0:00 Welcome to the definitive course on Claude skills. I built over 450 skills over the last year or so for my own business for the 3,000 plus students I've taught to use AI Live and for businesses I've consulted with. I pretty much doing this stuff all day every day. And essentially what I'm going to do is I'm going to transfer everything that I've learned and known through practice, research, and iteration and actually building these things. I'm going to transfer all of that knowledge to you today. So, even if you've never touched a claude skill in your life and you're just starting to use this stuff, we're going to go from these beginner routes all the way up to advanced level. So, I'm going to walk you through the route to get there so you can build full teams of AI skills running in your business to act like little employees working together. And no fluff, here's exactly what we're going to learn today. I'm
0:46 going to take you through the plan of what we're actually building so you can build along with me knowing the end result and the order we're going in. We're going to talk about what a claude skill actually is. Even if you sort of know what they are, going to help you conceptually explain them to other people or if you have no idea what they are. We're going to go through the basic definition and how it differs from what you already use with AI. We're going to talk about the anatomy of a skill and the different parts, how they all stack together to create a high performing file that runs inside Claude. We're going to talk about the map, the two types of skills, and the five different levels that again, you can mix and match and pair these together and understand where you are in the build process and how you're constructing your skills. [snorts] Once we've done that, we're going to actually start building and you can build along with me. We're going to
1:32 talk about the three key ways to build a Claude skill, and we're going to build together step by step. 1 2 3. Then we're going to talk about what separates a good skill or an average skill from a really great one that is high performing in your business. We're going to talk then about managing the skills, keeping them up to date, sharing them, using them maybe across multiple people. Then we're going to start to actually build into the levels. So we're going to start to connect skills to the outside world and outside tools so they can execute more work for you. We're going to learn how to link them together to create chains of them that run in teams. And we're even going to stitch them all together with an orchestrator, a manager that can manage these, like a line manager for you that checks in and out of your team of skills before showing
2:19 you how to package these all up in a neat plug-in. talk about what those are and then finish off with some other ideas about what you should be building, finding pain points in your business. And I'm even going to give you 101 of my best skill ideas for you to walk away with and build for yourself. So, in this course, we're actually going to be building a full team of SEO [clears throat] claude skills. These are going to act as a whole SEO department for your work and we're going to build everything live step by step. You can build along with me as well. If you don't want to do SEO or you think this might not be that useful for you, I still recommend watching and building along with me because this is really about building the skills and the knowledge for you to go and implement on your own things. So, there's two ways you could really watch this. You could actually build along with me or you could just watch and learn along the way, take notes for you
3:06 to then impart on your own ideas of what you want to build for your business, work, life, whatever it is. And while we go along, as well, this guide has all the prompts and steps that I'm using in this video. You can grab it below. There's a link in the description where you can get it for free. It's inside something called my AI recipe vault. If you go and grab the link below and sign up, uh you can get the complete Claude skills course guide. This is going to be your walk along with all the prompts, etc. in here for you to use along the way. So, there's some super powerful stuff in here if you want to build along with me. So, what is a Claude skill? Well, a skill is simply a set of instruction Claude keeps to run on command. It's essentially giving it a stepbystep checklist or a standard operating procedure that tells Claude
3:51 exactly how to complete a task. So you can write down a process once, give it to Claude in something called a skill.md file and then anytime you want Claude or any other AI platform to actually run that process or that task, it can go and look at the step-by-step instructions and follow that process. And the benefit of that is you give you the same output every time. So every time you ask Claude to do something, it follows the process to give you the same output. You can think of it as simply as a prompt. If you've got a really great prompt that gives you a great output every time, that's essentially a set of instructions of how to complete a task. You can take that prompt and build it into a skill. That's the most basic form, but you can go a lot deeper and wrap a lot of stuff up in these skills to complete a task.
4:37 And the beauty of this is you might think, well, I've already got prompts. Why do I need to build them into skills? Well, every time you want to re find a prompt or complete some work, you have to go find that prompt, copy and paste it in, actually repeat the process, and it's not that smooth in a very basic level. A skill actually can fire for itself. You can either tell it to run that process or it will naturally run these processes based off other work it was doing. So, a lot of the time you might build these standard operating procedures into your clawed workflows and you might not even know that the skill is running. It just might run automatically. For example, I might be asking about the performance of a particular website and it might just go and run my SEO audit skill automatically without me even thinking consciously to go and run that process. Equally, we're
5:22 going to build something called like a voice copywriting skill which applies which teaches it how to write in your tone of voice for your brand. Every time you output a document, for example, it might run that voice copywriting skill to make sure it's actually applying your tone of voice, your anti-AII writing rules to your output. Whether that's proposals for clients, LinkedIn posts, or general documentation around your business, work, life, whatever it might be, even just replying to an email. So, a skill is a set of instructions step by step that tells Claude how to complete a task wrapped up and ready to run that you can either run on command. You can just tell it to go and do that job and it will do it. Or it might just start doing it automatically because you roughly asked and described the job that you're doing in the rest of your work.
6:08 Next, I want to talk about what actually makes up a claude skill. The anatomy of a claude skill. It can be as simple as one file with just a wrapped up prompt, but it can also be a folder of information with several different files in there. And I want to break this anatomy down for you here. So, I'm actually going to show you in my clawed environment here an example of a skill. And you're going to start to see how we can start using these. To find a Claude skill, you go to customize on the left here, and then you'll see it will automatically jump into skills. It will give you a discover tab for others that are already built into Claude, or you can click yours, and it might show you some that you already have if you've been building them. You can actually scroll down and you can click on one and it will bring you into the actual the back end of the skill, what these
6:53 actually look like. Now, there are three main parts to a Claude skill. Number one, you'll get the overview, the description. This is exactly what it sounds. It literally describes what's going on here. So, this is an SEO blog writer skill. One of the ones we're going to be building today. It converts my YouTube transcripts, voice notes, rough content ideas into SEO optimized, publish ready blog posts in my voice. So, we already know what it's going to do. We also have built in here the trigger words. And these are really important in the description because you can see here any time that I would say something like turn this into a blog post, use the SEO content writer, turn this into an SEO article, write up write this up into a blog post from a transcript, Claude is just going to automatically do that job. I don't have
7:39 to specially invoke the skill with slash commands necessarily. I can just sort of be working naturally and say, "Hey, can you do this job?" Or anytime it realiz it should be doing that job, it's going to fire this skill. So, it's always reading this description about what the skill does and how to actually invoke the skill and make it run. Now, we also have this button here, the contents. And the contents is made up from two key areas. You have the skill.md. And a skill.md is literally just a markdown file. It's a text file of the instructions here. So, you can see on my screen here, we've then got the SEO blog writer. These are the instructions that it has to follow. It's got a five-stage step-by-step workflow. It goes through stage one. It goes through stage two of
8:26 drafting the article. Um, it has the article structure of exactly how I want these set up, the voice rules, what it shouldn't do. Has stage three wrap the SEO in production package. Again, how to wrap it all up. It does a voice audit to make sure it's running against my voice that I've trained Claude on on that skill. And it's also got an editorial part to act as a grumpy, grouchy newspaper editor without asking permission. And then it's got the output. So essentially what you can see here is that this is just a really long set of instructions. It's essentially a super duper detailed prompt. Now if I was to try to just use this prompt on its own, it would do an okay job, but it would it's quite a lot to just run in one prompt in Lord Chat. So what we actually package with it as well are
9:12 reference files. You don't necessarily have to have reference files for all of these, but these are a little extra knowledge base for it to run off. And generally, these are things like templates, checklists, other things it needs to refer to to actually get the job done. There might be a little bit of knowledge in there. Or it can actually have things like Python scripts that can actually run automations for you in here as well. But in this one, for example, what I love to package in my Claude skills are an article template. So it's got a template of how it should use for everything. It's got the metadata block, for example. It's got an internal links block, external links block, how the H2 body section should be formatted, how the headers should be formatted, so that all my articles essentially come out in
9:57 this uniform templated manner based on all the other prompts that I've got in there to get the quality up. We also have a checklist at the end because I think I had a review process at the end and it says, "Hey, can you go and review this as a grumpy editor?" This is the checklist that it would actually go through to rate itself against. So it can actually rank and rate its work against this secondary checklist which is which is its review cycle. So inside a claude skill we have three main areas. We have the description which I showed you at the start which is what the skill does and how it's triggered like how it should run. We have the skill.md file. This is the main set of instructions as to how the job gets done and what and in what order. And then we have the reference files. These are not fetched every stage of the process, but fetched
10:43 when a claude skill needs to do something specific like the final template, a review checklist, any extra supporting information. Um, and these reference files can also have scripts, essentially code that a skill can call when running the same process over and over again. You don't need to worry about too much these at the start of this, but it's worth knowing that, you know, you can have code written into these reference files as well. Now, sometimes people get a little confused about what should go in a skill.md file, the main instructions, and what goes in a reference file. For me, I like to think it as the decisions it needs to make along the way. So, you can think about how the jobs run, what order things happen in, what good light looks like, what it must never do. Essentially, as we said at the start,
11:29 the step-by-step instructions about how to complete a job. What goes in the reference files are the supporting detail. So, you can think of these as the long worked examples, a full style guide. So, you could say in the main skill.md file, hey, make sure to apply my full style guide at this stage. And then you could keep the full style guide in a reference file for it to go and reference at that point in the job along with any schemas, formatting, what good looks like, what bad looks like, those long worked examples are some great examples to go in the detail versus the main skill.md file. Often Claude when you're building these will know and it will understand where to put each thing. But I think it's important for you to understand this when you're building as
12:16 well so you can delineate between the two options that you have when you're building. Now the final area to talk about here is the description and the triggers. The triggers are really important because you want to make sure it's actually firing at the right time. So you could actually write into your description area at the top of a skill here. remember you could actually not have any trigger phrases in there and just make sure that it's only triggered when you actually command and select that skill to run versus actually making sure these trigger phrases are actually quite quite right. So in here I've got an example write up this transcript here is quite vague like I would probably actually looking at this remove this trigger phrase because I do a lot of things with transcripts and I don't want it to accidentally fire a skill. So, if you find your skills are accidentally
13:01 firing through the way you're working, then you might want to check your trigger phrases um to make sure they're actually aligned. Equally, I have multiple brands. So, I actually sometimes name the trigger phrases to actually make sure it's running for halfday Thursdays, which is my AI community that helps people gain more freedom of time in their life versus elliotprince.com, which is my personal brand. So, if I'm building similar skills between the two, I have to delineate in my description of like, hey, can you do this in the Halfday Thursday's brand, not the Elliot Prince brand, for example, in my trigger phrases. So, really important here, try to actually make sure it's putting trigger phrases in there that are words and ways of phrasing things you actually use naturally in your work, not that you kind of have to keep remembering and thinking like, how how do I invoke that skill again? And don't worry, of course,
13:48 we're going to get into all of this how they actually run in this course. So, let's get into how you actually use a Claude skill because if we're trying to build before we even know how to use them, then we're going to be in trouble. I showed you that SEO content writing skill and I talked about trigger phrases and how to run them on command. I want to show you exactly what that means. So, there's two ways you can actually run a Claude skill in your account. The first thing is you can hit forward slash here and then it's going to give you a list of all the skills that are available to you. Now you can scroll down this list and actually start running them or you can type in here and you can say SEO blog writer and it will come up with there you'll see the little description of how it runs and you can click that and it's going to highlight blue and you'll see now that Claude skill is
14:34 actually ready to run. It's 100% selected in my account. So if you want to make sure that you're definitely 100% using a clawed skill, you use that slash command. It's called a slash command method. Now the other way we can run this is by actually saying those trigger phrases. So when I was talking about trigger phrases in my in the previous section we actually were talking about just using them conversationally in your workflow. So I would say something like hey I want to turn this YouTube video transcript into an SEO blog post. Here's the transcript. Let's get to work. And because I used one of those trigger phrases, turn this YouTube transcript into an SEO blog post. Now, this wasn't an exact same phrase that I had in the main description there, but it's really close. It's close enough that Claude's
15:20 going to be like, "Okay, I know I need to use that skill." If I go and grab a YouTube transcript and we will paste it into the chat here, it'll come up pasted there and then run this, you will see that it actually starts to get to work. I haven't had to put a massive long prompt in there. The skill should invoke automatically because I've used the right phrasing. Here you go. You can see it in the description of what's going on. turning the video into an SEO, reading the SEO blog writer skill to convert this into this. So, you can see it's loaded the skill there and it's starting to use it. It's using the template structure, the SEO checklist, uh all of the reference files in there as well to actually complete this job for me. And that's the beauty of a skill. Once you have a defined process, you can just start using them conversationally in your work and it'll automatically then be able to run that
16:06 task. We could also set this to run on a schedule to go and scrape my recent YouTube video transcripts, bring them into Claude and actually create the blog post and then connect up to my website and go and publish that blog post for me automatically as well. We can teach you that whole process, which is kind of what we're going through in this video in many ways. Now, before we build as well, we'll just talk through some of the customize area of Claude because I want to show you how to actually add a skill to your account because some people might send you skills or you might download them or you might come across the files in many different ways because you can just share these files between people as well. So, if you've been sent a skill file, they're actually really easy to upload to your account. All you have to do is go to customize on the left hand side here. You go to add
16:53 and then you go to upload skill. You'll also see there's buttons like create a skill, create with Claude, record your screen, which we're going to go into. But to add one that someone sent to you, you click upload skill. You go and find a skill for example, add that to your account, click save, and then it's going to be ready to run in your environment. You can upload that, and you can invoke it just as I described before. You can see this one is actually got a whole bunch of reference files in there along with the description as well. And as well, you can turn them off and on. You can turn them off and on on the top right here. If you have them in your account, but you're like, "Ah, they just fire when I actually am not working on that sort of project." You can turn them off. You can also download them. So, you can go and download any skill in your account and send them to other people, as well as the ability to edit the text.
17:40 So, if you just want to actually change a trigger phrase or a description, you could just go in and type it yourself really easily. Or you could actually click the top three buttons here to click replace in chat or even remove it, delete it from your account, for example. So, it's very easy to sort of manipulate, move, upload these skills in your account. And now you understand the sort of basic fundamentals of a skill, how to add them, and download them from accounts. we can get on to actually going deeper into the types and levels of skills that you can build. Now, when you're building Claude skills, it's really important to understand that there's actually two types of skills that generally we're building and they deliver slightly different outputs. The first one I like to call deterministic skills and these always give the same
18:26 outcome. You know exactly how you want the process to finish. It might be something like an audit, an invoice, a monthly report. It's essentially comes out as a templated output. When you think about the SEO blog writer skill, I was giving it a messy input and it was giving me a templated output essentially for that skill. Now, I also have them for an invoice. An invoice, for example, I always know it needs to come out in the exact same PDF format with the headers in the same place. Really, we're just switching in and out the numbers and some of the details in there. So, it's actually really easy when you're building a deterministic skill. It's really easy to understand when they fail because it's pretty loud. If my invoice comes out and it's not in the right template or there's a column missing, there's the numbers are wrong. It's
19:12 really obvious to see and it's pretty easy to fix. These are generally the easiest types of skills to build because you say, well, this is the output I want in this template. Here's the type of information going in. And you're essentially just taking that information and placing it in that template for me. fairly simple but super super powerful in getting rid of some of the busy work in your life. Now on the other side we have what I call non-deterministic skills and these are skills that more carry like judgment, taste, voice and there's no set output that comes every time. So again you might think of this as like your your voice skill, your brand voice skill. It can be applied across many different situations and there's a lot of decisions and judgment and taste sort of building in there. thinking patterns, for example, that
19:57 need to be taken into account. And these can be harder to build because it's not as obvious when they go wrong or they require a lot more back and forth and training. Um, they could be thinking frameworks, second opinions on certain frameworks as well. It's really hard to actually define the perfect output because every output is slightly differently. And that means they they fail kind of quietly. They might look good the first time you've actually built them and run them, but there's no clear flag of an error or where they're slightly drifting. They have to take you have to keep a little closer eye on them and they tend to take longer to build and train and get into the place you want. So [snorts] really, when you're building, think about it in these two ways. Is this a deterministic skill where I just want this clear, defined output, or is this non-deterministic
20:42 which actually is applying more judgment, taste, and voice across others? They're sort of more non-deterministic ones. You can think of them more of like they're an art rather than a science. There's no perfect formula or perfect output. They always require tweaking and upgrading versus deterministic which is almost a mathematical or an equation or a scientific output that comes out the exact same time, the exact same way every time. Now, we also have five levels. So, we've just gone through the two different types. There's also five different levels of Claude skills which really comes down to complexity. We have level one which is essentially it's essentially a very simple wrapped prompt. We can just trigger it. It doesn't really require any extra input.
21:27 It doesn't require anything from us. It doesn't connect. It just hey can you just do this for me and it will run automatically. Very very simple. The level two is an ask. So it actually needs some more information for you. So, yes, it can run really easily, but you might need to give it more input. You think about the SEO blog writer skill. It's an ask because it needs a transcript. It's going to ask me some questions to actually get towards the final result rather than maybe an SEO audit skill where I can just give it, hey, can you audit this website? And it will go and run the audit. The level two needs more input, more of my human insight and expertise built in before it can actually deliver its great output. These are the two levels that most people sit at because it's very similar
22:13 to writing a normal prompt ready to go in Claude. It's what we've probably all got used to using AI and chat GPT from day one. Then we move into level three. Now level three is where we connect. We can actually allow this skill to reach outside of Claude using something called connectors. Now there could be some very simple explanations of this. You could I have a great one. It's a morning inbox triage skill and it's able to go into my Gmail. It knows I've trained it step by step what to do in there, where to file certain things away, what to do. It doesn't reply to my emails for me, but it could actually reply to emails for me if I wanted it to. This is more of a process of getting rid of all of the junk and stuff that doesn't need my attention and leading the key stuff that doesn't need my attention. So we can
23:00 actually train the skills to go and reach into our SAS tools either to retrieve information. So again, we can upgrade our level two skill by linking it to the YouTube API or connector, getting it to go and automatically fetch my transcripts and actually write the blog post. When we do that, when we connect it up to either go into a tool and pull information or push information to a tool, you could push information to we could push that blog post into a website or we could push it into notion or a word document for me to go and check and and make the final edits or it can actually drive a software tool as well. So, it can actually click around and use these tools for us as well. That becomes a level three, a connect skill. And again, I find people get to this level pretty quickly once they realize that
23:46 they can build in the right links and connections in their skills. Where people don't often make the jump to is level four. And this is a chain or a team of skills working together. So you can actually build your skills to hand off one to another either one 2 3 4 or to make decisions and go left and right etc. Now, when we build today, what we're looking at is a set of SEO skills. And there's a very defined process that an SEO department or an SEO specialist would actually go through step by step. They would run an audit. Then they would write an SEO strategy. Then they would start creating the website content. Then they might actually publish that SEO content. And then they might write a report and a commentary analyze and then
24:31 almost repeat that process. So we could actually get the skills to go one for another. I could get my SEO audit skill to then feed into my strategy skill to give it the information the strategy needs to write the strategy. Once we have the strategy, the strategy skill can then feed into the creator skill that's going to go and write a blog post, for example. And then you can get what are called chains of skills and it becomes like a complete production line that becomes a full workflow, not just one skill individually. And then we get on to level five, which is the highest level. And this is where you build a manager skill, an orchestrator skill, which when you have lots of skills chained together or or lots of them that can interact, you can build an orchestrator and manager that sits on top that knows what each skill does and
25:17 how they interact with each other and can actually you could give the the the orchestrator, the manager, the line manager of these skills, my overall goal or like say, "Hey, my website traffic is uh has gone down. Can you help?" and it would know which of its teams to go and invoke and how they link together and start working which is where we get this sort of team this whole department of little AI employees running in our business. So we so we will be getting to this level by the end of this course as well which is super exciting but I think it's important to say that there are two types of skill and they can be across all levels but the levels aren't a scoreboard. A level one skill isn't worse than a level five skill. They're just different. You can get a lot of value out of a level one skill that just runs automatically every morning for your business. Even more so than often a
26:03 level five skill, which a lot of people overengineer and they do it to prove that they built this whisbang amazing system. But it's actually not that well constructed and it's just over complicated compared to if I just had one of those skills running perfectly would be more leverage for certain areas of my work and life. So just know that just because you can build a level five management team of skills, it doesn't mean you actually have to. It's all about the pain and problem you're trying to solve in your business or your work. That's the real goal here to free up more time in your life and more wealth in your life. I mean, if AI is not doing that, then it's not really working for you. So, a level one skill that frees up two hours a week of work for me is a massive, massive win versus a level five skill that kind of needs a lot of troubleshooting and managing just to get
26:49 it working. So, with that all said, let's jump into actually building our first skill. We're going to go through the three main ways of building skills. And we're going to start with the easiest and simplest one, and this is taking a prompt and wrapping it up and making it fire on command. And before we actually start working in this window, we need to just go into our settings here. We need to go into customize and we need to go to skills in the middle here because we need to actually make sure something is turned on in our account. We need to go into the search skills and plugins. And I want you to type in creator because when creator comes up, you should have this skill in your account. Skill-creator by Anthropic. This should be on by default. If it's not on, go and find it and switch it on because this is actually a
27:35 skill inside Claude that's going to help build for you. It's designed by Anthropic and Claude to know the exact processes and requirements. You can see there's a ton of information in here. You can see all the reference files, the scripts, the graders, the analyzers, and just the general process of what it should actually do and some of the requirements such as here the three-level loading system, the data, the skill.md, and the bundled resources that we broke down earlier. This skill knows all of that and we can use this to actually build all of our and package all of our skills forever more. I'm not going to sit here and teach you how to type this by hand because it's much quicker just to use AI to help you build. So the first thing we need to do is we need to make sure our skill creator is turned on. Then we need to invoke it in our chat window here. So to
28:20 do that you can hit forward slash and we can type in creator skill creator. Select it and we know it's definitely turned on. Equally, you could just say naturally in your conversation, and I'm going to use an app called Whisper Flow here that allows you to like talk and dictate to your computer, can we build a claude skill? And just by saying build a claude skill or create a clawed skill, it's actually going to make sure that that's invoked. So, you could do it either way. For the purposes of this, I'm just going to use this this forward slash command here to get things working just so for an illustration. But often, I'll just say, hey, I want to build a cloud skill. and it will invoke the builder skill for us. Next, I'm going to tell it exactly what we want to do. So, I'm going to say, I'm going to give you a prompt for an
29:06 SEO audit skill. I want you to take this prompt and wrap it in a claude skill so that I can fire it and invoke it at any time. Very simple. I'm literally just telling it what I actually want it to do or how I want it to build this skill straight from off. Then, in this document that I've linked below, you're going to find this way one wrapper prompt. You're going to scroll down. It's going to say, "Hey, make sure the skill creator is turned on. Take this prompt and turn it into a clawed skill I can run on command. Everything's here for you to run." Then I'm going to copy this prompt. You can do the same if you want to build along with me. But you can see it's a really detailed prompt that I used to copy and paste for audits that actually now instead of me having to go and find this prompt and edit it, etc., we're going to build it into a claude skill. But essentially, you can see it's giving it a role at the top here. It's
29:52 got some input. It just needs the URL that I want it to run on. It's got the different steps that I want it to go through and then the output, the exact output that I want it to give in detail. So, I'm going to take this prompt. I'm going to copy it and I'm going to I'm literally just going to paste it into the Claude chat here. And then that's all we need to do. We can send that message. And Claude is going to go and analyze everything and build our skill for us. You'll be able to see in the chat here that it's understood the task it's got to go through. It's going to go and view that file. It's actually reading the skill creator skill here that we discussed. So, because we've invoked it here, it's reading the skill creator skill before building and working out the process that it needs to follow. So, what we're building here is a very simple, it's a level one skill. It's a pretty deterministic skill
30:37 because level one, I'm just going to give it a URL and nothing more. No extra expertise and it's going to go and do the job and give me a formatted output. So, it's a level one deterministic skill for us. Very simple. And this is the easiest way to build them if you've got prompts that you're like, I really like using this. This is dialed. I just want this ready to use on command in my Claude account. Now, it might take a few minutes to actually complete this build. But what we get out the other end is on the right hand side here, we get the breakdown of our skill and on the left, it's got a little bit more information about what it's done. It's called it the search visibility audit skill. I mean, we could have actually given it more direction of what we wanted it to call the skill, but I quite like that it's got the description and how it actually
31:22 runs. It's actually looked in my account and made sure it's not triggering in the same process as some of the other skills I have running in there already, which is really useful. You can actually go and turn other skills off or delete them if you if you are running into old skills that you don't want to run. It didn't run a live test audit against the skill, so it might be important. Sometimes Claude will actually run an all a test of its skill for itself, but I prefer to actually test a lot of these personally so I can actually see the output and give it feedback. So once we have our skill built, we can also see on the right hand side we can break down. Remember we had the anatomy here. So we've got the description and we can look at kind of explains it what it does. Uh that this one is doing our SEO and GEO audit here. You can see the trigger phrases it's got running and how
32:08 it would run. And then we've got the step by step about how it actually works and importantly the output that it's going to deliver. Now this one doesn't actually have any reference files. It didn't need any reference files particularly. We could certainly give it more templated structures, but because it was just a prompt that we really liked and ran, we didn't really need any upgrades. We didn't need any extra reference files for it to run. So in this case, we only have a description and the main sort of skill instructions running. Now, all we have to do is click save skill here. And you'll see it's saved that skill and it's ready to go. If we go back into our skills area, you can click manage in the bottom right or remember to access your skills. You can go to customize and then click yours. We could search for the skill.
32:53 Uh what was it called? The SEO visibility. And you'll see when you search for it here uh in your customize area, we'll see it come up and we can go and investigate and use and check the skill as it was built, which is pretty cool. It's ready to run in our account, which means now we can test out our first skill build. We could actually say, "Hey, can you run an SEO audit on my website?" https colon/corchavevel.vip. This is a website that I own that I that's actually not typed out at all. Right. I'm just going to type it out here. And you'll see I could invoke it. Remember, I could go forward slash uh search visibility audit. actually because I've just said, "Hey, can you run an SEO audit?" It should automatically invoke this skill for me as well because that was one of the
33:40 trigger phrases there. So, what we can see straight away is we've built a skill there by wrapping up a prompt. And then you can see we're now testing it and it's invoking it. You can see in the middle of my screen here reading the search visibility audit skill for this audit. And it's going to go and do that job. Okay. And here we go. the the skill is has finished working for us and it's created our search visibility audit on the right hand side here in this nice artifact. It's given us our nice verdict. It's given actually a review of what it actually did and the score in there as well where what's broken and what needs actually fixing on there. We've got things like GEO uh every page reads written by admin rather than having the actual author on there which was probably me a lot of the time when I
34:25 built this site. We've got our breakdown of different findings in this nice format and the evidence and citations. What else have we got? We've got the fix list. So, we've got literally step-by-step things that we actually need to go and fix. The long game of the strategy that we could be employing and some new pages worth upgrading and building. We've got 15 pages worth actually upgrading and building here. And some tips that need to be rewritten to get cited by AI search. So, this is incredible. This is super in-depth. This is a super valuable report. Anyway, like previously without a clawed skill like this, this might have taken me days to actually compile a full in-depth report like this. Now, because I've built my expertise into a prompt there to be fair, now we can just run the skill on any site. So I can either work more
35:11 quickly to do a lot more work for a lot less time or instead of spending a whole day on or two days on an audit, I can just run the skill, have that job done, get on to some other higher value work or go walk my dog on the beach, hand over work to AI. Or I could even say, hey, I'm going to do this at low cost or free or just as a value ad for a client now because it honestly cost me nothing to actually run this really except for a few few tokens. And it was now super easy to just wrap that skill up in that prompt and I can run it anytime I want automatically. Even set it on a schedule to audit a website once a month or every 6 months. So we've built our first level one skill here just by taking a prompt, giving it to the skill creator skill and wrapping it up. And now it pretty much just runs by me giving it a URL and it
35:57 just does the job for me automatically. Next, we're going to build our a level two skill actually, which is going to need extra expertise and input from us when invoking the skill. But also, we're going to talk about probably the best way to build skills. Now, the first one worked really well because I had a prompt that I'd spent a long-term dialing on my expertise rather than just going in and saying, "Hey, I can we can you just build me a prompt that does an SEO audit." It wouldn't have much direction and it wouldn't be very good. And when we talk about building the best skills, they're actually built in with either your expertise or some that you borrowed from other places. These are the best skills. You either want to bake in your expertise and processes of how you do things. Or if you're not actually sure, research and use other experts in
36:42 the industry. But these actually take a little bit more time to build than just saying, "Hey, wrap this up." We actually need to run this basically from a cold start as if we have nothing. And to do that, we're going to actually run the process. We're going to teach Claude the workflow that we want it to actually execute for us and then capture that process. Okay? And we're going to rebuild the SEO blog writer skill that can write SEO optimized content based on information you give it, not just making random things up and scraping other information from the website. It's actually going to be a useful SEO blog writer skill, which I don't really see many other people building them this way with content that actually ranks and isn't just 100% AI generated, if you will. Now, we're starting from a cold start. So, I'm not going to start by
37:29 saying, "Hey, can we build a skill?" I'm actually going to work through the process. So, I'm just going to start by actually giving Claude my pain and my pain point and then the outcome of what we're trying to actually do here. And I am going to voice these prompts in but I will put these prompts that you can then go copy paste in the uh main notes with this section as well. Uh so I want to take a transcript that I have and I want to turn it into a an SEO optimized blog post. Before we get to work, I actually want to research and create the process or the exact way that we should approach this task before going and executing it to make sure
38:14 we're delivering SEO content and GEO content to the highest possible best practice standards. Now, I have a couple of ways that we're going to do this. one, I want you to research the best practices at the moment, the most up-to-date information and build me the the SEO GEO blog writing standard operating procedure, the system that we can always follow and the template we should be following. I'm also going to give you a transcript of my expertise and some of the extra ways that I like to build my SEO articles, basically. So you can see here I've just voiced in what I want to do, the goal I'm trying to achieve. I don't want to build the skill or actually do the job yet. I just want to research and outline the best
39:00 practice and the process we should be following. And I'm also going to because this is quite a long task and it's multi-step and there's a lot of research. I'm actually going to invoke co-work mode. And co-work mode is great because it allows Claude to actually take multiple steps and actually complete tasks for you. And because I've got things like, hey, I want you to go off and research and create a report. It's great to do this in co-work mode rather than chat, which will just do one for one. It'll you'll ask it a question, it'll give you an answer back, maybe do a bit of research. Co-work's actually going to be able to do longer running tasks in our build here. So, that's a useful switch to turn on as well. So, the process here, and this is why this is one of the best ways to build a skill, is we're starting cold and we're building in expertise. We're not just letting Claude dictate what it should do. We're actually working back and forth to train it on
39:45 our process. So it might ask you a few questions in here in the primary goal of the SOP is it for a particular website or is it platform agnostic? I just want it works for any site. I just want it to work for any site in this scenario. How how should the SOP be delivered? It might offer you to actually deliver them in a clawed skill. I don't want to do this yet because I don't want to lock the process into a claude skill until we've actually finished it and got it really defined because we will be working back and forth to actually make some improvements. So, I'm just going to say markdown SOP and template dogs so that we're not locking anything in uh locking anything too aggressively before we've defined it. And it's just going to ask me for the transcript of my SEO process essentially because that's one of my my areas of expertise. So, I might
40:31 as well use that. The more you can use your own expertise and processes here, the better. So, I've actually pasted in here two transcripts from previous content I made, which is how to write content so good you can't stop ranking number one and where to put keywords on your website from I used to talk a lot about SEO on my my YouTube channel. But honestly, I would also go into potentially some of my frameworks and SOPs I have saved in my Google Drive, any templates I want it to follow, or even if you don't have transcripts, you could just brain dump your process. Even ask Lord say, "Hey, can you quiz me? Can you ask me five questions about my process so that I can give you my brain dump on the process? But essentially, because I've got the transcripts already on my processes, some from some of my teaching, I might as well use those because that is stuff based on, you
41:17 know, proven tactics and ways of working that actually run versus just trusting what other people are saying. Now, I'm also going to get it to go and research some of the most up-to-date practices because some of those videos are like a year old or two years old now. So, I just want to make sure that some of this stuff is the most upto-date based on the most recent updates, research, expertise out there. You'll see on the right hand side here as well, one of the beauties of using Claude Co-work is it can track its progress. You can see that it's actually mapped out all the steps it needs to do. Whereas Claude chat would just give you an answer straight back and it would just do one little bit of work at a time. Um, Co's mapped out exactly what it's going to do. It's going to research the best practices. It's going to research GEO best practices. It's going
42:02 to research my articles and fold in my expertise layer to then write the SOP, the templates, the QA checklist that you did see in the the skill that I showed you originally that we're rebuilding here. And then it's going to verify everything. So, this is a serious amount of work that it's actually going to cruise through for us here to build out this process. And I should say while we wait for this to build as well, the skills you build here, when you build a Claude skill in any area of the application, I'm in the desktop app here, but anywhere you build a Claude skill, if you build it in Claude chat or Claude co-work, they are available across both and then they will actually work in Claude code. On the other hand, if you are building skills in Claude code, they won't translate back into your Claude co-work and Claude chat area. So often I like to build include co-work because I know then I can build
42:48 them everywhere and use them everywhere wherever I'm working. Um particularly for this knowledge works type of stuff rather than the actual coding and building work type stuff. So you'll see here even though I'm in claude co-work mode if I go to that visibility uh search visibility audit skill that we built earlier it would be invocable in claude co-work even though we actually built it in claude chat. So you don't need to worry too much about where you're building this stuff because they translate across both platforms. It might take a little while cuz we've given it a lot of tasks to do here. You can see mine has worked its way through all of these tasks and given me some output about, you know, it's received both transcripts here. Took about 20 minutes to actually do this. You see, I've gone a little chilly today. So I put a little cardigan on, made myself a cup of tea,
43:33 and actually been reading a book. Sometimes instead of scrolling social media or getting distracted by other tasks, I'd like to go and read a book. I've been reading Simon Synynic start with Y at the moment while Claude does my work for me which is nice sometimes. Um it's given me the headline findings uh and three places evidence corrected by you in terms of it did some of the research and then it relied on some of my expertise before building out. It gave me some sources that it had checked as well so I could make sure these are actual uh good quality sources. I could have defined the sources if I knew them up front as well or some experts that I wanted to to to follow, but essentially we've got our SOP. Now, this thing is actually pretty beefy. Uh, stage one, stage two, stage three, like this thing goes super deep. Keyword fan out and
44:19 research. Um, you know, this thing is insane about how to go and create an SOP blog post. This I think this goes a bit deeper actually. Uh, sorry about the scrolling here. It's actually doing keyword research for us as well. Um, and then it's got like again you can start to see the building blocks of what we're building here. It's got the article template about how we want to template each article. The publish checklist to make sure this is actually done etc. And for some reason it produced them twice for me here. I don't know why it produced them twice but essentially it's built four documents here. The main SOP and then some supporting evidence and templates etc. Now, you might be tempted here to be like, great, turn this into a Claude skill, but I don't know this is actually working quite right for me yet. So, I would actually test this out. And
45:06 [snorts] to do that, I'm going to use this process now. So, I'm going to go and get a transcript, and we're going to run this process in this chat. I'm going to use this transcript from my one of my most recent videos, the three levels of clawed memory clearly explained. You could use all of this process or you could actually you could find your own brain dump some information into a transcript and give it it as well. However you want to generate a transcript is fine. I'm actually using a a quick tool here YouTube-transcript- io just because it's super quick to actually uh grab transcripts and it doesn't use any of we'll just grab that again. It doesn't use any tokens in claude [snorts] to actually go and do this. You can scrape transcripts with AI tools but this is also equally useful. So I'm just going to say now I'm going to say can we test out our SOP then I'm
45:53 going to give you a transcript about clawed memory and I want to write a blog post that targets the keyword claude memory. Okay. So you can see quick prompt here about let's test it out. This is the keyword I want to target and here's the transcript. And then we can run this and we're going to see how all of this performs before we actually lock this process into a skill. So we can see it's starting with stage one mine the transcript which is perfect. It's going to go and get all the information for me. Okay. So our SOP has finished running this processes. Give us some output on the left hand side here um about some of the processes it went through. It caught itself over optimizing on how often it had put memory in some of the headings and
46:40 stuff. Um and then it's got some information about confirmation it wants before it shipped. So, some gaps it couldn't fill cuz I didn't give it a site and and some other quotes that I put in there. It just wants to confirm everything. And then on the right hand side, we actually have the output. It's put it in a markdown file. I might actually say like, "Hey, I'd prefer you to output these in a uh a Word document for me so that I could edit them." Oh, and then we can actually see it. Right. One of the beautiful things here is it's actually pulled direct from my transcript. I like oranges that set in my claw memory for months. So, it's used my keyword early on. We've got the key keyword in the title. I'm not 100% convinced by the title. Like, I'd prefer it to be more on the nose for SEO purposes. What does Claude actually remember about you? We could probably
47:25 use AI memory earlier on in these subheadings. So, it's not got it 100% perfect, but like on the whole, this is a pretty well optimized, well-written article pulled from my transcript and bolstered by some research, which is cool. We got some comparison tables um and what it does and doesn't cover. Now, I think I'd like to actually just bake in a few more things. Like, I'm not going to lock this in a skill yet because I'm not 100% happy with the output it's actually given me. So, I'm going to give it some feedback about some edits I want to make to the uh to the article. Okay, I want to just make some edits to what's going on here. On the titles, I want them to be more on the nose. So, I would be something like clawed memory clearly explained brackets and how to set it up. Then I'd like to
48:13 make sure in our first H2 that we have the word the keyword claude memory in there. And I'd like it all to also to be like in the further down H2 I want it to be how to set up level one clawed memory on purpose. Um so we're just getting a bit more keyword rich in there. The other thing I think we've missed off the bottom is I'd like a short FAQ section that is the most commonly asked questions. the top five also asked questions about Clawude memory would be great to have in there with just like short one or two sentence answers to each questions. Um, can we make those edits to the article, please? So, I'm going to go and get it to actually do the next round of edits on this to get it a bit more closer to what I think is
48:59 a great SEO article versus it's just dropped the ball in a couple of places. So, we're working back and forth until we get the output exactly how we want it. Okay, brilliant. just gone and spent another minute or two just doing those edits. So, we can see clawed memory more in the title in the titles there. The title, the main H1, the header of the page is more on the money. And then on the right hand side here, we can see we've got the FAQs uh coming in nicely. So, I'm pretty happy with this now. I'm happy with this process. I'm like, great, our SOP and everything is been worked through and we've got everything in there. Now I'm happy that I've essentially trained Claude on how I want it to run this process each time. We're
49:44 going to lock this into a skill. Now there's two options here. You can go to the top of your screen if you're working in Claude co-work mode. Click this down button here and click turn into skill. And that's just going to put a prompt in the box on the bottom left here. Turn this into a skill. And we can learn from this that all we have to do is ask it to turn this into a skill. I'm going to give it a bit more direction. I want you to create an SEO blog writing skill that captures this process the way we've got the final output now and everything we've gone through in this task and chat. Uh it should be able to run on a transcript or a brain dump transcript, a video transcript or even any other long form inputs I give it. If the user hasn't given you any inputs or you need
50:30 more information, you should quiz the user before actually getting to work so that you have everything you need and quotes you can pull from the user. So essentially just giving it a little bit bit of direction, giving it a little bit of setup about how it should set up task as well before following this process out. And so every time you complete a task in claude really if this is something you're going to do again in the future, you can capture that process once you get the output right. [snorts] I find a lot of people think, "Oh, it took me so long to actually run that process. It would have been quicker if I just did it done it manually." That might be correct the first time you actually run a process in an AI tool, but once you actually capture this and say, "Oh, I've just taught Claude how I like to do that job, capture that in a Claude skill, and then you can just run it on command rather than having to run
51:16 the whole process again with multiple prompts and back and forth, etc. We can capture this moment of expertise and workflow at the moment that it hits perfectly for you." And that's the real unlock you start to get from Claude's skills like this. Yes, the first time might take a long time, but the from then on, we've captured the process to run. Okay. And it's gone through now. It's gone and analyzed everything then in this process. And we've got the SEO blog writer skill proposed. Turn a transcript, brain dump, video transcript, any long form input into a blog post in Elliot's voice. You could make this agnostic to different people as well. We're going to build a voice skill later for you as well. Um, so you can review all of this and the content included, but like I'm pretty happy that it's going to have exactly what we want in here. We might give it a quick read through some of the corrections. You can
52:02 see some of the house rule we discovered in actually running the process. This is so important actually, this house rules section because it it didn't know some of this stuff until we actually ran the process that it actually needed to do. So 0 m dashes, title case, H2s, numerals, on the nose titles, FAQs are standard. it's baked in all of those that feedback and actually running the process. A lot of people just bake the claude skill and then it will fall down because they haven't tested and refined the process of doing it before actually creating the skill. So this is why this is so important to actually do this way. So I'm pretty happy with this now and I'm just going to click save. That's it. We've now built from hardcore expertise. We built our SEO blog writer skill. Um, we can actually go and now go to our
52:50 customize area. We should be able to go to yours and then go SEO blog writer skill. You can see it's a brand new one. It's tagged up with what it does. Um, we can see something we have actually missed is we didn't give it the trigger keywords. So, turn a transcript into a brain dump. Um, we haven't actually put any trigger phrases in there, which is an interesting miss that we got. And then we have full SEO blog writer skill baked in there for us. So I might actually go and just do one last thing because I'm not quite happy with it now. I've actually looked at it. I might go and say I'm actually going to go and just I'm going to click this try in chat button. I'm just going to say SEO blog writer. Uh can we just make an edit to them? So we're going to edit a skill now. And you can do this very
53:36 conversationally. Can we just make an edit to this skill? It doesn't have any trigger phrases. Can we just put in five core trigger phrases into the description such as turn this transcript into a blog post, help me write a blog post, and then we should be able to just quickly, we could have done it in the previous chat as well, but I'm just going to show you you can pull any Claude skill and actually go and edit them really, really easily. Just if you if you've just missed something or you're not quite happy with it, uh you can conversationally edit these skills, which will happen a lot. I update mine a lot because things change. You spot new improvements, stuff just doesn't quite working work for some reason and you just need to go and update and just give it some more rules, some more information to actually start working with. And by the way, when you're using Claude Co-work as well, you'll see on the right hand side, this doesn't show
54:21 in Claude chat, but in Claude Co-work, you actually see because you might be running multiple step workflows, you can see in the context on the right hand side here that you can see what skills have been activated and even click on them to actually have a closer look at that skill as well. So when you start stacking these, you might see multiple skills being invoked on your D on your task board here on the right hand side. Okay, here we go. Then it's actually gone and made the edit. It's turned this transcript. We can see the description in the middle here has now the trigger phrases that write an SEO article, write this up for my website, etc. It's going to invoke the skill. So I can click update on the right hand side here and boom, it's updated the skill. We're all good to go. If we go and check the description now, we've got all of that put in there, which is beautiful. So that's how you go from a cold start. You
55:08 don't have a prompt. you don't have a process clearly defined to actually starting with grounding the the process in real expertise and research bringing an SOP or a process for actually completing this letting it do the work give it feedback and once you're happy once it's got that final perfect output then we capture it and we can always just make some slight tweaks if it doesn't capture it perfectly. So that's
…
