Loading video player...
The biggest problem with vibe coding as a non-engineer is that you don't actually know what makes code good. And if it isn't good, it means it's going to break all the time. It's not going to be scalable. It's not going to be testable. It's going to be ugly. And no one's going to want to use it. And I get it. A lot of beginners think clawed code on its own is enough to build things that are great. But here's the trap. By default, language models do not write good code. They write plausible code
based on what they've read. And plausible code turns into bugs, horrible UX, and things people don't want to use. So, in this video, I'm going to break down my sevenpiece clawed skill stack that allows you to package engineering expertise into your project. Now, before we start by getting into skill number one, I want to spend about 15 seconds giving you a rundown of where you should actually use Claude skills and not something else. So, a basic mental model is this. Custom
commands are basically just like reusable prompts. If you have a sequence of steps that you find yourself consistently prompting and copy pasting into the model you're using, just write that into a custom command. Now, sub agents are like mini instances of claude code that operate within their own context window. So, they are really good at solving very specific problems by having a context window that is for that specific problem. But skills allow us to give specific expertise
to our agent and then we can use that expertise inside of our main context window. So, I'll have a prompt below the video here if you want it that actually helps you decide if what you're intending to do should be a sub agent or if it should be a skill depending on what it is exactly that you are doing because they are useful in different situations. So, the first skill we're going to talk about is actually published by Anthropic and it is their front-end design skill. So, probably the
most complained about thing that I hear about from subscribers is that they really struggle with the aesthetics of their app and the user experience. So, basically, in any context where you aren't just cloning an existing design from someone like Airbnb, from a tool like Mobin, it can become very difficult to build something that actually looks really nice. And so, this front-end design skill is supposed to be a simple way for us to get around that. and it does it by giving the agent very
specific directions about what the skill of UI design is even about. So, we're going to start off by actually getting access to this first plugin. And we do that by typing /plugin and then going down to the clawed plugins official. Browse the plugins and then we can choose the front-end design plugin and we can install it here in our project. So, what we have here is a very basic thumbnail design tool that I have been using that I built with Nano Banana Pro. And obviously, it looks a little bit
stale. And so, I want to come down and I want to see if we can actually use this front-end design tool with a very simple prompt to make it pop. Now, these skills are typically automatically invoked by the system, but you can just tell it to use the skill as well. So, I'm going to give it a very simple prompt. I'm going to say, "Use this front-end design skill. Help me help me revamp the front-end design." It should use these libraries. The general feel should be
like tech forward yet kind of Spartan. So, kind of like a Tesla interior, but with pops of color and kind of ease on the eyes like you would get with a tool like Claude. All right, guys. So, that just finished up frontend design skill, and I ran through a prompt here. So, it's looking pretty nice compared with what it looked like before. There's still a few little things I think that I would want to tighten up here. If we go in and look at the actual like gallery,
they changed the entire gallery view. They did this nice little hover animation. Obviously, some of these buttons need to be kind of fixed. This should be inverted the other way. But if we were to like actually click on one of these, again, this looks really nice, really tight. And then nice editor workflow that it built out for us here with some nice little UI elements. So, for a very basic prompt, which again, I didn't dial in at all. This is ideally something that you would dial in, right?
This is a beginner or like a a starter prompt, but you can come in and you can make this your own, infuse it with your own design principles, um, all of that good stuff. And I really like what they did with the setting page. So, with this skill, that consistent struggle of trying to make UI that actually like conforms to your plan should be a thing of the past. And now we can move into our second skill which is actually the skill of writing skills but for code
specifically. So as you start realizing how powerful these skills are the next logical or natural question becomes hey can I build these things myself based on what I do and my structures and my processes. And like anything there are good ways to build the skills and there are bad ways to build the skills. So luckily the author of my favorite skill library superpowers has a skill that is specifically for writing skills that are good for claude code. I wonder if I can
say skills five times more in this sentence. And so what's interesting about how they approach it in this repository is they basically say creating skills is much like creating tests. And so it has this analogy from testdriven development and where that kind of maps onto this skill creation process. And so there's a lot of details in here. You can go and read it. But what we're going to do is we're going to hop in and we're going to actually create one. So I will clear out my
context. Again, you would want to go into your plugins and you would want to go over to marketplaces. In this case, it is the Ora superpowers marketplace. You would want to browse the plugins and then from here you'll be able to pick the superpowers library which is where this is going to actually live. Say for example that we want to create a skill to use inside of cloud code that is specifically for writing change logs. And so we want it to be that anytime we
maybe make a major feature addition or change something meaningful that it creates a change log which could be developer focused or customer focused and it creates that automatically based on this skill being called. So what we can do is we can use this superpower for writing skills to help us actually make this thing. Now in this case I already have one but I could just say hey I want to delete that one and start from scratch. And so what it's going to do is
it's going to go through and it is going to create me a fresh skill and delete the other one that I had. All right. So now that that thing is done and it created this change log generator skill for us, we can simply come through and we could ask it to create us a change log. And now we can see that this change log generator skill is being used. And this is obviously something that would happen automatically, but in this case, just to show you guys, we're going to
generate it manually. And I'm telling it to look through the last five commits from this project and then based on what it finds there, generate me a change log entry. And so there we have it out the other side, a really basic change log, which obviously we could ramp this thing up and amp it up and make it a lot better. But the point wasn't the change log. The point was we have this new skill where we can create Claude code specific skills very easily on our own.
So, the next one up on the list is actually a bit of a bonus skill because it's an entire library of my favorite skills. So, what we're looking at here is a repository called Obrapowers. And so, this is an entire library of Claude Code specific skills. And I know it sounds a bit crazy to say, but it really has changed my perspective entirely when it comes to how I work with Claude Code. and I'm going to show you a few of the skills that are the reason why that's the case. So, inside
of this repository, there's a few core skills that are really awesome. The first is the brainstorming skill. And so, what this does is anytime you set out to do some new work like creating a feature, adding functionality, making a major modification, it triggers this skill. And what this skill does is it really breaks down the process of really understanding the idea, coming up with different fundamental approaches to solving it and then presenting you some sort of design documentation
specifying how it thinks you should proceed. And from there you can use the next skill which is writing plan. So these things hook together where we go from the brainstorming stage, we approve the design and then we jump directly into writing out an actual stepbystep plan for implementing this new thing that we are building. Now there's a ton of other skills that they have access to that you can go in and check out. I love the systematic debugging skill, but I'm
going to hop in and show you guys how we can use these brainstorming and planw writing skills. So let's say for example we came in and we were like hey I want to build this new feature that uses the user's performance data for the thumbnails like the click-through rate the impressions and everything like that and it uses that to train future thumbnail generation so that we can get datadriven thumbnail generations from this tool that we've been building or updating.
And so we can see immediately what it does is it's actually triggering this brainstorming skill right off the bat. And so it's going to start off with asking us a few questions about the fundamental approach that we should be taking. So in this case, I want manual entries of this data. YouTube doesn't actually expose a lot of this data through their API. So outside of scraping it, this is going to be my best option. And then we're going to continue moving through and just answering the
base questions that it has. And so again, the reason that I love this skill is it's asking us a lot of really good questions that were clearly ambiguous based on what our ask was. So, I said, "Hey, I want to be able to use a user's performance data to trigger better thumbnails in the future." Well, there's a lot of things we need to hash out based on that, like, where is the data going to actually come from? How is this even going to get entered into the system and used? How should it influence
future generations of thumbnails? What are the actual criteria that determine a top performer? In this case, it's going to be the actual click-through rate. How many get sent through in each generation? And again, for all of these, it's giving us trade-offs, right? If we put too many images in, it's going to use a lot more tokens. So, it's going to be more cost costly. And so, we're just moving through and answering all of these questions one by one. So, once
this is complete and it's gone through, it's asked us to all the questions. Now, it's going to move through and make sure we are on board with the breakdown of what it is going to build. So, it goes from the feature overview, the core flows, the key constraints, making sure we're on board with all of that. Same thing with the changes that are going to be needed to our data model. How the actual API integration is going to work, how the UI is going to actually support
this thing we're building, how the settings and the user controls are going to change, and then how we are going to handle edge cases and errors. So, it's making sure that we're on board with all of this stuff. And then from there, we are now invoking automatically this writing plans skill. And so again, it's going to have a really structured way of moving through the codebase, documenting what we're going to do, and then actually writing a plan that we can implement against. So now that that
skill is complete, we have this comprehensive plan detailing everything that we would need to do in order to go build this feature. And when I say comprehensive, I mean it is actually quite long. It has all the testing in there, everything that we would need to go out and actually build this thing. So, if you're someone that is trying to vibe code with direction and a system behind you, this entire skill library I think is a bit of a nobrainer. So, next up on the list for anybody that ever
said prompt engineering is dead, clearly doesn't actually build stuff that needs good prompting. And so this prompt engineering skill is great if you are someone that's building apps that actually relies on calling model providers in order to do stuff. So what we're going to do here is we're going to ask it to use the prompt engineering pattern skill that we just looked at to analyze my Gemini calls and provide recommendations for how we can actually fix things. And so here it is. It's
calling the skill and then it should generate us a report explaining how we could optimize our prompting. So after this thing has run through, we can see that it has a bunch of recommendations for us about how we are actually running our main generation prompt and some of the issues that are there and the severity meaning like what we could do to actually or rather what the priority of the problem is and giving us very clear recommendations for how to improve
it. So my current prompt under the hood actually looks like that where it's relying a lot on this user prompt that comes through when in reality we could get much better outputs more consistently if we changed our prompt to something more like this more structured. And then that same approach applies to a few other areas. So this is all great because these are all things that would really meaningfully make this app better. And this is for something that's pretty straightforward, which is
a basic image generation workflow. But obviously, sky is the limit with something like this. So skills like this one are exactly why I say that skills themselves are one of the best releases since sliced bread because we can take serious expertise from other people and in this case it's how to prompt engineer properly and then we can extend our own agents with that knowledge in order to make all of the stuff that we build just that much better. And the next skill on
the list is really no different because we start getting into API design. So what we can do with this skill is that we can use it to help design our backend APIs in a way that does not stank. And so what we're going to do again is we're going to come down here into this project which doesn't actually have a REST API. And what I'm going to do is I'm going to ask it to plan out an approach to actually migrate it to a REST API. And so what we can see is that it's now using this specific plug-in for
back-end development and it's calling the API design principles skill. And so it's going to use this to actually explore my entire codebase and then decide what changes to make and how it would actually migrate this app to using a REST API. And so this one is super useful because as our agent or our sub agents begin any back-end API work, we could configure it to actually invoke these skills. And what that means is anytime it goes to do this backend API work, it's going to have a bunch of
great context and ways of working automatically injected into the definition of the agent at the time that it starts running. And this is why when people say that they think AI is going to replace coders and engineers, I kind of just laugh because we are always going to need people that can build out this type of expertise so that everybody else can actually use it and implement it in order to build things that are objectively better. So now we can see we have a pretty robust migration plan that
would take us from our current state where we are through to a fully implemented REST API full with all of the endpoints, the user models, all of the data models, everything that would need to be there in order to make this transition happen. And it is very very detailed. I think they even built a job Q architecture in. So a very awesome thing that we got out of this skill. So again, another highly recommended especially for vibe coders that want to be able to do things the right way. And
again, these skills can be invoked inside of your sub agents. And there's really a ton of awesome stuff that you can do. And so this next one up on the list goes handinhand with the last one and really doubles down on that idea of why we will always need people that are absolute experts in something to help create these skills so that if we are vibe coding, we're doing things well. and we're not doing stuff that is just dumb. And so what we're talking about is
Postgress table design. So this was one of those skills that really helped solidify for me how valuable these skills are really going to be because in this case we can again make sure that when we are working with Superbase for example that we aren't doing things that we really should not be doing that are going to create serious problems when we start to scale the database and get users and all of that stuff. And so what we could do to demonstrate this is we
could think about this plan that we just made for the rest API. And we could first clear out the context here and we could tell the system to use the Postgress table design skill and update our REST API migration plan based on the Postgress skills, right? And then we could link to that plan. And so now it's going to take that API migration plan and it's going to make sure that it's conforming to all of the best practices that are laid out within this skill. And so now that this
skill is finished processing, we can see that there were a lot of updates that were made to that migration plan when it comes to a bunch of different pieces of the database table design. And we can see that we have these other sections that were actually generated that are now at the end of this plan which we could move in and we could see exactly what was made. So pretty awesome. And again taking our project and elevating it beyond just some basic thing and starting to move it into that realm
where it would actually have some sort of place in a realworld application. All thanks to Claude skills. And so this is another skill that I think you'd be a fool not to use as a vibe coder. But we will move on to the last one, but definitely not the least one, the error handling skill. So error handling obviously very important. No denying that. But error handling well not only helps you, but it really helps the language models actually perform better. when you have much more contextual and
sensible errors that you can give it, you end up fixing things a lot faster and it's just a much better workflow overall. And so what this skill does is it actually helps us do that. So it prompts for some distinct philosophies on how to actually handle errors the right way. It categorizes the different types of errors. And then it actually gives language specific error handling advice, which is good for me because I use Python and JavaScript primarily. And that is what the error handling
conventions are that we have inside of this doc. And then it breaks down a bunch of patterns that are really helpful to know that you would potentially want in your app depending on what you're doing like circuit breakers and a bunch of other cool things. So the way that we could use this is again to hop back into our project. And so what we could ask it to do in this case is to look at our frontends hooks and then I want you to use the error handling skill to improve
error handling in those files. So these are all the files that are responsible for how we do like the actual character generation and the edits and storing characters and doing all of that stuff. And so now that this skill is running, we can see that it's already moving along and recommending us updates. So it's creating a shared error utility that all of these hook patterns are going to actually use. And then it's going to move through and it's going to actually start improving the patterns
that we're using to actually log these errors, which again is going to make our life a hell of a lot easier. So there you have it, guys. My seven skill claw code stack that I use on a weekly basis. Just remember, skills give your agents specialized expertise and so you would be a fool not to use them. Now, there's a ton of other skills out there that I didn't have time to cover, but even these seven alone will really improve the quality of what you build. So, get
out there and start building stuff with these skills and let me know how it goes. I've got a free Vibe Coding community in the description below where I post all of my YouTube files and prompts, templates, and workflows. You can go check that out and share the type of stuff that you have been building. That is it for this video. I will see you in the next
Get FREE access to my complete Vibe Coding course here: https://www.skool.com/tech-snack In this video, I break down the 7 essential Claude Code skills every beginner vibe coder needs to master. If you're building with AI but your apps keep breaking, have terrible UX, or just don't scale—these Claude skills will change everything. I'll show you exactly how to inject professional engineering expertise into your Claude Code projects so you can build production-ready applications that actually work. By default, language models write plausible code, not good code. That's why these claude code skills are absolutely critical for anyone serious about vibe coding. From frontend design to API architecture, database optimization to error handling—each skill packages real engineering expertise that transforms your AI-assisted development workflow. 📚 What You'll Learn: ✅ How to use the Frontend Design skill to create stunning, professional UIs without design experience ✅ The Skill Creator skill—how to build your own custom Claude skills for any domain ✅ Component Architecture patterns that make your code maintainable and scalable ✅ API Design principles for building robust backend systems with Claude Code ✅ Postgres Table Design best practices to avoid database disasters as you scale ✅ Error Handling strategies that make debugging faster and improve AI performance ✅ When to use Claude skills vs custom commands vs sub-agents for maximum efficiency ⌚ Timestamps: 0:00 - The Problem with Default AI Code 0:54 - Skills vs Commands vs Sub-Agents 2:18 - Skill #1: Frontend Design 5:09 - Skill #2: Skill Creator (Meta!) 9:20 - Skill #3: Component Architecture 11:54 - Skill #4: API Design Principles 14:42 - Skill #5: Postgres Table Design 17:06 - Skill #6: Error Handling 21:02 - Your Next Steps 🗄️ Resources: Prompt from the video: https://www.skool.com/tech-snack/classroom/0ca89703?md=2a3a4a6776c3493a9e8bfedddf181d2a FE Design: https://github.com/anthropics/claude-code/tree/main/plugins/frontend-design Writing Skills: https://github.com/obra/superpowers/tree/main/skills/writing-skills Prompt Engineering: https://github.com/wshobson/agents/tree/main/plugins/llm-application-dev/skills/prompt-engineering-patterns API Design: https://github.com/wshobson/agents/tree/main/plugins/backend-development/skills/api-design-principles Postgres Tables: https://github.com/wshobson/agents/blob/main/plugins/database-design/skills/postgresql/SKILL.md Error Handling: https://github.com/wshobson/agents/blob/main/plugins/developer-essentials/skills/error-handling-patterns/SKILL.md My #1 Skills Repo: https://github.com/obra/superpowers Join the free Vibe Coding community: https://www.skool.com/tech-snack All prompts, templates & workflows included 💪 Who Am I? My name is Sean... I'm a biotech consultant turned tech bootcamp bro-coder turned tech company sales engineer turned digital marketing entrepreneur turned back into a tech bro entrepreneur. Learning all I can about AI and trying to communicate some knowledge along the way. The singularity is near! 👇 My Other Social Accounts 📸 Instagram: https://www.instagram.com/seankochel/ 🐦 X/Twitter: https://x.com/IAmSeanKochel 👨💻 Linkedin: https://www.linkedin.com/in/sean-kochel/ 🎥 Facebook: https://www.facebook.com/realseankochel