Claude Code + Karpathy's System = $10,000 Skills

Skills are Cl code's most important

feature, but 99% of people are using

them incorrectly. Using Andre Kapathy's

system, I found a new way to build these

super skills that are self-improving,

have infinite memory on the kinds of

things that clients pay me thousands of

dollars for. And in this video, I'll

show you exactly how to build them for

any output, so you can stop wasting your

time, make more money, and unlock Claude

Code's full powers, even if you've never

done this before. And if you don't know

who I am, my name is Jack Roberts. I

built and sold my last tech startup with

tons of customers and now I build AI

startups and show you exactly the stuff

that works. So if you haven't already,

grab that beautiful coffee and let's

dive straight in. So let's talk about

how we use Claude Code and Carpathy

system to make skills that are worth

thousands and thousands of dollars and

will drive those cool things for your

business. So first thing to understand

about skills that most people get wrong

is that skills are like superpowers.

when so does super strength, flight, and

laser vision. These are the things that

we're going to use most of the time in

our business to accelerate everything

forward. But here's the difference is

that most people are just using basic MD

files and they're not configuring them

properly. And that essentially means

imagine having Superman that couldn't

fly, couldn't use laser vision or any of

these things. It wouldn't really be

worth doing. And the bad part is most

people don't realize this has happened.

And this brings on to the concept of

something called super skills. Now,

super skills themselves basically have

three things in common. And what that

effectively means is that every

conversation is indexed and recallable.

And Claude wakes up smarter every

session. So, say for example, you have a

YouTube intro skill and you give it

feedback, but then the next time you

open it back up, it just forgets

everything you said. That would be a

really bad skill. So, we wanted to have

this memory it can come back and use.

It's going to use actual tools that are

specific for the job. And crucially,

it's self-improving. It scores its own

output and it improves over time. So

every time you use a skill, it gets way

better. An example of that, for example,

would be something like this signal

dashboard here that improves time after

time after time. And we're going to

build something similar to this in this

video. I'll show you exactly how you set

this up. And I'm also going to show you

how you get your own memory operating

system that looks something like this

that connects everything from your

long-term memory, short-term memory, uh,

skills, and everything. So you get a

sense of what that actually physically

looks like. So when I'm talking about

super skills, what I'm effectively

meaning now, there are two types of

skills that we're going to use in Claude

code. One is going to be what I call the

utility skill. So that's going to be

something that's really easy and dead

straight. Now there are two types of

skills that we're going to be using

within Claude. Again, this can work in

any model. One is going to be what I

call the utility skill. So that's going

to be something that's really easy and

dead straightforward. It be something

like my bitly skill, right? My Bitly

skill, for example, just takes URL and

shorten it. I could say something like,

"Hey there. Um, just create a URL, a

shortened URL for glido.com." All right,

I give it a link and it just goes ahead

and shorten for me. Utility skills are

very wrong, very, very easy to get

correct. It's very difficult to get them

wrong. They're dead simple. There you

go. It's now used my bitly skill. I can

click this for example. I open it up and

then I see Glad the speech to text

service that I'm using to kick butt with

this. And then we have super skills that

are built on Karpathy's foundation, his

mental models for basically the four big

things holding you back with models. It

leverages data. It improves itself and

it's just significantly helpful for

driving your business forward. We call

these super skills. There's the stuff

that actually works and that's what I'm

going to show you exactly how to build

inside this video. So when you're

thinking of super skills, it's stuff

that listens, it remembers, and it

improves over time and it's actually set

up correctly cuz most people don't get

that correct. Okay? It could be stuff

like an insight a outlier analysis,

sponsorship, replying, anything that you

can think of. Now, here's the reason why

most skills don't work. And when we

understand that, we know exactly how to

improve skills that actually do. First

of all, they're just markdown files.

Most people I see, they just download

markdown files and they never use them

again or they don't work properly. And

they're generic. For example, a skill

that might help me with my YouTube is

going to be so completely different than

yours cuz it doesn't have that strategic

context. It's like a business adviser

trying to give you help and advice, but

doesn't know what the hell he's talking

about. lay out. This guy needs some

coffee. He's not thinking clearly. The

static, in other words, the second you

make it, it's exactly the same as it is

like a thousand times later. Businesses

evolve. Your needs change. It needs to

reflect those dynamic changes to it

properly. And again, it just forgets

everything every time you use a skill.

It's the exact same thing again and

again. And if you're wondering who Andre

Kopathy is, he's a guy that if you ever

want to feel bad about your CV, just

look at what this guy has done. All

right. So, he co-founded Open AI. Yeah.

Okay. We'll we'll take it. Ran Tesla AI.

done lots of really cool stuff and he

basically started the tweet that

authored part of the reason why I wanted

to make this video with this brand new

system. So he basically posted on a his

philosophy and mental models around the

big four limitations with models that

affect everything that we do whether

we're coding with it or in this case

using skills that can quietly hold you

back. This was then very kindly turned

into a skill that got over 88 8,000 uh

stars on GitHub and is part of the

foundation and one of the components of

what makes these super skills. So just

to explain the TLDDR what it is um it's

essentially instructing it's a very

simple markdown file but explains to

models that you need to think before you

code. Okay, keep it simple again don't

overengineer code surgical trenches only

touch what you must trace every line and

goal driven execution. The idea is these

are four guiding principles to

drastically improve the quality of

outputs because these are the four big

limitations essentially and this is

something you can use for skills

anything else. Now to get access to this

all you're going to do is shoot over to

this GitHub repo right here. You're

going to come over to code. You're going

to click on copy GitHub repo just being

fancy word of saying where we keep all

our files. Then you're going to open up

anti-gravity or claude code. Cool. So

I'm in the CL code section. I'm going to

say hey there go ahead and install this

repo for me. Okay. Then you're going to

come down and you're going to drop in

the GitHub repo. And all it's going to

do is literally install it into your

environment. Explaining what it is is

dead simple. Again, it's just four

principles. Think before coding. And

that tries to attack things like wrong

assumptions, hidden confusion, missing

trade-offs, simplicity first, don't over

complicate things, no bloated

abstractions, orthogonal edits, touching

code that you shouldn't. In other words,