📱

Get Our Mobile App

Take your business learning on the go!

Download on the App StoreGet it on Google Play

Claude Code Skills: Automate Everything You Do

aiwithbrandon47:55

Transcription

Hey guys, my name's Brandon and I have a confession. I'm addicted to Claude Code Skills. And by the end of this video, you're going to be addicted to Claude Code Skills, too, and know how to create skills to automate your own workflows.

And just in case you haven't heard of Claude Code Skills, it's a brand new feature released by Claude where you can teach it your custom workflows. And as soon as Claude understands what you want, it'll save it to a set of custom instructions. And then every time going forward, Claude will be able to do all the hard work for you and automate the entire process.

And personally, I've been using Claude code skills to automate everything in my life. All the way to helping me come up with titles and hooks for my YouTube videos. All the way to automating the boring task inside of my startup, Shipkit.AI. Basically, if there's something I'm doing more than once, you can bet I'm creating a Claude code skill to help me out because they let me move faster, more efficiently, and make my life a thousand times easier. And by the end of this video, you're going to be addicted and be creating them, too.

And most importantly, to help you wrap your head around creating amazing Claude code skills, in today's video, I'm going to be walking you through the six-step framework I've created called Master, which helps you make amazing Claude code skills to automate your own task. So, in the rest of this video, we're going to go through step by step what the heck is this six-step framework. Then at the very end, we're going to take everything and put it to practice and actually build out a claude code skill together so you can see just how easy it is. So if you've never created a cla code skill or you're looking to level up your automations, you're going to learn so much in this video. So without further ado, let's go ahead and dive into the six-step master framework for creating your own cloud code skills.

Okay, so let's break down the sixstep master framework, which the whole goal of this is to help us create claude code skills that are amazing. And what we're going to do in this part is break down everything step by step so it all makes sense. And I think the easiest way to understand it is to actually as we go look at an example of a real world use case. So let's break it down so that we understand how it all connects and so that you can create your own amazing well ststructured claude code skills.

So here's what we're trying to do inside of this process. The very first thing that we want to do is we manually want to use clawed code to help us perform a task. So in this example, if we're trying to write an email, what we could do is say, "All right, step one, I want to write a subject line. Here's how I like to create subject lines, and you know, Claude code, please create a subject line for, you know, introducing me to this other, you know, other software developer cuz I think we could collaborate together." So then, you know, Claude Code is going to come up with different tasks. So that's performing step one. We're telling Claude Code to do the work for us. But as you know, as we both very well know, AI is phenomenal but needs feedback. So that's where step two comes in, where along the way, we are going to be providing feedback to cloud code. So on upon every step, we're going to say, "Oops, nope, that wasn't right. I like to do this instead." Well, that was better. However, it still wasn't perfect. Try it again. So we're at every step providing feedback. And we're just doing this every step along the way. So as soon as we get step one done, we have a great title or subject line for our email. Fantastic. We'll now go to part two where we're going to start working on writing the body of the email where we're going to have like, "Hey person, I saw you did this. I would love to collaborate. Here's how I think we could probably work together." And it would be a win-win for everybody. And as we know, Claude Code is going to do a good job, but it's not going to write like us. So then we're going to say like, "Hey, I don't speak like that. I usually do, you know, exclamation marks. I'm usually I like to be conversational, really excited. like I have a ton of different ways that I like to write emails." So over and over and over we provide feedback once again and we go through this whole process and then maybe at the very end we're saying like hey you know I like the content of the email but I like to make it readable. So, like, can you add in some like bold text so it's easier to skim, you know, just like some maybe fine like last minute tweaks? And we would say, "All right, this is my three-step process for writing emails." And along the way, we've done the task and we provided feedback all in one cla session. And what's phenomenal is now we have set ourselves up to actually turn all of our key lessons learned into a systematized skill. And at this point, all you have to do is say, "Hey, Claude Code, please turn everything that we just did for writing emails into a new Claude Code skill. Ask me as many questions as you need so that you can convert my writing style and this entire process into a skill." And Claude Code is, they did a phenomenal job on making this very easy. It'll go, "Oh, yeah, of course. I can see what you did was broken down into three key steps. Let me go ahead and start planning out making this skill."

Now, what's super super important and you're going to see this in a minute is it we're going to basically when we're creating the claude code skill, we are going to do it in planning mode. What I mean by that is by using claude code plan mode, it's going to have a bunch of questions for us. So, it's going to say, hey, like, do you want this to apply to all emails? Do you only want this to apply to personal emails? You know, so we're basically going to train it by answering a bunch of questions about the skill itself. Then as soon as we cross out all of these, we have a very well-trained first version of our Claude code skill. And now this is where the magic of Claude code skills come in. Because now what we can do going forward is say, great, I need to start working on a new email. Please help me use the email writer skill and help me write an email to this new person. and we're going to go through the same process again. Is it going to be perfect attempt number two? The answer is no. There are still going to be a few issues where we detect new edge cases. We are in a different situation and we're going to update our subject line. We're going to update our body or the way we structure the email. And once we get done writing that email, we are going to enhance our claude code skill to say, "Hey, can you see all the changes I just asked for? Well, as you can see, we want to do rule A, B, C, and D differently. So, make sure next time you write an email, you pay attention to these special edge cases." So, then Claude Code is going to say, "Of course, I would be happy to update the skill for you." And the cool part is every single time you perform an action using your skills, you update your skills so that they get smarter and smarter every iteration. It's basically like training an employee to where eventually after having them do the task three to five times, they're dangerously good at doing their job. And going forward, boom, crank out an email, it's done in seconds versus done in, you know, 10 minutes.

So this is, you know, the first few skills. Then optionally, obviously, what you can do is roll it out. So what I mean by that is well at this point you have a skill that you can use yourself but if you want to really be an impact user of AI you can do multiple things you can roll it out to other people inside your business you can start to share it if you're you know growing your own startup you can say and basically transfer down these skills so that you can have like let's say your like community manager well now your community manager can help write emails just like you. So cool thing here is every time you perform an action, you're basically creating the standard operating procedures to do that task and then you can roll it out or share it with anyone else who you want so that they can start to do the job at your level of quality going forward. This just blew my mind the second I started using it which is why I use cloud code skills to help me do everything going forward.

So, with now that we've covered the basics of the sixstep framework, let's go ahead and take this into action. And I'm going to show you how I'm using Claude Code skills to help me work on my startup shipkit.AI to help me cut releases faster. So, instead of me manually clicking around for 20 minutes, I can now get it done in like two. So, let's go ahead and hop over to a real world demo. So, we'll break it down and then we'll actually start going through the six-step process together. So, without further ado, let's hop on in.

Okay. So before we just start automating everything, I want to give you a bit of context on the problem and workflow that we're trying to automate so it all makes sense. So here's the issue. I have a startup called Shipkit.ai where I teach developers how to build AI applications in days instead of months. Inside of Shipkit.ai, there's a ton of pre-built projects for AI chat, rag templates, ADK templates, and a bunch more projects to make it super easy for developers to launch real world applications stupid fast. In addition to that, there's a ton of courses walking them through the entire process from beginning to end. And we actually use AI to help with every step along the way to where AI is doing most of the work and you're just answering questions and then hey, AI builds it for you.

Now, here is the core issue that is sucking up a bunch of my time and that I want to automate. As you just heard, there was a ton of pre-built projects that people are using. Now, as developers are using these different projects, I'm adding new features, functionalities to make it as easy as possible for them. On top of that, every now and then, a user will find a bug. So, it is my job to continually update these projects and make them as easy as possible to use. So, every time a new fix or improvement comes in, I have to cut a new version of the project. I have to then update the website to say, "Hey, you know, here's a new feature for you guys that you can use." So realistically what this looks like because this is a multi-step process. Here's really what the process that I'm trying to automate with cloud code skills.

First I have to go in and either do a code fix or I have to add in new functionality. As soon as I make that fix I have to update the local project to keep track of the change so that the user can tell like cool my project number is version 2.4 4. But as you know me, the main contributor as I'm adding in new project versions, I might say, "Okay, cool. Well, now we're on version 2.6. So, I have to update this project. I have to add in new change logs. Everything down here." Then what else do we got to knock out? Well, every time I then update the change log, I have to create a new commit to say what happened wrong and what new functionality they get. I then have to use git to tag a new version. So I might have to say like get tag version 2.6. After that I have to then go into GitHub create a new release so that people can properly download and you know quickly update to the new version. Then finally I have to go update the actual shipkit website to add in a new entry to say like hey on version 2.6 you got some new cla code commands or new cla code skills. So these are this is a a process that I have to do every week to make you know basically shipkit best project out there for building AI applications. Now this takes up so much of my time even though it's dead simple it still takes up a bunch of time to do all of this.

So what you're going to see me do is in the rest of the video I'm going to be applying our master framework that we just covered and I'm going to show you exactly how we can apply this framework to automating this process. and you're going to see this thing's an absolute beast. So, without further ado, let's go ahead and hop over to Claude Code and start automating this process so you can see the master framework in action.

Okay, so now it's time for us to apply the master framework to the project so you can actually see everything in action. So, here is a quick to-do list of everything that we're about to do so you can see it all in action and laid out. So, at this point, we're going to say that, hey, I'm already working on a project, and I've made a ton of code changes. Just a quick heads up. So, all the code changes are for a new project template that I've adding to Shipkit. It's the number one most requested new type of project template where people are going to learn how to build background jobs so that they can have their NextJS applications kick off longunning task in the background, whether that's longunning AI workflows or just regular longunning jobs. And basically this was what people really really wanted. So that's what I've been working on building. So what I am trying to do is say, "Hey, I made a ton of changes on building this out so that it's going to be ready very soon." Speaking of, if you want to use Claude Skills, get $50 off when purchasing Shipkit. I'll have a link down in the description below. But outside of that out of the way, what we want to do now is we want to start actually documenting our process so that we can have Claude Code make a skill for us.

So, here is exactly what I'm going to do. I'm going to paste in a request. And here's basically what we want to start doing. We want to start updating our change log. So, specifically, here's our change log. We're currently on version one. And there is only, you know, there's no new things yet. So, here's exactly what we're going to do. We're going to kick off clawed code. And at this point, what it's going to do is start to look at all of the changes that we've made since the initial release, which was, you know, 1.0. And this is all testing because it's not fully public yet. But at this point, what it's going to do is say, "Okay, I will happily start to look through all of your recent changes. Let me go check out to see what the current version is. I'm going to go check to see what the update logs are. I'm going to also just check basically everywhere." And at this point, what I'm going to do is start to provide a little bit of feedback. So going back to our initial workflow, I manually do the work, then I start giving feedback every step along the way. So I'm just watching Claude Code do the work. And here's a few things that I think are interesting. I never saw it use the GH CLI tool to actually go check GitHub. So did you ever use the GH CLI tool to look at the most recent public release for this project to actually figure out what were the official release notes? So this is us providing feedback and uh it misspelled it ghi. So then we'll kick that off. And at this point, you know, it's already made some changes. So we're definitely going the right direction. It did auto increment it, which is awesome. That's exactly what we wanted. What is weird is normally I like the change log to go inside or go the other order. So that is another quick thing that I want to change. So at this point, we're basically just, you know, we're starting to figure out, hm, for doing task one, you kind of messed up a few things, and we're going to instruct it on what it needs to do differently. So it's going to go, ah, I can see. So it can see on the official GitHub page, what we had was just some like an basic boilerplate, you know, cuz this is an unofficial release yet. So hey, everything says example, but I see, however, I should use GitHub release in the future. So we'll just say great in the future. Make sure that you use GitHub release. Also, I do not like the way that you're updating the change log first. I like to have the order be ascending and you've made it descending. So, please fix that. Then, what we can also do is see all the other changes. So, it's saying, "Hey, add complete uh transcript workflow to trigger.dev file upload job management summaries." So, I like the change log. Like, that's a that's a pretty good message. So, I'm very happy with the way it created the change log. So, at this point, yeah, you can see it flipped the order. That is fantastic. So, at this point, we have updated the change log to accurately reflect what we're working on.

So, what we want to do now is create a new commit. So, we're going to say, fantastic. Can you please use git add and get commit to store the new updated shipkit change log and push it to GitHub? So we'll kick that off. Now this is, you know, something I normally have to do where I'm manually having to create the commit. I'm manually having to update the shipkit JSON. So this is what I normally would do manually. So we are crushing it. We are cruising on the manual side and we're cruising on the analyze and get feedback. And at this point it's trying to add a new git commit message and I'm totally fine with it doing that. And then very very soon we should say success. I've committed and pushed the changes. Everything is looking good. Fantastic.

So now what we can do is move over to the next step. So we can say great. Now can you please tag the current new version using the following command and then what we're going to do is just tell it that it should do get tag and then the name of the new version. So, and I'll just say, and just for context, this tag version is the updated version you just applied to the change log in shipkit.json. So, now it's going to then kick that off. So, let's see. It's trying to actually point out the the file. So, we're going to say fantastic. Now, what it's going to do is it's going to be like, of course, I will happily tag it. So, now it's going to update and actually run the tag command. So you can see it's running the tag command for me and I'm happy for it to do that. So yes, please proceed. And then it's going to push the tag. I didn't have to tell it to do it. It just knew to do it. If obviously if I didn't want this, I would correct it, but I do. So fantastic. So at this point, we have now tagged the new version and we have applied and pushed the change to GitHub. So if we actually hop over to GitHub real fast and we go over to ShipKit and look at this exact project. Let's see. I think I went a little too deep. Let's see. GitHub Shipkit. And then we'll actually go over to the project. Yeah, wrong repository. Yeah. So, you can now see if we go inside of here, we have a new tag. So, our new tag is version 1.1.1. That is or 1.0.1. That is perfect.

So, now what I'm going to do is say, great, you did that perfectly. Can you please use the GH CLI tool to create a new release for this new tag? Do not actually do it yet. Just show me what you would do for the title of the change log and the, you know, the actual body of the change logs so I can review your work. So, I'm going to kick that off cuz I don't want it to actually run the command yet. And this is where we as, you know, developers who are doing this workflow constantly, we probably have a few examples that we could automatically pass in. So, that's exactly what I'm going to work on. So I'm going to go over to maybe an existing tag that I've just worked on recently. So we'll just go over to the rag application and then I will go copy another release just so it can have reference files. So I will actually and heck I'll just tell it that cuz it's like you could always pass screenshots if you have access to other existing reference material that way or you can just say hey by the way if you want to see examples so you better understand what to do please use this. So, what I'm going to tell it real fast is say that is looking great. However, I would like you to match the style of how I normally create titles and bodies. So, please use the same CLI tool to look up the most three recent releases for the and then we'll pass it in rag SAS application or we'll say repo just so you can get a feel of what I like. And if you have any questions, please let me know. if not update your change log to reflect the new style. Fantastic.

So, this is a great meta lesson real fast to where when you're actually going through and giving feedback or manual instructions, you can always add in existing reference material. So, if this is your first time doing it, you could, and we were going back to the email example, you could say like, "Hey, here's seven previous emails that I've used in the past. Feel free to go ahead and steal inspiration for these." Or if you're giving feedback, you could once again say, "Hey, here are three example references of what I've done in the past. Update what you have done to reflect this new approach." So that's what it's doing now. It's going to say basically, am I cool to start searching through using GitHub to look at the most previous releases for your other project? And you can see it's grabbing the three previous examples, doing a phenomenal job. It's understanding what the title style is. It's understanding what the body is. How I usually like to use headers and make it very easy to read what was updated, if there was any big impacts. Now that it understands exactly what it wants, it's going to say, "All right, great. Here is the updated version of the body." So, hey, title, we have a new one. It's following the exact format. It does a high-level overview of what was in the change. then it does a really good job of breaking things down into what's new. So this is perfect. So I'm going to say yes, that looks fantastic. Can you please use the GitHub CLI tool and create a new release for the new tag? And if you have any questions, please let me know. Fantastic.

So at this point, it's going to proceed to go through the process and we are almost done. We are on the very last code step and eventually we're going to hop over to updating this on ShipKit. So, we're very very close to being done with the manual process. And it's going to ask if it's cool to start creating new GitHub releases. The answer is yes. So, very soon what we should see is for Shipkit over on the new worker repository, we should see it's made a new release. So, tada, it has done it. That is fantastic. So, it is successful.

So, here's what we're going to do next. Great. We are now on the final step. Can you please use the Superbase MCP server to go forth and look at the shipkit project and specifically the change log table and can you please create a new change log for this project. If you have any questions, please let me know. And just as a quick update while that's going to load, what we're trying to do now is we want to have and basically update our change log. And we're using Superbase and I have an MCP server running. So what's awesome is Cloud Code can actually directly write to my database and create new changes. So that's exactly what we're about to run in just a second. So let's hop on back. And what we're going to do is kick this off. But I'm instantly going to pause it just because I had yet to actually hook up any MCP servers. So that was a quick mistake. So what I'm going to do is kill this very, very quickly. Then I'm going to restart it because I had to add back in my cloud code, you know, MCP servers to where I was adding back in the superbase MCP. So this is just hooking up to my Shipkit project is basically what's happening. So now whenever I type in MCP what would happen is it's going to say tada you are connected to everything. Perfect. That's exactly what we want. Now what we can do is go to resume and we can actually you know kick back off where we left things where we have our most recent instruction. So we can say proceed now. Now what it's going to do is it's now going to start to use the MCP server to start looking at shipkit projects and the change log table. So at this point, it's like hey are you cool with me listing the tables? Yeah, sure. Please go ahead look at the different tables, look at the different schemas. We can actually look in here to see what it's seeing. So, it's seeing all of my different tables inside of Shipkit. I hit control O to do that, by the way. Now, it's going to say, all right, are you cool with me actually making a query to your database? Personally, when using MCP, I never allow execute SQL. I always manually review the SQL command. So, I'm going to say, yeah, you can do it, but not permanently. just for this one command, you're good to select my most previous entries. So, at this point, what it's going to do is it's going to say, let's see, it's going to go through the process, and it's basically trying to look through and just find other examples. So, I'm just going to keep letting it do its thing to find how we've done change logs for previous workflows. So, at this point, it's saying, "All right, I'm just going to do a few more tasks to figure out, you know, what your style is." And you know at this point it understands the whole format. So fingers crossed it's actually going to start to create a new change log for our new project. So we're going to give this just a second cuz I think we are almost there. And it's now saying all right here is what I'm going to do. I am going to run this SQL command that's going to insert. Here's the new title. Here is the body and everything else. So that looks fantastic to me. So, I'm going to say yes, you can proceed. And sorry, it's spazzing out. I shouldn't have resized. But now, what it's doing is it's writing that new entry. And it's going to say, "Tada, we have done that." So, if we actually hop back over to, in our case, shipkit, and we go to change log. When we zoom in now, what we should see is let's refresh. Give this a second. So, now you'll see change log. We have the, you know, thing like we never had to click inside shift kit. We never had to click inside GitHub. We did the entire process through Cloud Code. So guys, give yourself a pat on the back cuz this is knocking out a manual task. We provide feedback along the way. Now, at this point, we are done with the first two steps. And we're now over to the systemize into a skill. This is one of the most important steps to make sure you get things set up properly right here between, you know, setting up the skill and training it.

So, here's exactly what you're going to do inside of Cloud Code, the exact same instance. What you want to do is make sure you hit shift tab to turn on plan mode. Plan mode inside cloud code basically does a phenomenal job of asking you questions to make sure it fully understands what you want it to do before taking action. Awesome tool. So we're going to say fantastic. Thank you so much for helping me go through the process of, you know, documenting my code changes, updating the project change log, committing everything, tagging it, also committing a new GitHub release, and then finally updating Shipkit with the specific change log. Can you please help me make a new clawed code skill that will systematize this entire process? Please ask me as many questions as you have so that we can be on the same page before creating this new skill. Fantastic.

Now, at this point, we're going to kick it off and I'm hoping you guys can kind of see the power of this because this session is filled with gold. It understands our request, what we wanted to do correctly, incorrectly, and you know the entire seven or six steps to actually do the task we want at hand. And the second we get this done right, we are good to go to actually use this every time going forward. So, I'm hoping you can see how fast we're going to be able to move the second this is done.

Okay. So, it's going to say, should the skill automatically determine the version bump? So, I'm going to basically say this is where you shine as a developer because this is where you get to decide exactly what you want and the more context you provide, the better. So, I'm going to say yes, you should always default to a patch. But if the user says they want to do a minor or major update, obviously you should listen to that. But always default to looking at the specific shipkit.json file, looking at the cart version and incrementing the patch. Okay, so that's what we're going to do. Oops, I put that in the wrong spot. So let's copy this over. Paste that in. Then what we're going to do, hit enter.

Now, it's going to say, "All right, so how should the skill generate the GitHub release title and body?" So, we're going to say what we should do is always use the in our case, show me a draft review first, just because before we create the release, I just want to make sure that like, do I like that name? So, that's what we're going to ask for here.

Then, should this skill handle multiple templates at once or do things one at a time? So we're going to say support batch mode. So I'm going to say I would like to support batch mode because there are numerous times when I have worked on updating all of the templates by providing features across the board. So you should also handle that approach too. So basically the ideal workflow would be I would say please go to work on project work SAS or chat SAS and then you would just take it from there. Okay. And then we're going to do that.

And then it's going to say what should happen when superbase change log entry creation. We're going to say I think I would always like for confirmation first just to make sure but you know you can obviously do it your own what you want to do. So once you have answered all of the questions at this point it's going to maybe ask with a few more follow-up questions or if it has enough context it'll just proceed. So this is the best part about plan mode is it makes sure it fully understands what you need before taking action.

All right. So now obviously we introduced a new concept of updating multiple projects at the same time. So it's going to say all right I have a few questions on this. So when it comes to batch mode with working on multiple projects. Should the skill process all specific templates and show me all release drafts at once for review or handle them one at a time? So we're going to say handle sequentially. Should the skill automatically detect when templates have been changed since they're last released or should I always explicitly specify which templates to process? So I think I will always specify like hey you need to work on project A and project B. Then finally when analyzing commits for change log should the skill look at all commits since the last release? And we're going to say yes all commits. And finally where should cloud code skills live? though it's going to say do you want this new skill to live in your current project directory or do you want this skill to apply to all of your projects and this is very specific to the current project. So we're just going to say yeah this is only for the current project. Fantastic.

So at this point we have done a phenomenal job guys of systematizing and training the new skill thanks to plan mode. Now we are in the home stretch where we're going to basically be able to go through enhancing through iteration. So this is will be the very final step. It's asking a few final questions. And at this point, I'm just going to say we'll do create release. Uh we'll do shipkit create release. Yeah, like this. And then change log. We'll just say auto detect submit. Okay, fantastic. So at this point, if it ask anything else, I will just say, hey, you're good to go. So it's now going to present with me the entire plan to make the new skill. So this is an awesome opportunity for us to at this step to start providing feedback on the final results. So maybe it misunderstood something, but 90% of the time for stuff like this, I'm just going to say I just want to start using the template and I'll adjust from there. I don't want to spend another 3 minutes reviewing everything manually. Just like, hey, you got a good idea of it. Probably this was not complicated. It was seven steps with some tweaks. So I'm pretty sure you absolutely crushed it.

So, what it's going to do is the second you start to make Claude Code Skills, especially if this is your first skill, what you would expect to see is in whichever project directory you're working in, you will see inside the Claude folder, you will see a brand new skills folder. Inside of this skills folder, you will see all of the specific skills that you have created so far to help you accomplish your work. So, this is going to take a few seconds, but we will eventually have the ship kit release skill. So, we'll pause and then we'll review it together as soon as it's done.

All right, guys. So, we just created our first skill. However, I want to point out a major mistake made by Claude Code. So, you if you ever run into the same issue, you can fix it yourself. So, here is what went wrong. You'll see that it says, "Hey, I created this new skill for you. It does everything that you asked me to do." And what you can do is basically specifically call out this skill by just running forward slash shipkit-refs. So what basically happened is claude code mistaked this as a command for claude code instead of a skill for claude code. There is obviously there's it's getting a little complex now in cla code. However, we can easily fix this and you can tell if something went wrong if inside of skills you directly see a markdown file like this. you know, clawed skills are not in markdowns. They're supposed to be in files or sorry, in folders like this. So, here is what I did and here's what you can do to fix this very easily. So, what you can do is say, "Hey, please use a tool or an MCP tool called Context 7 and look up how to create cloud code skills and can you just make sure that the brand new skill that we created follows best practices. If you don't have this set up, by the way, you can always set up context 7 like this. Super super straightforward to do. This is the most powerful MCP server for AI development. Definitely recommend you just grabbing that snapshot, adding it to your project. But once you do that, it's going to go fantastic. I'll of course look look this up for you and I know exactly how to create Claude Code skills now. And I'll review yours. And after reviewing yours, I can see yeah, this was a mess up. Claude code skills are not standalone markdown files. What should be happening is we need to you know change it. So it it gave me some recommended actions and I was like yeah just keep it a skill and make sure you fix all the errors you pointed out. And here is what we will end up with. So here's our news file where it has a name and a description which makes it super easy for claude code to at all point look up. Oh, okay. I can see the current user is trying to automate and work on and deploy a new release for a shipkit template. Fantastic. I know that I need to trigger this skill. So skills get triggered more implicitly rather than like doing forward slash like you normally would with cloud code. And what I mean by that forward slash like this is how you can call commands like all these are commands but you can't do forward slash and then call you know shipkit release. It doesn't skills don't work like that. So now that we have that set up, you know, the rest of the skill includes everything that's needed for actually, you know, working on performing all the instructions. So now what we can do is take this for a spin and we will actually delete this old one. And things are looking great. So fantastic. So we're now going to take this for a spin. I'm going to make some code changes and we'll test it out so you guys get to see the final step of enhance and roll out. So let's go ahead and hop over and make these changes.

All right. So, I just cranked out a bunch of code changes for you guys. So, now we get to try out our skill one more time. And just a quick reminder, we are now officially in the enhance through iteration phase because we already have the skill. Now, we're going to iterate. So, here's what we're going to do. We're going to open up a new instance of Claude. So, what we can do is open up Claude. And just quick side note, if you're interested in how I'm just like quickly opening up new instances of claude code and how I'm basically keeping my hands on the keyboard to be a much more efficient software engineer, I have a free resource that you can grab down below. It's a Claude code cheat sheet. 100% recommend you checking out. A lot of developers inside of my school community and Shipkate community have absolutely been loving it. I'm sure you guys will too. So just it'll be one of the first links down below. But back to the task at hand. So what is interesting is at this point when we're trying to use this skill, what you can't do is do forward slash and say shipkit release. You cannot do claude code commands. You have to make the request that basically you know is implicit to say hey I'm trying to do this task and claude code will go oh I know what you're trying to do. You're trying to run this skill. So here's what we're going to do. I just made a bunch of changes to the and then I'll call it out. It is the worker SAS project. Actually, you can probably just call it out this way. Worker SAS. Fantastic. Can you please help me cut a new release for this project? And if you have any questions for me, please let me know.

Now, at this point, we're going to kick it off. And ideally, what should happen here in just a second is it'll go, okay, I can see, oh, I did it perfectly. I can see you're trying to cut a new release. Let's examine the current state. Boom. At this point, it's going to start detecting the skill. So, it knows exactly what instructions it needs to do, and it's starting to do it, guys. Like, this is the crazy part. So, it's going to say, "All right, I can see you're at version 1.0.1." Now, what we're going to do is start working on making improvements. So, here's what it's going to do. At this point, it's trying to make a request to start looking at the latest release. So, we're going to say, "Sure, go ahead. Yes, look at the latest release." And you can see, all right, cool. GitHub is at 1.0.1. Now, it's going to start to check to see what the difference is. So here is a issue potentially is it's looking at the old version. So it's looking at version here's what's going on. It's looking at version you know 1.0.1 and comparing that to 1.0. But basically we have to correct it. So this is where we iterate and we're going to say basically this is where I'm going to say hey here's what you're doing wrong. Okay so what you're doing wrong is I have changes that I have pushed and I have not cut a new version for them yet. So, you were supposed to look at the actual changes I've made locally in additional commits that I've made and compare it against our latest version. You're not supposed to be comparing release against release. What questions do you have for me so that you can update your process and actually do what you're supposed to do. Fantastic.

So, this is the iterate stage. So, this is exactly how you create dangerously effective cla templates or claude code skills. It's a mouthful. So at this point you can see okay I can see you know you're absolutely right I messed up and what we have to do is in the back of our mind we know we need to update the skills. So as soon as we get done going through this task we need to say hey look at what you did wrong now help me do it again. So here is what it's going to do. It has a few questions for us based on the changes we made. This is just a new patch. Fantastic. So it is nice. It's like hey this is kind of a bigger change. So like a minor might be effective, but this project hasn't been released yet. So I'm just doing small patches. So ideally, I'll actually reset everything to version 1.0 afterwards whenever it fully goes live here hopefully in the next 2 weeks. But right now, we're just still practicing. So fantastic. So at this point, we'll release it. Now, here's what we have to watch non-stop is we have to make sure that it's following the process step by step. So, we have to correct it because going back to our task from a second ago, it is skipping a lot of the steps that it was supposed to do. So, this is where you have to come in and start correcting it. So, I'll show you going back to our scratch folder. So, uh let me show you what it's supposed to do. So, you know, at this point, I'll just copy it in and basically just say like, hey, here is the process you supposed to follow. So I'm going to say so here is the process you were supposed to follow. First you were supposed to update the change log. Then you were supposed to look at making a new commit after updating the change log. Then create a tag then start working on the GitHub release. You're kind of going out of order. So please realign and update based on these instructions. And if you have any questions please let me know. So this is one of the once again we are just iterating and refining and the more step by step the process the better. So it's like oh yep I can see exactly what's going on. So what we should see it do here in just a second is start to update our shipkit.json with our new version and you know it's basically like training a new employee where they're like ah I think I know what to do or no I don't. So you can see yep uh app update transcript viewer multiiformat downloads eventbased tracking. Yep that's exactly what we worked on. It did create the new version. At that point, it's now trying to actually push the changes. Now that it made a commit, then it's going to tag it, which is exactly what

We want. So, we're going to say yes. So, at this point, we are almost in the home stretch of like, you know, we're cruising at this point. Now, it's going to push it. That's also what we wanted it to do. Now, it's going to start to work on trying to create that new release. So, you can see it's going in the right order now that it's been corrected. So now before I create the release, just one chance for you to review it. And this new release looks fantastic. So we're going to say looks great to me. Fantastic. So now we are in the home stretch. And ideally what it's going to do after making that GitHub release, we can actually go check it out. And while we're doing that, what was pretty cool is just so you can understand the changes that it's talking about, it's these. So I should be able to like upload a file for the project. Actually, it's not fully fully fixed. Still still working on it. Oh, it's not running. That's why. Silly me. But long story short, you can upload videos and it gets the transcript.

So at this point, it's going to try to actually cut the release. So we're going to say yes, proceed. And if we head over to GitHub, so GitHub, we should start to see in just a second a new release for our new project. So you can see version 1.0.2. It worked. It properly did it. Now what it's going to do is now that that's done, we can say great. And the next step is that you are supposed to use the superbase MCP to update the shipkit change log. That's the very next step you need to work on. So at this point, we can kick it off. And what I want to show you at the very end as soon as this is over is how we can iterate effectively cuz we're going to go back to using the plan tool to update the skill. So it's a very, like I said, it's a very iterative process. So we're going to allow it to properly understand who is the author for the new change. Then it's going to start to insert a new change log. And we can see in this change log what it's trying to do is make a new event and it has the proper values. I just want to make sure that it's attaching it to the right project which is worker SAS and it is perfect. So we're going to say you can insert it. Now if we hop back over to our change log over here and start refreshing you can see ta it did the full process end to end which is awesome. Like that's this is what the skill is supposed to do.

So, here's what we're going to do now that we're finally at the end. We're going to turn on plan mode. So, you can do that by hitting shift tab a few times. And we're going to say, "Great. So, when going through this skill, you made numerous errors and did not follow the to-do list in the appropriate order and you just missed out on steps. What questions do you have for me so that you can properly update the skill going forward so that it works properly next time?" And at this point, we can probably also just call out the exact skill. So this is the skills right here. Boom. So now we can kick it off. And this will start to iterate and once again take us through the plan cycle. And this is the like you will do this. Every time it makes a mistake, you will go through this. And after like two to three attempts, this thing is going to work like an absolute charm. So when you're, you know, future you is trying to do this, you just say, "Hey, I'm trying to make a change." And just boom, boom, boom, boom, it's all done.

So, as you can see, it's starting to ask a few questions. And let's see. So, first off, it did clearly articulate what it did wrong, which is great because it's going to make sure it doesn't do that in the future. Let's actually make this full screen so you guys can see it a little bit easier. So, we'll go down to the bottom and zoom in. So, let's see what errors did it make. So, it properly it tried to do things in the wrong order. Yeah, it basically just it followed a ton of things that were in the wrong order. So, when it comes to step two, analyze changes. The skill says to use GitHub releases to view the latest tag to get the most recent release. But in this case, there was no GitHub release yet, only the tags. So when it comes to step two, that is not true. There were GitHub releases. I can actively see them right now whenever I hop over to GitHub. So I think you might have just been using the command wrong, but the GitHub releases and tags were both there. And obviously it was your job to compare the changes we made against that latest tag/release. Next, when it comes to step five, you know, the skill says to show a draft review before making the changes, but you corrected mem. I'm not sure what you're asking on step five, to be honest. Yeah, just basically simplify your questions and you should tell me what you're planning on changing on updating the skills so I can review it before you actually make the change. Now, this is just a quick way to like, you know, just have the AI to like cut things down to be a little bit more simpler for us just because sometimes it gets a little in the weeds and doesn't clearly articulate what it's trying to do. So, we'll give this a second.

Okay, so here's the final quick things is it's saying all right, so when it comes to step two, it's going to fix the GitHub releases. It's saying it used the wrong command. So that is totally fine. So we're going to say, yep, your step two fixes. Sounds great. Then when it comes to step five, think we're on the same page. Use your best judgment based on what I've said so far because I usually do like to view the draft. But this was an odd edge case just because you were going out of order. So just use your best judgment based on the correct approach that we ended up implementing towards the end. Outside of that, I'm pretty happy with the changes that you're suggesting. So now we'll just kick it off. And at this point, what should start to happen is Claude code is going to go, okay, fantastic. So I'm going to use best judgment. And you can always just say use best judgment because this is iterative. like you don't it doesn't have to be perfect today cuz when you do it again next time we're going to make it better again. So this is this is the hard part about refining a broken process. So here is the updated change. So here's what it's going to do. So it's going to add a new requirements so that it fully understands all the changes that were made. From there it's going to figure out the GitHub release. Then it's skipping some steps. I'm just going to say looks great. Proceed. Then we're going to kick that off. And now what's going to happen is we actually hop over to our skill. What we should start to see here in just a second is it's going to make all of those necessary changes to where it's going to update the steps to appropriately reflect what we wanted it to do. So, we're going to say yes. And from that point forward, we are setting ourselves up to, you know, do this one over and over and over again. And finally, definitely not going to like we don't have to show this one. But at this point, you guys are set up for success. So, if after going through iterate and it works perfect, boom, you guys can start to share it with other developers on your team so that they can move faster. If you have employees who are under you, you can share it with them so that they can, you know, start to work and perform tasks at your level and to your liking or if you are, you know, hiring new people to help you out. Boom. You already have working standard operating procedures that you can just say, "Hey, go do this for me next time." Or if you want to be a one agent man army, hey, no need to roll it out. you can just crank out with all the skills, all the work yourself using AI because you're going to move 10 times faster by automating stuff that takes time that you shouldn't even have to do in the first place.

So yeah, so that is a quick wrap on getting your skills iterated and getting yourself and your cloud code skills working like a charm. So that's a wrap for this video guys. I hope you enjoyed seeing how you can create your own claude code skills using the master framework. If you have any questions, feel free to drop a comment down below or you can hop over to the free school community that I've created for you guys. We have weekly coaching calls every Tuesday and would love to see you on there. It's completely for free. Also, if you are looking to bring your own AI projects to life in days instead of months, you're definitely going to want to check out shipkit.ai. It's a project that I've created for you guys where we include a bunch of pre-built AI projects. We have courses to help you level up as an AI developer. Basically, if you're looking to create real world AI applications in 2025, it is the project for you. But enough of that. Without further ado, I have a ton of other AI related content right here on this channel. So, I definitely recommend checking out whichever videos are popping up right now on the screen. And with that out of the way, I cannot wait to see you guys in the next video. Hope you have a great day. See you in the next one.