So let's say there is a sub aent. Let's say this is the sub

sub aent. Let's say this is the sub aent. Okay.

aent. Okay.

This is a sub aent.

Simple. Okay. So this sub agent can be Claude codes junior agent or not just like let's say junior agent a kind of

sub aent. So now this parent agent has

sub aent. So now this parent agent has some team available. Okay now this is you can say a kind of senior person and he's working with some independent

developers. The developer can be junior,

developers. The developer can be junior, senior, whatever. But this claude code

senior, whatever. But this claude code parent agent, senior data engineer or let's say senior engineer is working with some more engineers. Okay. And we

are just calling them as sub sub agent.

Now what will happen? We have provided all of the tools that means all of the built-in tools. Okay. To this parent

built-in tools. Okay. To this parent agent because that is by default.

Perfect. We have provided all of our skills that we have created to our parent agent so that it can perform our work. Yes. Now

work. Yes. Now

every time every time let's say here I'm using haiku model or let's say sonnet because for parent model we usually use set. Let's say set 4.6 which is the

set. Let's say set 4.6 which is the latest set version. So here I'm using set 4.6. Simple. What I will do now? I

set 4.6. Simple. What I will do now? I

am performing everything. Parent agent

is performing everything. Everything is

fine. But sometimes we need to perform some parallel tasks.

We need to perform some independent tasks.

We need to perform some tasks which should not be done by senior person. For

example, I will say um could you please create a documentation for the development that we have done. Would you

ask this thing from a senior data engineer or senior engineer? Obviously

no. You will say like I'm already busy with so much of work. Why do I need to just work on the documentation? Um I can ask someone else. the person can just write the documentation or create the

documentation. So here this parent agent

documentation. So here this parent agent will do the same thing. Let's say I want to just create the readme file or let's say documentation or let's say I just

want to scan my repository or just small small things. I do not need to disturb

small things. I do not need to disturb this parent agent every time I can ask this sub agent not I like claude code will directly spawn one sub

agent to do those small small tasks.

Those tasks can be big as well but for now just for the simplicity we are using this analogy so that you can understand the context. Okay. So now this person

the context. Okay. So now this person this sub agent will do the work. Okay.

And do you know the benefit of it? The

benefit is I can use some you can say affordable or you can say some less expensive models. Let's say I can use

expensive models. Let's say I can use haiku here in the sub agent. Let's say

this is a sub agent. I can use haiku because I know the work that we need to perform is not very very big. Haiku

let's say 4.6 or 4.5 I don't know. So I know that the work that we want to perform is not very very big. We just need to scan the

very big. We just need to scan the repository. We just need to create the

repository. We just need to create the documentation and that's it. Why do I need to use sonnet 4.6?

Obviously and that's how corporate work, right? That's how does corporate work.

right? That's how does corporate work.

They have senior developers, they have junior developers. So if you want to get

junior developers. So if you want to get the work done from those senior developers, you need to pay them hourly, right? And if you know that that work

right? And if you know that that work will take 2 hours. So instead of paying to the senior developer or let's say instead of paying like the cost for 2 hours for a senior

developer, you can just pay the junior developer, it will save cost. Same thing

is happening here as well. Now instead

of paying this glad code sonet 4.6, we can pay haiku 4.6 Plus, it's not just that. Plus, whenever

there's a person, there's a senior data engineer, senior engineer, senior engineer, engine engineer would have access to everything, would have access to your um ad groups, would have access

to your all of the repositories, would have access to your all the things and would have all the permissions and this clot code parent has all the access, all the tools, all the skills, all the

things, everything. But this this sub

things, everything. But this this sub aent can and cannot have all the things.

I can define the permissions because I know this is a junior developer. I do

not need to provide all of my skills because it can do the wonders if I just provide all of the skills. So what I will do? I will just might provide only

will do? I will just might provide only one skill which is not very risky. I

will simply say hey you just need to fetch the data that's it and do not touch the data like this. So let's say

this agent will have only one skill. Okay, only one skill.

And not just this, I will also not do not provide all of the access to this particular person. I will only provide maybe just read access. I will

not provide right access. Maybe this

person can just do something. So I do not want to just provide all of the access especially let's say um bash access because bash access is like risky okay because it can just do anything on

your system so I can restrict all of these things as well. So let's say tools. So I'll simply say you have read

tools. So I'll simply say you have read and after read you have let's say crep.

So that can just search the file then let's say globe.

That's it. You have just these three tools access and that's it.

That's it.

This will help us to do that. Simple.

So this is the agent.

Okay. understood this thing. So this

kind of system, this kind of you can say architecture will ask you to create one sub agent so that you can just allow this sub agent just a few things and not

everything. So this is your sub agent.

everything. So this is your sub agent.

Okay, this is the architecture that we have.

Okay, what are the benefits? Let's see.

And how this is different from skills and we will just talk about that as well because skills is totally different thing. Skill is just like the

thing. Skill is just like the orchestration that you want to perform.

Who will utilize this skill depends upon the agent and who will have the permission to just use all the tools depend upon the tool or basically depend on the um sub agent or agent. Make

sense? Because skill is totally different thing.

Yes, you can define orchestration in agent as well but you should not. Do you

define orchestration in cloud code?

Obviously not because it utilizes the skill and that's it. So here it will simply spawn one sub aent and this sub aent is specialized in doing some work which is let's say calling the API

skills and that's it. That's it simple.

So now because your understanding is like in a good shape so let's try to read what is it in the documentation because in the documentation it's a little bit tricky but I think you can

now understand it. So now let's try to do a

understand it. So now let's try to do a quick scan what is written here. Here it

is written that sub agents are specialized AI assistants. So they are specialized in a specific area. In our

example, it can be API extract. It can

be anything. It can be like scanning the repository and giving us some suggestions, scanning my codebase, giving some suggestions, writing the logs for my coding, everything that

handles specific types of task. Each sub

agents runs in its own context window.

This is very important. That means what will happen here when we just use your claw code directly parent agent whatever

we are just um whatever messages we are sending to to claude code that is being stored in the chat history make sense just to see the context just to see all the results

right but now whenever cla code will spawn one sub agent it will not share the entire chat history no it will be a fresh agent and it will only pass on the

message the latest message that we have simply asked to do. Simple. Rest of the history will not be shared. So this will save a lot of context window as well

because this is an isolated new window.

Simple. Okay.

So it runs in its own context window with a custom system prompt obviously specific tool access that I just already told you and independent permissions obviously when claude encounters a task

that matches a sub aent description it delegates it simply delegates a task to that sub aent and that's it which works independently and return results. Wow.

Yes.

And what is the benefit of it? We can

have multiple sub aents and each sub aent will run in its own context window.

Make sense? Okay. And we can also run them in parallel. I will say hey um I want to do this. I want to do that. I

want to do do this. Claw code will just see all of the sub agents. What are the sub aents do do I have? Okay. This sub

aent can do this work. Uh this sub aent can do this work. This sub aent can do this work. Okay. All of you just do the

this work. Okay. All of you just do the work. That's it. Simple. Delegate.

work. That's it. Simple. Delegate.

Simple. But if I do not have any sub agent for that task, Claude God will say okay let me do it because I am the senior developer. Let me do it for you.

senior developer. Let me do it for you.

Simple simple.

Okay. Now let's try to read some important points. These five points are

important points. These five points are the summary of everything that we have discussed. Why do we need to use sub

discussed. Why do we need to use sub aents? First of all, preserve context.

aents? First of all, preserve context.

Why? I have already mentioned this thing. By keeping exploration

thing. By keeping exploration implementation out of your main conversation. So it will not use your

conversation. So it will not use your main conversation main chat history.

simple enforce constraints because I do not want every agent or you can say every task to use all of my skills to use all of the tools let's say I want to

perform one task where I want to just provide very restricted permissions I can create a sub agent for that so that while performing that task it will not utilize all of the tools that is available to clot code because all of

the tools are very risky all of the skills are very very risky right so I can limit that simple then reuse use configuration obviously because I can just use my all of the sub aents across

projects. I do not need to recreate

projects. I do not need to recreate them. Specialized behavior this is very

them. Specialized behavior this is very obvious and this is the most important thing because I want to save cost. So if I want to do

the work with some cheap models like haiku as compared to sonnet or opus I can do the work because if I know if I just want to scan the repository generate the documentation why do I need

to use sonnet? Solid should be used for like writing the code, writing the better code with documentation within the code, not like outside the code.

Simple. So this is the perfect explanation of claude codes sub aents.

Okay. And how do we create it? It's very

similar to um skill. We just need one YAML and some description. But

description is very very brief. It's it

should not be like very very detailed because we are not orchestrating anything. We are simply saying you are

anything. We are simply saying you are the agent and you are created for this particular area or this particular task or this category of task and that's it.

How you need to do it, what you will do, I do not know because you are the agent and you need to explore all of your tools, all of your skills and that's it.

So this is a kind of creating one small agent out of your parent agent and providing all of the access like tools, skills, whatever. But here you do not

skills, whatever. But here you do not write the orchestration steps. Why?

Because you do not want your agent to do the specific steps. Tuck, tuck, tuck, tuck. No, you create skills for that for

tuck. No, you create skills for that for orchestration. Agent can use those

orchestration. Agent can use those skills. You can allow those skills to

skills. You can allow those skills to that particular agent. Simple.

So in short, it is a like lighter version of your clawed code which doesn't have access to all of the tools because you can restrict them. It

doesn't have access to all of the skills because you can restrict them. It should

not be the same model that you are using as a parent model because you can just restrict that as well using the model parameter. This is amazing thing

parameter. This is amazing thing especially here and how we can just create it. Um I can just show you the

create it. Um I can just show you the example by the way it is very similar to that but when you see in documentation you will feel much much much more confident. So this is the simple example

confident. So this is the simple example here name description tools all of the built-in tools read glob and model here I can say haiku. Okay, there are some

other parameters as well. Let's say

memory here we have memory. If you want to persist the memory for that particular sub agent because see it is an isolated uh agent and if I want that agent to persist its own memory I can

pass this parameter. It is optional. It

is not required but I can pass it. Then

skills. I can pass the skills as well like what you can do. Simple, right? So

we we will be creating like at least one to two um sub agents so that you can just grab the concept. And in first agent we will not provide skills. We

will simply say you have access to just read and grab and globe and that's it.

And in the second agent we will provide skills as well because in the skills we need to write the data as well. Simple.

And choose a model. Yeah we also know that. And tools. Okay. So we can also

that. And tools. Okay. So we can also disallow tools as well. So we can specify hey you do not have write and edit and this is very important when you are just trying to perform one task

which is very very risky. It can just change the code. So you do not want that right. So you can also say disallowed

right. So you can also say disallowed tools. You do not have write permission

tools. You do not have write permission or edit permission. Simple. You can read you can grab you can globe. You can run commands as well using bash command just

to let's say run the python files. But

you cannot edit those files. You cannot

you can say change the code within that.

So this way you can also say disallowed tools. Wow. Simple. Very good.

tools. Wow. Simple. Very good.

Okay. And yes. So let's try to create one sub agent then only you will understand better. But I hope so that

understand better. But I hope so that this architecture provided you some understanding like what is this thing?

This is nothing but just a lighter version of this clot code and that's it.

That's it. Okay. Simple. Why people feel confused between skills and sub agent?

Because the definition like the style of creating it is very similar. But it is not. It is like way too different.

not. It is like way too different.

Make sense? Okay. So now let's open our VS code. So um one thing is same which

VS code. So um one thing is same which is creating everything inside folder because it is so so so convenient. You

just need to explore your docode and everything is written there. Simple. Do

not need to mess up with your entire code base. It's very clean. So we need

code base. It's very clean. So we need to create one folder called agents.

Simple agents.

Okay. I just opened Docker by mistake.

Okay. So now in the agents you need to create MD file again. Oh really? Yeah.

Just MD file. So again you can create subfolders as well similar to skills.

Let's say you have one agent called um codebase reader code reviewer. Let's

create code reviewer because I want to create one sub agent which will review our code and it will write the information into the directory like what should be improved but it should not

have access to change the code but it can give me some suggestion. I will read that description. I will read the

that description. I will read the suggestions and if I think that I should just make these changes, I can make the changes. Simple. Maybe in my code I

changes. Simple. Maybe in my code I missed some edge cases in my code. Maybe

I just missed some missed some type because in Python we use type hands a lot. So it can suggest me those things,

lot. So it can suggest me those things, right? Simple thing. So what I will do?

right? Simple thing. So what I will do?

I will create a dedicated folder for let's say code reviewer. Simple code reviewer. Inside

reviewer. Simple code reviewer. Inside

inside this I can create um code reviewer MD simple because this is the MD file for this. And here again first of all we

this. And here again first of all we need to describe name just grab that particular thing from documentation to save some time not to save some time obviously to do not need to make any

errors and all. So yeah both the things and here we can just paste it like this and I think I can also provide some description after this yaml I think so

just to provide like who are you and who should use you or all of these things but it description is like self-defined so let me just see yeah use a playright

tool to navigate so it is just like providing some more context to um help the parent agent because parent agent will only call this sub

agent if it feels like right it cannot directly call it until unless you just provide some hints in the command but yeah we just need to provide as much

context as we can simple okay here the person is using MCP server but you do not need to worry about MCP server for now MCP server is like totally a different thing and it is an amazing thing and we'll just talk about MCP as

well at the end do not need to worry so here the person is using skills so perfect okay makes sense so here we have applied and name will be let's say

code reviewer.

Okay. Research agent with directed capabilities. No. So I think there was

capabilities. No. So I think there was one very nice description for code reviewer. Let me just search

reviewer. Let me just search code. Yeah. Code reviewer. Reviewer is a

code. Yeah. Code reviewer. Reviewer is a code for quality and best practices. And

let me just paste it here.

And here I can also pass the description as well like the detailed comments.

You are a code reviewer blah blah blah blah blah uh perfect okay so I have said that you are a sub

agent and your area is code reviewing you are a code reviewer as you review code update your agent memory with patterns conventions and recurring issue you you discover so basically I'm saying

that you can just do this thing you are a code reviewer and invoke analyze a code and provide specific actionable feedback quality whatever whatever you want to Right? Simple. So here because I want to use a cheap model. So I can

simply add one more thing which is called model.

Model GPT4. So model will be haiku.

Simple. And I can also say memory.

Memory will be project. Now what is this project? So basically there are three

project? So basically there are three types of memory. User project and one more I think system. So project memory means it will just keep the memory till the project level. It will not mix the memory with the system level or you can

say user level. User means like the systems root. But I will simply say

systems root. But I will simply say project project level. Okay. And for

that I can also create one more folder here.

It's called agent memory.

Simple. So it can just keep the memory if it feels like. So this is my sub agent. Now do you know what I will do? I

agent. Now do you know what I will do? I

will ask claude code to do the work for me. And I will say um could you please

me. And I will say um could you please um review my code written inside the skills folder and give me some suggestions like for all the code files that I have written here like all the

data scripts. Yeah, I have scripts here.

data scripts. Yeah, I have scripts here.

So it can just give me some suggestions.

So let's ask this and let's see if it feels like to call this sub agent. Let's

test it. Okay. So, I want you to um review my code located

inside the claude and then skills directory.

I want to review my code in all the scripts located inside the directory.

Provide some suggestions as well. Let's

see if it automatically calls the sub agent or what.

Uh I think it is it has not invoked that particular thing.

No, it has not invoked that particular thing that your agent sub agent basically.

Okay. So, it has provided me some context but it has not invoked. Let me

just confirm.

Okay. Yeah, it has simply let me just enforce it using the prompt and let's see if it works. Let me just open a new session.

Um use sub aent for that.

because I just want to test it if my sub agent is working fine or not.

Uh let's wait agent. Yeah. So agent is invoked. Very

agent. Yeah. So agent is invoked. Very

good. So now agent is doing the work for me. Very good. So now it has read

me. Very good. So now it has read everything. And now because I didn't

everything. And now because I didn't provide any write or edit permission. So

it cannot edit any kind of you can say my code and it cannot do anything but it will simply provide me the suggestions

and areas of improvements in my code and that's it and it will also use haiku model because I have defined haiku model here so that is also the thing make

sense so it has provided me this amazing summary and overall solid with proper structure and documentation minor issue issues found hardcoded windows path okay So it is saying that you have hardcoded

sub path which you can just use uh path library path li library or os module okay and you have also imported os module but you didn't use it and all of these

things so everything everything everything is there but it has just figured out so many issues with my code repetitive code and incorrect KPI calculation and blah blah blah blah blah

so you can just use this thing use this particular sub agent and this time it invoked that agent because I obviously provide some hints in my prompt but when you're doing this work recursively it will know that we have the sub agent

registered for that as well which is very common sense now I can even enhance the capability of the sub agent I can also say you have permissions

for write and edit as well write and edit as well and I can say you are got it as your review code

pattern suggestion Write detailed, not detailed.

Write feedback in a constructive and actionable manner focusing on improving code quality

at the location.

Claude and then agents then code reviewer and then logs.

Okay. And

I can say at the location what quantity in the folder with current date and time.

Perfect. So now let's use the same round again and let's see what it will do and let's open a new session

control V and that's it. So now this time I hope that it will also write some suggestions in my folder because I want it to do that particular thing. Make

sense? Okay. So now read read and okay now I'll use a code review agent. Wow.

So now it is using code reviewer agent to review these scripts because it knows like I wanted to just do this thing. So

now this is good thing. Now it has invoked code reviewer agent. Very good.

So this way you can actually perform some specific tasks with restrictions and with specific guidance as well.

Simple. Okay. And now I have allowed write and edit command. So it can write the file because in order to write the file it needs write permission. So this

is the way we will just perform all of these things. Simple. And now let's see

these things. Simple. And now let's see what it will do. Synthesizing get

current time stamp and blah blah blah fishing.

So it keeps on just writing these random words just to make you feel happy. This

is a very nice concept by the way. You

improve your vocabulary by reading these words. Okay. Okay. So now let's wait

words. Okay. Okay. So now let's wait until this task is done and then we can also perform one new task. So it has written this particular file which I asked it to do. Very good. I will say

yes. Just do this. And this is the logs.

yes. Just do this. And this is the logs.

Very good. So see code reviewer and it has created this particular file. So

this is the summary of my code which is actually very detailed man. Very

detailed.

Wow.

Now I I can create one more sub agent which can utilize this information and that

particular agent will have let's say all of the permissions to change the code in my files and everything then that particular sub agent can perform these changes in my code. There can be a

dedicated sub agent for that. Make

sense? Okay. So this is this is the way that you can just use to create your sub aents.

Okay, makes sense. So this is the way you should use or you can say include your sub agents. Simple now and this is the agent memory. So this agent memory is for you know that for code reviewer.

So inside the code reviewer it has just created memory.mmd file and this is the

created memory.mmd file and this is the memory for this particular sub aent. So

whenever I will be not I basically clawed code will spawn this sub aent.

This sub agent can use this memory. So

it would know what I did in the past because once the work is done it will simply say okay bye-bye now just take rest but whenever it will be invoked or spawned next time it will use the memory

MD file to see the history like what I did in the past simple and now it has some feedback visualization pattern so it is just for that particular u model

okay how it needs to just carry on the conversation for the future use cases then pattern code duplication and this is the Make sense? So it is just for that

Make sense? So it is just for that particular agent sub agent. So this way you and I didn't create this folder code reviewer. It is automatically created

reviewer. It is automatically created inside agent memory because I just created this agent memory folder and that's it.

Make sense? Okay. Now I will also show you that I can also create one sub agent which will do the orchestration work. So

I know that I have these three skills.

Um I have three skills. let's say fetch API migrate and visualize right so what I can do I can create one sub agent which can orchestrate the work for me

which can actually behave as a data engineer right or let's say orchestrator just to be more specific so this orchestrator can fetch the data from the

API and migrate the data as well from the data stored from the stored in the uh you can say folder I will specify just this thing to the parent agent and parent agent will see

do I have any sub agent for this task?

Yes, I have the sub agent. So sub agent will be invoked with the fresh memory context window and that sub agent will invoke or basically use the skill fetch

API and migrate and that's it.

Are you ready to do this? Are you ready to do this? Let's see how will it will it work. So let's try to create one more

it work. So let's try to create one more sub agent. And by the way, if you just

sub agent. And by the way, if you just search agent here, if you click on agents, you can even create agent directly from the terminal. Oh, so you

can say continue terminal. But I do not like this way. I simply like creating my own agent using everything from scratch.

I don't need to dependent to be dependent on any particular you can say feature because this feature will be available in the extension but this feature might not be available in the

terminal or in any other third party um software where I'm just integrating cloud code. So try to become more and

cloud code. So try to become more and more independent when it comes to the coding and try to use less and less modules from the vendor. This is pro

tip, okay? Because if you use more and

tip, okay? Because if you use more and more modules, you have to be like 24/7 keeping an eye on the upgrades and you have to change the code. No, only use

modules when you feel like first it will take a lot of time to rebuild it.

Second, it is actually solving a problem. Simple. Otherwise, write your

problem. Simple. Otherwise, write your own code. Okay, that's why we have

own code. Okay, that's why we have programming languages. Okay, makes

programming languages. Okay, makes sense. Okay, now let's create one more

sense. Okay, now let's create one more sub agent. So I will simply create one

sub agent. So I will simply create one folder and I will say orchestrator.

Simple. Okay. Within that file I will create orchestrator MD. Okay. And here I will

orchestrator MD. Okay. And here I will simply copy this thing and orchestrator. And let me just close

and orchestrator. And let me just close all of the unnecessary files. So this is orchestrator descriptions.

Um, orchestrates the execution of orchestrates the execution of let's say data fetching

from APIs and then data migration to the new folder using

skills.

Simple tools I have provided read, grap, globe, bash, write, edit. Do I need to provide any other tool? I think these are enough. IO is fine. Memories

are enough. IO is fine. Memories

project. Perfect.

The So, how did we just write this thing? You okay? So we can say you

thing? You okay? So we can say you you are an orchestrator agent which coordinates execution of various skills

to achieve a larger goal. Your primary

your primary responsibility is to fetch the data from APIs and then migrate to a new folder.

You will use your skills.

Perfect.

Perfect. I think this particular context is enough. Let's see if it if it will

is enough. Let's see if it if it will work. And I can also say sales skills.

work. And I can also say sales skills.

And if you want to pass multiple values in the form of list in the YAML because this is a YAML format, you have to use something called as tab and then hyphen.

So the hyphen will be like skills one.

So first skill is fetch API and then you have the second skill called migrate.

Yeah these two skills I want to provide.

So in the YAML it is equivalent to a list. So I have passed a list of the

list. So I have passed a list of the skills. Simple. I can also show you if I

skills. Simple. I can also show you if I can find a good example here. So this is like just a YAML way of dealing with it.

And here I can just show you where is skull.

Let me just search.

Yeah, here. See? So this is the way you can also click or hit tab. But ideal way is using double space. But nowadays you

can even use tab. It was throwing some errors in the beginning when we used to use tab with the YAML files but now I think it is it has understood that we

will not use double space we have to use tab so now let's try to ask claude code I want to

orchestrate my skills fetch API

and migrate migrate to fetch the data and

store it in the folder. Let me just try to invoke it and

folder. Let me just try to invoke it and let's see if it will automatically call it or I have to just put some hints in the prompts.

Uh let's see.

Okay. So I will help you orchestrate the fetch API this in a folder since you have the orchestrator agent. Oh, I can use it to coordinate this workflow. So

it identified that I have this particular agent. Okay. So now what is

particular agent. Okay. So now what is it saying? Before I proceed, I need a

it saying? Before I proceed, I need a few details. API endpoint target folder.

few details. API endpoint target folder.

Okay. So now this parent agent is like just trying to ask some questions just for the sub agent. So that sub aent can utilize that particular thing. Make

sense? But we know that everything is written in the skills itself. So it

doesn't need to worry about anything.

Alternatively, if you already have these details configured in your existing scripts like this thing, I can review those and use orchestrate agent to run the workflow. Would you like me to check

the workflow. Would you like me to check your existing files? So I can say all of the

um URLs and info is there in the skills sub agent is good to go.

So it is just trying to make the process smoother before simply invoking the orchestrator agent without any context.

So it is taking care of that particular thing as well. This is the beauty of this sub aent architecture. It will not directly invoke or make the tool call and that's it. No, it will make sure that orchestrator agent that sub agent

has everything because that is running in the isolated context window that has nothing to do with the parent agent after it is invoked. Make sense? So it

is actually saving your tokens before simply running your sub agent because sub agent can do a lot of things right.

So now what it has done fetch API allow write to fetch API py. So we already have fetch or we do not have fetch API.

Okay I think we didn't have fetch API py. Yes. So it can just create fetch api

py. Yes. So it can just create fetch api py. So I remember we didn't have the

py. So I remember we didn't have the skill. So now it has created the scripts

skill. So now it has created the scripts in the fetch API as well which is good.

It will save our tokens in the future use case as well because we were just running the command in the terminal.

allow right migrate py I think I have migrate py convert to park py so I want to orchestrate the skills okay

let's say yes let me just see what it has done micro migrate py so it has just written migrate py okay so in the skills

Perfect.

And dot CSV.

So it has created a new file but I was expecting it to use my existing file convert to pocket. py. So let me just check maybe I didn't mention it very very specifically in skills and this

behavior is expected because when you are just creating the skills you have to iterate it through multiple times and that is totally fine

and let's wait and I know it will just do the work and let me just see in the skills if I just go to skills skill.md

file and if I just read it you need to run this migrate. Ooh,

I see what happened. I see what happened. Okay,

happened. Okay, I see what happened. So, do you know what? First of all, this work is done.

what? First of all, this work is done.

In our skill, in our skill, we mentioned the wrong file name. So, when it asked for migrate py, there was no file called migrate. py. So, it created the

migrate. py. So, it created the migrate.p file for us so that it can

migrate.p file for us so that it can just use it for the future use case.

Wow, so nice of him or her. So this was the thing. Okay, I was like why I didn't

the thing. Okay, I was like why I didn't use my existing file. So now I know why I didn't. But yeah, it's fine. That's

I didn't. But yeah, it's fine. That's

how you just enrich your skills. Okay,

so now let me just check if it has done the work. So I will simply go to data

the work. So I will simply go to data and I can see the latest data for today.

Very good. All the CSV files are there in the migrate. I should see today's data. Yes, I have all but this is just

data. Yes, I have all but this is just CSV files. Yes, because it didn't it

CSV files. Yes, because it didn't it didn't run this particular file convert.pocket. This is my mistake.

convert.pocket. This is my mistake.

Okay. So, totally up to me. Okay. So, I

can just correct it right now. Um,

convert to park because obviously it doesn't know that it needs to convert that particular file in the you can say park. We just

informed that you just need to migrate the data in the migrate folder. Simple.

So that makes sense. Okay. So that is something which is up to me. Okay. So I

have corrected it. So it is fine. So

that's how you can just create these sub aents which will take care of independent tasks. Now do you know what

independent tasks. Now do you know what what is the difference that we have simply faced here. Our parent claude code agent didn't run these particular

skills. Our sub agent ran these skills.

skills. Our sub agent ran these skills.

And we're using haiku model.

Wow.

And I can also ask clawed code multiple tasks in one go. Do this, do that, scan my codebase, um run the orchestrator a region, blah blah blah. It will spawn two different regions automatically

automatically and it will do the work. But for that we have much more formal way. By the way and to be honest it is not recommended

in the beginning stages because when you spawn multiple parallel agents every agent is running it in its own context window. So that can utilize a

context window. So that can utilize a lot of tokens. Make sense? So you should not spawn multiple parallel agents. You

should behave smartly. You should create sub agents obviously because sub agents can do the dedicated work but you should not just spawn 10 agents together. Hey

do the work do the work do the work.

What you need to supervise them right and just for this particular task because obviously in the industry not as you like in the industry they are running like like they are spawning

multiple agents parallelly right. So for

that particular concept they have introduced something called as agent lead. Yes. So agent lead takes care of

lead. Yes. So agent lead takes care of that particular all of the sub aents. By

default agent lead is turned off and I would also recommend you to just gain the knowledge around it but do not try it directly because you are in the learning phase. You have to just make so

learning phase. You have to just make so many mistakes now before directly jumping onto the agent leads because agent lead which will just spawn

multiple agents together. That's it. And

more the number of instances, more the number of tokens and the more more the number of you can say your API calls and your cost. Simple but still obviously we

your cost. Simple but still obviously we are mastering clot code. So we will investigate and we will learn about agent lead as well. But it is turned off by default. Okay, let me just tell you

by default. Okay, let me just tell you this thing. But it is used to spawn

this thing. But it is used to spawn multiple agents in parallel because your claw code can create team leads as well.

Okay. Now those sub aents will directly report to the team lead. They will not directly report to the claude code which is the parent model. Make sense? So they

are simply trying to delegate the tasks and they are just simply trying to replicate the things that we see in the corporate. they are also doing the same

corporate. they are also doing the same thing which is like obvious because they are just AI is just trying to um you can say not replace but they are just trying

to mimic the same thing right because that's the only agenda simple so they are also doing the same thing simple so now let's try to learn about the team leads or basically agent lead or lead

agent what is this thing okay let's try to so now let's try to talk about agent team leads okay and this is very similar to those

corporate team leads. Okay, so now what is the difference? Okay, Angela, what is the difference? Because we have

the difference? Because we have understood sub agents and amazing amazing explanation. I do you know what

amazing explanation. I do you know what I also loved this explanation. When I

was learning this thing, this was not that much clear to me. But now this explanation I would say the best is given to you. So now let me just explain one thing before jumping onto the team agent because you will see the

difference like what is the major difference and let's talk about the okay this thing. So let's say we have this

this thing. So let's say we have this agent. Okay. Now I created this agent.

agent. Okay. Now I created this agent.

Yes, this agent. Okay, I created this agent

this agent. Okay, I created this agent and this agent is running in the

isolated context window. That means it is running in the isolated context window. Let's say this is a sub agent

window. Let's say this is a sub agent one and this is running in the isolation. Wow.

isolation. Wow.

He does not worry about anything. Okay?

He doesn't have to worry about any teammate, anyone. He just reports to his

teammate, anyone. He just reports to his manager or let's say senior data engineer.

He is the most sincere employee. Okay?

And he works by himself. And this guy doesn't have any kind of middleman. Okay? So this guy directly

middleman. Okay? So this guy directly talks to the clot code. Wow. Okay.

Simple. Similarly, we have another agent that we just created and you already saw. Let's say this agent. And this

saw. Let's say this agent. And this

agent will also directly talk to the cloud code to the main agent. Okay.

Simple.

To the main agent. Okay. Now this is isolated. This isolated. They cannot

isolated. This isolated. They cannot

talk to each other. No, this is not allowed.

This is not allowed. You cannot talk to each other. Okay.

each other. Okay.

But there's no icon for this. So let's

say they cannot talk to each other. Why?

Because each sub agent is running in the independent context window.

This agent is different. This agent is different. Yes, different. But they

different. Yes, different. But they

cannot even talk to each other. Simple.

Okay. Now,

now when we have the concept of agent team or basically agent lead or lead agent whatever you want to say, there are so many names but the official name which is written in the documentation

which is agent team. Simple. So now in the agent team

team. Simple. So now in the agent team what will happen? First of all we will be having the claude code which is the parent or main agent.

Simple. Okay. This is there now along with this claw code. Okay, main agent.

Now this main agent what it will do? It

will obviously spawn multiple agents like this. Let's say one

like this. Let's say one like this two. Okay, let's say one more. Let's

two. Okay, let's say one more. Let's

create one more. Let's say this one.

Okay, there are three agents, three sub aents.

Now what will happen instead of spawning multiple sub aents claude code this parent agent okay senior agent will

create a kind of list okay we'll create a kind of list shared list so this is like shared tasks list simple so it will

create this thing and we need to achieve this thing we do not care who will do which work okay obviously We do care about this thing. We have dedicated

people. But here what will happen? This

people. But here what will happen? This

will be delegated to these three agents.

Okay.

I am hungry.

So 1 2 and three. Simple. Now these three agents can talk to each other as well.

Wow. Wow. This is different. Okay. They

can talk to each other. Let's say like this.

Okay. Or let's create just two-way. It

is very obvious. So they can just talk to each other.

This is the difference.

And they can share the results as well.

They can share the context window as well. Wow. This is the only difference.

well. Wow. This is the only difference.

Okay. And they all communicate through this particular shared journey.

Okay. They do their work. They do their work. They do their work. But they also

work. They do their work. But they also they also talk to each other. So what

can happen? We can just create multiple sub aents and this clawed code will just create a shared task list. Okay? And it

will say u we have sub agents to do the work. So now what will happen? This sub

work. So now what will happen? This sub

sub aent will say um I can just take this task. I can do this. This sub agent

this task. I can do this. This sub agent will say um I can do this that do that task. I can just take this task. And the

task. I can just take this task. And the

third one will say the same thing. Now

because these two let's say agents have shared skills as well. So they will delegate the task between them. They

will say hey you do this thing I will do that thing. Perfect. Oh wow. This is

that thing. Perfect. Oh wow. This is

actually autonomous but it has some restrictions because this is recently launched and this is in the beta version and even the claw team is saying that hey this is in the beta version and we

know there are some issues with this particular architecture. We are working

particular architecture. We are working on it. So you can just play around it.

on it. So you can just play around it.

You can just test it but if you want to implement the real world solution you should be cautious simple and cautious as in like in the good way in the healthy way that we are just working on it and this will be implemented very

soon. So let me just show you run agent

soon. So let me just show you run agent teams. So see this is the notice agent teams are experimental and disabled by default. So you cannot even directly use

default. So you cannot even directly use it. Enable them by adding clawed code

it. Enable them by adding clawed code experimental agent teams to your settings dojson. Now what is the

settings dojson. Now what is the settings dojson? So inside your VS code

settings dojson? So inside your VS code inside this docloud folder you can create settings do.json file and you will see that settings.json will be just working with MCPS. Okay. So you can

create settings.json and you can enable

create settings.json and you can enable it. I won't enable it. Okay. You want to

it. I won't enable it. Okay. You want to enable you can do it. Okay. Just test

it. Okay. And agent teams have known limitations. They know this thing around

limitations. They know this thing around session resumption, task coordination, shutdown behavior like when one sub agent will shut down, when one agent will just resume the you can say context

window. So they have these things but

window. So they have these things but this is really nice. This is something which will actually perform all of the things autonomously and this is actually

very good. Okay, obviously maybe they

very good. Okay, obviously maybe they would need to just do a lot of improvements in their LLMs because everything is LLM. Brain is LLM.

Currently they have Opus 4.6.

In future they will launch Opus 5.0 maybe and then that LLM can handle this thing because see what is happening under the hood. Why did I show that very

basic light version of claw code in the beginning? I wanted you to realize that

beginning? I wanted you to realize that everything is actually a kind of tool call but the efficiency of tool call depends upon the architecture that

you're using plus the brain because brain will say call this tool with these arguments with these particular things.

Make sense?

And now unlike sub agents which run within a single session and can only report back to the main agent, you can also interact with the individual teammates the same thing that I just

mentioned directly without going through the lead. So they do not need to just

the lead. So they do not need to just report back to the manager. They can

also talk to each other. Hey bro, what are you doing? Um I'm simply ingesting the data since yesterday. Um okay, let me just help you out and I can just also do the work. Let me just take some of your tasks. Simple. Now the lead is not

your tasks. Simple. Now the lead is not actually worrying about who is doing the task. They simply know that how a task

task. They simply know that how a task are being done. That's it. This is

called team work right that's why they have just called it as like agent teams. So they have just created a team around it. And now this parent agent agent is

it. And now this parent agent agent is actually working as a team lead right.

So team what what team lead will do?

Team lead will say I want to do this work. Okay let me just do it. And it

work. Okay let me just do it. And it

will just share all the tasks and it will also do some task. Team lead also perform some task some senior level task but all the junior level things all the dedicated task will be delegated

simple simple okay and now we can just look at this beautiful diagram so sub agent this is the main agent they just spawn sub agent okay spawn sub agent spawn sub

agent and this is the work that they need to do and that's it they simply report back it to the main agent that's it but with the agent team they just create a shared task list and then they

just create teammate instead of sub agent because they are not like now sub agent they are like teammates they communicate directly okay with the main agent through the share task and also

communicate with each other wow this is amazing I know like what are they just trying to do they're just simply trying to run a loop within the team agents and they're just simply trying to call agent

from the main graph so it is actually a complex one but it is doable see when you have funding of billions of dollars you can just build anything. It's not

impossible. It's not impossible at all.

When you have like great developers, you have great funding, you can build this thing. It's not something very good like

thing. It's not something very good like wa what are they building man? LLMs are

good because obviously we do not know how to build LLMs like Sonnet but these graphs are like not very very big thing.

If you know how to build graphs, if you know how to work with um nodes, orchestration, edges and all those things and if you're a data engineer, I would say graph building is the easiest

thing that you can do. Why? Because who

who are we? Who are we data engineers?

We build pipelines. We build graphs. So

yeah, so this is about agent lead. This

is a new thing. Obviously, we just need to gain the knowledge around it conceptually but not practically. Okay,

very good. So this is all about your agents and sub agents and this new feature called agent team in cloud. So

now let's talk about this important this important pillar of clot code. So now we know that we can work with clot code. We

have seen how we can just create new new skills orchestrate work and everything.

Anjula being a data professional or being an being a developer let's say we sometimes let's say um directly get on

boarded on the project yes can claude code help us in that particular area as well I would say yes for example let's take a very simple

example and let's take a data engineering example so let's say you are a developer where is Rohan let's say Rohan one is

cute little guy. So let's say this developer okay is newly onboarded on the project simple okay now the project or basically

organization is building a project around Apache airflow which is a very in demand technology right now because it's not now like it has been like in demand since the beginning but now it is being

even more in demand why because of its latest upgrade to Apache Airflow 3.0 0 and everyone knows the power of it. So

yeah and if you want to learn Apache Airflow 3.0 I have created a dedicated course on Apache Airflow 3.0 as well.

Simply search Apache Airflow and Anlamba on YouTube. You will find that as well.

on YouTube. You will find that as well.

So everything is available on YouTube.

Everything is available on this channel.

That's why I simply saying I'm simply saying hit the subscribe button so that you can find all the resources. So let's

say this person this developer is working on a project and project is built on airflow. Simple. Okay. Apach

Airflow. This is the project. Now what

we need to do or let's say what this person can do. So this is a project.

This person is working on this project and he is directly on boarded. And this

is the scenario as well in the industry.

You will rarely be onboarded on the fresh project on the raw project because each project right now in the industry is either migration project or let's say

um optimization project or enhancement project or blah blah blah simple right so let's say you are also onboarded on the project where you are working with

the teammates senior developers junior developers and they're like proper team for it so you are working in the in this project simple. Okay. So let me write

project simple. Okay. So let me write let's say project.

Now the thing is in this project you know claw code right. So how can you utilize claw code

right. So how can you utilize claw code skills here like your code codes code claude code skills here in the apache

airflow project. How that is a question.

airflow project. How that is a question.

It is very easy. It is very very easy and it will actually boost your performance in the project. You can

actually write a lot of code, a lot of DAGs automatically.

Yeah, I will just show you. Don't worry

at all. And I will also show you the step-by-step guide like how should you start, how should you get the context first of all whenever you are onboarded on the project. Okay, so let's try to see that particular thing. Okay, this

would be very helpful and I will also tell you some amazing commands, okay, which are newly added, newly added because cloud code is growing rapidly.

So this is my VS code editor and inside this VS code editor I have created this folder called airflow project. Simple.

Don't worry you will find this folder with this particular repository. So you

can also try and test your commands of clot code with this project. Simple. Now

I will simply open this folder. So

everyone knows that airflow is used to build DAGs. DAGs means directed asyclic

build DAGs. DAGs means directed asyclic graphs. Even if you do not know about

graphs. Even if you do not know about airflow do not need to worry. We are not learning about airflow. You can treat it like one of the technology that we use in data engineering. Okay, simple, very

good. So now let's say you are a data

good. So now let's say you are a data engineer onboarded on the airflow project. This project can be a DBT

project. This project can be a DBT project. This project can be let's say

project. This project can be let's say any project simple but just for the simplicity just for the explanation I have used airflow project because this is one of the most relevant

topics. So inside the airflow project

topics. So inside the airflow project when you are onboarded you already have some DAGs. See data fetch data report.

some DAGs. See data fetch data report.

Wow. Yes you already have some DAGs. So

now obviously being a new developer on boarded on the project you first need to get the context. What is this thing? Um

what the things are already being written in tags. So traditionally you will go to each code file and you will just try to read the file. You will try to understand it and it will take a lot

of time. And in this particular D

of time. And in this particular D folder, do you know what I have built?

Assets. Assets is the most recent advancement in the Apache Airflow era.

Okay. Yes. Apache Airflow 3.0 era basically. So even if you know airflow,

basically. So even if you know airflow, even if you are a senior or let's say mid-level data engineer, you might not know what is an asset and what is this

thing written?

It's possible, right? It's possible

because no one like not everyone knows about assets. If you are upgraded, if

about assets. If you are upgraded, if you are updated with the um knowledge that you have around the Apache Airflow world, you know like this is an asset. I

can just go through the code. But just

imagine that you when you do not even know what is an asset, how will you just proceed further obviously right? So let's imagine that

obviously right? So let's imagine that you do not know what is an asset. That's

why I just built this code here in this project so that you would also feel that hey I need to understand what is this thing. So what I will do I will simply

thing. So what I will do I will simply go to clot code and I will say I have an airflow project

at claude airflow by the way it will automatically fetch it but when you just provide some context you can save some tokens okay um project at this location

simple I want to get the context and I want to understand what's going on

in the DAGs in brief. Simple. So create

a readme dot MD file for me for that project inside the

Airflow project. Simple

Airflow project. Simple project folder. Simple. Hit enter.

project folder. Simple. Hit enter.

So now what it will do? It will first of all try to read your DAX. See

it will try to read those DAXs. Okay.

Then oh I just provided dotload. I just

provided wrong destination. So that is why it automatically corrected that thing. Okay. Sorry. Sorry. So that is

thing. Okay. Sorry. Sorry. So that is fine. So I simply write dotcl then slash

fine. So I simply write dotcl then slash then this thing but it is available on the root directory. So it corrected it automatically. That is the power of

automatically. That is the power of cloud code. So now it has read both the

cloud code. So now it has read both the files. See now it will try to generate a

files. See now it will try to generate a readme file for me so that I can actually analyze it. It gives okay it is ready by the way. Let me just say yes.

And you can see readme.md.

Very good. Now I can just read this readme file. Okay. I can just close this

readme file. Okay. I can just close this window. And that's it. In order to

window. And that's it. In order to render it, I have already installed an extension which is this one that can render the readme file. You can also download it from the extensions. Simply

search markdown extension and you will just find it. Okay. So this is airflow project data pipeline orchestration project using Apach Airflow. Perfect.

This is project overview. This is the first DAG. This is the second DAG. DAG

first DAG. This is the second DAG. DAG

uses asset based orchestration. Wow.

Wow. Asset based orchestration means data level orchestration or data first orchestration. Obviously we are not

orchestration. Obviously we are not learning Apache Airflow here but yeah even if you do not know about these things you can get the context directly from your project. Wow. And project

structure is this one. Amazing. Amazing.

It has literally created the project structure as well. Obviously you can see that thing in the repository but you have also got some oneliner summarization as well. Right. Simple.

And then DAGs overview. Now it has provided amazing man. First of all, this has

amazing man. First of all, this has provide this has understood the code and now this has literally translated that technical code that you might not understand because of upgrades in the

Apache Airflow 3.0 because of assets because of everything now you can understand the code using your natural language. See it is saying this is the

language. See it is saying this is the workflow prepare storage that means for um your asset fetch API data then transform data then materialize asset.

Wow. And this is the exact thing I'm doing in the DAG exact thing. And I can even show you see if I just go to data fetch, you will see that thing here.

There are 1 2 3 4. Four tasks. Wow, man.

Then these are the tasks. 1 2 3 4. It is

also explaining you the functions, the tasks that I'm creating with all the you can say necessary information.

Amazing. Then second tag. Wow. Key

features. Modern approach uses airflow's asset object for cross stack dependencies. Very good. Loose coupling.

dependencies. Very good. Loose coupling.

Yes, that's the advantage of asset.

Automatic triggering. Yes. Data pipeline

extract transform load configuration is this one. And it has also defined you

this one. And it has also defined you the ascent definition which is very important for this entire tag. And this

is produced by this tag. And do you know what? I'm really happy with this

what? I'm really happy with this particular area because I was not expecting that it will tell me which DAG is producing the asset and which DAG is

consuming the asset. But but this is amazing because the latest asset bundles or you can say asset um features available in the Apache Airflow is very

very new and this has also mentioned the sample data flow and this is amazing and it has also told you like how to run these things. Oh man, amazing.

these things. Oh man, amazing.

Amazing.

This is amazing. This is amazing. So

that's how you can literally scan the entire project and can get a very simple readme file out of it so that you can understand what's going on in the

project. Now being a developer, you do

project. Now being a developer, you do not need to contact the senior developers. Hey, can you please explain

developers. Hey, can you please explain me the project? Hey, what is this asset?

Hey, what is this code? What is this function? And why did you use this

function? And why did you use this function? Nothing. Now you just need to

function? Nothing. Now you just need to sit on your system, run some cloud code command because now you are educated in this particular era of claude code.

Simply understand everything and the people the team around you would think this person is just onboarded. This

person didn't even ask me anything like what is this function? What is that function? How how this person is like

function? How how this person is like able to work directly on the project?

Because this person knows how to use clot code. Simple. Simple. And this

clot code. Simple. Simple. And this

person doesn't copy and paste your code to chat GPT to understand each part with hallucinations. This person uses CL

hallucinations. This person uses CL code. Simple. And this readme file is

code. Simple. And this readme file is amazing. I'm telling you just go to any

amazing. I'm telling you just go to any project you won't find any readme file with this much of simplicity and detail. Simple.

This is amazing. This is amazing.

This way you can actually work with any data engineering or any technology with claude code because claude code has

models. Haiku, Sonnet, Opus. They are

models. Haiku, Sonnet, Opus. They are

well trained on all of the frameworks on all of the technologies. They can

literally understand your code better than them who built them. I'm telling

you this thing.

Wow.

So whenever you are onboarded on any project, simply use simply follow this exercise.

This is very useful when you have complex DAGs. Currently we just have two

complex DAGs. Currently we just have two DAGs. Just think about the scenario when

DAGs. Just think about the scenario when you'll be having like let's say hundreds of DAGs. you will and not just you

of DAGs. you will and not just you everyone would spend at least a month to understand what's going on and people start judging you that hey this person is not performing this person is not doing that thing this person is not

doing that thing I know this happens with the consultancy based companies I know this thing and I literally um hate that particular pattern of corporate but it is what it is it is what it is but

now shock those people okay give a shock because now you know how to perform you know how to become better.

Not becoming better than anyone, you need to become better than yourself.

Okay. So this is about how you can integrate claw code with any technology.

Now you can replace airflow with dbt and dbt is like one of those technologies which like every company is using right now. You can integrate with that thing.

now. You can integrate with that thing.

Just uh think about apache spark when you have so many notebooks. Simply use

this thing. Simple

simple this way you can integrate claude code with any project. Claude code is yes that developer who knows how to work

with data engineering pipelines data sets everything. Now you can even ask

sets everything. Now you can even ask claude code that hey I want to create one more DAG which is dependent on this particular DAG. Could you please do it

particular DAG. Could you please do it for me? It will build a DAG for you. You

for me? It will build a DAG for you. You

already know how to do that. You can

even create a skill as well for that particular thing to do it recursively.

Simple. Simple.

That's how clawed code is not replacing you. Clawed code is actually making you

you. Clawed code is actually making you much more productive and sooner you will become more productive, better position you will

gain in your organization because of obviously after some time everyone would know clot code as well. But the ones who already know clot code will gain some importance within the organization right now which is the best time because when

this thing is like becoming common every everyone would know cloud code. Everyone

would know like how to do these things.

Timing is everything bro. You need to just learn cloud code right now and you need to show that you can say efficiency productivity in your organization right

now because after a few months I would say everyone would use clot code. It's

not a big deal. It's not rocket science.

It's very easy to use. But it's like how efficiently you can integrate it with your day-to-day activities.

Simple. Okay. Nice. So now what you can do, this is all about your cloud code integration with project. Now I will just tell you about some commands that

are very very handy. Okay. The first

command that I want to just talk about is very new by the way. It is recently added again. So this command is called

added again. So this command is called btw.

btw means by the way. By the way, clot code. Really? Yeah. Yeah. BTW. And by

code. Really? Yeah. Yeah. BTW. And by

the way, why did I just open this amazing um browser? Because

this browser is the best browser to give the related web pages. And

claude code btw.

So claw code settings settings I think it is in the references.

If I go to references btw command where is that references interactive mode checkpointing let me just search it for

you. Okay so finally I have found this

you. Okay so finally I have found this command btw. So what is this command? So

command btw. So what is this command? So

btw means um I don't know if the full form is like by the way but I think the way it behaves it is a kind of by the way command. So let's say you are having

way command. So let's say you are having a conversation with clot code and the context window is really really big and you want to keep that context because you are doing your work in the middle of

your conversation let's say you and me are talking okay and we are having a very great conversation and in the middle of that you say by the way um do

you eat this thing you can ask any side question right which is not directly relevant to the chat and you do not want to include that question in the chat

because model can use that as a context and you do not want that. It is

basically a side question like you and Claude code are talking to each other.

you both are like collaborating to do the work in between you want to say bro um what does this function do like you are doing your work you're fetching the

APIs you are doing something um could you please explain what does this function do out of nowhere so you want to ask a side question and you do not want to pass the entire chat history for that side question because that is

irrelevant so you can use by the way command I don't know if the full form is by the way I'm just simply assuming because that behaves like this but I think it will be like this. So let's say

use by the way command to ask a question quick question about your current work without adding to the conversation history. So let's say

history. So let's say this is useful when you want a faster answer obviously but do not want to clutter the main context. Yes. Derail

cloud from the log running task.

Exactly. And this is an example. By the

way, what was the name of that config file again? Obviously you can just ask

file again? Obviously you can just ask anything. anything.

anything. anything.

This can be related to the conversation that you are having in the middle. You

can say, "Hey, can you please repeat that thing? I forgot." Or you can simply

that thing? I forgot." Or you can simply ask a different totally different question.

Simple. And the advantage is it will not use the context in the form of tokens because every time we are asking anything, the entire chat history is going to the conversation so that it can just remember your conversation. Makes

sense. But this is amazing. Okay. And

what are the advantages? available while

cla is working. That means if the claude is working, you see you you see like um thinking or blah blah blah. Those simple

words, you can even run / the way when claude is working. Wow.

So you can run by the way command even while claude is processing a response.

That means even if that particular clawed code agent is busy with something the side question runs independently and

does not interrupt the main run. Wow.

So it is like let's say um you ask me to do something and I'm doing that work t in the in the middle you said um um aren listen um could you please tell me what is that thing? So I'm doing my work. I

will say um yeah yeah it is it is based on that particular function and that function just prints this command and that's it. Yeah. So I'm just resuming my

that's it. Yeah. So I'm just resuming my work. Simple.

work. Simple.

Simple. I know that the question that you just asked has nothing to do my do with my work that I'm currently doing.

Simple. And obviously no tool access because side questions are just like short answer that I want to just get and cla cannot read files run commands.

Obviously because I'm doing some work.

You cannot say um okay can you like can you just please do this work while typing? Obviously no. I will simply say

typing? Obviously no. I will simply say I'm currently doing my work but I can answer some quick questions. simple and

single response. There are no follow-up tones. If you need back up and forth

tones. If you need back up and forth like back and forth um kind of question and answer thing, use a normal prompt instead. If you want to just continue

instead. If you want to just continue the conversation like let's say you are asking me a question and I'm just doing my work and you will say um do you know what this print command is doing? So I

will say yeah this command is doing that thing. Now you'll say oh okay how did

thing. Now you'll say oh okay how did you create that file? So I will just look at you. I will I will say bro I'm just doing my work. If you want to ask something, let me complete this work or

just ask someone else. Just create a new session and that's it. And low cost that means context is not being provided every time to the side question that

means cost is low. Simple. Yes. So let's

try to fun let's let's try to have fun with this command by the way command which is newly added. So let's use it.

So let's say I am just using clot code.

Okay. And I can just use a past conversation and I can simply say um by the way it is saying

no command. Are you serious? Are you

no command. Are you serious? Are you

serious? Are you serious? No matching

commands. Do I need to upgrade my claw code, bro? Let me just check.

code, bro? Let me just check.

Let me just check. So it was in the references. Checkpointing. Yeah, not

references. Checkpointing. Yeah, not

checkpointing. Interactive mode.

Uh by the way okay type slash in the cloud code to see all available commands. Type followed by any letters.

commands. Type followed by any letters.

Not all commands are visible to every user. Okay. Some depends on your

user. Okay. Some depends on your platform plan or this. For example,

desktop only appears on Mac and OS.

Upgrade this this clot code also ship with bundled skills like simplify, badge, debug. Okay. In the table below

badge, debug. Okay. In the table below add directory ask site question. Let me just open this in the new tab.

Let me simply try to access this command to the terminal. And you do not need to open the terminal. I'm just simply testing it out because I also didn't use this command. It is like recently

this command. It is like recently recently launched while creating this video. So I was like let's include this

video. So I was like let's include this plot. Let me just search here. Oh yeah

plot. Let me just search here. Oh yeah

by the way. So it is available here but it is not available in the extension because I would need to might um reinstall the extension because it is not upgraded and it is just recently launched this by the way command. So I

can just use by the way command in the terminal because terminal is the like latest version every time. Simple simple

so I can say by the way and now it is asking me to just put the question.

Simple you can also just open terminal if you want to perform this activity.

Okay, that's why I love like working with terminal whenever I feel like because terminal is like terminal best.

Okay, so by the way, I will simply say um by the way and this is a new session so it doesn't have any kind of context to our um conversation but still just to

show this I can say um what is an asset in airflow DAX? Wow.

DAX? Wow.

So it will simply answer that to that question and currently I didn't ask it to perform any other task but you can ask it to do the thing so that it will have some context and it will not use it in your particular command or question

and why assets matter it has just returned everything to me and detailed answer is there very good so you can simply use assets like this

press space enter escape to dismiss.

Okay simply say escape done and let me just say exit. So if you just want to perform this thing, you might need to reinstall this particular extension. But

you can also use terminal which is a better way because here you do not see that command but you can see it in the terminal. Simple. So this is about the

terminal. Simple. So this is about the btw command. Wow. Now there's one more

btw command. Wow. Now there's one more command which is newly added. Okay. What

is that command? That command is called voice. Okay.

voice. Okay.

Voice. Is it here? No. Okay. Let me just search voice. command

search voice. command not good uh using voice mode. Okay.

So here because see whenever we work with the claw desktop application obviously we have the button of mic we can enable it and we can speak it because sometimes we need to explain the stuff in hurry and we cannot write

everything. So they have created this

everything. So they have created this command for claw code as well. Um I

don't know if they have enabled it for me or not. I can simply check. So here I can say let me just open this

extension. So I will write voice. Now

extension. So I will write voice. Now

maybe this extension is not updated.

Let's test it here.

Clot and then voice. No. So it is not available here as well. Maybe this will not this will only be available for the

ones who have the subscription because this feature was primarily added into the claw desktop application pro or that particular version max version. But I

can do a quick little research because I have never tried this command but I know what this command will do. So let me just tell you uh do they have any documentation? I don't know like if I

documentation? I don't know like if I search on Google they they show us the videos bro just show us the web pages first of all you are search engine.

So let's read this thing. Cloud code now supports a native voice mode / voice allowing you to interact with the terminal via speech for coding refactoring. Basically what what uh this

refactoring. Basically what what uh this command will provide. You can write / voice and when you run this command you can start talking and whatever you will

be just speaking it will be directly passed as the prompt to clot code and it will do the work. So instead of writing everything every time you can simply speak and this is the only difference

and I don't think so it is like a very very gamechanging feature but sometimes if you're just working at home you can just speak because obviously if you're working at office you cannot speak every time simple but yeah so it is currently

rolling out to 5% see I was so sure that it is not available to everyone right now so it is currently rolling out to 5% of pro team or enterprise so I am not using the pro version I am not using the

team version I'm not using the enterprise version. I'm using the API

enterprise version. I'm using the API version, console version. So I do not have access to this command. And by the way, I do not need this access. This

like it is good but currently while recording this video, it is fine. Okay.

So this is it about the voice and your PTW command. Now let's talk about one

PTW command. Now let's talk about one more important one which is very important. By the way, whatever we are

important. By the way, whatever we are doing here, whatever we are doing here, like creating the skills, let's say I am just

running the skills. Uh, let me just close this terminal.

Yeah, let's shut it down. So, let's say whatever I'm doing, let's say I'm just using this clot code. So you have you already know that I can run this skill by simply typing

fetch API migrate visualize all of these things. I can also write the prompt. I

things. I can also write the prompt. I

want to do this. I want to do that.

Simple. Let's say I want to perform this task or let's say these tasks recursively. I want to perform these

recursively. I want to perform these task daily at 9:00 a.m. sharp.

That means in short, I want to schedule my um work that I'm just doing with my clot code. In clot code, I simply pass

clot code. In clot code, I simply pass the prompt, right? Simple. or I simply use the skill and then I simply write my NLP

like natural language simple now I can even schedule my prompt that whatever prompt I'm writing here in the box I can schedule it so that that prompt will be

running automatically at let's say 9:00 a.m. 10:00 a.m. or any time that you

a.m. 10:00 a.m. or any time that you specified in the schedule. So that's

called something called a scheduled option which is again used with loop command.

Let me just search if it is available here. These are the new new commands by

here. These are the new new commands by the way. Loop command clot code.

the way. Loop command clot code.

How do how the agent loop works. Let's

see if we have it here. Agent SDK. This

is the agent SDK. I don't think this is like clot code.

How cloud code works. run prompts on a schedule. Yes, I think this is the one.

schedule. Yes, I think this is the one.

Yeah, this is the one. So now let's try to explore this documentation.

So you got the context. We want to run our prompts on a schedule. Simple. If

you are a data engineer, you can just create it like when we schedule our pipelines. Simple because obviously

pipelines. Simple because obviously every skill, every orchestration is a pipeline in itself and we want to schedule it so that we do not need to run it again and again. So what you can

do? You can use loop command and the

do? You can use loop command and the chronululing. Chrome is basically the

chronululing. Chrome is basically the syntax that we use to schedule anything star blah blah blah. Okay. So what we need to do we need to simply write loop

command then time using chron syntax and the prompt and that's it. Just these

three things. That's it. Okay. And

schedule task require claw code version 2.1 2.1.72 or later. Check your version with a clawed version. Let me just check mine

because I installed it like way back.

Let me just check.

Let me just check claud. You can also check your version.

So I have 2.1.74 and here they are saying 2.1.72.

Okay, we are good. Very good. Otherwise

I would have to download the latest version but we are good. So schedule

tasks let Claude rerun a prompt automatically on an interval. For

example, let's say you have some weekly deployments that you need to do. So you

can create a skill around that and you can say run this skill every week on let's say Thursday at 3 p.m. and that's

it. Wow. Or let's say you are a senior data engineer and you do the PRs as the first thing in the morning like all the pull request

analyzing all of the code everything you can literally build a skill using your GitHub commands basically you can just provide all of the access and then you can say analyze all of the PRs for me

and just give me the brief of all of the PRs in the morning and just run it at 8:30 a.m. So that the moment I will

8:30 a.m. So that the moment I will reach the office I will just read all all of the things. I not need to spend my hours to just go through those things. Wow. You can literally do this

things. Wow. You can literally do this thing.

Amazing. Okay. So this is the kind of schedule that we build. Make sense? So

schedule a recurring prompt with loop.

So as I just mentioned you can simply say loop and then whatever your chrome syntax is and then your prompt. So here

you can see loop 5 minute check if the deployment finished and tell me what happened. Simple simple and this is

what happened. Simple simple and this is a very good example loop and 30 minutes.

30 minutes means like in every 30 minutes but I would prefer writing the cron syntax because chron syntax is like more professional and then check the build simple prompt

and then check the build that is prompt and then every 2 hours. So basically you are trying to schedule something using NLB

but I would recommend use loop command then use cron syntax because it is more professional and then use your prompt and that's it.

Simple okay and if you do not provide everything if you just simply use loop and your prompt then it will run that particular piece in every 10 minutes.

Wow. Okay. So supported units are S S for seconds, M for minutes, H for hours.

Simple and then you can just see. So

this is a very basic one if you see and then set one time reminder. If you just want to set the reminder, manage schedule task, you can literally do everything with loop command. And see

here they just mentioning about the chron syntax.

Chrome syntax is the best syntax that you can just use. Simple and that's it.

So this is these are some examples if you do not know how to use chron syntaxes. So if you just use this thing,

syntaxes. So if you just use this thing, it means like every hour on the hour basically in every hour. Okay. And then

seven like every hour at 7 minutes past.

And then this like every day at 9:00 a.m. local. Local means like your

a.m. local. Local means like your systems time stamp. Okay. Simple. And

how we can just disable because this is very important. Let's say we want to

very important. Let's say we want to disable the loop. Set claw code disable chron equals to 1 in your environment to disable the scheduleuler entirely. That

means all of the loops that you have created so far, you can simply created this. You can simply create this

this. You can simply create this particular variable in your environment to disable the scheduleuler entirely.

The chron tools and the loop become unavailable and any already scheduled task stop firing. See environment

variables. Let's see it. Environment

variables. So these are the environment variables in the clot code. Now

if we want to set the environment variable we have set the environment variable in the env right but if we want to pull those environment variables

directly in the cloud code we can just put them in the settings dojson. So all

the environment variables can also be configured in the settings dojson file.

This is useful as a way to automatically set the environment variables for each session. Wow. So every time whenever we

session. Wow. So every time whenever we want to use cloud code it will go to settings.json JSON file to get the

settings.json JSON file to get the environment variables. Simple and you

environment variables. Simple and you can just simply put entropic API key, entropic off token, entropic custom mode, all of the things are here. All of

the things like all of the environment variables that you can just set simple now let's try to run this thing and now let's see how does it behave. So here

before directly implementing the stuff let's try to discuss what are we just trying to do because obviously we can just schedule anything but what exactly we areuling so first of all I will just

show you when we have a skill okay let's say when we have a skill let's bring the skill API calls I want to schedule only

this skill simple only this skill we'll simply Okay, schedule simple. Okay, so first of all,

schedule simple. Okay, so first of all, I will schedule only this thing.

Then once I know like how to schedule it, then what I can do? I can simply set the timer like how frequently I want to schedule it or basically let's say in 10

10 minutes 20 minutes 30 minutes and I will I will also show you how you can just delete the schedule a specific schedule because obviously we have an environment variable where we can

disable all of the schedules but how to disable only a specific schedule how to disable only specific job in the data engine language so I will also show you that thing as well and it's not just

aboutuling ing a skill. You can also schedule any random prompt. What do I mean? Because see under the hood, we are

mean? Because see under the hood, we are not scheduuling a skill. We areuling a prompt. So let's say here I do not have

prompt. So let's say here I do not have any skill. Okay,

any skill. Okay, I just want to schedule a prompt. prompt

can be let's say this uh update readme.md file in every

readme.md file in every 30 minutes simple I want to schedule this prompt simple I can do

this thing I can do this thing simple so you can literally schedule one skill because when you'll be justinguling a skill that will also a prompt you will Okay, I want to invoke my API skill

skill API call skill. Simple. When you

just invoking a prompt which has nothing to do with the skills, then claw code with will simply run that prompt

in the or you can say on that particular moment. Simple

moment. Simple cloud code will say this is my prompt.

This is the time I just want to run this prompt. and then rest of the stuff will

prompt. and then rest of the stuff will be handled by the LLM with the clot code. Simple. So now let's try to build

code. Simple. So now let's try to build something. So now what I will do I will

something. So now what I will do I will open my clot code and let's say I want to um use loop command obviously loop and

then I will use chron syntax. I can use chron syntax but just for simplicity I can simply say 3 minutes because we will be just keeping it very very quick. So

let's say 3 minutes and I want to just provide a prompt. I will simply say invoke my skill

to fetch the API data simple and let's hit enter now what it will do it will create one chron function for it so now

let's wait first of all see it is using chron create function and it will just pass all of the required parameters and it is saying Auto expires after 3 days that means it will be automatically

expired after 3 days. But I can also delete this particular schedule automatically or you can say manually because it will be automatically deleted after 3 days. But if I want to manually

delete this before that I can utilize this job ID. Okay. And I can use chron delete function. Chrome delete and then

delete function. Chrome delete and then simply job ID and that's it. Simple.

Okay. Now this is scheduled. This is

scheduled. This is done. Okay. And you

can also see schedule task.log. lock.

Okay, if I just open it, it will just open the session. It has nothing to do with your development. Nothing. If I

just go to fetch API, if I go to data and after 3 minutes, I should see one new data because it will just trigger that particular thing obviously and it will be invoked in every 3 minutes.

Okay, so let's wait and let's see what it will do and I will also show you how you can just delete this job because otherwise it'll be just keep on calling the you can say API. one prere request that you should have your system should

be on while performing these things because obviously without the server without the system it cannot perform anything it needs a terminal so that is a drawback for now not a drawback that is a common sense you cannot schedule

anything even using your task manager taskuler on your normal system so let's wait for 3 minutes and let's see how does it perform the work and then and

you do do you know what you can even create another schedule so let's say I will Okay. Um loop and then I will write

will Okay. Um loop and then I will write or see baking. So I think it is it has simply yeah it has simply triggered the skill.

Wow this is amazing. This is amazing. So

now automatically I didn't do anything.

See it has automatically triggered that particular skill for me. Simple. And now

it is just simply creating that particular script for me. And obviously

because skill will just use that particular script which is fetch API and in the scripts fetch API.

Oh again my mistake I have not put this fetch API data inside this or I would I didn't mention it in my skill maybe. So

that is kind of skill issue because we didn't refine or enrich it but that is fine. I can even say um I want to use my

fine. I can even say um I want to use my existing script

which is inside the uh scripts folder.

Simple.

So I can also argue with it while running this thing and it will automatically save this information in the um chat history so that next time it can

improve the you can say your calling or invocation. So it is saying that

invocation. So it is saying that fetch API data. No, I don't want to use this thing. I want to use maybe I didn't

this thing. I want to use maybe I didn't use the exact scripts use case. So if I just show you in the migrate I have the scripts here in the visualize I have the scripts in the capital S. So maybe that

will be like typo but that's fine. So

run the API data using uh script fetch script using the env. I want you to use

fetch API. py

fetch API. py inside um cloud skills

then fetch API and then scripts folder. Now I hope it will be

scripts folder. Now I hope it will be able to find that particular script and that's how we can just train our clot code with some iterations.

So now it will just simply run this and that's it.

Okay. Yeah. Now you can just run it.

Login error that is fine.

Perfect. So this is done. Simple. Now it

will again trigger this thing in 3 minutes obviously. So now I don't want

minutes obviously. So now I don't want that right. So now because I do not want

that right. So now because I do not want that what I can do I can simply delete the job using what chron ID this. So let

me just quickly get the job ID and I will simply say chron delete and this thing.

So now this will simply delete this particular chron job using my job ID.

Done. Cancelled the recurring job.

Perfect. The every 3 minute API data fetch task has been stopped. Now it will not fetch any data. Nothing. Why?

Because I have deleted it. And I can also confirm it. It should not trigger anything after this fetch API data. I

just have four files. Two for yesterday, two for today. So that is fine. Okay.

That's how you can just do this thing.

You can even create some you can say random prompts. Let's say I will say loop

prompts. Let's say I will say loop in every 3 minutes update

uh your let's say readme.md inside

airflow project. Let's say this one readme.md

readme.md inside airflow project folder. Simple. So it

will simply schedule this task as well.

And here I didn't ask it to schedule any skill. I simply asked it to schedule my

skill. I simply asked it to schedule my prompt. Simple. Okay. So now this is

prompt. Simple. Okay. So now this is scheduled.

But if I want to cancel it, if I want to just delete it, I can simply run this command gro delete and this job ID because I don't want to run this. simple

but I'm just showing you you can just schedule it like this. Now this feature is also available in the claw desktop pro version but in the form of UI you can schedule the task using buttons

because obviously cloud desktop application is for those who do not want to write everything in the form of code and they just want to um you can say create some schedules just for their day-to-day activities that they can

operate from the claw desktop application. But if you are doing

application. But if you are doing something just for your code and obviously when you do something any any like do anything or something for code you need to open clot code within the

code editor then you have to know all of these things. Simple. So this is the

these things. Simple. So this is the best example how we can just take care of chron jobs and deletion. When you

have multiple chron jobs, obviously it is um like it is next to impossible to remember all of the job ids until and until or unless you just store or create

a log of all the job ids. And you know what you can even ask claude whenever you are just creating any chron job. You

can simply say create one log file for me where you will store all of the scheduled jobs and just add a brief about it like this.

Chrome job or chron ID is for this task.

This chron job is for this task and that's it. That way you can just manage.

that's it. That way you can just manage.

But if you do not want to manage all of these things, the simple thing that we just discussed right now, how we can just delete it using the environment variable, then you can simply create

settings.json file and simply store that

settings.json file and simply store that variable here. Uh

variable here. Uh where did we just see that particular thing to cancel the command? I think

here disable. Yeah. So then you can simply put that environment variable in the settings dojson file like this and then you can simply say equals to 1 I

guess uh equals to 1. You can just check or double check the value it was zero or one I don't know but you can simply create that variable inside the settings

dojson and it will work. Simple. This is

like scheduuling your prompts and skills for your clot code. Simple. Okay. So now

let's talk about this amazing feature plugins and marketplace. The good news is you have nothing to do with this particular thing plugins and marketplace. Most of the time yes you

marketplace. Most of the time yes you can just create your custom plugins but what exactly the plugins and marketplace are under the hood. Basically whenever

you want to use the packages right packages means skills which might be created by the third party providers or which will be created by the community

or which will be created by the anthropic itself. So all of those skills

anthropic itself. So all of those skills packages um all those you can say utilities can be shared through these two ways plugins and marketplace. So

first of all what they do they create a marketplace okay and in that marketplace they add more and more plugins and you are already aware of plugins and marketplace because you have installed

two marketplaces right and within that marketplace you got access to so many plugins skill creator is also one of the plugins right but if you just want to

explore more and more plugins you can simply open your clot code and search like basically command slash and then plugin you will Manage plugins. You will

see everything here that you have and there's a long list. See, you can install any plug-in. Any plug-in, any plug-in, any plug-in, these are what these are kind of packages, skill, um

any kind of utility, anything. And these

things are available because you downloaded or basically installed marketplaces in your cloud code. Now you

can install any plug-in. Simple. Let's

say you want to connect um to any MCP server. MCP servers are also available

server. MCP servers are also available through marketplace. See Figma, Figma is

through marketplace. See Figma, Figma is a kind of MCP that is available. We will

just talk about MCP. Don't need to worry at all. But everything MCP um skill

at all. But everything MCP um skill everything will be available through the plugins and the marketplace.

Simple. So then you have let's say um anything like plug-in dev, grappile, notion. Notion is a very good example.

notion. Notion is a very good example.

So what is notion here? Notion is

basically a plug-in that is available and under the hood do you know what is that thing this is actually the MCP server but that MCP server is well

integrated with claude code so that is why this behaves like an MCP server but now it becomes plug-in make sense but actually it is one of the

MCP that we have here in the clot code so everything that you need to access through the marketplace plugins or the third party providers. You access it through the plugins. See, there are so

many things. You can literally install

many things. You can literally install anything. The moment you will install

anything. The moment you will install it, you can easily access it. No, no

need to worry at all. Simple. Okay. Very

good. And you can even see the documentation if you want to. So, if you just go to plugins and marketplace. Uh

where is the plugins and marketplace?

Okay. So, plugins and yes, you can even create your custom plugins because you know how to create skills. What is skill plug-in? What is an agent plug-in? All

plug-in? What is an agent plug-in? All

of these things are plugins, right? So,

plugins let you extend clawed code with custom functionality. So, we know how to

custom functionality. So, we know how to create our custom plug-in. We know how to create skills, right? But we can also use a skills which are created by the other people. Simple. And this is the

other people. Simple. And this is the documentation for that. If you just want to just create a plug-in, you can create a plug-in directory and so on. But do

you know what? Most of the time you will not create your custom plug-in. You will

be just using third party plugins because there are amazing plugins available. Yes, amazing plugins

available. Yes, amazing plugins available and amazing skills available.

So if you want to just search for the skills or plugins that are available in the market, you can simply go to let's say skill hub this is or let's say

skills MP. This is a very good place

skills MP. This is a very good place where you can just use the skills which are created by the third party or some developers some community people. So you

can see there are so many skills available. So you can just use any skill

available. So you can just use any skill from here. And there's one more site I

from here. And there's one more site I think smithyri and here also you can find so many skills. Smri and skills MP are two most

skills. Smri and skills MP are two most popular ones but there are many many many and even on GitHub you will see so many skills. So these are the skills

many skills. So these are the skills which are already developed by the people. So you can use their skill

people. So you can use their skill instead of just creating it from scratch. You can just pull it you can

scratch. You can just pull it you can just use it and then you can just refine it. Simple

it. Simple make sense? So these are the things that

make sense? So these are the things that you can just use and these are very well integrated in the form of MCP as well but we will just talk about MCP. Don't

need to worry. Okay. So this is all about the plugins and the marketplace.

Nothing to do much with that but yeah you should be aware of this thing. So

now let's talk about my personal favorite topic in this particular clawed code or aenti era which is called MCP

model context protocol. First of all let me just give you a quick um info. MCP is

itself an independent topic that we need to learn separately. Yes. So this is a quite harsh truth. So it is not that is

directly related to claude code only.

MCB is the standard that we have in the AI industry. So it is not something okay

AI industry. So it is not something okay MCB is only available for clot code. So

it is just a component within cloud code. Not at all. We enhance the

code. Not at all. We enhance the capabilities of cloud code with MCP but I will obviously explain some basics of

MCP but you need to learn MCP separately. Now I have not created any

separately. Now I have not created any video or tutorial on MCP. But if you want a dedicated video on MCP because without MCP you cannot build powerful AI

agents in today's era. So if you want a dedicated video just do let me know in the comment section. I can create a dedicated video on MCP as well. But in

today's lecture we will see very basic and very overview of MCP and we'll see how we can just integrate MCP within our

clot code. Simple.

clot code. Simple.

Let's do it. So basically what is this MCB model context protocol? Because

everyone is talking about MCB and do you know what MCP was also built by entropic who built clot code? Hm. Okay.

Now um MCP is not a very straightforward thing. First of all, it is like very

thing. First of all, it is like very twisted thing. But let me just twist

twisted thing. But let me just twist that twisted thing to convert it into a straight line. So the thing is let's say

straight line. So the thing is let's say you have clot code. Okay

because you first need to understand the thing with respect to clot code. Okay.

Do we have some tools available within clot code? Yes. What are those tools?

clot code? Yes. What are those tools?

Uh these tools.

Okay. These tools. Okay. Makes sense.

Perfect. Simple.

Okay.

Now, this claude code has some built-in tools. Yes.

tools. Yes.

Do we have anything else? Yes, we have skills. Yes. Very good. Very good. Very

skills. Yes. Very good. Very good. Very

good. We have skills. Very good. Let's

say we have these skills as well. Okay.

Perfect.

Now, let me just replace this particular thing. Okay,

thing. Okay, simple. So, we have these skills as

simple. So, we have these skills as well. Perfect. Now,

well. Perfect. Now,

our claude code is obviously very powerful because it has our custom skills and skills can contain the Python scripts. Any skill basically any script,

scripts. Any skill basically any script, right? So, that means we can write any

right? So, that means we can write any custom logic in our skill because in the skill we can run our custom scripts.

Simple because skill is what a kind of tool that we have created which is a kind of orchestration tool right very good. Now our claw code is obviously

good. Now our claw code is obviously very very powerful and our cloud code can just make tool calls which are built in and some custom tools that we have created and we just call them as skills.

Wow.

Now we can even enhance the power of claude code by this thing. Let's say let me

connect code with uh you can say external tools. Let's say

this is tool one. Okay,

this is one tool. This is second tool.

This is third tool. Okay, I want to connect my cloud code with these tools.

Okay, what are these tools? This tool

can be your Gmail tool. That means you can connect your cloud code with Gmail.

So that claude code can access your Gmail and clot code can read all the emails. It can reply to those emails and

emails. It can reply to those emails and all of the things that you do with your Gmail. So obviously for that you need a

Gmail. So obviously for that you need a tool. So does cloud code provide any

tool. So does cloud code provide any tool for Gmail? No. Can you create a skill for that particular tool? Yes. But

you would need to write the logic all of the code for that. Yeah, that will be a pain because obviously you would not need to just create a script. You would

need to create your own function. kind

of yes. Okay. Very good. Now what if I say that I have tool created here. Who

created this tool? Any community. Any

community. This tool is there. Okay. So

Anala you just mentioned that we can use the skills which are created by third party community. So there will be a

party community. So there will be a skill for Gmail as well. That's a very good question. Yes, there will be a

good question. Yes, there will be a skill for it. But as I said, MCP is not just used for clot code. We can use MCP anywhere.

Oh, okay. So, we are becoming more future proof. Whenever we would need to

future proof. Whenever we would need to migrate from clot code, we can just use MCB there as well. So, your question is very good that we can also use skill which are created by the third party.

Simple, good. But here we do not want to use skill. Let's say we just want to use

use skill. Let's say we just want to use tool direct script. Simple. Because

skill is not just a tool. It is just like orchestration.

Simple. So now I have these tools. Okay.

So one is Gmail tool. Okay. Second one

can be let's say your browser tool which can access your browser. Let's say you want to do some shopping on Amazon and you just want to delegate that task to

clot code and you can say hey just do the shopping for me. Here's a list I want to buy. Just do it. Okay. At least

put them in the cart. do not make the payment but at least put them in the cart. I I will just review it and I can

cart. I I will just review it and I can just make the payment. Okay, this is that tool. Third tool is let's say for

that tool. Third tool is let's say for your um day-to-day management. Let's say

this tool manages your file system. By

the way, file system can be done by cloud code directly. So let's say this tool is for uh Google search. Okay,

let's say you want to just apply some Google search, you want to search something on the web, it can do that.

Okay, these are the tools available. Now

claude code can connect to these tools that means claude code will not need to write any logic because it is not a tool that means

built-in tool it is not a skill it is just a code that claude code can run oh okay treat these tools as rest APIs

okay treat them similar to rest APIs so whenever you hit an API you give the response you get the response back this is the standard standard way to standard

way of talking to any application.

Simple. Now MCPs are exactly like that.

You give them the response. Okay? Or

make you can say you make a request and you get the response back. Simple. This

is a scenario when your MCP or anyone's MCP server is hosted somewhere. It is

called HTTP MCP server. Okay. So we have two types of MCP servers. One is hosted, one is local. But in today's era, most

of the MCB servers are local MCB servers because um they are slowly you can say migrating to uh you can say migrating their architecture to the hosted one.

But for now most of the MCP servers are local MCP servers. What does that mean?

That means they will simply share the code. They will say bro this is my

code. They will say bro this is my Python code. This is my function and you

Python code. This is my function and you can use it. So clawed code will just take the code and it will run it locally

on your machine. Okay, this is the best way to run the MCB models, MCB servers.

Okay, makes sense. So now

yes, the future of MCP servers is on the hosted side on the HTTP side because streamable HTTP um mode of MCP server is newly added

just a few months back. But now I can say not all of the service providers have migrated their architecture to hosted one. So most of the MC servers

hosted one. So most of the MC servers are available through local execution.

Okay. Now how claw code can use those particular MCP servers? Because we know that these tools exist. Okay. And I can say this is MCP server.

Now how claw code can use this? What is

the URL? Is there any repository? What?

Okay. So yes, there is something called as configuration. So clawed code can

configuration. So clawed code can connect to these external tools using MCP config. Okay, kind of config. This

MCP config. Okay, kind of config. This

is a kind of JSON RPC2. So this is a kind of config.

So let's say this is the one.

Uh let's use this color. Okay, perfect.

So this is config JSON.

Okay, simple. So this is the JSON that it can use and it can connect to this particular MCB server. Wow. Yes, this is

the architecture.

Now as I just said that MCP server can be ano can be a hosted one and can be a local one as well. Most of the MCP

servers are usually um available through local execution using uvx or npx. npx is

for NodeJS and JavaScript and UVX is for like Python functions. But you do not need to worry about anything because you just use the code and you just execute the code. Simple.

the code. Simple.

So paradigm of MCP server means anyone can use these tools using this JSON config. Anyone not just clot code. So

config. Anyone not just clot code. So

now what happen? What will happen in the case of your MCP server? So let's say this is a client. This can be any

client. It can be clot code.

client. It can be clot code.

Okay. It can be clot code and it can be cursor.

Okay. Yes, cursor. I hope you would have heard about cursor. Cursor is an agentic ID available. Cursor. It can be claw

ID available. Cursor. It can be claw desktop as well because that is also a client. That is why it is very popular.

client. That is why it is very popular.

So this is also available. And by the way, Anthropic is very smart. They just

launched MCP just for claw desktop and then they just started promoting it here. You can just use it in the gentic

here. You can just use it in the gentic world blah blah blah. So yeah

claude and they have VS code as well.

Now VS also supports MCB. Okay. After

cursor because cursor is the kind of competitor. So these can be the clients.

competitor. So these can be the clients.

Okay. These are what? Clients.

Okay simple.

Now these clients can connect to any MCP server. Let's say

this is our MCP server.

Okay, this is our MCP server and let's put it here. Let's put it here. Let's

say these are the tools.

Simple.

Okay, makes sense. So this is our server.

Now this is my server. I can connect to this using that config that we just talked about. Simple that we just talked

talked about. Simple that we just talked about. Simple. So this JSON is very

about. Simple. So this JSON is very important.

This is the main thing because in this JSON if your MCP server is hosted on cloud, okay, hosted somewhere, then obviously we need to use a different

JSON and if it is a kind of you can say normal local execution, then it will be a different one. But more or less the plug-in will exactly the same. Now let

me just tell you a very interesting thing. We often considered MCP server as

thing. We often considered MCP server as a kind of USBC port. USBC port. Why? So

let's say you want to connect to these tools to your cloud code.

This JSON, this configuration is actually USBC port. The same way you connect your laptop with let's say webcam if you do not have webcam in your

s in your laptop or let's say if you want to connect to the HD one. So what

you will do you will simply insert the USBC port and you will connect it. Let's

say you want to um use a mic. Then also

you will see you you will use USBC port and you will connect the mic. So this

tool is your webcam. This tool is your um you can say your mic. This tool can be your let's say hard disk. So tools

are different but you connect all of these tools through the USBC port. So

this is your USBC port.

This one okay this one. This is the USBC port. Make sense? Make sense? So every

port. Make sense? Make sense? So every

time you just need to insert the JSON config and that's it. You can

insert the same JSON in cloud code, same JSON in VS code, same JSON in cursor, same JSON in cloud and it will work

simple. Second thing, let's say I have

simple. Second thing, let's say I have another MCB server that I want to connect because I can connect to multiple MCB servers and don't worry, I will just show it to you. I want to connect to different MCB server. So what

I can do? I can simply let's say let's remove let's say in this MC server I just have one tool let's say. Okay. So I

can connect to this MCP server as well using the JSON config and same config literally same

using USBC simple.

So this way I can connect to as many MCB servers as I want using this JSON config. Now this JSON config is very

config. Now this JSON config is very important. Simple very good. We will

important. Simple very good. We will

learn it step by step. I know this is a very detailed topic in itself but I will try my best to just to give you a very quick overview but yes if you want a dedicated course on MCB server do let me know we will just create one okay so now

let's see how we can enhance the capability of our clot code okay so this is our clot code let me just open anything

um any file let's say this one just to pop up this clot code so let's say this is our clot code simple now I know that my clot code has tools like let's say

built-in tools and my cloud code also has something called as skits. Now I

want to enhance the capabilities of my claude code. So do you know what I will

claude code. So do you know what I will do? I will add MCP server. Now claude

do? I will add MCP server. Now claude

code already provides us a long list of MCP servers in the form of plugins. In

the form of plugins. If I just show you plug-in.

If I go to manage plugins, you will see if I just scroll down, you will see all of these things. These are these things are what these things are actually

plugins. Plugins means your MCP server.

plugins. Plugins means your MCP server.

So I can connect to anything. Let's say

I want to connect to notion. I will

install it. Then I might need to authenticate myself using oath. Right?

Once it is done, boom, boom, I can directly work with notion because clawed code will have access of notion. Simple my notion simple.

notion. Simple my notion simple.

Okay. Then there are like so many tools.

Versal is there if you want to deploy something. Slack is there if you want to

something. Slack is there if you want to just connect your messages with the clot code. Done. You can do that. There's a

code. Done. You can do that. There's a

long list literally long list like you can literally Firebase is also there see so if you just want to search for a specific uh one you can simply search it

now I personally like using playright MCB so let me just search playright so this is there this is a kind of MCB server which adds the browsing

capabilities to our agent this is built by Microsoft so if the playright MCP is there it can browse my browser and it can do whatever I want. Simple. So I can

enable it. Simple restart. Perfect. So

enable it. Simple restart. Perfect. So

now what will happen? Now if I just search let's say MCB servers I have this one available. I

have one for pencil. I let me just first of all remove it because I don't know why I just used it. So I will say remove

pencil MCP from my root location.

Simple because I just want to remove it.

Simple. So it will say yeah just run it.

So I have removed this pencil MCP. Okay.

Now if I say MCP um I have not refreshed my session. Let

me just create a new session.

MCP servers it is still there I don't think so um if I just see the terminal I I'm so sure like it will be gone so I can just double check don't need to worry so currently I have one MCP if you see play

right let's suppose we just have one MCP server now this MCP server is connected simple it's connected now

there are two things this can install MCP server in your local uh you can say location as well.

Local means like your root location.

So you will be having choice whenever you want to install any MCP server because I installed this particular plug-in before that's why it didn't ask me at all. Plus this is a like kind of

plug-in but let's say if you want to add any specific plug-in any like any plug-in simple. So this way you can

plug-in simple. So this way you can literally go and search any plug-in which is directly related. But do you know what? You can obviously work with

know what? You can obviously work with the integrated ones. Let me just show you how you can just use the external ones because these are already there for you. This is very easy. But the beauty

you. This is very easy. But the beauty is we can connect to the external MCP.

So let me just show you. So if you just want to let's say find any MCP for any work. So you can literally type let's

work. So you can literally type let's say there's a very cool MCP server or you can say the service that we use duck duck search for searching the you can

say latest news and all of these things.

So I can say duck go search MCP. Um Ana, why are you searching it

MCP. Um Ana, why are you searching it here? Why can't you why can't you just

here? Why can't you why can't you just search it here? Because there's no MCP server for duct go search. See

d go search. No DDG. No, no, there's no but I want to use it because I want to allow some services, some tools to my agent to

make some you can say search for the latest news. I will simply say duck go

latest news. I will simply say duck go search MCP. Hit enter and then most

search MCP. Hit enter and then most probably you will find the GitHub repository for any MCP server because that's where people just host their MCP servers. Just a quick guideline, you

servers. Just a quick guideline, you should always use a trusted MCP servers.

You should always read the comments. You

should always see the stars because see MCP server can do anything on your system, right? So you should always use

system, right? So you should always use it with your common sense. So I will simply go here dug search and I can just explore. I can see the stars are like

explore. I can see the stars are like good. Okay. And this is the

good. Okay. And this is the uh configuration. Perfect. So if I just

uh configuration. Perfect. So if I just scroll down a little bit, I will see this particular JSON. What is this JSON?

This is the exact JSON that we use to add any MCP server to my agent. Okay,

this is the exact JSON. Now what this JSON is saying? First of all, this command is common MCP servers. Within

that we have the MCP server name which is called DDG search. Perfect. Then the

command how can we run this server? So

this command uses UVX. UVX basically is a package built on top of UV. With the

help of this package, we can immediately download this package which is a Python package available in Pi and it will run that package locally on my system

whenever I will call this MCB server.

Simple. Simple. Yeah, it is that simple.

That is why it is called as USBC plug-in. I do not need to worry about

plug-in. I do not need to worry about this package management. I do not need to worry about any kind of you can say installing the package nothing. I just

need to copy this JSON and I just need to go to my VS code and here I can just put that MCB. Now

how can I just add that MCB server to my CL code? There are two ways. First way

CL code? There are two ways. First way

is I can directly create one file called MCB.json JSON and put the content there

MCB.json JSON and put the content there or I can ask claude code to do the work for me. What is the difference? If I ask

for me. What is the difference? If I ask claude code, let's say I want to add this MCP in my project.

I will provide this thing. I'll simply

say hit enter. Now what this will do?

This will ask me like I want to just use this allow reading from this. So what

this can do or you can say but this will do this will create a settings dojson file and it will put that particular MCP

server there. Okay I can say yes

server there. Okay I can say yes then let's wait and this has see this has added this file okay domcb.json JSON

but where where in my project see if I say um allow yes so if I see my structure I will seem

mcb.json JSON here. See? Wow.

mcb.json JSON here. See? Wow.

Wow. It has automatically done that.

Now this MCB server is only available in this project which is a good thing.

Sometimes what people um like people can make mistake in this particular area.

They can make the MCP servers available throughout their system which is not a mistake. It totally depends upon the

mistake. It totally depends upon the requirements. Sometimes you need to make

requirements. Sometimes you need to make available all the MCP servers throughout the project so that any project can use it. But sometimes some MCP servers are

it. But sometimes some MCP servers are strongly aligned towards one project. It

should not be shared among different MCP servers or basically different projects using MCP servers. In that particular scenario, you should always ask cloud code that I want to add MCP server in

this particular project only.

Simple. What what what will happen after that? That MCP server will only be added

that? That MCP server will only be added in this particular project. So as you can see that it is saying done I have created a mcbp.json file in your project

root with the duckgo search MCB server obviously and it is saying when you use cloud code text it will detect this server. Why? Because whenever I will

server. Why? Because whenever I will make any call it will literally go here and it will see if there is there any MCB server to perform this task. If yes

then I will use it who cla code simple and now if I open this this is the file and this

file has this thing wow yes now let's say you want to run this MCP I can just show you don't worry we will just add one to two more MCP

servers so that you can just get the exact structure of MCP server so this is very simple how to we how we can just

add the MCP Simple. Okay. Now here

clawed code is acting as a client.

MCP server is this one which will run locally on your system. So whatever

communication you are having with this particular MCP server is through standard input output. What what did I say? By the way this is a technical

say? By the way this is a technical stuff regarding MCB but consume this knowledge. Stdiodio standard input

knowledge. Stdiodio standard input output. That means this this MCP server

output. That means this this MCP server will create subprocess to talk to the server. This is the most secure way

server. This is the most secure way because you are not sending any request outside. You are not receiving the

outside. You are not receiving the response from outside. This simple like communication between your system and you or your client. Simple. Okay. So

whenever you see command like uvx npx that means you are running MCP servers using stdiodio standard input output.

Simple very good now uvx means that it will consume python and you already have python installed but some mcp servers use npx that means that relies on nodejs

to execute the mcp server. So if you ever use any MCB server command JSON configuration where npx is written like

this like this then in that scenario make sure that you have NodeJS installed on your system otherwise that program will not run

simple so simply go on Google type download NodeJS and that's it if you are using npx if UVX then it's fine because UVX will consume your UV or you can say

your Python environment. Simple.

Very good. This is a very basic example.

Very basic example. Okay. Now, let's say you want to add another MCP server.

Obviously, you can add it, right? So, if

you want to add another MCP server, let me just show you a very good example.

Okay, very good example.

Um, you can also go to, let's say, Gmail MCP server.

And I think there's a very popular guy who has created this MCP server. Um,

let's open this one. Let's see.

Um, I don't know about this guy, but I know there's a guy like who has created a very popular Oh yeah, this one. Gong resi. Yeah, this

is a popular one. So now if I want to use this MCP server Gmail o MCP server what I can do I can just simply search for the command that I need to install

in my um you can say MCP server. So see

this is the command. Now this is using npx. Now npx is what npx is a kind of

npx. Now npx is what npx is a kind of um python or you can say nodejs equivalent of uvx.

Okay. So you need to download NodeJS in your system if you want to use Gmail MCP server. But if you are able to find any

server. But if you are able to find any MCP server for Gmail using UVX then you are good. If not download NodeJS it's

are good. If not download NodeJS it's just a matter of like few minutes and that's it. Simple. So you can just add

that's it. Simple. So you can just add this as well. So all of these MCP servers are local MCP servers. Now let

me just show you a very good thing.

I need water bro. But let's complete the lecture first of all. So now let me show you a very cool example. I want to show you one thing. Um yeah play right. So

now you know let's say these all are the MCP servers that we see here. Right.

Let's say plugin and if you go in plug-in you will see a long list right? Uh loading plugins.

Why it is taking so long bro? Let me open terminal

bro? Let me open terminal uh claw.

Perfect. And use this. Yeah. So now if you have installed any MCP server and if you open the terminal it will ask you that do you want to use this MCP server.

I will say use this MCP server. So yeah.

So now whatever question I I I will ask it will always refer to MCP.json if it will feel like it can use that MCP server it will use it. Okay. So we are not using it right now but I want to

show you something. So simply say plugin. Okay. So I will see the plugins.

plugin. Okay. So I will see the plugins.

Okay. So here are all of the plugins.

All of the plugins. GitHub. Oh, so we have GitHub plug-in as well. Yes. GitHub

plug-in means it can actually manage everything using your GitHub account.

all of the PRs that means pull requests all of the issues all of the bugs everything okay so you can also install this MCB server in your cloud code connect your GitHub account and just

rely on cloud code for everything and it is amazing thing okay and if you just want to search anything you can just search it so the intent behind opening this terminal is there are so many so

many so many you can say MCB servers but let's say we do not want to use these integrated MCB servers So for example I am currently using or let's say I have

recently downloaded the MCP server for playright right. So now let's say this

playright right. So now let's say this feature would not be available. Okay

this feature would not be available. Okay so now if I go to Google or I search playride mgb server then

I can just go to their GitHub repo. So

here I can just scroll down and see this is their command. Wow. So even if you are not able to find MCP server for playright in clot code, you actually do

not need to be dependent on the connectors or plugins that we have in clot code. Why? You can easily find MCP

clot code. Why? You can easily find MCP servers everywhere because MCP servers are not just for your claw code. It is

for everything. So you can literally copy this. Okay, you can literally copy

copy this. Okay, you can literally copy this. I can just even show you but I

this. I can just even show you but I obviously it won't work because I do not have the node I have NodeJS. Yeah, I

have NodeJS. I can confirm I can yeah I have NodeJS. So I can even add it because I want to show you like how you can add multiple servers. So what you need to do simply add a comma after DDG

search and Ctrl +V and here you need to remove MCP servers because MCP server is just for the heading. That's it. And

here remove this as well. Uh not really.

Okay.

Okay. This this

MCB servers. Okay. This is open. This is

closed. This is closed here. Yes. This

is closed here. And this is closed here.

Okay. And this is closed here. Is it

good? MCP servers. Okay.

Okay, this is fine. Perfect. And oh

yeah, this is extra.

Now this is good. Perfect. So this way you can even add I think now it is good. Yeah.

Yeah. Now it is good. So this way you can add all of these MCP servers like this. So now I have MCP server for DDG

this. So now I have MCP server for DDG search. This is playright. So just to

search. This is playright. So just to distinguish I will simply say playright own that means this is my personal p play right MCP server. So now if I just search MCP

it will give me plugin DDG search and I think this has not taken care of this thing. Let me just open a new session

thing. Let me just open a new session and the best way is use a terminal plot and then

MCB. So I have MCP servers DDD search

MCB. So I have MCP servers DDD search and I have pencil maybe it is not removed. So the second thing I have

removed. So the second thing I have playright. So now this one is not

playright. So now this one is not actually picked up. I don't know why. Um

okay command C wrapper to execute npx.

Okay makes sense. So now they're saying that they need command C. Okay like

basically back/ C wrapper to execute npx because by default it cannot run your npxbased MCP server. But that is not a big problem. So you can literally

big problem. So you can literally replace this entire command with this one. So what this will do, this will

one. So what this will do, this will literally run your command with C c / C.

So whenever you want to run any MCP server with npx with cloud code, you have to use this thing. Make sense? So

now let's try to confirm it. Okay. So I

will simply say exit claude and use this MCP server. Yes. Then MCP.

And now you can see DDG search playrite own. That means that is my personal MCP.

own. That means that is my personal MCP.

Okay. And then I have pencil which is still there. It's fine. Then I have

still there. It's fine. Then I have built-in MCB which is playright playright. I can even disconnect it

playright. I can even disconnect it because I now have my own MCB server running here, right? But just for the simplicity, let's keep it. It's not a big deal. But you can just disconnect it

big deal. But you can just disconnect it using the plug-in here. Okay, that's not a big deal. Makes sense. Okay, makes

sense. Now, what you can do, you can use these particular MCP servers. Yes, you can use these MCP

servers. Yes, you can use these MCP servers. Now, I will go to clot code. I

servers. Now, I will go to clot code. I

will say I want to get the latest news on let's say

claw code using um MCP.

Okay, let's see what it will do. It

should just run the So it is saying do you want to proceed with MCD search?

Yes. So it is able to identify my MCP called DDG search. So it will just use DDG search. It will download the

DDG search. It will download the package. It will run the code. It will

package. It will run the code. It will

get the response. It will give it to me.

Simple. So it has would you like me to fetch more detail.mm.json? What is it saying? Okay. Key updates. Okay. So it

saying? Okay. Key updates. Okay. So it

has just given me the update for cloud code review launch March 9. New

multi-agent code review. Automatically

analyze AI generated code and flags logic errors. boost engineering product

logic errors. boost engineering product by 200% and blah blah blah. Opus 4.6

release. Yes. So these are all of the latest news that I have got using DDG search because claude code doesn't have any kind of built-in tool for making the

realtime search related to news. Wow.

Now I will say um I want to I want you to open

Amazon and tell me the price of or let's say um just let's say I want you to open Amazon for me. That's

it. Okay. So now what it will do? It

should open one of the windows. Let's

wait. Uh

I can help with that but I can cl first.

What are you trying to accomplish related to cloud project? Amazon is

outside the scope of the cloud code is web. So I will simply say open Amazon

web. So I will simply say open Amazon for me using MCP if you have any for this simple

yes yes. So now it will just open. Oh

yes yes. So now it will just open. Oh

yeah it has opened it this thing. See

where is the screen? Okay this is the screen. See this window is automatically

screen. See this window is automatically opened by plot code. Very good. And you

know what? When when I was just testing MCP server for playright um few days back, I asked it to just um put some

stuff that I wanted to buy in the cart.

Worked. It worked. It literally searched everything in the browser in the search bar of Amazon.com. Um search for that stuff. Same color, same everything in

stuff. Same color, same everything in the cart. Wow. Wow. So you can also do

the cart. Wow. Wow. So you can also do these things using these MCP servers. So

these are called stddio based MCP servers that means local MCP servers.

But now as I just mentioned that you can also access the remote servers. Remote

servers are the future for sure. For

sure. And now I can also show you what are the remote servers. And a very good example of remote server is tavly search.

Tele search is a kind of tool which allows you to make search similar to Google search. Simple. So what you can

Google search. Simple. So what you can do you can literally go to te search mcp server and one thing that you would need in te

mcp server you would need an api key and tely allows you to create one free API key with thousand free requests. So you

can directly go to tili tei or tei te search and create your account. um

create your API key for free and then you can just use this MCP server for free. Simple. Okay, good. So I will just

free. Simple. Okay, good. So I will just click over here on the te.com and here you can see that this is a tely MCP server. But do you know what when you

server. But do you know what when you will be looking at the command to add this MCB server you will be like what what is this command? Let me show you.

So remote MCB server and this is the command. You will say this is a URL bro

command. You will say this is a URL bro this is not a MCP server this is a MCP server but this is an hosted one

this is a hosted MCP server and in the hosted MCP server we always use HTTPS hypertext transfer protocol secured okay so that is why it is the future

because when we have the hosted remote or remote or hosted servers we can just treat it like a normal API call that is fine Fine. But Anel Lamba, how can we just add this in our MCP list?

That's a very good question. So, you can just copy this URL. And by the way, you need to complete this URL by your API key. I have my API key, but you can just

key. I have my API key, but you can just put your API key in place of this. Okay.

So, you will go to um cloud code and you will say I want you

to add this MCP in my MCP do.json. So, do you know what what

MCP do.json. So, do you know what what it will do? It will write one more mcp server and it will simply say server

name type is http http and then url that's it. It will not write command. It

that's it. It will not write command. It

will not write arguments. Why? Because

this is not running locally. This is

simply running on a server outside of your machine. Simple. Let's see.

your machine. Simple. Let's see.

Let's see.

Uh yes, see it is here. It is here. It

has added this thing. I think it forgot to add type. Type is very important by the way. But still I can simply confirm

the way. But still I can simply confirm it by going to the terminal. This is my assumption because maybe they it because this is a client they can just change their inside code that they have

written. So I simply say claude

written. So I simply say claude and I will write MCP.

So I have got okay so I think there's one error does not add to MCPS server schema. I was so sure I was so sure

schema. I was so sure I was so sure about it. So we need to say type and

about it. So we need to say type and type is HTTP. Okay. Now if I just refresh it.

Now if I say claude and uses MC yes MCP perfectly is there see DDG search play

right on and te perfect. So this way you can just add your tely search API as well basically MCP server but do not forget to replace that tely API key with your API key. Okay and it is free you

can just create your account. Okay just

create your account and everything is sorted. Perfect.

sorted. Perfect.

This is your MCB server. So in short, you can enhance the capabilities of your claw code by pulling the external tools.

Now what is the advantage?

Now how DDG search actually performed the search for me? DDG search has some tools available within that particular

MCP server extract and there's one more tool that it has. So I didn't write any code by myself. I simply used the code that was written in the MCP server by

the owner of this MCP server. I simply

ran the code. Simple. It saved a lot of time. Plus it is approved by the

time. Plus it is approved by the community because obviously whenever you are just building something for the community a lot of iterations would happen. A lot of improvements would

happen. A lot of improvements would happen. A lot of edge cases would have

happen. A lot of edge cases would have been taken care of. So obviously and now when I'm using the MCP server for a reputed brand Microsoft they have written amazing functions to user

browser simply use this now Microsoft also doesn't provide the hosted MCB server because hosted MCB servers are the future but now they're simply using their old MCB server that runs locally

and it's fine it's fine and then teav is a you can say AI based application so they just hosted their MCB server so in the TV as well they have four to functions four to five tools and I'm

directly using the code and here and I'm not even running the function locally why because I'm simply sending my request to this hosted version when I

will be having API key simple that is the power of your MCP servers where you do not need to worry about the tools you do not worry about the code

you simply create a client wow now let me just tell you one very interesting thing clawed code handles MCB servers

very maturely because it is code first approach. Developers use it. But when

approach. Developers use it. But when

you use claw desktop application for MCB servers, it behaves differently.

Wow. In claw desktop, you can run your local MCP servers without any kind of effort. Why? Because

effort. Why? Because

um it was made like that. Okay. Then

when companies started pushing their MCP servers on a hosted server then claude directly connected those MCP servers

into their claw desktop application and they just call it as connectors connectors is just a fancy word behind the scenes it is just an MCP simple but

now if you want to add your custom MCP let's say te is not here or it is here I don't know let me just Manage plugins. If I search

Manage plugins. If I search no te is not there. So if you want to use tavly you cannot use it like this in claw desktop application. Why? Because

it is not allowed like this. Now this is a technical stuff that I'm just going to say. Now claw desktop what it will do.

say. Now claw desktop what it will do.

It cannot make a connection directly to the hosted MCB server. Okay. And I'm

just talking about claw desktop. Cloud

code can work. It is fine. Okay, I'm

just talking about the claw desktop.

Claw desktop will use a proxy. Okay,

local proxy. So for that proxy, we have a special function called npx which is the command and the um your package name

is MCP remote.

Okay, so we use that package which deres all of the traffic from local URL to that hosted one. So this is a kind of workaround that we use till the time

claw desktop has that particular you can say compatibility and it is still not there. We still use that particular

there. We still use that particular package but we do not we developers do not need to worry about this thing.

Developers will not use claw code in claw desktop application. developers use claude code

application. developers use claude code directly in the code editor. But why I'm telling you this thing because obviously you can just explore cloud desktop application. You might get the

application. You might get the subscription from the cloud desktop application because see it is worth it because if you just compare the cost um cloud subscription will actually be a better thing if you are just using cloud

code on a daily basis and using it for hours 8 hours 9 hours 10 hours cloud subscription is a better place. And when

you are using cloud desktop subscription, you can actually use a cloud desktop application as well. But

my duty is just to tell you the raw stuff that developers use because cloud desktop application is also built for the non-developers, non-technical people where they just want to connect to

Google, they want to connect to Gmail, they want to connect to whatever without writing any code, without seeing what is happening behind the scenes.

Simple, very good. So this is a thing and similar to this like this is like pure claw code thing if you want to add any MCP server in the VS code okay in the VS code then also you can just do it

it's not a big deal but we are not using those clients those are just like the ID clients that we use but this is like pure cloud code it has nothing to do with VS code cursor whatever because I

do not know like you are using which ID so I cannot just tell you MCP through VS code or MC um that cursor Simple.

So this is your MCP servers with clot code. And I know now you will be

code. And I know now you will be thinking like hey we can just connect to this MCP. We can just do this work. We

this MCP. We can just do this work. We

can just automate this thing. And

exactly that is a stuff being happen today. People are actually automating so

today. People are actually automating so much of things using MCP servers because they do not need to write any code to automate the stuff. They simply need to

use the tools available within that MCP server.

Just for example, let's say you can even automate your emails, you can even automate your Slack messages, you can even automate your group chat, whatever you have. Let's say you are going on a

you have. Let's say you are going on a vacation, okay? And you are going out

vacation, okay? And you are going out for at least a week and you already know if you're working in 9 to5 corporate, by the end of the next week, you will have like thousands of messages and you

cannot like read each and every message, but you cannot skip it. Maybe some you can say important message will be written in that particular chat. So in

that particular scenario, you can Just use clot code. You can connect your slack, your teams, whatever you want to connect. Boom. Simply ask clot code,

connect. Boom. Simply ask clot code, what is a summary? What is happening? Is

there anything important for me? Is

there any to-do for me? Simple. Okay.

That's how we work. Okay. And you can see one more thing here. Dot playright

mcp. This is like for console that we just used. Okay. So this is your MCP

just used. Okay. So this is your MCP under the hood. And if you now want to explore MCP in detail, feel free to go to the official documentation which is

model context protocol. So this is the thing. Okay. So you can just read about

thing. Okay. So you can just read about it in detail. HTDIO, SSE, streamable HTTP. These are three basic kinds of MCP

HTTP. These are three basic kinds of MCP servers that we have and you should know the reasons behind behind MCP servers because I can just simply say plug and play. I can simply plug the URL and I

play. I can simply plug the URL and I can get the access to all of the functions and tools available. I do not need to change the configuration. I do

not need to build a different dedicated client. Nothing. Just MCP JSON and

client. Nothing. Just MCP JSON and that's it. You can even create a client

that's it. You can even create a client using Python. But that all all of these

using Python. But that all all of these things we will just we can if you want we can cover all of these things in the dedicated MCB server masterass if you want to study MCB server in detail like

from scratch. Okay. But yeah, this is

from scratch. Okay. But yeah, this is everything about MCB servers and cloud code. Now let's talk about this amazing

code. Now let's talk about this amazing feature in claude code hooks. This was

added I think few months back in clot code. Why it was added and what is the

code. Why it was added and what is the use of hooks? Let's talk about hooks because this is very important. In

short, let's say we are using clot code.

We all know that. Okay. So let's say this Rohan is using clot code.

Perfect.

So now this clot code does so much of things.

This claude code will let's say perform so many tool calls skills whatever while performing so many things claude

code has inner checkpoints. What do I mean by inner checkpoints? For example,

claude code manages all of these things in the form of events.

If cloud code wants to make any tool call, that is a kind of checkpoint. That

is a kind of area where clot code knows hey this is a different event. Once the

tool call is completed then next tool call if it wants to call multiple tools.

Simple after that after making all of the these tool calls there are some more events as well. Okay. What is the benefit of these events? Earlier there

was no benefit but then claude code team made the best use of these events. So

what they did they introduced hooks.

Okay. So now let's say this developer this Rohan knows that I have these events. I will just show you all of the

events. I will just show you all of the events. Okay. So Rohan knows that I have

events. Okay. So Rohan knows that I have all of these events. So now Rohan is a developer. Rohan knows how to perform

developer. Rohan knows how to perform how to write scripts how to perform the automation.

Now Rohan knows that every time that every time one event is invoked or event is completed I can run something in between

for example let's say this claw code made a tool call okay so this tool call is invoked this tool

call is made okay makes sense now this is an event okay this is an event.

Okay. Now this event so there will be a point before this event. Make sense? Before this event and

event. Make sense? Before this event and there will be a point after this event.

Common sense.

So now Rohan can make use of these two points. So now Rohan if want if if Rohan

points. So now Rohan if want if if Rohan wants to run something at this point.

Let's say at this point Rohan can do it and what he can do anything.

He can run any Python script. He can run any bash command. He can run any PowerShell script. Anything. So this is

PowerShell script. Anything. So this is called pre pre-invocation and this is post right. So now Rohan is smart. Rohan

right. So now Rohan is smart. Rohan

knows that sonnet like this claude makes a lot of tool calls and perform so much of things. So let's say before making

of things. So let's say before making any tool call I want to perform something I want to do something. So

Rohan can generate any script for example we are taking Python as an example but you can run any script bash commands um terminal commands then

powershell scripts anything. Now Rohan

can run this particular script on every tool call or basically before any tool call because this is a called of pre-event.

Okay. Now you can think about this thing like this. Let's say you are making any

like this. Let's say you are making any tool call. Okay. And you want to send

tool call. Okay. And you want to send some alerts to yourself. You want to send some messages to yourself. You want

to notify yourself on every tool call.

Because when you automate so much of things, you need to be aware of like what's going on behind the scenes, right? So you can perform so much of

right? So you can perform so much of logging. So you can say every time you

logging. So you can say every time you are doing something, you need to make this post event command or let's say pre-event if you want to do that. Now

what's so special in this particular thing? The special thing is no LLM is

thing? The special thing is no LLM is involved here in this particular step because this step will not invoke any LLM. This script will not be run by LLM.

LLM. This script will not be run by LLM.

No, this will be a pure local invocation. Pure local execution.

invocation. Pure local execution.

Claude code will just execute this script and that's it. Or execute any command that I will provide and that's it. Cloud code will not simply send that

it. Cloud code will not simply send that command or send that script to LLM or to sonnet to run that thing or use it as a

context. No, this will be pure local

context. No, this will be pure local step that it will perform. Simple.

Oh, okay. So, you can think of it like a kind of step or a kind of local node that we have added in our execution.

Simple make sense. So, now let's try to read the documentation for this. So now

what is written here? If you just go uh here I think yeah extend cloud code and then we have hooks. Yeah, perfect.

And I think first we should just go to reference. Yeah, perfect. So first of

reference. Yeah, perfect. So first of all, this is the thing that I was just talking about. These are all of the

talking about. These are all of the events.

First of all, let me just zoom in. First

of all, the event is session start. That

means whenever we start the session.

Second step is user prompt summit.

Usually we do not perform anything here.

Then we enter the agentic loop where we make tool calls. So this is one of the events pre-tool use. Then permission

request that means if we have set and I have also set it that whenever it wants to do anything you can just ask me but I have not set it on the tool usage I have simply set it on the you can say editing

my file system and that's it but some people just try to make permissions or you can say some people try to just um restrict permissions on some tool usage when you work in an organiz organization

because no one like not not everyone should have the access to all of the tool usage right so you can just do that as well. And then tool executes simple.

as well. And then tool executes simple.

Then post tool use. Then sub aent if you have any sub aent and if cla code is spawning any sub aent it will create an event for that as well. Simple. And

finally task completed and then stop.

Stop means like everything is done. So

these are some of the events and you can just see the long list here. These are

all of the events. I can make use of these events. Wow. Wow. Literally. Wow.

these events. Wow. Wow. Literally. Wow.

So now what is a hook? The same thing that I mentioned. Hook is basically any action you want to take on these events.

Simple any action.

And what is the syntax? If you want to define a hook. So this is the very basic syntax. So let me just define this

syntax. So let me just define this particular thing or let's say explain because I can define in VS code. So

first of all you will say hooks.

Whatever hooks you want to create, you can create multiple hooks. Then this is a category of book which is called pre-tool use. Here you just need to

pre-tool use. Here you just need to write the event name like on which event you want to perform your action. I want

to perform my action on pre-tool use.

That means whenever it will be using my tool I want to run something. I want to do something. Then comes the matcher.

do something. Then comes the matcher.

Matcher means what type of event?

Because tool can be of any type. It can

be of bash command. It can be of create, edit, read, write, so many things, right? So you simply need to write some

right? So you simply need to write some tool type. Simple. So now this hook is saying

type. Simple. So now this hook is saying that whenever cloud code is running any bash command, okay, any kind of bash tool, so you need to run my hook, you

need to run my thing. And what is the thing? What is the action? This is the

thing? What is the action? This is the action. So here the person has created

action. So here the person has created the bash script. Okay, it's called sh which

bash script. Okay, it's called sh which is a kind of you can say shell scripting. So the person has created the

scripting. So the person has created the script and the person has saved the script inside hooks folder. We can save it anywhere but this is a standard. And

then what will happen whenever the claw code will make any tool call using bash before making that tool call has to run this command.

Simple, simple.

This is pure automation. Pure

automation, simple. So let's see some other examples

simple. So let's see some other examples as well. This is a very good example. We

as well. This is a very good example. We

will also try to create our own example and we will use Python for that because Python can do literally everything including bash scripting using subprocess. So it's not a big deal. We

subprocess. So it's not a big deal. We

can we will use Python. So this is another very good example. So this is a this is an example because whenever we run anything with bash command right now you will say hey what is um inside this

script um so the let let me just show you what is inside this script. So

basically this person has applied a kind of security or you can say a kind of guardrail on his repo or working directory. Whenever we run the bash

directory. Whenever we run the bash command we can literally perform anything using bash. So person has restricted this command rm rf. So if the person finds or basically cloud code

finds okay anything related to bash command it will directly run this script simple and this script is saying hey if any command contains these type of

characters to remove anything deny I do not want to just perform that particular thing simple so this is that script you can just add some guardrails in your

working tree simple and this is the architecture for that particular execution so first of or Claude will run this command. Why? Because you would

this command. Why? Because you would have asked, "Hey, please delete this thing. Hey, please delete that that

thing. Hey, please delete that that thing." Claude will say, "Hey, um, let

thing." Claude will say, "Hey, um, let me generate the command that I want to run." Okay, it generated this command

run." Okay, it generated this command bash rmrfmp build. Then it was going to the tool call. Which tool call? Bash

tool call. But this pre-tool use fires.

Why? Because there there's a hook for this particular task. So now this is fired. So now what it will do? it will

fired. So now what it will do? it will

just perform its task and it will say if if there is any tool call related to bash which is there in in our case then simply run this particular script and

the decision is blocked that means it cannot perform that tool call but if let's say I'm performing any python operation then obviously this will also um fire it but that command will not

contain any kind of this particular character let's say rm rf then obviously it doesn't need to just um block that particular tool. It will simply continue

particular tool. It will simply continue the execution. So this is a kind of you

the execution. So this is a kind of you can say some people use it for guardrails, some people use it for automation, some people use it for logging, some people use it for summarization and it's totally your

choice what you want to do on these events. Simple. Some people even use it

events. Simple. Some people even use it for let's say notifying them. For

example, let's say you are working with clot code and some bigger um executions where you have long long skills. Um then

obviously cloud code takes some time to execute everything and sometimes cloud code would need your input in the middle of the execution and let's say you are working on 10 different things then

obviously you do not need to keep the clot code up and running in front of your eyes obviously it'll be running behind the scenes but it will be running in the background right so what it will

do it can just simply run one bash command whenever it needs your input and there's a special event for that it's called notification.

Similar to pre-tool use, there's a special um you can say event called notification. So whenever it needs your

notification. So whenever it needs your attention, whenever it needs your input, you will simply run that particular command. It's a bash command or you can

command. It's a bash command or you can say PowerShell command and it will simply um give you the notification.

Hey, I need your input. Hey, I need your attention and simple. So people use it for notification purpose as well. So

that your tasks are not stuck. They are

not blocked. it will simply notify you immediately and it totally depends if you want to automate that thing using your Windows notification or you can simply create one Python script to

generate some sound from your system or you can even connect your slack let's say you're working in an organization and you are wearing headsets um or let's say your microphone is not connected anything so you cannot hear the sound

and you should sometimes simply u mute the notifications especially system notifications then what you can do you can simply connect your slack or teams and it will simply send you the DM. Hey,

I need your attention. Simple. So, you

can also do that thing as well. So,

there are so many things. Okay, it

totally depends up to you what you want.

So, the idea is very simple. Idea is

very simple. These are the events. I can

literally perform anything on these events. Simple. This is the whole idea.

events. Simple. This is the whole idea.

Okay, this is the whole idea. These are

the events. I can literally do anything that I want on these events. I just need to create my hook file and that's it.

Simple. Now let's see how can we create our hook file. So if you just go to VS code and if you go inside cloud.clude

and here yeah let's create here. So if I just go here I can create um one hooks folder first of all because inside the hooks folder I will

create my Python file or Python script basically. So let's create our script.

basically. So let's create our script.

Let's say I want to run my Python script. Okay. Okay. And this Python

script. Okay. Okay. And this Python script what it will do? It will simply um do something. Okay. And just to test it, so what we can do, we can simply create very simple Python file. So what

I will do? I will simply write one text file whenever my tool call is invoked.

Simple like you can literally write anything but we need to test it right.

So what I can do I can create one Python file and I will name it as let's say prescript.

py. And here I can say import OS and let's use path. So

file path uh sh not really prescript py

uh yeah sorry txt. So here I can say with open file path not in the reading mode we

want to write it or let's say append it.

Perfect.

Simple. So this is a very simple script that we have created where we are simply trying to um create a text file whenever our tool will be invoked basically.

Simple and it will be just created in the this folder in the hooks folder.

Simple. I can also decide any other directory but let's say I want to just write it here just to test it. So this

is my hook. Now how to configure the hook? So basically there are two ways. I

hook? So basically there are two ways. I

will just tell you both. the the way is just one but you can just go to that way through two different path. So the thing is we need to create one settings dojson

file not settings.local.json

this is not available across all the work trees settings dojson file. So when

you create settings dojson file, you have to configure this thing like this.

Let's say like this. You just need to paste this JSON. Uh where is that JSON?

Let's say like this. Simple. But do you know what? Um this settings do.json file

know what? Um this settings do.json file

is actually created automatically by cloud code to perform the other task tasks as well. So yes, you can create it manually, but I would recommend you to create this settings do.json file for

the very first hook through cloud code.

Why? Because when you will ask it to just create the uh settings dojson file, it will keep it in the records and it will not purge that file. Otherwise, you

can see that you will create settings.json file and it will just

settings.json file and it will just remove it automatically after some time just to sync the all of the properties because it has not registered your settings.json file. So that is why this

settings.json file. So that is why this is my personal recommendation. Okay?

Otherwise you can just create it manually as well. Okay? Because I faced this problem. I wanted to share it with

this problem. I wanted to share it with you. You might not face it. But if you

you. You might not face it. But if you face this problem, simply delete the settings dojson file and ask cla to create the hook for you and automatically it will write it in the

settings.json file. Simple. Okay. So let

settings.json file. Simple. Okay. So let

me just ask claude. I want you to create one pre tool hook

for me which will run the prescript py file

located at dotcl hooks. Simple

hooks. Simple then it will automatically create the file settings.json file. Uh okay it is

file settings.json file. Uh okay it is reading my file. Okay

so it is just trying to create it where it has created settings dot oh it has simply called as hooks.json.

Um by the way now I'll create hook.json

file bro.

It should be called as settings dojson.

Cloud code doesn't even know its own documentation.

Bro, now it is just updating settings.local.json.

settings.local.json.

you need to create a separate file called settings.json.

settings.json.

That is why when you work with cloud code, you will feel like those bigger models sonnet and um opus models are actually very good in

terms of coding. So you should only use um opus model um not oppus uh this haiku model just for like low-level tasks or you can say which require which do not require much technical knowledge because

obviously they have created multiple models on the basis of the work that we perform. But yeah we are simply creating

perform. But yeah we are simply creating this particular demo. So we do not need to just use that high class model. But

it is a personal recommendation if you're working in the real world. If you

are dealing with some coding problems and you need that particular you can say um knowledge from LLM then obviously you

should just go with the set at least.

Okay. So I think it just tested this prescript as well. Um not actually because this I think hook was automatically called. Let me just delete

automatically called. Let me just delete this file because obviously this is also one of the tool calls. So that is why.

So but that is fine. No need to worry.

So let's see what it has created. So it

has created hooks pre-tool use hooks type command and command is this one.

Python and the location or basically path of my file. Now just a quick tip for anyone who can face some errors.

Sometimes clot code run your um python scripts directly through the terminal.

In that scenario um if you're using something in your python file let's say here I'm simply using import OS nothing else but sometimes I can use multiple libraries and those libraries are only

available in your virtual environment.

Then in that particular scenario what you can do you can simply go to settings.json JSON and here in the

settings.json JSON and here in the command instead of just writing Python you should copy the path of virtual environment like this copy path and just put it here because it will then use

your virtual environment simple and you can obviously add an escape character if you are on Windows but it's fine. Okay.

So now this way as well and you can simply provide the full path of your Python script whenever you are just giving the virtual environment. So this

is a tip when you are using so many things simple. So this is preused. So

things simple. So this is preused. So

now I want to test it. So I know that this particular Python file this my script should run every time whenever I'm making any tool call right any tool

call by the way I can just simply define what type of tool call let's say edit um write um read but I want to just test it on any tool call so I will simply open

one new chat and I will say um create one test dot txt file for me

with hello world in it. So now what I have asked I have simply asked to create one test file for me. Why? Because I

know whenever I ask something to create or do whatever it will do something it will make tool call and whenever it will make tool call this particular script

should run right. So let's test it.

So okay I will say yes. And I can see that prescript.txt txt is already there because it created this file called

where it created test test.txt in your working directory. Where is that file?

working directory. Where is that file?

Okay, here. So it has created test.txt and obviously it would have used tool called write right. So now my script automatically ran and it created it

generated this file which is called this is a prescript. Wow. This way you can even create a postcript as well. Let me

show you.

Let's say post script.

Perfect. And here I will say this is a post script. Simple. And let me just add

post script. Simple. And let me just add this hook as well.

Now I can extend it on my own. Nothing

to worry.

Simple. Now let's ask it for one more time just to create that thing because this time I should see two files.

Ctrl C. Okay, let me just delete the previous one just to test it.

Um, let's go to CL code.

Perfect.

So, let's see if we can see two files this time.

Okay. Okay. Prescript is there because that is called automatically. The file

already exists. Hello world in it. No

action needed. Wow. So, let me just say create a new file, bro. But it ran my prescript because that automatically runs before the script before the tool call. Simple.

call. Simple.

Create test two.

Perfect. So I can see press script already because tool call is recently made and my script is already there. Now

if I approve it, I should see postcript as well.

Uh postcript is not there. Why? Why? Why?

Why? Oh, I see. because I have not changed the path.

Okay, perfect. Let's test it for one more time.

Create test three. Okay, let's see this time.

test three. Okay, let's see this time.

Okay, prescript is already there obviously and now I will approve it. I

should see postcript. Perfect. So, both

of my Python scripts are done. Simple.

So this way you create all of these particular hooks and hooks will automatically be triggered whenever you are associating a hook with your event.

Simple and let me just now just show you some more examples so that you can just try on your own. There's there's like very one very good example and let me

just check. I think it was there in the

just check. I think it was there in the um in the example in the build with build with code and then where was it? Let me just find it for

you.

Hooks create. Oh yeah, here let me show you. I

create. Oh yeah, here let me show you. I

found it. Let me just show you.

Uh create getting started extend code books. So they have provided this

books. So they have provided this amazing example. You can just try it. So

amazing example. You can just try it. So

what this example will do this is I think hooks reference. Let me find it.

Yeah, here I finally found it. It is

inside build with code and automated with hooks. So these are some of the

with hooks. So these are some of the how-to tutorials in this particular section. So here you will see that they

section. So here you will see that they have simply created one hook which is very nice. So which will simply send you

very nice. So which will simply send you some notification whenever it needs your approval or anything in your tool call and it will just use this particular PowerShell command to send you the

notification which is not a very good thing by the way. You should instead connect your personal text messages or chat messages because most of the time you would also keep your Windows system

on DND. But yeah, if you just want to

on DND. But yeah, if you just want to test it, you can just create one hook for this. You can just play around with

for this. You can just play around with it. You can just see so many examples

it. You can just see so many examples here. Most of the time what people

here. Most of the time what people prefer is running the bash commands or PowerShell scripts. Okay. But you can

PowerShell scripts. Okay. But you can simply run Python scripts because Python can do literally anything. But sometimes

when you just want to perform some additional things, you can just go with the PowerShell scripts as well. Simple.

And obviously Linux commands. So here

you can just see some examples, some how to tutorials. Simple. But now I think

to tutorials. Simple. But now I think you have got the intent like what is the end goal that we want to achieve. We

simply want to run our script on a specific event and that's it. Because

clawed code allows us to utilize the events. Simple. And we can create hooks

events. Simple. And we can create hooks around that. Okay, simple. I think it

around that. Okay, simple. I think it will be very very helpful whenever you want to automate literally everything.

Automation plus logging is very important. So it will do both with the

important. So it will do both with the help of hooks. Plus it will not use LLM.

It will not use a context. So I think you can smartly integrate hooks whenever you want to automate anything obviously because you do not need to pass that particular context in your LLM. This

will be like pure local execution and that's it.

Simple. So this is your hooks in claw code and the best part is now we have covered everything in claw code.

Everything everything means because see obviously everything means now you are wellversed. Now you are capable enough

wellversed. Now you are capable enough to integrate cloud code in your workflows, cloud code in your day-to-day activities, claude code in your

day-to-day job, whatever you do. But now

you can confidently work with clot code.

Yes, there will be so many so many upgrades in the future in the AI obviously in the data engine obviously.

So if you want to just stay updated just hit the subscribe button right now and just hit the like button and drop your lovely comment on this video and just click on the video coming on the screen.

I will see you there. Happy learning and happy