YT YouTube2Text

YouTube2Text → Transcripts → CLAUDE SKILLS FULL COURSE 2.5 HOURS (Beginner to Expert)

CLAUDE SKILLS FULL COURSE 2.5 HOURS (Beginner to Expert) — Transcript

by Eliot Prince · 31,595 words · 4,366 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00Welcome to the definitive course on
0:01Claude skills. I built over 450 skills
0:04over the last year or so for my own
0:07business for the 3,000 plus students
0:09I've taught to use AI Live and for
0:11businesses I've consulted with. I pretty
0:13much doing this stuff all day every day.
0:15And essentially what I'm going to do is
0:16I'm going to transfer everything that
0:18I've learned and known through practice,
0:20research, and iteration and actually
0:22building these things. I'm going to
0:23transfer all of that knowledge to you
0:25today. So, even if you've never touched
0:27a claude skill in your life and you're
0:29just starting to use this stuff, we're
0:30going to go from these beginner routes
0:32all the way up to advanced level. So,
0:34I'm going to walk you through the route
0:36to get there so you can build full teams
0:39of AI skills running in your business to
0:40act like little employees working
0:42together. And no fluff, here's exactly
0:44what we're going to learn today. I'm
0:46going to take you through the plan of
0:47what we're actually building so you can
0:49build along with me knowing the end
0:51result and the order we're going in.
0:53We're going to talk about what a claude
0:55skill actually is. Even if you sort of
0:57know what they are, going to help you
0:58conceptually explain them to other
1:00people or if you have no idea what they
1:01are. We're going to go through the basic
1:03definition and how it differs from what
1:05you already use with AI. We're going to
1:07talk about the anatomy of a skill and
1:10the different parts, how they all stack
1:12together to create a high performing
1:14file that runs inside Claude. We're
1:16going to talk about the map, the two
1:18types of skills, and the five different
1:20levels that again, you can mix and match
1:22and pair these together and understand
1:24where you are in the build process and
1:26how you're constructing your skills.
1:27[snorts] Once we've done that, we're
1:29going to actually start building and you
1:30can build along with me. We're going to
1:32talk about the three key ways to build a
1:35Claude skill, and we're going to build
1:36together step by step. 1 2 3. Then we're
1:40going to talk about what separates a
1:42good skill or an average skill from a
1:44really great one that is high performing
1:46in your business. We're going to talk
1:48then about managing the skills, keeping
1:51them up to date, sharing them, using
1:53them maybe across multiple people. Then
1:55we're going to start to actually build
1:57into the levels. So we're going to start
1:59to connect skills to the outside world
2:02and outside tools so they can execute
2:04more work for you. We're going to learn
2:05how to link them together to create
2:07chains of them that run in teams. And
2:09we're even going to stitch them all
2:11together with an orchestrator, a manager
2:13that can manage these, like a line
2:15manager for you that checks in and out
2:17of your team of skills before showing
2:19you how to package these all up in a
2:21neat plug-in. talk about what those are
2:23and then finish off with some other
2:25ideas about what you should be building,
2:27finding pain points in your business.
2:28And I'm even going to give you 101 of my
2:30best skill ideas for you to walk away
2:33with and build for yourself. So, in this
2:35course, we're actually going to be
2:37building a full team of SEO
2:39[clears throat]
2:40claude skills. These are going to act as
2:42a whole SEO department for your work and
2:44we're going to build everything live
2:45step by step. You can build along with
2:48me as well. If you don't want to do SEO
2:50or you think this might not be that
2:51useful for you, I still recommend
2:52watching and building along with me
2:54because this is really about building
2:55the skills and the knowledge for you to
2:57go and implement on your own things. So,
2:59there's two ways you could really watch
3:00this. You could actually build along
3:02with me or you could just watch and
3:04learn along the way, take notes for you
3:06to then impart on your own ideas of what
3:08you want to build for your business,
3:10work, life, whatever it is. And while we
3:12go along, as well, this guide has all
3:14the prompts and steps that I'm using in
3:16this video. You can grab it below.
3:18There's a link in the description where
3:19you can get it for free. It's inside
3:21something called my AI recipe vault. If
3:23you go and grab the link below and sign
3:25up, uh you can get the complete Claude
3:27skills course guide. This is going to be
3:29your walk along with all the prompts,
3:31etc. in here for you to use along the
3:33way. So, there's some super powerful
3:35stuff in here if you want to build along
3:36with me. So, what is a Claude skill?
3:39Well, a skill is simply a set of
3:42instruction Claude keeps to run on
3:44command. It's essentially giving it a
3:46stepbystep checklist or a standard
3:48operating procedure that tells Claude
3:51exactly how to complete a task. So you
3:54can write down a process once, give it
3:57to Claude in something called a skill.md
4:00file and then anytime you want Claude or
4:02any other AI platform to actually run
4:05that process or that task, it can go and
4:07look at the step-by-step instructions
4:08and follow that process. And the benefit
4:10of that is you give you the same output
4:13every time. So every time you ask Claude
4:15to do something, it follows the process
4:17to give you the same output. You can
4:19think of it as simply as a prompt. If
4:21you've got a really great prompt that
4:23gives you a great output every time,
4:24that's essentially a set of instructions
4:26of how to complete a task. You can take
4:28that prompt and build it into a skill.
4:30That's the most basic form, but you can
4:32go a lot deeper and wrap a lot of stuff
4:34up in these skills to complete a task.
4:37And the beauty of this is you might
4:38think, well, I've already got prompts.
4:40Why do I need to build them into skills?
4:42Well, every time you want to re find a
4:44prompt or complete some work, you have
4:45to go find that prompt, copy and paste
4:47it in, actually repeat the process, and
4:49it's not that smooth in a very basic
4:51level. A skill actually can fire for
4:53itself. You can either tell it to run
4:55that process or it will naturally run
4:57these processes based off other work it
5:00was doing. So, a lot of the time you
5:01might build these standard operating
5:04procedures into your clawed workflows
5:06and you might not even know that the
5:07skill is running. It just might run
5:09automatically. For example, I might be
5:12asking about the performance of a
5:13particular website and it might just go
5:15and run my SEO audit skill automatically
5:18without me even thinking consciously to
5:20go and run that process. Equally, we're
5:22going to build something called like a
5:23voice copywriting skill which applies
5:26which teaches it how to write in your
5:27tone of voice for your brand. Every time
5:29you output a document, for example, it
5:32might run that voice copywriting skill
5:34to make sure it's actually applying your
5:36tone of voice, your anti-AII writing
5:38rules to your output. Whether that's
5:40proposals for clients, LinkedIn posts,
5:43or general documentation around your
5:45business, work, life, whatever it might
5:47be, even just replying to an email. So,
5:49a skill is a set of instructions step by
5:52step that tells Claude how to complete a
5:54task wrapped up and ready to run that
5:56you can either run on command. You can
5:58just tell it to go and do that job and
6:00it will do it. Or it might just start
6:02doing it automatically because you
6:04roughly asked and described the job that
6:06you're doing in the rest of your work.
6:08Next, I want to talk about what actually
6:09makes up a claude skill. The anatomy of
6:11a claude skill. It can be as simple as
6:14one file with just a wrapped up prompt,
6:16but it can also be a folder of
6:18information with several different files
6:21in there. And I want to break this
6:22anatomy down for you here. So, I'm
6:24actually going to show you in my clawed
6:26environment here an example of a skill.
6:29And you're going to start to see how we
6:30can start using these. To find a Claude
6:32skill, you go to customize on the left
6:34here, and then you'll see it will
6:37automatically jump into skills. It will
6:39give you a discover tab for others that
6:40are already built into Claude, or you
6:42can click yours, and it might show you
6:44some that you already have if you've
6:45been building them. You can actually
6:47scroll down and you can click on one and
6:49it will bring you into the actual the
6:52back end of the skill, what these
6:53actually look like. Now, there are three
6:56main parts to a Claude skill. Number
6:59one, you'll get the overview, the
7:00description. This is exactly what it
7:02sounds. It literally describes what's
7:05going on here. So, this is an SEO blog
7:07writer skill. One of the ones we're
7:08going to be building today. It converts
7:10my YouTube transcripts, voice notes,
7:12rough content ideas into SEO optimized,
7:14publish ready blog posts in my voice.
7:17So, we already know what it's going to
7:19do. We also have built in here the
7:22trigger words. And these are really
7:24important in the description because you
7:26can see here any time that I would say
7:28something like turn this into a blog
7:30post, use the SEO content writer, turn
7:32this into an SEO article, write up write
7:35this up into a blog post from a
7:36transcript, Claude is just going to
7:38automatically do that job. I don't have
7:39to specially invoke the skill with slash
7:42commands necessarily. I can just sort of
7:45be working naturally and say, "Hey, can
7:47you do this job?" Or anytime it realiz
7:48it should be doing that job, it's going
7:50to fire this skill. So, it's always
7:52reading this description about what the
7:54skill does and how to actually invoke
7:58the skill and make it run. Now, we also
8:00have this button here, the contents. And
8:02the contents is made up from two key
8:04areas. You have the skill.md.
8:08And a skill.md is literally just a
8:10markdown file. It's a text file of the
8:13instructions here. So, you can see on my
8:15screen here, we've then got the SEO blog
8:17writer. These are the instructions that
8:19it has to follow. It's got a five-stage
8:21step-by-step workflow. It goes through
8:23stage one. It goes through stage two of
8:26drafting the article. Um, it has the
8:28article structure of exactly how I want
8:30these set up, the voice rules, what it
8:32shouldn't do. Has stage three wrap the
8:35SEO in production package. Again, how to
8:37wrap it all up. It does a voice audit to
8:39make sure it's running against my voice
8:42that I've trained Claude on on that
8:44skill. And it's also got an editorial
8:46part to act as a grumpy, grouchy
8:48newspaper editor without asking
8:50permission. And then it's got the
8:52output. So essentially what you can see
8:54here is that this is just a really long
8:57set of instructions. It's essentially a
8:59super duper detailed prompt. Now if I
9:01was to try to just use this prompt on
9:03its own, it would do an okay job, but it
9:05would it's quite a lot to just run in
9:07one prompt in Lord Chat. So what we
9:10actually package with it as well are
9:12reference files. You don't necessarily
9:14have to have reference files for all of
9:15these, but these are a little extra
9:17knowledge base for it to run off. And
9:19generally, these are things like
9:21templates, checklists, other things it
9:23needs to refer to to actually get the
9:25job done. There might be a little bit of
9:27knowledge in there. Or it can actually
9:29have things like Python scripts that can
9:31actually run automations for you in here
9:34as well. But in this one, for example,
9:36what I love to package in my Claude
9:37skills are an article template. So it's
9:40got a template of how it should use for
9:42everything. It's got the metadata block,
9:44for example. It's got an internal links
9:47block, external links block, how the H2
9:50body section should be formatted, how
9:52the headers should be formatted, so that
9:55all my articles essentially come out in
9:57this uniform templated manner based on
10:00all the other prompts that I've got in
10:01there to get the quality up. We also
10:03have a checklist at the end because I
10:05think I had a review process at the end
10:06and it says, "Hey, can you go and review
10:08this as a grumpy editor?" This is the
10:10checklist that it would actually go
10:11through to rate itself against. So it
10:13can actually rank and rate its work
10:15against this secondary checklist which
10:17is which is its review cycle. So inside
10:21a claude skill we have three main areas.
10:23We have the description which I showed
10:25you at the start which is what the skill
10:28does and how it's triggered like how it
10:29should run. We have the skill.md file.
10:32This is the main set of instructions as
10:35to how the job gets done and what and in
10:37what order. And then we have the
10:38reference files. These are not fetched
10:41every stage of the process, but fetched
10:43when a claude skill needs to do
10:45something specific like the final
10:46template, a review checklist, any extra
10:49supporting information. Um, and these
10:52reference files can also have scripts,
10:55essentially code that a skill can call
10:58when running the same process over and
11:00over again. You don't need to worry
11:01about too much these at the start of
11:03this, but it's worth knowing that, you
11:04know, you can have code written into
11:06these reference files as well. Now,
11:08sometimes people get a little confused
11:09about what should go in a skill.md file,
11:13the main instructions, and what goes in
11:15a reference file. For me, I like to
11:18think it as the decisions it needs to
11:20make along the way. So, you can think
11:23about how the jobs run, what order
11:24things happen in, what good light looks
11:26like, what it must never do.
11:28Essentially, as we said at the start,
11:29the step-by-step instructions about how
11:32to complete a job. What goes in the
11:35reference files are the supporting
11:37detail. So, you can think of these as
11:39the long worked examples, a full style
11:42guide. So, you could say in the main
11:45skill.md file, hey, make sure to apply
11:47my full style guide at this stage. And
11:50then you could keep the full style guide
11:52in a reference file for it to go and
11:54reference at that point in the job along
11:56with any schemas, formatting, what good
11:59looks like, what bad looks like, those
12:01long worked examples are some great
12:03examples to go in the detail versus the
12:06main skill.md file. Often Claude when
12:08you're building these will know and it
12:10will understand where to put each thing.
12:12But I think it's important for you to
12:13understand this when you're building as
12:16well so you can delineate between the
12:18two options that you have when you're
12:19building. Now the final area to talk
12:21about here is the description and the
12:24triggers. The triggers are really
12:25important because you want to make sure
12:27it's actually firing at the right time.
12:29So you could actually write into your
12:31description area at the top of a skill
12:33here. remember you could actually not
12:35have any trigger phrases in there and
12:37just make sure that it's only triggered
12:39when you actually command and select
12:41that skill to run versus actually making
12:44sure these trigger phrases are actually
12:46quite quite right. So in here I've got
12:48an example write up this transcript here
12:50is quite vague like I would probably
12:52actually looking at this remove this
12:54trigger phrase because I do a lot of
12:56things with transcripts and I don't want
12:58it to accidentally fire a skill. So, if
13:00you find your skills are accidentally
13:01firing through the way you're working,
13:03then you might want to check your
13:04trigger phrases um to make sure they're
13:07actually aligned. Equally, I have
13:08multiple brands. So, I actually
13:11sometimes name the trigger phrases to
13:13actually make sure it's running for
13:14halfday Thursdays, which is my AI
13:16community that helps people gain more
13:18freedom of time in their life versus
13:20elliotprince.com, which is my personal
13:22brand. So, if I'm building similar
13:23skills between the two, I have to
13:25delineate in my description of like,
13:27hey, can you do this in the Halfday
13:29Thursday's brand, not the Elliot Prince
13:31brand, for example, in my trigger
13:33phrases. So, really important here, try
13:35to actually make sure it's putting
13:37trigger phrases in there that are words
13:38and ways of phrasing things you actually
13:40use naturally in your work, not that you
13:43kind of have to keep remembering and
13:44thinking like, how how do I invoke that
13:46skill again? And don't worry, of course,
13:48we're going to get into all of this how
13:50they actually run in this course. So,
13:52let's get into how you actually use a
13:54Claude skill because if we're trying to
13:56build before we even know how to use
13:57them, then we're going to be in trouble.
13:59I showed you that SEO content writing
14:02skill and I talked about trigger phrases
14:03and how to run them on command. I want
14:05to show you exactly what that means. So,
14:07there's two ways you can actually run a
14:10Claude skill in your account. The first
14:12thing is you can hit forward slash here
14:15and then it's going to give you a list
14:16of all the skills that are available to
14:18you. Now you can scroll down this list
14:20and actually start running them or you
14:23can type in here and you can say SEO
14:25blog writer and it will come up with
14:27there you'll see the little description
14:28of how it runs and you can click that
14:30and it's going to highlight blue and
14:32you'll see now that Claude skill is
14:34actually ready to run. It's 100%
14:36selected in my account. So if you want
14:38to make sure that you're definitely 100%
14:40using a clawed skill, you use that slash
14:43command. It's called a slash command
14:44method. Now the other way we can run
14:47this is by actually saying those trigger
14:49phrases. So when I was talking about
14:50trigger phrases in my in the previous
14:54section we actually were talking about
14:57just using them conversationally in your
14:59workflow. So I would say something like
15:01hey I want to turn this YouTube video
15:03transcript into an SEO blog post. Here's
15:06the transcript. Let's get to work. And
15:08because I used one of those trigger
15:10phrases, turn this YouTube transcript
15:12into an SEO blog post. Now, this wasn't
15:14an exact same phrase that I had in the
15:17main description there, but it's really
15:18close. It's close enough that Claude's
15:20going to be like, "Okay, I know I need
15:21to use that skill." If I go and grab a
15:23YouTube transcript and we will paste it
15:26into the chat here, it'll come up pasted
15:28there and then run this, you will see
15:30that it actually starts to get to work.
15:32I haven't had to put a massive long
15:34prompt in there. The skill should invoke
15:37automatically because I've used the
15:38right phrasing. Here you go. You can see
15:40it in the description of what's going
15:42on. turning the video into an SEO,
15:44reading the SEO blog writer skill to
15:46convert this into this. So, you can see
15:48it's loaded the skill there and it's
15:50starting to use it. It's using the
15:52template structure, the SEO checklist,
15:54uh all of the reference files in there
15:56as well to actually complete this job
15:58for me. And that's the beauty of a
16:00skill. Once you have a defined process,
16:02you can just start using them
16:03conversationally in your work and it'll
16:05automatically then be able to run that
16:06task. We could also set this to run on a
16:09schedule to go and scrape my recent
16:11YouTube video transcripts, bring them
16:13into Claude and actually create the blog
16:16post and then connect up to my website
16:17and go and publish that blog post for me
16:19automatically as well. We can teach you
16:21that whole process, which is kind of
16:22what we're going through in this video
16:23in many ways. Now, before we build as
16:26well, we'll just talk through some of
16:28the customize area of Claude because I
16:30want to show you how to actually add a
16:32skill to your account because some
16:34people might send you skills or you
16:36might download them or you might come
16:38across the files in many different ways
16:40because you can just share these files
16:42between people as well. So, if you've
16:44been sent a skill file, they're actually
16:46really easy to upload to your account.
16:48All you have to do is go to customize on
16:51the left hand side here. You go to add
16:53and then you go to upload skill. You'll
16:56also see there's buttons like create a
16:58skill, create with Claude, record your
17:00screen, which we're going to go into.
17:02But to add one that someone sent to you,
17:04you click upload skill. You go and find
17:06a skill for example, add that to your
17:08account, click save, and then it's going
17:10to be ready to run in your environment.
17:12You can upload that, and you can invoke
17:14it just as I described before. You can
17:16see this one is actually got a whole
17:18bunch of reference files in there along
17:21with the description as well. And as
17:22well, you can turn them off and on. You
17:24can turn them off and on on the top
17:26right here. If you have them in your
17:28account, but you're like, "Ah, they just
17:29fire when I actually am not working on
17:30that sort of project." You can turn them
17:32off. You can also download them. So, you
17:34can go and download any skill in your
17:36account and send them to other people,
17:38as well as the ability to edit the text.
17:40So, if you just want to actually change
17:41a trigger phrase or a description, you
17:43could just go in and type it yourself
17:45really easily. Or you could actually
17:47click the top three buttons here to
17:49click replace in chat or even remove it,
17:51delete it from your account, for
17:52example. So, it's very easy to sort of
17:54manipulate, move, upload these skills in
17:57your account. And now you understand the
17:59sort of basic fundamentals of a skill,
18:01how to add them, and download them from
18:03accounts. we can get on to actually
18:05going deeper into the types and levels
18:08of skills that you can build. Now, when
18:10you're building Claude skills, it's
18:12really important to understand that
18:13there's actually two types of skills
18:15that generally we're building and they
18:18deliver slightly different outputs. The
18:20first one I like to call deterministic
18:23skills and these always give the same
18:26outcome. You know exactly how you want
18:28the process to finish. It might be
18:31something like an audit, an invoice, a
18:33monthly report. It's essentially comes
18:35out as a templated output. When you
18:38think about the SEO blog writer skill, I
18:40was giving it a messy input and it was
18:43giving me a templated output essentially
18:45for that skill. Now, I also have them
18:47for an invoice. An invoice, for example,
18:49I always know it needs to come out in
18:51the exact same PDF format with the
18:53headers in the same place. Really, we're
18:55just switching in and out the numbers
18:57and some of the details in there. So,
18:59it's actually really easy when you're
19:00building a deterministic skill. It's
19:02really easy to understand when they fail
19:05because it's pretty loud. If my invoice
19:07comes out and it's not in the right
19:08template or there's a column missing,
19:10there's the numbers are wrong. It's
19:12really obvious to see and it's pretty
19:13easy to fix. These are generally the
19:15easiest types of skills to build because
19:17you say, well, this is the output I want
19:19in this template. Here's the type of
19:21information going in. And you're
19:23essentially just taking that information
19:24and placing it in that template for me.
19:27fairly simple but super super powerful
19:29in getting rid of some of the busy work
19:31in your life. Now on the other side we
19:33have what I call non-deterministic
19:35skills and these are skills that more
19:38carry like judgment, taste, voice and
19:41there's no set output that comes every
19:43time. So again you might think of this
19:45as like your your voice skill, your
19:47brand voice skill. It can be applied
19:49across many different situations and
19:51there's a lot of decisions and judgment
19:53and taste sort of building in there.
19:54thinking patterns, for example, that
19:57need to be taken into account. And these
19:58can be harder to build because it's not
20:00as obvious when they go wrong or they
20:02require a lot more back and forth and
20:04training. Um, they could be thinking
20:06frameworks, second opinions on certain
20:08frameworks as well. It's really hard to
20:11actually define the perfect output
20:13because every output is slightly
20:15differently. And that means they they
20:17fail kind of quietly. They might look
20:19good the first time you've actually
20:20built them and run them, but there's no
20:22clear flag of an error or where they're
20:26slightly drifting. They have to take you
20:27have to keep a little closer eye on them
20:29and they tend to take longer to build
20:30and train and get into the place you
20:32want. So [snorts] really, when you're
20:34building, think about it in these two
20:36ways. Is this a deterministic skill
20:38where I just want this clear, defined
20:40output, or is this non-deterministic
20:42which actually is applying more
20:45judgment, taste, and voice across
20:47others? They're sort of more
20:49non-deterministic ones. You can think of
20:50them more of like they're an art rather
20:52than a science. There's no perfect
20:54formula or perfect output. They always
20:56require tweaking and upgrading versus
20:58deterministic which is almost a
21:00mathematical or an equation or a
21:02scientific output that comes out the
21:04exact same time, the exact same way
21:06every time. Now, we also have five
21:10levels. So, we've just gone through the
21:11two different types. There's also five
21:13different levels of Claude skills which
21:16really comes down to complexity. We have
21:19level one which is essentially it's
21:21essentially a very simple wrapped
21:22prompt. We can just trigger it. It
21:25doesn't really require any extra input.
21:27It doesn't require anything from us. It
21:30doesn't connect. It just hey can you
21:32just do this for me and it will run
21:34automatically. Very very simple. The
21:37level two is an ask. So it actually
21:40needs some more information for you. So,
21:42yes, it can run really easily, but you
21:44might need to give it more input. You
21:46think about the SEO blog writer skill.
21:49It's an ask because it needs a
21:51transcript. It's going to ask me some
21:53questions to actually get towards the
21:55final result rather than maybe an SEO
21:57audit skill where I can just give it,
21:59hey, can you audit this website? And it
22:01will go and run the audit. The level two
22:03needs more input, more of my human
22:05insight and expertise built in before it
22:08can actually deliver its great output.
22:10These are the two levels that most
22:11people sit at because it's very similar
22:13to writing a normal prompt ready to go
22:15in Claude. It's what we've probably all
22:17got used to using AI and chat GPT from
22:19day one. Then we move into level three.
22:23Now level three is where we connect. We
22:26can actually allow this skill to reach
22:28outside of Claude using something called
22:30connectors. Now there could be some very
22:32simple explanations of this. You could I
22:35have a great one. It's a morning inbox
22:37triage skill and it's able to go into my
22:39Gmail. It knows I've trained it step by
22:42step what to do in there, where to file
22:44certain things away, what to do. It
22:46doesn't reply to my emails for me, but
22:47it could actually reply to emails for me
22:49if I wanted it to. This is more of a
22:51process of getting rid of all of the
22:53junk and stuff that doesn't need my
22:55attention and leading the key stuff that
22:57doesn't need my attention. So we can
23:00actually train the skills to go and
23:02reach into our SAS tools either to
23:04retrieve information. So again, we can
23:06upgrade our level two skill by linking
23:09it to the
23:12YouTube API or connector, getting it to
23:15go and automatically fetch my
23:16transcripts and actually write the blog
23:19post. When we do that, when we connect
23:20it up to either go into a tool and pull
23:22information or push information to a
23:24tool, you could push information to we
23:27could push that blog post into a website
23:28or we could push it into notion or a
23:30word document for me to go and check and
23:32and make the final edits or it can
23:35actually drive a software tool as well.
23:36So, it can actually click around and use
23:38these tools for us as well. That becomes
23:40a level three, a connect skill. And
23:42again, I find people get to this level
23:44pretty quickly once they realize that
23:46they can build in the right links and
23:48connections in their skills. Where
23:50people don't often make the jump to is
23:52level four. And this is a chain or a
23:55team of skills working together. So you
23:58can actually build your skills to hand
24:00off one to another either one 2 3 4 or
24:04to make decisions and go left and right
24:06etc. Now, when we build today, what
24:09we're looking at is a set of SEO skills.
24:11And there's a very defined process that
24:12an SEO department or an SEO specialist
24:14would actually go through step by step.
24:17They would run an audit. Then they would
24:20write an SEO strategy. Then they would
24:22start creating the website content. Then
24:24they might actually publish that SEO
24:26content. And then they might write a
24:28report and a commentary analyze and then
24:31almost repeat that process. So we could
24:33actually get the skills to go one for
24:35another. I could get my SEO audit skill
24:38to then feed into my strategy skill to
24:40give it the information the strategy
24:42needs to write the strategy. Once we
24:43have the strategy, the strategy skill
24:45can then feed into the creator skill
24:47that's going to go and write a blog
24:49post, for example. And then you can get
24:51what are called chains of skills and it
24:53becomes like a complete production line
24:56that becomes a full workflow, not just
24:58one skill individually. And then we get
25:01on to level five, which is the highest
25:03level. And this is where you build a
25:05manager skill, an orchestrator skill,
25:07which when you have lots of skills
25:09chained together or or lots of them that
25:11can interact, you can build an
25:13orchestrator and manager that sits on
25:15top that knows what each skill does and
25:17how they interact with each other and
25:19can actually you could give the the the
25:21orchestrator, the manager, the line
25:23manager of these skills, my overall goal
25:25or like say, "Hey, my website traffic is
25:27uh has gone down. Can you help?" and it
25:29would know which of its teams to go and
25:31invoke and how they link together and
25:33start working which is where we get this
25:35sort of team this whole department of
25:37little AI employees running in our
25:39business. So we so we will be getting to
25:41this level by the end of this course as
25:43well which is super exciting but I think
25:45it's important to say that there are two
25:47types of skill and they can be across
25:49all levels but the levels aren't a
25:51scoreboard. A level one skill isn't
25:53worse than a level five skill. They're
25:55just different. You can get a lot of
25:57value out of a level one skill that just
25:59runs automatically every morning for
26:01your business. Even more so than often a
26:03level five skill, which a lot of people
26:05overengineer and they do it to prove
26:07that they built this whisbang amazing
26:09system. But it's actually not that well
26:12constructed and it's just over
26:13complicated compared to if I just had
26:14one of those skills running perfectly
26:17would be more leverage for certain areas
26:19of my work and life. So just know that
26:21just because you can build a level five
26:24management team of skills, it doesn't
26:25mean you actually have to. It's all
26:27about the pain and problem you're trying
26:28to solve in your business or your work.
26:30That's the real goal here to free up
26:32more time in your life and more wealth
26:35in your life. I mean, if AI is not doing
26:37that, then it's not really working for
26:39you. So, a level one skill that frees up
26:41two hours a week of work for me is a
26:44massive, massive win versus a level five
26:46skill that kind of needs a lot of
26:48troubleshooting and managing just to get
26:49it working. So, with that all said,
26:51let's jump into actually building our
26:54first skill. We're going to go through
26:55the three main ways of building skills.
26:57And we're going to start with the
26:58easiest and simplest one, and this is
27:00taking a prompt and wrapping it up and
27:02making it fire on command. And before we
27:05actually start working in this window,
27:07we need to just go into our settings
27:09here. We need to go into customize and
27:11we need to go to skills in the middle
27:14here because we need to actually make
27:15sure something is turned on in our
27:17account. We need to go into the search
27:20skills and plugins. And I want you to
27:22type in creator because when creator
27:24comes up, you should have this skill in
27:26your account. Skill-creator by
27:29Anthropic. This should be on by default.
27:31If it's not on, go and find it and
27:33switch it on because this is actually a
27:35skill inside Claude that's going to help
27:38build for you. It's designed by
27:40Anthropic and Claude to know the exact
27:42processes and requirements. You can see
27:43there's a ton of information in here.
27:45You can see all the reference files, the
27:47scripts, the graders, the analyzers, and
27:50just the general process of what it
27:51should actually do and some of the
27:53requirements such as here the
27:55three-level loading system, the data,
27:56the skill.md, and the bundled resources
27:58that we broke down earlier. This skill
28:01knows all of that and we can use this to
28:03actually build all of our and package
28:05all of our skills forever more. I'm not
28:07going to sit here and teach you how to
28:09type this by hand because it's much
28:10quicker just to use AI to help you
28:12build. So the first thing we need to do
28:15is we need to make sure our skill
28:17creator is turned on. Then we need to
28:18invoke it in our chat window here. So to
28:20do that you can hit forward slash and we
28:23can type in creator skill creator.
28:26Select it and we know it's definitely
28:28turned on. Equally, you could just say
28:30naturally in your conversation, and I'm
28:32going to use an app called Whisper Flow
28:34here that allows you to like talk and
28:36dictate to your computer, can we build a
28:38claude skill? And just by saying build a
28:41claude skill or create a clawed skill,
28:43it's actually going to make sure that
28:44that's invoked. So, you could do it
28:46either way. For the purposes of this,
28:48I'm just going to use this this forward
28:49slash command here to get things working
28:52just so for an illustration. But often,
28:54I'll just say, hey, I want to build a
28:55cloud skill. and it will invoke the
28:57builder skill for us. Next, I'm going to
28:59tell it exactly what we want to do. So,
29:01I'm going to say,
29:03I'm going to give you a prompt for an
29:06SEO audit skill. I want you to take this
29:08prompt and wrap it in a claude skill so
29:11that I can fire it and invoke it at any
29:13time. Very simple. I'm literally just
29:15telling it what I actually want it to do
29:17or how I want it to build this skill
29:19straight from off. Then, in this
29:20document that I've linked below, you're
29:21going to find this way one wrapper
29:23prompt. You're going to scroll down.
29:25It's going to say, "Hey, make sure the
29:26skill creator is turned on. Take this
29:28prompt and turn it into a clawed skill I
29:30can run on command. Everything's here
29:31for you to run." Then I'm going to copy
29:33this prompt. You can do the same if you
29:35want to build along with me. But you can
29:37see it's a really detailed prompt that I
29:39used to copy and paste for audits that
29:42actually now instead of me having to go
29:44and find this prompt and edit it, etc.,
29:46we're going to build it into a claude
29:47skill. But essentially, you can see it's
29:49giving it a role at the top here. It's
29:52got some input. It just needs the URL
29:54that I want it to run on. It's got the
29:56different steps that I want it to go
29:58through and then the output, the exact
30:00output that I want it to give in detail.
30:03So, I'm going to take this prompt. I'm
30:04going to copy it and I'm going to I'm
30:05literally just going to paste it into
30:06the Claude chat here. And then that's
30:08all we need to do. We can send that
30:10message. And Claude is going to go and
30:12analyze everything and build our skill
30:14for us. You'll be able to see in the
30:16chat here that it's understood the task
30:18it's got to go through. It's going to go
30:20and view that file. It's actually
30:21reading the skill creator skill here
30:23that we discussed. So, because we've
30:25invoked it here, it's reading the skill
30:27creator skill before building and
30:29working out the process that it needs to
30:31follow. So, what we're building here is
30:32a very simple, it's a level one skill.
30:36It's a pretty deterministic skill
30:37because level one, I'm just going to
30:39give it a URL and nothing more. No extra
30:41expertise and it's going to go and do
30:44the job and give me a formatted output.
30:46So, it's a level one deterministic skill
30:50for us. Very simple. And this is the
30:52easiest way to build them if you've got
30:53prompts that you're like, I really like
30:55using this. This is dialed. I just want
30:57this ready to use on command in my
30:59Claude account. Now, it might take a few
31:00minutes to actually complete this build.
31:03But what we get out the other end is on
31:05the right hand side here, we get the
31:07breakdown of our skill and on the left,
31:09it's got a little bit more information
31:11about what it's done. It's called it the
31:13search visibility audit skill. I mean,
31:16we could have actually given it more
31:17direction of what we wanted it to call
31:18the skill, but I quite like that it's
31:20got the description and how it actually
31:22runs. It's actually looked in my account
31:24and made sure it's not triggering in the
31:27same process as some of the other skills
31:28I have running in there already, which
31:30is really useful. You can actually go
31:32and turn other skills off or delete them
31:33if you if you are running into old
31:35skills that you don't want to run. It
31:36didn't run a live test audit against the
31:39skill, so it might be important.
31:41Sometimes Claude will actually run an
31:42all a test of its skill for itself, but
31:45I prefer to actually test a lot of these
31:47personally so I can actually see the
31:48output and give it feedback. So once we
31:51have our skill built, we can also see on
31:53the right hand side we can break down.
31:55Remember we had the anatomy here. So
31:57we've got the description and we can
31:58look at kind of explains it what it
32:00does. Uh that this one is doing our SEO
32:03and GEO audit here. You can see the
32:06trigger phrases it's got running and how
32:08it would run. And then we've got the
32:09step by step about how it actually works
32:12and importantly the output that it's
32:14going to deliver. Now this one doesn't
32:16actually have any reference files. It
32:17didn't need any reference files
32:19particularly. We could certainly give it
32:21more templated structures, but because
32:23it was just a prompt that we really
32:25liked and ran, we didn't really need any
32:27upgrades. We didn't need any extra
32:29reference files for it to run. So in
32:31this case, we only have a description
32:32and the main sort of skill instructions
32:35running. Now, all we have to do is click
32:38save skill here. And you'll see it's
32:40saved that skill and it's ready to go.
32:42If we go back into our skills area, you
32:44can click manage in the bottom right or
32:46remember to access your skills. You can
32:48go to customize and then click yours. We
32:51could search for the skill.
32:53Uh what was it called? The SEO
32:55visibility. And you'll see when you
32:57search for it here uh in your customize
32:59area, we'll see it come up and we can go
33:00and investigate and use and check the
33:02skill as it was built, which is pretty
33:04cool. It's ready to run in our account,
33:06which means now we can test out our
33:08first skill build. We could actually
33:10say, "Hey, can you run an SEO audit on
33:14my website?" https
33:18colon/corchavevel.vip.
33:21This is a website that I own that I
33:23that's actually not typed out at all.
33:25Right. I'm just going to type it out
33:27here. And you'll see I could invoke it.
33:29Remember, I could go forward slash uh
33:31search visibility audit. actually
33:33because I've just said, "Hey, can you
33:35run an SEO audit?" It should
33:36automatically invoke this skill for me
33:38as well because that was one of the
33:40trigger phrases there. So, what we can
33:42see straight away is we've built a skill
33:44there by wrapping up a prompt. And then
33:46you can see we're now testing it and
33:48it's invoking it. You can see in the
33:50middle of my screen here reading the
33:51search visibility audit skill for this
33:54audit. And it's going to go and do that
33:56job.
33:58Okay. And here we go. the the skill is
34:00has finished working for us and it's
34:02created our search visibility audit on
34:04the right hand side here in this nice
34:05artifact. It's given us our nice
34:07verdict. It's given actually a review of
34:11what it actually did and the score in
34:13there as well where what's broken and
34:15what needs actually fixing on there.
34:17We've got things like GEO uh every page
34:20reads written by admin rather than
34:22having the actual author on there which
34:24was probably me a lot of the time when I
34:25built this site. We've got our breakdown
34:27of different findings in this nice
34:29format and the evidence and citations.
34:32What else have we got? We've got the fix
34:33list. So, we've got literally
34:34step-by-step things that we actually
34:36need to go and fix. The long game of the
34:38strategy that we could be employing and
34:40some new pages worth upgrading and
34:42building. We've got 15 pages worth
34:44actually upgrading and building here.
34:47And some tips that need to be rewritten
34:49to get cited by AI search. So, this is
34:51incredible. This is super in-depth. This
34:53is a super valuable report. Anyway, like
34:57previously without a clawed skill like
34:58this, this might have taken me days to
35:01actually compile a full in-depth report
35:03like this. Now, because I've built my
35:05expertise into a prompt there to be
35:07fair, now we can just run the skill on
35:09any site. So I can either work more
35:11quickly to do a lot more work for a lot
35:14less time or instead of spending a whole
35:16day on or two days on an audit, I can
35:18just run the skill, have that job done,
35:20get on to some other higher value work
35:22or go walk my dog on the beach, hand
35:24over work to AI. Or I could even say,
35:26hey, I'm going to do this at low cost or
35:28free or just as a value ad for a client
35:30now because it honestly cost me nothing
35:32to actually run this really except for a
35:34few few tokens. And it was now super
35:37easy to just wrap that skill up in that
35:39prompt and I can run it anytime I want
35:41automatically. Even set it on a schedule
35:43to audit a website once a month or every
35:466 months. So we've built our first level
35:48one skill here just by taking a prompt,
35:51giving it to the skill creator skill and
35:53wrapping it up. And now it pretty much
35:55just runs by me giving it a URL and it
35:57just does the job for me automatically.
35:59Next, we're going to build our a level
36:02two skill actually, which is going to
36:04need extra expertise and input from us
36:06when invoking the skill. But also, we're
36:09going to talk about probably the best
36:10way to build skills. Now, the first one
36:13worked really well because I had a
36:14prompt that I'd spent a long-term
36:16dialing on my expertise rather than just
36:19going in and saying, "Hey, I can we can
36:21you just build me a prompt that does an
36:22SEO audit." It wouldn't have much
36:24direction and it wouldn't be very good.
36:25And when we talk about building the best
36:27skills, they're actually built in with
36:29either your expertise or some that you
36:31borrowed from other places. These are
36:33the best skills. You either want to bake
36:35in your expertise and processes of how
36:37you do things. Or if you're not actually
36:39sure, research and use other experts in
36:42the industry. But these actually take a
36:44little bit more time to build than just
36:45saying, "Hey, wrap this up." We actually
36:48need to run this basically from a cold
36:50start as if we have nothing. And to do
36:53that, we're going to actually run the
36:54process. We're going to teach Claude the
36:57workflow that we want it to actually
36:59execute for us and then capture that
37:01process. Okay? And we're going to
37:03rebuild the SEO blog writer skill that
37:06can write SEO optimized content based on
37:09information you give it, not just making
37:11random things up and scraping other
37:13information from the website. It's
37:15actually going to be a useful SEO blog
37:16writer skill, which I don't really see
37:18many other people building them this way
37:20with content that actually ranks and
37:22isn't just 100% AI generated, if you
37:25will. Now, we're starting from a cold
37:26start. So, I'm not going to start by
37:29saying, "Hey, can we build a skill?" I'm
37:31actually going to work through the
37:32process. So, I'm just going to start by
37:34actually giving Claude my pain and my
37:37pain point and then the outcome of what
37:40we're trying to actually do here. And I
37:42am going to voice these prompts in but I
37:44will put these prompts that you can then
37:46go copy paste in the uh main notes with
37:49this section as well. Uh so
37:53I want to take a transcript that I have
37:58and I want to turn it into a an SEO
38:02optimized blog post. Before we get to
38:04work, I actually want to research and
38:07create the process or the exact way that
38:10we should approach this task before
38:12going and executing it to make sure
38:14we're delivering SEO content and GEO
38:17content to the highest possible best
38:19practice standards. Now, I have a couple
38:22of ways that we're going to do this.
38:23one, I want you to research the best
38:26practices at the moment, the most
38:28up-to-date information and build me the
38:31the SEO GEO blog writing standard
38:36operating procedure, the system that we
38:38can always follow and the template we
38:39should be following. I'm also going to
38:41give you a transcript of my expertise
38:44and some of the extra ways that I like
38:46to build my SEO articles, basically. So
38:50you can see here I've just voiced in
38:51what I want to do, the goal I'm trying
38:54to achieve. I don't want to build the
38:56skill or actually do the job yet. I just
38:58want to research and outline the best
39:00practice and the process we should be
39:01following. And I'm also going to because
39:03this is quite a long task and it's
39:05multi-step and there's a lot of
39:06research. I'm actually going to invoke
39:07co-work mode. And co-work mode is great
39:10because it allows Claude to actually
39:12take multiple steps and actually
39:14complete tasks for you. And because I've
39:15got things like, hey, I want you to go
39:17off and research and create a report.
39:19It's great to do this in co-work mode
39:21rather than chat, which will just do one
39:22for one. It'll you'll ask it a question,
39:24it'll give you an answer back, maybe do
39:26a bit of research. Co-work's actually
39:27going to be able to do longer running
39:28tasks in our build here. So, that's a
39:30useful switch to turn on as well. So,
39:32the process here,
39:35and this is why this is one of the best
39:36ways to build a skill, is we're starting
39:38cold and we're building in expertise.
39:40We're not just letting Claude dictate
39:42what it should do. We're actually
39:43working back and forth to train it on
39:45our process. So it might ask you a few
39:47questions in here in the primary goal of
39:49the SOP is it for a particular website
39:53or is it platform agnostic? I just want
39:55it works for any site. I just want it to
39:57work for any site in this scenario. How
40:00how should the SOP be delivered? It
40:02might offer you to actually deliver them
40:03in a clawed skill. I don't want to do
40:04this yet because I don't want to lock
40:06the process into a claude skill until
40:07we've actually finished it and got it
40:09really defined because we will be
40:10working back and forth to actually make
40:12some improvements. So, I'm just going to
40:14say markdown SOP and template dogs so
40:16that we're not locking anything in uh
40:18locking anything too aggressively before
40:21we've defined it. And it's just going to
40:23ask me for the transcript of my SEO
40:27process essentially because that's one
40:28of my my areas of expertise. So, I might
40:31as well use that. The more you can use
40:33your own expertise and processes here,
40:35the better. So, I've actually pasted in
40:37here two transcripts from previous
40:39content I made, which is how to write
40:41content so good you can't stop ranking
40:42number one and where to put keywords on
40:44your website from I used to talk a lot
40:46about SEO on my my YouTube channel. But
40:48honestly, I would also go into
40:51potentially some of my frameworks and
40:53SOPs I have saved in my Google Drive,
40:55any templates I want it to follow, or
40:57even if you don't have transcripts, you
40:59could just brain dump your process. Even
41:01ask Lord say, "Hey, can you quiz me? Can
41:03you ask me five questions about my
41:05process so that I can give you my brain
41:07dump on the process? But essentially,
41:09because I've got the transcripts already
41:11on my processes, some from some of my
41:13teaching, I might as well use those
41:15because that is stuff based on, you
41:17know, proven tactics and ways of working
41:19that actually run versus
41:22just trusting what other people are
41:24saying. Now, I'm also going to get it to
41:26go and research some of the most
41:27up-to-date practices because some of
41:28those videos are like a year old or two
41:30years old now. So, I just want to make
41:32sure that some of this stuff is the most
41:34upto-date based on the most recent
41:36updates, research, expertise out there.
41:38You'll see on the right hand side here
41:39as well, one of the beauties of using
41:41Claude Co-work is it can track its
41:43progress. You can see that it's actually
41:46mapped out all the steps it needs to do.
41:48Whereas Claude chat would just give you
41:50an answer straight back and it would
41:52just do one little bit of work at a
41:53time. Um, Co's mapped out exactly what
41:56it's going to do. It's going to research
41:58the best practices. It's going to
42:00research GEO best practices. It's going
42:02to research my articles and fold in my
42:05expertise layer to then write the SOP,
42:07the templates, the QA checklist that you
42:09did see in the the skill that I showed
42:11you originally that we're rebuilding
42:13here. And then it's going to verify
42:14everything. So, this is a serious amount
42:16of work that it's actually going to
42:17cruise through for us here to build out
42:20this process. And I should say while we
42:22wait for this to build as well, the
42:23skills you build here, when you build a
42:25Claude skill in any area of the
42:27application, I'm in the desktop app
42:29here, but anywhere you build a Claude
42:30skill, if you build it in Claude chat or
42:32Claude co-work, they are available
42:34across both and then they will actually
42:35work in Claude code. On the other hand,
42:37if you are building skills in Claude
42:39code, they won't translate back into
42:41your Claude co-work and Claude chat
42:43area. So often I like to build include
42:46co-work because I know then I can build
42:48them everywhere and use them everywhere
42:49wherever I'm working. Um particularly
42:51for this knowledge works type of stuff
42:54rather than the actual coding and
42:55building work type stuff. So you'll see
42:57here even though I'm in claude co-work
42:59mode if I go to that visibility
43:02uh search visibility audit skill that we
43:04built earlier it would be invocable in
43:06claude co-work even though we actually
43:08built it in claude chat. So you don't
43:10need to worry too much about where
43:12you're building this stuff because they
43:13translate across both platforms. It
43:15might take a little while cuz we've
43:16given it a lot of tasks to do here. You
43:18can see mine has worked its way through
43:20all of these tasks and given me some
43:22output about, you know, it's received
43:24both transcripts here. Took about 20
43:26minutes to actually do this. You see,
43:27I've gone a little chilly today. So I
43:30put a little cardigan on, made myself a
43:31cup of tea,
43:33and actually been reading a book.
43:34Sometimes instead of scrolling social
43:36media or getting distracted by other
43:37tasks, I'd like to go and read a book.
43:39I've been reading Simon Synynic start
43:40with Y at the moment while Claude does
43:42my work for me which is nice sometimes.
43:44Um it's given me the headline findings
43:48uh and three places evidence corrected
43:50by you in terms of it did some of the
43:52research and then it relied on some of
43:54my expertise before building out. It
43:56gave me some sources that it had checked
43:58as well so I could make sure these are
43:59actual uh good quality sources. I could
44:03have defined the sources if I knew them
44:04up front as well or some experts that I
44:06wanted to to to follow, but essentially
44:08we've got our SOP. Now, this thing is
44:10actually pretty beefy. Uh, stage one,
44:13stage two, stage three, like this thing
44:16goes super deep. Keyword fan out and
44:19research. Um, you know, this thing is
44:21insane about how to go and create an SOP
44:25blog post. This I think this goes a bit
44:26deeper actually. Uh, sorry about the
44:29scrolling here. It's actually doing
44:31keyword research for us as well. Um, and
44:34then it's got like again you can start
44:36to see the building blocks of what we're
44:37building here. It's got the article
44:39template about how we want to template
44:41each article. The publish checklist to
44:44make sure this is actually done etc. And
44:46for some reason it produced them twice
44:48for me here. I don't know why it
44:49produced them twice but essentially it's
44:51built four documents here. The main SOP
44:53and then some supporting evidence and
44:55templates etc. Now, you might be tempted
44:58here to be like, great, turn this into a
45:00Claude skill, but I don't know this is
45:02actually working quite right for me yet.
45:04So, I would actually test this out. And
45:06[snorts] to do that, I'm going to use
45:08this process now. So, I'm going to go
45:09and get a transcript, and we're going to
45:10run this process in this chat. I'm going
45:12to use this transcript from my one of my
45:16most recent videos, the three levels of
45:17clawed memory clearly explained. You
45:19could use all of this process or you
45:21could actually you could find your own
45:23brain dump some information into a
45:25transcript and give it it as well.
45:26However you want to generate a
45:28transcript is fine. I'm actually using a
45:30a quick tool here YouTube-transcript-
45:33io just because it's super quick to
45:35actually uh grab transcripts and it
45:37doesn't use any of we'll just grab that
45:40again. It doesn't use any tokens in
45:42claude [snorts] to actually go and do
45:43this. You can scrape transcripts with AI
45:45tools but this is also equally useful.
45:48So I'm just going to say now I'm going
45:50to say can we test out our SOP then I'm
45:53going to give you a transcript about
45:55clawed memory and I want to write a blog
45:58post that targets the keyword claude
46:02memory. Okay. So you can see quick
46:04prompt here about let's test it out.
46:07This is the keyword I want to target and
46:10here's the transcript. And then we can
46:11run this and we're going to see how all
46:14of this performs before we actually lock
46:16this process into a skill. So we can see
46:18it's starting with stage one mine the
46:22transcript which is perfect. It's going
46:23to go and get all the information for
46:25me. Okay. So our SOP has finished
46:27running this processes. Give us some
46:29output on the left hand side here um
46:31about some of the processes it went
46:33through. It caught itself over
46:35optimizing on how often it had put
46:38memory in some of the headings and
46:40stuff. Um and then it's got some
46:42information about confirmation it wants
46:44before it shipped. So, some gaps it
46:47couldn't fill cuz I didn't give it a
46:48site and and some other quotes that I
46:51put in there. It just wants to confirm
46:52everything. And then on the right hand
46:54side, we actually have the output. It's
46:55put it in a markdown file. I might
46:57actually say like, "Hey, I'd prefer you
46:58to output these in a uh a Word document
47:01for me so that I could edit them." Oh,
47:03and then we can actually see it. Right.
47:04One of the beautiful things here is it's
47:06actually pulled direct from my
47:08transcript. I like oranges that set in
47:09my claw memory for months. So, it's used
47:12my keyword early on. We've got the key
47:14keyword in the title. I'm not 100%
47:17convinced by the title. Like, I'd prefer
47:19it to be more on the nose for SEO
47:22purposes. What does Claude actually
47:24remember about you? We could probably
47:25use AI memory earlier on in these
47:28subheadings. So, it's not got it 100%
47:31perfect, but like on the whole, this is
47:33a pretty well optimized, well-written
47:35article pulled from my transcript and
47:37bolstered by some research, which is
47:39cool. We got some comparison tables um
47:42and what it does and doesn't cover. Now,
47:45I think I'd like to actually just bake
47:47in a few more things. Like, I'm not
47:48going to lock this in a skill yet
47:50because I'm not 100% happy with the
47:52output it's actually given me. So, I'm
47:54going to give it some feedback about
47:56some edits I want to make to the uh to
47:57the article. Okay, I want to just make
48:00some edits to what's going on here. On
48:02the titles, I want them to be more on
48:04the nose. So, I would be something like
48:07clawed memory clearly explained brackets
48:10and how to set it up. Then I'd like to
48:13make sure in our first H2 that we have
48:15the word the keyword claude memory in
48:18there. And I'd like it all to also to be
48:21like in the further down H2 I want it to
48:24be how to set up level one clawed memory
48:27on purpose. Um so we're just getting a
48:30bit more keyword rich in there. The
48:33other thing I think we've missed off the
48:34bottom is I'd like a short FAQ section
48:37that is the most commonly asked
48:40questions. the top five also asked
48:42questions about Clawude memory would be
48:44great to have in there with just like
48:46short one or two sentence answers to
48:49each questions. Um, can we make those
48:51edits to the article, please? So, I'm
48:53going to go and get it to actually do
48:55the next round of edits on this to get
48:57it a bit more closer to what I think is
48:59a great SEO article versus it's just
49:02dropped the ball in a couple of places.
49:04So, we're working back and forth until
49:06we get the output exactly how we want
49:08it. Okay, brilliant. just gone and spent
49:10another minute or two just doing those
49:12edits. So, we can see clawed memory more
49:16in the title in the titles there. The
49:19title, the main H1, the header of the
49:21page is more on the money. And then on
49:23the right hand side here, we can see
49:24we've got the FAQs uh coming in nicely.
49:27So, I'm pretty happy with this now. I'm
49:29happy with this process. I'm like,
49:31great, our SOP and everything is been
49:34worked through and we've got everything
49:36in there. Now I'm happy that I've
49:40essentially trained Claude on how I want
49:42it to run this process each time. We're
49:44going to lock this into a skill. Now
49:46there's two options here. You can go to
49:48the top of your screen if you're working
49:49in Claude co-work mode. Click this down
49:52button here and click turn into skill.
49:55And that's just going to put a prompt in
49:56the box on the bottom left here. Turn
49:58this into a skill. And we can learn from
50:00this that all we have to do is ask it to
50:02turn this into a skill. I'm going to
50:04give it a bit more direction. I want you
50:06to create an SEO blog writing skill that
50:11captures this process the way we've got
50:13the final output now and everything
50:15we've gone through in this task and
50:17chat. Uh it should be able to run on a
50:19transcript or a brain dump transcript, a
50:22video transcript or even any other long
50:26form inputs I give it. If the user
50:28hasn't given you any inputs or you need
50:30more information, you should quiz the
50:32user before actually getting to work so
50:35that you have everything you need and
50:37quotes you can pull from the user. So
50:40essentially just giving it a little bit
50:41bit of direction, giving it a little bit
50:42of setup about how it should set up task
50:45as well before following this process
50:47out. And so every time you complete a
50:49task in claude really if this is
50:51something you're going to do again in
50:52the future, you can capture that process
50:54once you get the output right. [snorts]
50:56I find a lot of people think, "Oh, it
50:57took me so long to actually run that
50:59process. It would have been quicker if I
51:00just did it done it manually." That
51:03might be correct the first time you
51:04actually run a process in an AI tool,
51:07but once you actually capture this and
51:08say, "Oh, I've just taught Claude how I
51:11like to do that job, capture that in a
51:13Claude skill, and then you can just run
51:15it on command rather than having to run
51:16the whole process again with multiple
51:18prompts and back and forth, etc. We can
51:20capture this moment of expertise and
51:22workflow at the moment that it hits
51:24perfectly for you." And that's the real
51:26unlock you start to get from Claude's
51:28skills like this. Yes, the first time
51:29might take a long time, but the from
51:31then on, we've captured the process to
51:33run. Okay. And it's gone through now.
51:35It's gone and analyzed everything then
51:37in this process. And we've got the SEO
51:39blog writer skill proposed. Turn a
51:42transcript, brain dump, video
51:43transcript, any long form input into a
51:45blog post in Elliot's voice. You could
51:47make this agnostic to different people
51:48as well. We're going to build a voice
51:50skill later for you as well. Um, so you
51:53can review all of this and the content
51:55included, but like I'm pretty happy that
51:57it's going to have exactly what we want
51:58in here. We might give it a quick read
52:00through some of the corrections. You can
52:02see some of the house rule we discovered
52:04in actually running the process. This is
52:06so important actually, this house rules
52:08section because it it didn't know some
52:10of this stuff until we actually ran the
52:12process that it actually needed to do.
52:14So 0 m dashes, title case, H2s,
52:17numerals, on the nose titles, FAQs are
52:20standard. it's baked in all of those
52:22that feedback and actually running the
52:24process. A lot of people just bake the
52:26claude skill and then it will fall down
52:28because they haven't tested and refined
52:30the process of doing it before actually
52:32creating the skill. So this is why this
52:34is so important to actually do this way.
52:37So I'm pretty happy with this now and
52:39I'm just going to click save. That's it.
52:41We've now built from hardcore expertise.
52:44We built our SEO blog writer skill. Um,
52:47we can actually go and now go to our
52:50customize area. We should be able to go
52:52to yours and then go SEO blog writer
52:55skill. You can see it's a brand new one.
52:57It's tagged up with what it does. Um, we
52:59can see something we have actually
53:01missed is we didn't give it the trigger
53:03keywords. So, turn a transcript into a
53:05brain dump. Um, we haven't actually put
53:07any trigger phrases in there, which is
53:10an interesting miss that we got. And
53:11then we have full SEO blog writer skill
53:14baked in there for us. So I might
53:17actually go and just do one last thing
53:19because I'm not quite happy with it now.
53:20I've actually looked at it. I might go
53:22and say I'm actually going to go and
53:25just I'm going to click this try in chat
53:26button. I'm just going to say SEO blog
53:28writer. Uh can we just make an edit to
53:31them? So we're going to edit a skill
53:34now. And you can do this very
53:36conversationally. Can we just make an
53:37edit to this skill? It doesn't have any
53:39trigger phrases. Can we just put in five
53:41core trigger phrases into the
53:43description such as turn this transcript
53:46into a blog post, help me write a blog
53:47post, and then we should be able to just
53:50quickly, we could have done it in the
53:52previous chat as well, but I'm just
53:53going to show you you can pull any
53:54Claude skill and actually go and edit
53:56them really, really easily. Just if you
53:59if you've just missed something or
54:00you're not quite happy with it, uh you
54:02can conversationally edit these skills,
54:04which will happen a lot. I update mine a
54:06lot because things change. You spot new
54:08improvements, stuff just doesn't quite
54:10working work for some reason and you
54:12just need to go and update and just give
54:13it some more rules, some more
54:15information to actually start working
54:16with. And by the way, when you're using
54:18Claude Co-work as well, you'll see on
54:19the right hand side, this doesn't show
54:21in Claude chat, but in Claude Co-work,
54:23you actually see because you might be
54:25running multiple step workflows, you can
54:27see in the context on the right hand
54:28side here that you can see what skills
54:30have been activated and even click on
54:32them to actually have a closer look at
54:34that skill as well. So when you start
54:36stacking these, you might see multiple
54:38skills being invoked on your D on your
54:40task board here on the right hand side.
54:42Okay, here we go. Then it's actually
54:43gone and made the edit. It's turned this
54:45transcript. We can see the description
54:47in the middle here has now the trigger
54:49phrases that write an SEO article, write
54:52this up for my website, etc. It's going
54:54to invoke the skill. So I can click
54:56update on the right hand side here and
54:58boom, it's updated the skill. We're all
55:00good to go. If we go and check the
55:02description now, we've got all of that
55:04put in there, which is beautiful. So
55:06that's how you go from a cold start. You
55:08don't have a prompt. you don't have a
55:09process clearly defined to actually
55:11starting with grounding the the process
55:14in real expertise and research bringing
55:17an SOP or a process for actually
55:19completing this letting it do the work
55:21give it feedback and once you're happy
55:23once it's got that final perfect output
55:26then we capture it and we can always
55:27just make some slight tweaks if it
55:29doesn't capture it perfectly. So that's
55:31the second way we're going to build a
55:32clawed skill. Next, we're going to jump
55:34into the third way to build a Claude
55:37skill, and that's to show it rather than
55:40describing it. This is a feature called
55:42record a skill. And essentially, what
55:44we're going to do is we're going to
55:46literally record our screen within
55:48Claude and show it clip by click, step
55:50by step, how we do a task. Now, why
55:53would you do this as opposed to the
55:55other way? Well, sometimes you might not
55:57be able to complete a whole workflowing
55:58floor. that last one I did, I was able
56:00to work conversationally in the chat
56:02back and forth to get the output we
56:03want. But let's say actually I'm not
56:06quite sure how to describe it or some of
56:08the job requires me to click around
56:10other tools around the internet because
56:12remember Claude can actually use use
56:14your computer for you. You can use
56:16things like the Claude in Chrome
56:19extension, which I'll put a link in the
56:21notes as well and in the description for
56:23the Claude in Chrome extension, which
56:24allows Claude to actually use your web
56:27browser for you. Or you might actually
56:30have connectors that are set up, but you
56:32actually want to show Claude how you
56:33would use that connector. So, how do we
56:36do this? Well, next we're going to build
56:38a reporting skill because this used to
56:41be a job that used to take me hours and
56:44hours to do because I'd be clicking into
56:46things like Google Search Console,
56:48Google Analytics, even other SEO tools
56:50or manually checking results on uh a
56:53website search results. And there was a
56:55lot of manual pulling things,
56:57downloading reports, etc. analysis that
56:59actually AI can do now. So to run this
57:02process, I'm literally going to screen
57:04share and show Claude and commentate
57:06exactly what I do. It's going to record
57:08it and then lock that process down for
57:10me. So to activate this, you're going to
57:12click the plus button in the middle.
57:13First, you're going to select co-work
57:15mode here. The second thing you're going
57:17to do is click this plus button in the
57:18middle. We're going to scroll down and
57:20we're going to find record a skill. And
57:22this is exactly as it sounds. It's like
57:24creating a Loom video or something with
57:26inside Claude. [snorts] Record a skill.
57:29It's going to follow our screen clicks,
57:31typing, and voice are recorded and then
57:34sent to Claude and turned into
57:35repeatable skill. So, it says, you know,
57:37don't put passwords or secrets or
57:39sensitive information in here because
57:40that's going to be captured by Claude.
57:43But essentially, I'm going to start
57:45recording and then follow this process.
57:47You can see my screen is glowing orange
57:49here and it's capturing what I'm saying
57:52uh for the for the purposes of the tape.
57:53This is me just explaining to a viewer
57:55how to use uh Claude. So this task we
57:59are going to show you the SEO reporting
58:02process of how I collect data, analyze
58:04it and write commentary and compile SEO
58:08reports for clients. Now the first thing
58:10I will do is I'll open my Brave browser
58:13or my web browser and I will go to
58:15Google search console. In Google search
58:17console I will select the website I want
58:19to analyze. This one we're going to do
58:22VIP but we could do this for any website
58:24within my Google search console. Next, I
58:27would click performance and I would
58:30select more and I would compare and I
58:34would do the last 28 days yearover-year.
58:37We do year over year so that the data
58:40isn't tainted by seasonality etc. in
58:42most of our websites. Then what I would
58:44do is I would either scroll down and
58:46start reading the queries, looking for
58:48the pages that have been losing a lot of
58:52attention, pages that have been gaining
58:54a lot of clicks, uh any changes in the
58:57data, looking for
58:59the good options and the bad options
59:01that have happened on the site. I would
59:03go through the different pages, see
59:04what's been performing well and what's
59:06been dropping off. And then I might even
59:08export the data and download it to a CSV
59:12to start analyzing and looking through
59:14the data or even giving that to Claude
59:16to analyze. I would also go to Google
59:18Analytics and I would start looking at
59:20the same sort of information here. Um
59:23trying to understand the site's
59:25performance, what's up, what's down,
59:27what's stayed the same. I might also run
59:30some of my key search terms through a
59:32live Google search to just check
59:34performance directly. Once I have all
59:36this data, I would open up a Google doc
59:39because I want to create a word
59:41document. And in my word document, I
59:43would title it website name plus SEO
59:46report as my H1. I would then have five
59:50bullet points with the good, the bad,
59:53key insights, opportunities, anything
59:55else notable. Uh, and then I would write
59:58a one pager explanation of all of this.
1:00:00I'm always trying to frame everything in
1:00:02a positive light. If there's something
1:00:04bad that's happened, we're reframing as
1:00:06an opportunity to regain. We're also
1:00:08looking at the easiest, highest leverage
1:00:11things we can do and outlining what
1:00:13we're going to do for the next 30 days
1:00:15in our report for the client. Then once
1:00:18I've written this document, I would be
1:00:20then going into my Gmail, creating an
1:00:24email to the client with the subject
1:00:27name month SEO report, month plus SEO
1:00:31report. I'd write dear client name and
1:00:34then I would give like a really short
1:00:36two lines about what's going on. And
1:00:38then I would say hey the rest of the
1:00:40report is attached below. Let me know if
1:00:42you have any questions. And then I would
1:00:44send that email to the client. Okay. So
1:00:46you see there I've just hit done. It's
1:00:47actually outlined that I took 199 steps
1:00:50including all the clicks and
1:00:54commentaries and all the different
1:00:55things I did there. only seemed like a
1:00:57short process, but it's captured a whole
1:00:59lot there in that 291 second process.
1:01:02It's captured the whole transcript. So,
1:01:05I've basically narrated my exact
1:01:07standard operating procedure for doing
1:01:09this. What's cool is it's taken
1:01:11screenshots of the whole process here as
1:01:13well. So, you can see it's taken a
1:01:14screenshot of every screen. It's track
1:01:17it's tracked every uh key stroke I've
1:01:21made even all the backspaces and
1:01:22mistakes and it's going to match that
1:01:24all to my commentary before creating
1:01:27this process for me. So you can see the
1:01:28detail it goes into. It's checked
1:01:30exactly what websites have been on and
1:01:32everything. So it literally is like
1:01:34creating a Loom video of a a procedure
1:01:37that you're running or something you're
1:01:39doing and Claude being able to then take
1:01:40all that information and create the
1:01:42instructions to create a skill. Here you
1:01:45can see it's working its way through to
1:01:47suggest the skill it's going to build.
1:01:49Now what's cool is yes, we can use
1:01:53Claude in Chrome. It's like a Gentic web
1:01:55browser to actually go and like repeat
1:01:58some of those clicks that it's done for
1:02:00me. So, you'll see in my web browser
1:02:02that this Claude and Chrome, if I log
1:02:04in, this is an add-on you can get for
1:02:10Google Chrome here that it allows Claude
1:02:12to actually use your web browser on your
1:02:15behalf. Now, it can either summarize a
1:02:17page, tell you what's on a page, or
1:02:19actually navigate the web for you. I'll
1:02:21put a link in the description as well
1:02:23where you can find out more about how to
1:02:25use Claude and Chrome specifically. But
1:02:27just know that because I've done clicks
1:02:29around the web, the skill that Claude
1:02:31has made here might decide to go and use
1:02:33that Claude and Chrome feature for
1:02:35itself. It can drive it from the
1:02:37application here. Or it might decide,
1:02:39hey, Elliot has a Gmail connector. I'll
1:02:41just use a Gmail connector to complete
1:02:43that task. So, it is very clever about
1:02:45how it actually use this uses this. You
1:02:48can see no Google search MCP is
1:02:50connected. So I'm going to use Claude in
1:02:52Chrome for that step. Further down, it
1:02:54actually says, hey, when we're drafting
1:02:56the client email, use the Gmail
1:02:58connector. So it's figuring out the most
1:03:00effective way to actually get this job
1:03:02done for me, not just doing exactly what
1:03:05I would do. It's kind of mixing both. So
1:03:09we're happy with this. We should work
1:03:10pretty well. We can click save skill for
1:03:12my SEO client report. And now we should
1:03:14have that ready to go. SEO client
1:03:16report. So, we've got our third skill
1:03:18built. That was really quick and easy to
1:03:20actually do. Got our third skill built.
1:03:22We can actually run this SEO client
1:03:24report for So, we'll run it on
1:03:25Coravevel.vip again, which is the
1:03:28website that um
1:03:31we ran that initial site visibility
1:03:34audit on here. You can see a ski website
1:03:36about Corvel Ski Resort basically. And
1:03:38what's cool is actually read my memory
1:03:40there. It knows I own that website. So,
1:03:41it's not going to have uh any problem if
1:03:43I say, you know, go and do something on
1:03:46this site or give me some more
1:03:47information because it knows it's my
1:03:48property, which is handy. And it's even
1:03:50actually asking me how in depth do we
1:03:52want to go.
1:03:54It can even say decide to build a
1:03:56dashboard for me. I don't want the full
1:03:58dashboard at the moment. Um, but
1:04:00actually that might be super useful for
1:04:01a client uh to visualize something. I'm
1:04:04just going to say, hey, do the report
1:04:05and email and then give me the last
1:04:08let's do last 28 days yearon year. You
1:04:10can start to see the power of these this
1:04:13record a skill feature which goes and
1:04:15executes for you. It's gone out and
1:04:17pulled the step by step on the right
1:04:18hand side here. We can see the skill is
1:04:20invoked and now it's actually using
1:04:21Claudin Chrome web browser to go and run
1:04:24this process. It's going to ask me if it
1:04:26can actually use that. We should see in
1:04:28Claudin Chrome we can see here actually
1:04:29yeah it's controlling my web browser.
1:04:32This is the actual tab that it's
1:04:33running. When Claude and Chrome is using
1:04:35your web browser you'll see it's glowing
1:04:37orange. So you can see now we've said
1:04:39remember the first thing was go to
1:04:40Google search console. It's gone to
1:04:42Google search console. The screen is
1:04:44glowing orange. You can see actually my
1:04:45hands are off this and that little
1:04:47orange cursor in the middle of the
1:04:48screen there. This one is clicking
1:04:50around for me and collecting this
1:04:52information on my behalf. So it's
1:04:54automating this whole process because I
1:04:56showed it how to do the clicks once and
1:04:58then it's going to just go and repeat
1:04:59every step if it needs to. If it had a
1:05:01connector it could go and just use the
1:05:03connector and do it sort of through the
1:05:05API but it didn't. So, it's doing the
1:05:07manual clicks that I did for it here as
1:05:09well. So, we can just leave that to run.
1:05:12Okay, check it out. Now, it has
1:05:15literally all I said was SEO client
1:05:17report. It's gone and run the whole
1:05:18process. It's just finishing off uh
1:05:20writing the report for us here. It's
1:05:23actually run the skill and you can see
1:05:25on the right hand side, it's used our
1:05:27Gmail and Google Drive connectors
1:05:28because remember I wanted it to create a
1:05:30Google Doc and uh an email. You see, if
1:05:33we go into my Google Docs here, we've
1:05:35got the full SEO report here, the good,
1:05:39the bad, the key insights, the
1:05:40opportunities, and anything else notable
1:05:42and what we're going to do over the next
1:05:4430 days. Beautiful. The full commentary
1:05:46there. We should also check it in my
1:05:48Gmail draft. Dear Elliot, it's trying to
1:05:50write to me because I'm the website
1:05:52owner. Uh, it's put in the full or the
1:05:55it's put in the couple of lines. We
1:05:56might touch this up and make it like a
1:05:58little bit more friendly. Um, but the
1:06:00full report is here and then it's
1:06:02actually linked to my main Google doc
1:06:04report. Beautiful. It's worked like an
1:06:06absolute treat. So, we've gone there
1:06:09from zero process. I just clicked
1:06:11through. I selected the record a skill
1:06:13feature in Claude. You have to be
1:06:14running in Claude Co-work for that to
1:06:16actually build what was a connected
1:06:19skill, a level three skill. Remember the
1:06:21levels of skills? We had just a process
1:06:23that runs like a wrapped prompt with not
1:06:25much input. We had level two which
1:06:27needed a bit of bit more expertise or
1:06:30input from me either answering questions
1:06:32or a full transcript or a level three
1:06:34which was the process we used to use
1:06:37recorder skill and we've built it in
1:06:39with a connector as well. So it's using
1:06:41connectors to actually go and complete a
1:06:43job there which is just beautiful. Chef
1:06:46chef's kiss of skill building. So, after
1:06:49ascending those three levels of Claude
1:06:51skills, I just want to pause for a
1:06:52minute, take a break from the the
1:06:54building um to just talk about the six
1:06:56rules, like what takes a skill from
1:06:58being good to great because the next
1:07:02steps are actually going to get a little
1:07:03bit more tricky. We might start
1:07:05stitching well, we're going to start
1:07:06stitching these skills together to work
1:07:08as a team and managers and stuff, but I
1:07:10really want to give you some rules to
1:07:11make sure you're actually the base
1:07:13you're building is really great. Now,
1:07:15rule number one is one skill or one job.
1:07:18The common structural mistake is that
1:07:20people try and get a skill to do too
1:07:24many things. They try and build one SEO
1:07:26skill which does everything from
1:07:29auditing to reporting to writing to
1:07:31whatever or it just tries to cover
1:07:34multiple tasks in one go to give
1:07:37multiple outputs. If you're struggling
1:07:39and you're like it's just not giving me
1:07:41the output I want. This is just not
1:07:43quite right. Then my advice is always to
1:07:45split the job down and focus your skill
1:07:48on completing one thing really well. So
1:07:50if our SEO skills were struggling a
1:07:53little bit, I might actually break it
1:07:54down and just say, "Hey, your job is not
1:07:57to be a full audit skill. It's just to
1:07:59do the technical or it's just to do the
1:08:01content audit skill." I do have some of
1:08:03my skills broken down to go to like more
1:08:06granular level, but only if I'm
1:08:07struggling to get them to work. So I'll
1:08:09start off by giving it one job and
1:08:11making sure it can complete that. If
1:08:13it's still struggling to do that, I
1:08:14might think, how can I break this job
1:08:16down into even smaller tasks to get more
1:08:20consistent, highquality outputs? So, one
1:08:22skill equals one job, not trying to do
1:08:25everything at once.
1:08:28Now, rule two is again, if they're not
1:08:30quite working properly, trying to keep
1:08:31that main file, that main skill.md file
1:08:34as tight as possible. Some people
1:08:36recommend about 500 lines. There's no
1:08:38exact hard math to it, but just know the
1:08:41longer and more in-depth a skill goes,
1:08:44then the more likely it is to drift and
1:08:46start losing focus and run into
1:08:48problems. So again, if it's starting to
1:08:50to drift and forget things and and go
1:08:52wrong, think like, well, have I got a
1:08:55skill file, the main instructions area
1:08:57that's, you know, thousands of words
1:08:59long and trying to do multiple tasks? If
1:09:01so, can we chunk that down or can we
1:09:03move some things into the reference
1:09:05files? for example, a template that's
1:09:07built into the main instructions. Let's
1:09:09move it out of the main instructions,
1:09:11put it into a reference file for it to
1:09:13reference when it's running that
1:09:14particular of the task. So, it's not
1:09:16trying to compute all of the extra
1:09:18instructions in one go. Rule three, this
1:09:20is a little bit more technical, but
1:09:21write code for the mechanical bits. So,
1:09:24I don't always rely on the LLM, the AI,
1:09:28to do all the work. I like to get it to
1:09:30build the system to run some of the
1:09:32work. That could be Python scripts. That
1:09:34could be little bits of software, other
1:09:35bits of code. The easiest way to
1:09:37describe this is I will never get Claude
1:09:40to do maths directly. What I'll do is
1:09:42I'll get it to build the spreadsheet
1:09:44that does the maths. So, we get it to
1:09:46build the spreadsheet with the formulas
1:09:48in there to actually run the equations
1:09:51and put the output because you might
1:09:52have noticed that LLMs aren't the
1:09:53greatest at maths. So, all it has to do
1:09:56is move data into a spreadsheet. The
1:09:58spreadsheet takes care of it. The same
1:09:59can be said of converting files like
1:10:03doing arithmetic, checking formats,
1:10:05renaming things in bulk. Get it to
1:10:07actually ask it to write the code or the
1:10:10scripts to actually do some of this
1:10:12stuff. If you're finding that the skill
1:10:14is taking up way too many tokens, we can
1:10:16use remember we can have these scripts
1:10:18as our reference files as well. Rule
1:10:21number four, build from work you've done
1:10:23well. And this was this is what we've
1:10:26done through this process. The first
1:10:28example was a skill that I'd got dialed
1:10:30and I knew worked really well. So, it
1:10:32was easy for me to take that process and
1:10:34bake that into a skill. The second one,
1:10:36again, proofbased creation. I actually
1:10:38ran the process with Claude. I got it to
1:10:41a decent output. I then taught it the
1:10:43final tweaks that I wanted it and then I
1:10:46captured that process. So, if you try
1:10:48and build a skill that you've never
1:10:50actually done properly for yourself and
1:10:51you've never actually shown Claude how
1:10:53to do with the right instructions and
1:10:55you try and just say, "Hey, make me a
1:10:57skill that writes LinkedIn posts," it's
1:10:59not going to be very good cuz it's got
1:11:01no direction. It's had no expertise.
1:11:03It's just going to build something that
1:11:04looks good but you don't actually know
1:11:06performs very well. And then you'll get
1:11:08into the problem of trying to like edit
1:11:10and upgrade them. And instead, you might
1:11:12just want to start from scratch and be
1:11:14like, "Let me do good work and get a
1:11:16great output." and then capture the
1:11:18process to help do this as well. Rule
1:11:21number five, three to five examples. You
1:11:23could make three examples. So, for
1:11:25example, to upgrade my blog writing
1:11:27skill, I might actually put in the
1:11:28reference files, and I should do this.
1:11:30We haven't done in this process, but a
1:11:32great way to upgrade it would be to put
1:11:33three examples of what a perfect great
1:11:36blog post looks like. Not for it to like
1:11:39pull the content from, but so it can
1:11:41look at what the A star standard is and
1:11:44compare its work to that. That's how you
1:11:46get away from these like flat, they look
1:11:48good versions but missing something to a
1:11:51version that you'd actually ship. It can
1:11:53actually use the other examples baked in
1:11:55there to model the example. We could
1:11:57also use it when we were actually
1:11:59building the skill and showing it the
1:12:00the what we were trying to build. We
1:12:02could have actually said, "Hey, here's
1:12:04three examples of high ranking blog
1:12:05posts." We probably should have done
1:12:07that when we were building the blog
1:12:08writer skill in the chat there. Here's
1:12:10three examples of great. We could have
1:12:12done the same for the recorder skill. We
1:12:15should have actually said, "Hey, here's
1:12:17three examples of really great
1:12:19highquality reports that I want you to
1:12:21make sure you're hitting these marks."
1:12:22That would have just given it more
1:12:24direction and clear goal of what you're
1:12:26trying to hit with your final output.
1:12:28So, that's a rule five that I probably
1:12:30should have followed a bit more in this
1:12:31video. And then stop them fighting each
1:12:34other. Just be checking that your
1:12:36descriptions and your trigger words
1:12:38aren't fighting each other and you're
1:12:39trying to run multiple skills kind of to
1:12:41do the same job by mistake. That's a
1:12:44little problem that some people run
1:12:45into. And one final little tip here as
1:12:47well is you can really test the output.
1:12:50If you want to see how good or not your
1:12:51skill is, run it with this, run the
1:12:53skill or the job. Tell Claude to write a
1:12:56blog post with the blog writing skill
1:12:59and then get it to write one from a
1:13:00transcript without it and compare the
1:13:02two to see, you know, is your skill
1:13:05actually better and actually doing a
1:13:06good job versus just cold AI. And then
1:13:09finally before we move on to the next
1:13:11build is just three little tips here
1:13:13that all of your descriptions, you'll
1:13:15see this a lot when Claude is building.
1:13:16They have to be under 1,024 characters
1:13:20in that little description area of a
1:13:21skill. You might get it say, "Oh, I
1:13:24can't upload this skill or can't save
1:13:25the skill because my description is
1:13:27above 1,024 characters." If that's the
1:13:30case, ask give Claude that error message
1:13:32and say, "Hey, you you made a mistake."
1:13:33Cuz it does make that mistake. Um, just
1:13:36know there is a reserved word. You can't
1:13:38use the word Claude in the name of your
1:13:40skill. That is like blocked off by the
1:13:43system. So, you can't name a a skill
1:13:46something with Claude in it. And then if
1:13:48there is a colon or quote inside the
1:13:49description, it might break the file.
1:13:51These are just some little things that
1:13:53you might run into errors along the way
1:13:54while you're building. But next, let's
1:13:58jump into the next level of building a
1:14:00Claude skill. Now, we're actually going
1:14:02to rebuild a level three skill again
1:14:05where we're embedding connectors cuz I
1:14:07just want to show you again how you can
1:14:09go and use connectors, set up certain
1:14:12connectors, and actually quickly bake
1:14:13them into a skill cuz we kind of did it
1:14:15on the fly with recorder skill. We let
1:14:17Claude work it out for itself. I'm just
1:14:19going to show you once again how to do a
1:14:21level three skill with a connector
1:14:23because that's going to give Claude
1:14:25hands. It's going to allow Claude to
1:14:27reach and pull in data, real numbers
1:14:30from a real source without you, you
1:14:32know, you know, exporting and
1:14:34downloading anything. Whether that's in
1:14:36this example, keyword data, there's no
1:14:38copy and pasting. Nothing goes stale.
1:14:40It's always live and able to pull from
1:14:42them. We're also going to be able to
1:14:44push the finished work. So, this is
1:14:46going to be a two-way skill that's also
1:14:48going to be able to push things into
1:14:50like we saw Google Drive, Gmail, um, in
1:14:53nice structured output where, you know,
1:14:56some of your team members might actually
1:14:57look in in project management software,
1:14:59for example. Okay. And this obviously
1:15:02gets really powerful because now we're
1:15:03not just working in Claude. We're
1:15:04actually able to pull stuff and push
1:15:06stuff out and complete full jobs in
1:15:09software tools essentially. So, the
1:15:10first thing we need to do is we need to
1:15:12make sure our connectors are added.
1:15:14There are multiple like hundreds or
1:15:16thousands of connectors. Now, if you go
1:15:18into your customize area on the left
1:15:20hand side here, you can click
1:15:22connectors. You can see some of the ones
1:15:24I've already got set up here. Uh if we
1:15:26click yours, you can see a huge list
1:15:28going on here. Now, some might already
1:15:31be built in. One of the great ones for
1:15:33this sort of SEO or marketing work or
1:15:36sales work is Appify. This is great
1:15:38because it can actually scrape the web
1:15:40for you. So, if you want to add
1:15:42connectors, you could actually go and
1:15:43click discover. We could type in the one
1:15:45we're looking for. In this case, it
1:15:47would be ampify.
1:15:49That's which allows it to pull live web
1:15:52data. We could click this and we could
1:15:54click the connect button in the top
1:15:56right. In Appify, you could create a an
1:16:00account and it has access to 60,000
1:16:03different scrapers across the web,
1:16:05including SEO data, etc. We can click
1:16:07settings in the bottom left here and
1:16:09then API integrations. It's going to
1:16:11give you an API key. So when we copy
1:16:15that API key, we can go back into Claude
1:16:17and in our configure area when we're
1:16:19setting this up and connecting it, you
1:16:20put your API token in here. You copy
1:16:22paste that in, click save, and ampify
1:16:25will be set up. So that's how you're
1:16:27going to basically set up connectors.
1:16:29Some of them are much easier. You might
1:16:31even just be able to go into customize,
1:16:34uh, find one you like the look of. So,
1:16:36let's see if there's any other SEO
1:16:38options in here while we're here. Sorry,
1:16:40we need to select connectors.
1:16:42Um, and let's say there are some other
1:16:45SEO ones in here. There's Semrush.
1:16:47Semrush is an SEO tool. We could connect
1:16:50that to Claude. You can see continue
1:16:52connecting. I've got a Seamrush account
1:16:54that I'm logged into. And then we're
1:16:57connected to Semrush as well. So, you
1:16:59[snorts] can pick and choose which ones
1:17:01you want. You can use the appy one for
1:17:03this demo as well if you want because it
1:17:05gives you uh $5 free credit every month
1:17:08to actually go and test some of this
1:17:09stuff out. So with the connector So with
1:17:12the connector setup, we've got in our
1:17:14main document here, we've got another
1:17:16SEO prompt that we can actually copy and
1:17:19take that back into Claude and paste in
1:17:21here. You can see it's a bit it's a
1:17:22little bit funny in the way it's
1:17:24actually pasted in there, but I just
1:17:25want to break out a couple things in
1:17:27here. We've got things like notion URL.
1:17:29So, it's going to know to actually
1:17:30connect to our notion connector, which
1:17:32is another one here, but you could use
1:17:33any other project management tool. You
1:17:35could use Google Drive for Google Docs,
1:17:37etc. You can see I've got it as tools.
1:17:40You can use the appy for information and
1:17:42data to start with. It can also use I've
1:17:44got some other connectors baked in here
1:17:46as well. Data for SEO if it needs to and
1:17:49also something called firecraw which
1:17:51helps read web pages directly. But
1:17:53essentially, I put that prompt in here.
1:17:56And we're going to do the same. We're
1:17:57going to go for skill creator and just
1:18:01say, can you turn this uh prompt into a
1:18:04SEO strategy skill for me? So, this is
1:18:06an SEO strategy skill. It's going to be
1:18:08able to use connectors as our level
1:18:11three skill. What we're going to do
1:18:13eventually as well is we're going to
1:18:14link these all up so that our audit is
1:18:17able to then fire into our strategy
1:18:19skill, then into our blog skill, and
1:18:21then into our report skill, etc. And
1:18:23we're going to layer a voice copywriting
1:18:25skill over the top as well. So, you can
1:18:27see the layers that we're building up in
1:18:28this department and why I wanted to walk
1:18:30you through this process as well. But
1:18:33you could see basically my point here
1:18:35was that it's really that easy to
1:18:37actually build connectors into skills.
1:18:38You just tell it what connector you want
1:18:40it to use. We could actually go further
1:18:42and say when if we were building a very
1:18:44specific one, this is going to be a bit
1:18:46business in like agnostic and I can use
1:18:49it on any platform. But if we had
1:18:51something specific and I wanted it to
1:18:53point at a very specific uh Google doc
1:18:55or spreadsheet or notion database or
1:18:58notion page, you can actually give
1:19:00Claude a direct link to a particular
1:19:02resource for it to be able to work with
1:19:04on a connector as well. So, it's just
1:19:05going to clarify a few questions there
1:19:07with the little also ask questions. Uh,
1:19:09and then go and build the skill for us.
1:19:14Right. Then we built our skill. Then
1:19:16here we've got some feedback on the left
1:19:17and it actually says this is what I love
1:19:19as well about turning a skill or a
1:19:22prompt into a skill is Claude will
1:19:24analyze it and it's probably got a
1:19:26memory of how we've been working so far
1:19:28today as well. And it's gone in the six
1:19:29places I went beyond your prompt which
1:19:32are the ones worth arguing with. So,
1:19:34it's given us some more depth to our
1:19:36prompt. That's where prompts fall down.
1:19:37They can really only be so long because
1:19:39they just become unwieldy and un
1:19:41unusable otherwise, whereas a skill can
1:19:43bake more into its process. So, it's
1:19:45gone and actually given us some upgrades
1:19:47on the left hand side here and marked
1:19:49out any possible failures. So, again, we
1:19:52have this skill built. We should be
1:19:55pretty confident in it because we built
1:19:57it off a really in-depth prompt that I'd
1:19:59been testing and run. So, it's got the
1:20:01all the steps and the output ready to
1:20:03go. So, we can click save that. But
1:20:06you'll get this problem. Now, I really
1:20:07wanted to trigger this problem and I'm
1:20:09happy it did because quite often when
1:20:11you're building, you'll get this field
1:20:13description in skill.md must be at most
1:20:1620 1,024 characters as I said before.
1:20:19So, actually what I want to do is I'm
1:20:21going to take this error message in here
1:20:23and I'm just going to paste it in the
1:20:24chat because Claude has gone way too
1:20:26in-depth here. And then it's going to go
1:20:28and edit the skill file for us there
1:20:30nicely. So we'll click save skill and
1:20:32then we should be able to
1:20:36uh what we strategy
1:20:38SEO 6 month SEO strategy skill. Here we
1:20:41go. We've got that build. We should be
1:20:43able to check it out. The description
1:20:44works. Sorry. And for some reason didn't
1:20:46actually package the uh files properly
1:20:49in here, the reference files. So I've
1:20:51just gone and asked it, hey, make sure
1:20:53you include the reference files and
1:20:54updated the SEO skill there. And you can
1:20:56see on the right hand side here now uh
1:20:58we've got all the assets the strategy of
1:21:01what a uh a strategy should come out
1:21:03looking like in terms of the template.
1:21:05We've got some forecasting scripts built
1:21:07in here, some tool routting. It's built
1:21:09all this logic in there for us which is
1:21:12super cool. So now we have our strategy
1:21:14skill built out. We've got our original
1:21:17audit, remember the first one we built.
1:21:19We've just built a strategy which is the
1:21:21second in the process. We've then got
1:21:23our blog writer which is going to create
1:21:25the content and the pages for us. And
1:21:27then we've got our reporting skills. So
1:21:28we've got four skills here. Now here's
1:21:31where we get a little more complex
1:21:33because we're going to level up to the
1:21:35fourth level here which you can see the
1:21:37four skills. Audit, strategy, blog
1:21:39writer, reporter. We're going to make
1:21:41them all work together cuz at the moment
1:21:42they stand alone and we're like, okay,
1:21:44do a bit of auditing, do a bit of
1:21:45strategy, do a bit of blog writing. But
1:21:47we can stitch these together to work as
1:21:49a team. So each one knows what the other
1:21:52does and what information it needs to
1:21:54pass along the way. So our audit skill
1:21:56is going to find what's broken and worth
1:21:58chasing. Uh it's going to pass that
1:22:00information to our strategy skill, which
1:22:02is going to actually turn them into like
1:22:03a plan of how to address them, which is
1:22:06going to then give individual options to
1:22:08our blog writer to create with us, which
1:22:10will need a bit more input from us. uh
1:22:13and then we can then the blog writer can
1:22:15inform the reporting that watches how
1:22:17they're performing. So we're essentially
1:22:19building a nice production line. It is a
1:22:22little bit dumb because it still
1:22:23requires just 1 2 3 4 which by level
1:22:26five we're then going to add the
1:22:27intelligent sort of manager over the top
1:22:29of these as well as adding some brand
1:22:31voice in there as well. But let's turn
1:22:35these into a chain. And we're going to
1:22:37do this really simply. We're going to
1:22:39start off doing it one to one and then
1:22:41we're going to build up the chain of
1:22:43these skills to hand off information.
1:22:45So, I'm just going to start really
1:22:46simply and again I'll record all these
1:22:47prompts for you to use in the in the
1:22:49documentation and the the guide to
1:22:51follow along with. I have my and I'm
1:22:54actually going to make sure my search
1:22:56visibility audit skill and my six-month
1:23:00SEO strategy skill. Just going to
1:23:02actually make sure to name them directly
1:23:04to make sure that we're building
1:23:05correctly here.
1:23:08I want these skills to link together to
1:23:11work as a team. So when I kick off a
1:23:14search visibility audit, I would give it
1:23:16a website and then I want the audit to
1:23:18go and do its job and then push its
1:23:21findings. It should link and push its
1:23:23findings to inform the SEO 6 month
1:23:26strategy skill which should run off the
1:23:28back of the search visibility audit when
1:23:30it requests it. And that's really is
1:23:33quite simple as that. We need to start
1:23:35thinking in my mind which pushes to
1:23:37which. Now, our search visibility audit
1:23:39could also push straight to our blog
1:23:41writer. Or our report skill could also
1:23:43link back to our six-month SEO strategy.
1:23:46If we're making updates, you could say
1:23:47every month I want the report skill to
1:23:49feed back to my six-month strategy to
1:23:51update our strategy. So, they're not
1:23:53necessarily going in that nice defined
1:23:55order. We might we are going to loop
1:23:57them back around to all intermingle as
1:23:59well. So now this is a really
1:24:01interesting point it brings up which is
1:24:03why we don't need to prompt too
1:24:04aggressively because it'll ask the
1:24:05questions that it needs information on.
1:24:07When the audit finishes should the sixth
1:24:09month strategy skill get triggered? Now
1:24:12in some cases you might autochain and be
1:24:14like yeah this next process always
1:24:17happens. So don't even ask me just
1:24:19actually keep going. But honestly I
1:24:22might not always want a strategy
1:24:23written. I might just be running an
1:24:25audit
1:24:27and then want to just be like oh hold on
1:24:29a minute. So, I'm going to get it to do
1:24:30offer then wait. So, it's always just
1:24:32going to offer rather than auto run, but
1:24:34you could get it to auto run as well.
1:24:36I'm just going to ask a couple questions
1:24:37just to fix anything that it miss or
1:24:39where it might fall down before then
1:24:42actually going and updating both of
1:24:43these skills to intermingle. The search
1:24:46visibility audit is going to know where
1:24:48it's pushing to next. And the sixmonth
1:24:50SEO strategy knows that that could be an
1:24:53intake process for it as well. And now
1:24:55the only problem we'll face when we're
1:24:57actually linking all these sorts of
1:24:58skill is obviously they need historic
1:25:00data to work on particularly if I was to
1:25:03do let's say we go further and we create
1:25:05the report skill and it's reporting on
1:25:07performance and then feeding back
1:25:09information to like our strategy skill
1:25:11to run updates it needs to have the
1:25:13underlying historical information of
1:25:15what's gone on. So either you need to
1:25:17run this in a clawed project with the
1:25:19project information always updated or
1:25:21you run it in co-work on a specific
1:25:24folder for that client website or your
1:25:26website where you're going to hold all
1:25:28of the outputs. So if we're running in
1:25:30co-work on a folder, you can see we can
1:25:32add a folder in here. Um I'm just
1:25:34working in one that's just for this SEO
1:25:37skill build. every time I run a
1:25:39visibility audit or a six-month
1:25:40strategy, a blog post, a report, the
1:25:42outputs are either going to be put in
1:25:44there or we're going to store them in
1:25:46somewhere like notion. So, we have a
1:25:47database of all this ongoing work for it
1:25:49to refer back to. So, we can keep a
1:25:51strategy updated, we can use these
1:25:54skills in and out independently as well
1:25:56as keeping them informed through this
1:25:59workflow. And when we add our manager
1:26:01orchestrator skill on top. Okay, so we
1:26:03got it. Both uh files have been built
1:26:05and verified. And hey, this is what I
1:26:07want to see. Proposing them as a pair so
1:26:09they know that they're a team that they
1:26:11work together. Now, so starting to
1:26:12create this SEO whole department, our
1:26:14whole team of little AI employees that
1:26:17are going to operate this for us. Okay,
1:26:19so we've got our finished updates here.
1:26:22It's going to let us just review them
1:26:23quickly. I just will check through to
1:26:26make sure that they're actually working
1:26:28here. So, we'll check search visibility
1:26:30audit. Obviously, we need to test these
1:26:32as well. We would go and test that these
1:26:33actually work. I'm not going to do this
1:26:35live now because that's we've already
1:26:37seen them working. And then we can see
1:26:39down at the bottom here, it's helping.
1:26:41You can see some of them. Hand off save.
1:26:43You want me to run a six month strategy
1:26:45off these findings. So, it's actually
1:26:47giving us a handoff format as well. Not
1:26:49just our audit. It's actually giving us
1:26:51a handoff that we can use on our next
1:26:53skill. So, I'm going to update that
1:26:55skill. And I'm going to review this one.
1:26:57And the intake. Look for an audit
1:26:59handoff before anything else. and it's
1:27:01going to know exactly what it should
1:27:03look for and how it should go. Awesome.
1:27:05Great. Now we've built those two
1:27:06together. We've clipped them together
1:27:08like Lego bricks. We might also just
1:27:10actually go and put some instruction
1:27:12into the other skills as well here and
1:27:14just say my blog I also want to update
1:27:17my blog writing skill which should be
1:27:20able to feed off of my uh SEO strategy.
1:27:25So the SEO strategy will have action
1:27:28points in there and it should be able to
1:27:31action or the blog skill should be able
1:27:34to action the first blog article off of
1:27:37the strategy skill to get the process
1:27:39running. We also have my SEO
1:27:42report skill which
1:27:45will run and it should be able to then
1:27:48invoke the SEO strategy to skill to
1:27:51review our current strategy and make any
1:27:53tweaks along the way.
1:27:55uh based on the report as well as
1:27:57potentially the blog skill to make any
1:28:00tweaks to a blog post we have although
1:28:03that is probably rare. Um it might also
1:28:06invoke the SEO audit skill if it finds a
1:28:09particular problem that we need to
1:28:10investigate further. We might need to
1:28:12loop back to the SEO audit skill to be
1:28:15able to do a very particular piece of
1:28:17auditing based on a problem that the
1:28:19report has found. So, we're just going
1:28:20to give it a little bit more instruction
1:28:22about how the blog writer skill and the
1:28:25SEO uh reporting skill might actually
1:28:28invoke based on some stuff it's found.
1:28:30So, if it finds something really
1:28:31problematic in a piece of a website's
1:28:35like chuntering downhill and and it's
1:28:37not obvious why, it might say, "Hey, I'm
1:28:39going to offer to actually run the SEO
1:28:41audit skill, the first one, and push
1:28:44this information back to it to work in a
1:28:46team there." So, it's actually designing
1:28:47a loop. You can see it runs in a loop
1:28:49essentially. We might get to the end of
1:28:51the SEO six-month SEO strategy, file a
1:28:54report, and then run the audit skill
1:28:55again just on a schedule, which would be
1:28:57a really nice way of working. So, it's
1:28:58actually asking me some questions in
1:29:00here about how does this work. I can't
1:29:02run automatically. It's flagged. The
1:29:03blog writer needs more input. It's a
1:29:05level two skill, which needs a lot of
1:29:06input from me. So, I'm going to say,
1:29:09hey, you need to stop and read the brief
1:29:11that you've got from the strategy, and
1:29:13then tell me what you need from me. The
1:29:15report can invoke three things. is a
1:29:16strategy, a target audit or a blog fix.
1:29:19What should it be allowed to do on its
1:29:20own? I actually like I don't want it to
1:29:22invoke automatically.
1:29:24I just want it to recommend what it
1:29:27should do next based on the options. And
1:29:29should the monthly report grade
1:29:30performance against the six-mon
1:29:31strategy? Um, yeah, I think it should
1:29:33grade against the plan. Absolutely. It
1:29:35should be able to reflect back on the
1:29:37plan and report how we're tracking
1:29:38against the the laidout strategy. So,
1:29:40again, we're making these smarter.
1:29:41They're a bit dumber. They ran
1:29:43independently before. Now they're
1:29:44getting smart and learning how they
1:29:45should work together. Claude's going to
1:29:47be going through and proposing the next
1:29:49skill. We can go through and obviously
1:29:51check that we're happy with it. So we
1:29:54should have our visibility audit update
1:29:56with the different modes. It's actually
1:29:58added a mode in here which is interested
1:29:59the targeted investigation based off a
1:30:02particular problem that we're bringing
1:30:05to it from the report or the user. In
1:30:07fact, so we can go through check all of
1:30:09these obviously make sure you're happy
1:30:11with with them. Um, and then obviously
1:30:13run a test to make sure that they
1:30:15actually all run nicely. But it should
1:30:17present them to you one by one here just
1:30:19to start ticking off the list and make
1:30:22sure everything is running and linking
1:30:24together as you want them to. But
1:30:25remember, because we built these skills
1:30:27out well in the first time with our
1:30:29expertise and we know we got the first
1:30:31run of the skill dialed, now that we're
1:30:33linking them together, they should work
1:30:35really well. problem I often see people
1:30:37run into is that they try and build them
1:30:39all at once and they're like, "Build me
1:30:40a full suite of SEO skills or build me a
1:30:43full suite of um operations skills or
1:30:46something or finance skills and they
1:30:48don't time take the time to build each
1:30:49one individually and get that perfectly
1:30:51dialed before starting to get them to
1:30:53work together. So, they end up in just a
1:30:55complete mess. And again, if you're
1:30:57trying to capture too much stuff in one
1:30:59skill as well, if you're finding like,
1:31:01ah, they're starting to lose the plot a
1:31:02little bit, you need to think about
1:31:04going back and then breaking them down
1:31:05into smaller parts. So, actually, I
1:31:08might find that when I'm running this
1:31:09search visibility audit, we have full
1:31:12audit mode and we have targeted
1:31:14investigation. If I find that actually
1:31:16it's not quite working how I want it to,
1:31:19I might actually say, well, actually two
1:31:20modes. I might go and say, this isn't
1:31:22quite working right. Let me go split
1:31:24these skills up into I have a full audit
1:31:26skill and I have a targeted
1:31:28investigation skill. I'm not saying that
1:31:29two modeed skills don't work. I'm just
1:31:31saying if it stops working that is your
1:31:34best option to go and shrink them down
1:31:35into individual tasks. So it's thrown up
1:31:38the last two for review which were the
1:31:40new ones to update the SEO blog writer
1:31:42and the client report the ones at the
1:31:44end of our flow.
1:31:47It's nice. I like this put the do not
1:31:49use LinkedIn and do not use for
1:31:50newsletters to make sure governing
1:31:53principle. But this one I don't think
1:31:54it's changed a huge amount. It can push
1:31:57it into the report skill. But the report
1:32:00skill should be the one with the bigger
1:32:02one where it sits in the loop here. So
1:32:05that's nice that it's got the loop
1:32:06information. Uh we can update that skill
1:32:08as well for the sake of it now. And then
1:32:11we have our full set updated. And it's
1:32:14even written down the full loop
1:32:16architecture which is really handy.
1:32:18We've got the different skills, how they
1:32:20actually work, etc. This is more
1:32:22documentation at this point, but it
1:32:24actually is going to come really in
1:32:26handy when we move on to the level five
1:32:28skills, which is our manager, our
1:32:30orchestrator skills. So, so far we've
1:32:32gone from level one just wrapping a
1:32:34prompt up, small inputs. Level two
1:32:37giving more expertise into the skill and
1:32:40the output. Uh, level three connected.
1:32:43Level four multiple skills sort of
1:32:45linking together in a chain. And now
1:32:48we're going to put a manager on top of
1:32:50them. Just want to say where we are in
1:32:51the process of this video because we're
1:32:53going to do this. We're going to build
1:32:54our manager skill. Then we're going to
1:32:55wrap everything up into what's called a
1:32:58plugin. So, it's all neatly contained.
1:33:00And that's going to be a nice breaking
1:33:01point. After that, we're going to go and
1:33:03do things and talk about how to create
1:33:05brand voice skills and anti- AI writing
1:33:08slop skills, etc. to run on top of
1:33:11almost everything we do so that every
1:33:13output is coming out in the right voice.
1:33:15You can imagine in a process like this
1:33:17be amazing to have a brand voice skill
1:33:19for my website running on top of all of
1:33:21the outputs being created. Now let's
1:33:24jump into our level five manager or I
1:33:27call them orchestrator skills because
1:33:29what we have at the moment we have a
1:33:30production line. We have to remember
1:33:32where we are in the production line. It
1:33:34sort of runs from one end to the other.
1:33:37it's a little bit hard to jump in or
1:33:39remember all the different things that
1:33:40these skills can do or particularly if
1:33:42you're a new user to them. It's like
1:33:44where do I even start? Um so we need to
1:33:47actually give them a line manager for
1:33:49them to report to and from me as well.
1:33:52Uh so that we can have a little bit more
1:33:53flexibility and management of how
1:33:55they're uh working. So our manager skill
1:33:58is going to do exactly this. It's going
1:34:00to know what every skill is, what it can
1:34:03do, where it sits in the process, and
1:34:05it's also going to be able to problem
1:34:07solve and pain solve. So, I can go to my
1:34:09manager skills like, "Hey, my website
1:34:10traffic has dipped." And it can then
1:34:12start suggesting and invoking skills to
1:34:14help get us out of that process. So, a
1:34:17manager skill can help with our
1:34:18decision- making. Uh, it can help us
1:34:20jump in and out of different processes
1:34:22as well. And here we have the prompt
1:34:25here for you. It's actually four skills
1:34:27we bought at the moment because we
1:34:29haven't actually built our voice skill
1:34:32quite yet. We don't really need to add
1:34:34that in here quite now. We can easily
1:34:35add that on top of our orchestrator
1:34:38skill. Uh but I'm going to copy this
1:34:39prompt. And you can see it knows I've
1:34:41got to slightly change the names of
1:34:43actually what we called them here uh as
1:34:46we built. Um and it's got it should know
1:34:48what each does. It should state take a
1:34:50goal in plain language, tell me its
1:34:52plan, run the skills, report back, and
1:34:54ask me anything. So, we're going to copy
1:34:56this prompt and we're going to go back.
1:34:58We might as well run it inside the chat
1:35:00that we've actually just built this loop
1:35:02into. Um, and I'm just going to tweak
1:35:04this prompt a little bit. So, we have
1:35:06cuz we've been working with them in the
1:35:08chat here. I'm also just going to make
1:35:10sure we're naming the skill based on
1:35:12what it's actually ended up being
1:35:14called. If you don't like the names of
1:35:15the skills, of course, you can go and
1:35:17just get Claude to edit and change the
1:35:19name of the skills. SEO client report.
1:35:21Bada bing, bada boom. Pop that in there.
1:35:25um build me an orchestrator skill that
1:35:28manages them and we're going to run this
1:35:30process. So you can see our level five
1:35:32skill is the the manager sitting on top
1:35:35of all of this process for us which is
1:35:37really nice. Um and then we can start to
1:35:39see how some of this works once we've
1:35:41built out this or this orchestrator
1:35:43skill and this fairly simple across a
1:35:45four skill process. I actually have a
1:35:47bigger set which is about 15 SEO skills.
1:35:50And in that case, it becomes way more
1:35:51powerful to have an orchestrator and a
1:35:54manager skill when you're like, I've
1:35:55just got so many different little
1:35:56employees running around doing stuff in
1:35:58my claude. I need something to actually
1:36:00manage them. Or you might have them
1:36:02working across different departments. I
1:36:04think I've actually got in my account.
1:36:05While that builds, let me show you
1:36:07another example of them. We go into my
1:36:10uh customize here. If I search for
1:36:14orchestrator, you can see actually I
1:36:16have my marketing orchestrator, the
1:36:18marketing strategy that roots you to the
1:36:20right skills when someone doesn't know
1:36:22where to start or has a vague marketing
1:36:23request or or needs a multi-step
1:36:26workflow. Ask clarifying questions,
1:36:28diagnose the situation, and recommends a
1:36:29skill sequence. Triggers on help me with
1:36:31my marketing, where do I start, etc. So,
1:36:33you can start to see how you can build
1:36:35these out. Then, in the contents of it,
1:36:37it knows when to use this skill. It's
1:36:40thinking of it as a fractional CMO in a
1:36:42box. So again, we're we're building an
1:36:44SEO department today. And I could
1:36:46actually link my SEO department to this
1:36:49uh marketing or orchestrator, which has
1:36:51some stuff already built into it. This
1:36:53is a bit of a different setup uh with
1:36:56brand voice, which I'm going to show you
1:36:57how to build, keyword research,
1:36:58positioning angles, direct response
1:37:00copy, a newsletter writer, a content
1:37:03atomizer skill. So there's tons of stuff
1:37:05in here, and it understands how all the
1:37:07skill dependencies work. one upon
1:37:10another. Basically, it's even mapped
1:37:12everything out here for us. This is a
1:37:13really good example of an of an
1:37:15orchestrator skill. And here we go. We
1:37:17got the proposed skill coming out. I
1:37:18love it. So, it's called SEO department.
1:37:20Told you we're building a whole SEO
1:37:22department here. Take plain language SEO
1:37:24golf site and work out how to improve
1:37:26the SEO or which skills to invoke. We
1:37:28could actually see some of the content
1:37:31in here. The school this rule this skill
1:37:34lives or dies by. If you run the same
1:37:36skills in the same order every time, you
1:37:38failed. you are then a chain with extra
1:37:40steps. A chain does not need a manager.
1:37:42Uh so it's got all the specities about
1:37:44when and how to run each skill, how it
1:37:46should work as well, how they all route
1:37:48to each other and making sure you know
1:37:50suppress gates that you've already
1:37:52answered. Don't keep pushing back on the
1:37:53same things. It's got its reporting
1:37:55process um and some particular ways that
1:37:58someone might invoke this skill as some
1:38:00anti-AI patterns in there. So we can
1:38:03save our orchestrator now and the SEO
1:38:05department is up. It's up and running.
1:38:07So cool. So we've built our whole SEO
1:38:10department. Now we can test it before
1:38:12showing you how to wrap and file this in
1:38:14a plug-in basically. But essentially I
1:38:16could say, "Hey SEO department, you can
1:38:18see here it's ready to go." Or I could
1:38:20use it in natural language. I could say,
1:38:21"Hey, uh, can you help me improve the
1:38:24SEO traffic for and we're going to put
1:38:28in, uh, correl.vip, which is the site
1:38:31we've been sort of working on all
1:38:32along." Um, https. Let's get that right.
1:38:36We'll use the same folder that we've
1:38:37been building in all along and we'll see
1:38:39how this operates now. Now, instead of
1:38:42me thinking, do I need to audit? Do I
1:38:44need keyword research? Do I need to
1:38:45write blogs, etc., we've got our SEO
1:38:48department skill to actually start
1:38:50figuring out what the next steps are.
1:38:52And if you're running it on top of files
1:38:53that already exist, of course, it can
1:38:55start looking at like, oh, I've got a
1:38:57strategy. Um, I've got a report here,
1:39:00like what are we meant to be doing next,
1:39:01etc. You can see it's starting to read
1:39:03my claude memory, searching through the
1:39:05project, reading project info to get a
1:39:07landscape of where we are actually in
1:39:09this project, which starts to make sense
1:39:11now. Before we were like, oh, did we
1:39:14have an SEO strate where in the SEO
1:39:16strategy are we? Now it can remember.
1:39:18It's using memory to remember where we
1:39:20are in a project, in a strategy, what's
1:39:22been done, what we might need to do. It
1:39:24knows that it's a six-month SEO
1:39:26strategy. So, it would know when we last
1:39:28ran the strategy skill to be like, well,
1:39:30that was 5 months ago. Maybe we should
1:39:31update the strategy or the report, etc.,
1:39:34etc. And now it's actually going to, you
1:39:35know, work with me and be like, what are
1:39:37we actually doing here? You can see the
1:39:38SEO department skill running on the
1:39:40right. It's going to start, what is the
1:39:41binding constraint? How much access do
1:39:43you have? Can you connect and fix it?
1:39:45Like, what are we actually working on
1:39:47here? And then it's going to give us a
1:39:48plan. So, it's really interesting as
1:39:50well. It knows that Corvel's revenue
1:39:52happens December to April. It's
1:39:54currently September. So it's actually
1:39:55being proactive in the feedback and
1:39:58we're in a really critical window. So
1:40:00anything that needs to be indexed and
1:40:02ranking by December has six to eight
1:40:04weeks to ship. That's a genuine
1:40:05dependency. We need to be urgent. So we
1:40:08need to be clever about what we're
1:40:09doing. And this is the strategy we're
1:40:11going to run. So it is going to run the
1:40:13full audit and then hit a decision point
1:40:15and why and telling me why it's not
1:40:17running some of the other SEO skills we
1:40:19built and where we're starting from. So,
1:40:21I'm actually just going to say, "Hey,
1:40:22can you just run the audit and strategy
1:40:23in one shot for me? See if we can get
1:40:26the output." That's going to take quite
1:40:27a while. So, I might actually let the
1:40:29orchestrator go and actually run its
1:40:31process and run this the audit and the
1:40:34strategy straight off the bat for me.
1:40:36Um, because we can go and do some other
1:40:38stuff to help improve this as well. It
1:40:40might fire some questions our way as we
1:40:42go. You can see here on the right hand
1:40:43side, I've only ever invoked SEO
1:40:45department. It's gone and invoked search
1:40:47visibility audit to start our audit.
1:40:49Super cool. We've got our level level
1:40:51five SEO orchestrator, our SEO manager
1:40:54of our whole department running for us.
1:40:57Now, while our SEO department gets to
1:40:59work for us, I just want to talk about
1:41:01plugins because plugins are the last
1:41:04extra secret area of Claude Skills.
1:41:07We've gone through the five different
1:41:08levels up to that full team of
1:41:10orchestrated managers and workings. Now,
1:41:12we're talking about plugins, which in my
1:41:14mind, the easiest way to describe it is
1:41:15a filing system. Because at the moment
1:41:18in our Claude account, if we go into our
1:41:21customize area and we go into skills and
1:41:23we click yours, it's like it's just a
1:41:26long list of skills. Yeah, we've got the
1:41:29SEO skills that we've just created uh
1:41:32today, but like there's honestly, this
1:41:34is my account as well. There's a mess of
1:41:36just so many different skills in here
1:41:38that we really need to start sorting
1:41:40out. We also have connectors, which
1:41:42again are in an independent list. But
1:41:44the third tab along here is plugins.
1:41:47Now, a plug-in is essentially a filing
1:41:49system for all of your skills and your
1:41:52connectors and ways of working. So, for
1:41:54example, I've got this one here. It's my
1:41:56relentless content plugin that I built.
1:41:59It's got a short description, and again,
1:42:00it's a chain of skills that atomize,
1:42:03ideulate, write, publish, and convert.
1:42:06It's got the plugin has got a sort of
1:42:08contents area which is filled with um
1:42:11information about all the different
1:42:13skills and connectors, etc. in here. So
1:42:15you can see all the different skills
1:42:16that are actually packaged into this
1:42:18file and then the main overall
1:42:20information about what this does and how
1:42:22it might actually run. It's got a readme
1:42:25file about like what this actually does.
1:42:28So it's got the outline of the pipeline,
1:42:30the full map of all the skills. So you
1:42:32might be thinking, well Elliot, we just
1:42:33built an orchestrator skill, but it's
1:42:35still a bit chaotic in where they
1:42:37actually sit in Claude. This just brings
1:42:38it all together nicely under one hood.
1:42:41We've also got skills in here as well
1:42:43that we can actually see what's in here.
1:42:45And we would have connectors if we had
1:42:47connectors built to actually run on some
1:42:49of these skills which I'm going to show
1:42:50you how to set a plugin up. So when you
1:42:53come into your plug-in area, you'll just
1:42:55have SEO department. And then when we go
1:42:57back into our main area of Claude, you
1:43:00should see here as well to access them,
1:43:02we can click plus. We can go to plugins
1:43:04and we can see our different plugins and
1:43:06we can say oh let's go to our relentless
1:43:08content plugin and just invoke that
1:43:10skill as well. It's all nicely filed
1:43:12away essentially rather than just being
1:43:14loosey goosey scattered around your
1:43:17claude account. And they also make them
1:43:18really easy to share. If I wanted to
1:43:20share all of these skills nicely wrapped
1:43:22up, I could just put them in a plugin
1:43:23and send the whole plugin to someone and
1:43:25just send them their whole SEO
1:43:26department in a box rather than lots of
1:43:29individuals. And that might mean they
1:43:30miss or drop one or one gets deleted and
1:43:33it just disappears and then doesn't work
1:43:35properly. So think of them as a nice
1:43:38your plugins are a nice filing system
1:43:40that keeps everything nicely together.
1:43:42It doesn't change how any of the skills
1:43:44work or your ability to actually fire
1:43:45them. You can still run them and they
1:43:47make them easy basically to hand over to
1:43:49someone else. So how do we actually go
1:43:51and build this? It's actually super
1:43:53easy. All we've got to do is just ask
1:43:56and say, "Can we turn my four SEO skills
1:43:59plus my SEO department skill into a
1:44:03plug-in with all of the skills included
1:44:05with their reference files as well as
1:44:07the connectors required to run the SEO
1:44:10department. Call it the SEO department
1:44:12plugin. And then I might just actually
1:44:14list out the skills just to make sure we
1:44:17are actually nicely packaged up. What do
1:44:20we have?" Uh, search visibility was the
1:44:23other one we wanted in there that we can
1:44:25actually run. I I mean it would go and
1:44:27find the right ones anyway because we
1:44:29just built one today, but I just like to
1:44:30do a double sure job that we're actually
1:44:32selecting the right ones here. And we're
1:44:34just simply asking in natural language.
1:44:36You'll see a lot today. It's just about
1:44:37communication. What you like, what you
1:44:39don't like. Can you just turn this into
1:44:40a plug-in for me? And Claude will get to
1:44:42work. So over here on our um testing our
1:44:46orchestrator and our SEO strategy and
1:44:48chaining and everything you can see our
1:44:50what has it done? It has completed
1:44:54so much on the left hand side here our
1:44:57audit handoff. So it's actually created
1:44:58the handoff file for handing off with
1:45:01all the different pages. It's checked
1:45:03the trust signals. Everything that the
1:45:06audit has picked up has now been fed
1:45:08into the SEO strategy skill which is
1:45:12getting to work to actually create an
1:45:14action plan for us for us. Now over
1:45:16back, sorry to jump around back and
1:45:18forth, but we're actually building the
1:45:19plugin here. What you'll see is really
1:45:22interesting. It's actually making sure
1:45:23the connectors, so all the connectors
1:45:25that it specified during the build that
1:45:28we made has picked them out. Which
1:45:29connectors should the plugins declare?
1:45:31It's going to declare all of them.
1:45:33Should the plug-in be written
1:45:34specifically or or generic enough to
1:45:36hand to someone else. I'm just going to
1:45:38give it as a sharable template here. So
1:45:40essentially when you in if I was to give
1:45:42this plugin to someone else and they
1:45:44would upload it, it would actually know
1:45:46that all of these skills need these
1:45:48connectors and either if it's got them,
1:45:50it would be able to fire them or tell
1:45:52the user to make sure they are connected
1:45:53in their account and set up. So a new
1:45:55install of this plugin would go and tell
1:45:57them to go and connect these connectors
1:45:59to make this work. Again, that's another
1:46:01one of the benefits here. If I just give
1:46:03you a skill, um, it doesn't really know
1:46:07what connectors are needed straight off
1:46:09the bat. It might say, "Hey, do you have
1:46:11this?" And you might not actually
1:46:12install them. Whereas, we'll bring up
1:46:13front of the start about what the
1:46:15processes are we run and what connectors
1:46:17we need to actually make them run as
1:46:19well.
1:46:20And it's all going to be built into the
1:46:22instructions and the overarching
1:46:23architecture and scaffolding of the uh,
1:46:25plugin here. Okay. So it actually took
1:46:27just a couple minutes to finish uh
1:46:29writing this. It's again got us three
1:46:32things. It's changed not copied some
1:46:34name spacings in there. Genericized for
1:46:37sharing. So since you picked the
1:46:38template Elliot references became the
1:46:41operator. It just pulled out some of the
1:46:43real specity in my build and just used
1:46:45like example.com and things. Um and it's
1:46:48named a couple of the uh connectors that
1:46:51need to be used in there. One thing to
1:46:53decide once in school installed all five
1:46:55skills exist twice on your account and
1:46:57as a plug-in skill the bare SEO
1:46:59department becomes ambiguous worth
1:47:01removing removing. Okay, makes sense. I
1:47:03need to remove just the the naked skills
1:47:06in the account. It's got a bit of a
1:47:08build there and then it's got the you
1:47:10can see now you can see the depth that
1:47:12we've got in. We've got the main plug-in
1:47:14area. We've got each of the skills with
1:47:16their references with their different
1:47:18assets in here, the MCP, the connector
1:47:21details, and a read me of how this all
1:47:23works. Like this is a full SEO
1:47:26department [clears throat] in a box.
1:47:27Now, all I have to do is click save
1:47:29plugin, and it should then hook that up.
1:47:32I can go to manage, and we've got my SEO
1:47:35department plug-in. We've got the
1:47:37contents just broken down. We've got the
1:47:39skills that are included in here with
1:47:41all of their files and information and
1:47:43the contents and connectors they need to
1:47:46run. And we also have the connectors as
1:47:49well. So if I gave someone this plug-in,
1:47:52they would come in here and they'd be
1:47:53like, "Oh, I need to connect all of
1:47:55these connectors to make this actually
1:47:57work." Just straight away outlined as a
1:48:00nice filing system. So there we go. Our
1:48:02SEO department plug-in is built and
1:48:04ready to rock and roll with our five
1:48:06skill in-depth setup that we built
1:48:08today. You'll see as well on my we were
1:48:10running our uh practicing our
1:48:13orchestrator to run our SEO visibility
1:48:17audit and our
1:48:19SEO ports and our strategy and stuff
1:48:22which you can see it's actually working
1:48:24and pulling all of this information for
1:48:26me. Super nice. One thing that it's
1:48:29actually realized is I need to top up my
1:48:31she uses data for SEO. I need to top up
1:48:33my data for SEO to get some more data in
1:48:36there to make the strategy fully data
1:48:39informed, which is super super simple to
1:48:42do. I'm not going to do that right now
1:48:43because we're talking about just the
1:48:44build, but you can see exactly how we
1:48:46went. Built all the skills up, linked
1:48:48them all together, put an operator on
1:48:50top of it or an orchestrator on top of
1:48:51it, and then just told it to build it
1:48:53into a nice plug-in for me to keep
1:48:55things neat and tidy and nicely kept
1:48:57together. So, I just want to take us in
1:48:59a little bit of a different direction.
1:49:00Now, we're stepping a little bit away
1:49:02from the SEO skill build that we've been
1:49:04doing cuz that's complete. Now, we can
1:49:05say, "Okay, done that. I want to teach
1:49:08you how to build a voice skill." And
1:49:10your voice skill is something that
1:49:12allows Claude to learn how you write and
1:49:14interact and talk and apply that to
1:49:16everything it outputs. You think about
1:49:18go back to the SEO blog writer skill or
1:49:20even just the report skill. If it knows
1:49:22how I like to write and talk, it can
1:49:25apply that voice to all of my work as
1:49:27well as getting rid of some of those
1:49:29like annoying AI patterns and AI slop
1:49:31basically. And to be honest, my voice
1:49:33skill, although I don't use it directly
1:49:35to write all of my like newsletters and
1:49:38LinkedIn posts, for all the reports,
1:49:40general outputs, uh, tidying up a
1:49:43transcript, putting some final touches
1:49:45through it, it will actually run it
1:49:46through my voice skill to tidy things up
1:49:48and improve and write in my voice. Um,
1:49:51and it's actually much easier than you
1:49:52think to build. We can take a prompt
1:49:55here that's actually going to analyze.
1:49:56This is my analysis voice analysis
1:49:58prompt to help extract how you write and
1:50:02construct. So you can run this in
1:50:04several different ways. We'll go and
1:50:05start a new task here. Uh you can run
1:50:08this prompt. I'll just paste it in here.
1:50:11You can run this prompt by copy and
1:50:13pasting writing. So I'm actually going
1:50:15to build this for the let's say the
1:50:18corvel.vip VIP website that we've sort
1:50:21of been building with today and doing
1:50:23our analysis on and stuff because this
1:50:25has got tons of articles that I've
1:50:27written myself in the brand voice for
1:50:29this website. So what you can do is you
1:50:32could go and say uh find some of your
1:50:35best performing articles and you could
1:50:36literally just go and copy paste all the
1:50:38writing on there. Take that and paste it
1:50:42into your prompt in here's my writing.
1:50:44The other option you could do you could
1:50:46have a transcript as well. Transcripts
1:50:48work really well for this because that's
1:50:49you talking supernaturally.
1:50:51Um, or we could go and say instead of my
1:50:54writing, you could give it specific web
1:50:56pages or I want you to work out how how
1:51:00I actually write. We build that writes
1:51:02so we can build a skill that writes in
1:51:04my voice. I'm actually going to tweak
1:51:06this because I actually want to do it
1:51:07not for my voice. I already have that
1:51:09set up. I want to do it for corvel.vip's
1:51:12brand voice. So, I'm just going to tweak
1:51:14the intro and just, you know, you can
1:51:15tweak prompts if they're not quite for
1:51:17what you're doing. I want you to help
1:51:20build a voice profile for my brand,
1:51:23corvel.vip.
1:51:26You can find the website at corvel.vip.
1:51:29And I want you to go and pull the top
1:51:31articles from there uh for our analysis
1:51:34here. I'll just tidy that up because
1:51:36it's
1:51:38uh HTTPS
1:51:40want to help build a brand voice
1:51:43profile.
1:51:45Uh and then it's going to go through.
1:51:47You can see so I've given it where it's
1:51:49actually going to pull that information
1:51:51from before you analyze. Make sure that
1:51:54they're not, you know, weird, badly
1:51:56written, etc. how to extract it,
1:51:58vocabulary structure,
1:52:01what I do that most people don't, and
1:52:03then give me anything evidenced and
1:52:04inferred. So, we can run that. You can
1:52:06either paste the files in there, set
1:52:08codework to run on top of a folder of
1:52:10structures, or just point it at a
1:52:12website to go and crawl and pull all of
1:52:14the the best content from a website, for
1:52:16example. And this is stage one. This is
1:52:18analyzing, okay? We're not building
1:52:21anything yet. We're just trying to
1:52:22analyze our voice profile from as much
1:52:24content as possible. the more the
1:52:26better. You could, if you're building a
1:52:28personal voice profile, you could be
1:52:29pulling in uh LinkedIn posts, email
1:52:33newsletters, general emails you've
1:52:35replied to customers to, you know, your
1:52:37LinkedIn um your LinkedIn, your other
1:52:39social media posts, your YouTube
1:52:41transcripts if you're me, for example.
1:52:43Whatever you've got that is written or
1:52:45naturally voiced by you is going to be
1:52:47best here. If you've got stuff that's
1:52:49already got AI writing in it, then
1:52:51that's not going to be ideal because
1:52:52then we're not really detecting any new
1:52:54patterns. So, it's going to be
1:52:56unhelpful. It needs to be human written
1:52:58stuff. You might even give it some of
1:52:59your WhatsApp messages or your email
1:53:01replies. Okay, check it out. Now, it's
1:53:03built me my brand voice profile. It's
1:53:06actually done five pages, which is
1:53:08plenty. It excluded something because uh
1:53:13it didn't appear to be my writing
1:53:15apparently, which is great. I think I
1:53:16might have she's using AI for that one.
1:53:18So, that's fair enough. Uh, but it's
1:53:20cool that it managed to distinguish my
1:53:22writing versus other people's writing on
1:53:24the site and then pulled out the
1:53:26sentence mechanics that is evidenced in
1:53:27there, the vocabulary that we use, the
1:53:30structure of the sentences, all of this
1:53:32sort of stuff. It's actually pulled out
1:53:34for us to build this report as well. Uh,
1:53:36it just has five questions before it
1:53:38actually goes and builds anything next.
1:53:40Um, correct. The top and tail are mine,
1:53:43so it was right about that.
1:53:46uh mine, but I was written to a partner
1:53:48brief. Switches between I and we
1:53:50deliberate for I. I for experience, we
1:53:52for the brand is a is a decision I
1:53:54probably unconsciously made on the site,
1:53:56but I know it exists. Should the profile
1:53:57cover the whole site or just only the
1:53:59editorial articles? Um, let's only say
1:54:02editorial rather than any just bog
1:54:04standard business listings that we have
1:54:06on there that aren't particularly
1:54:08creatively written. And once we've had
1:54:10all of that confirmed, again, the more
1:54:11information you give it, the better. We
1:54:13can actually use the next prompt, which
1:54:14is the build prompt. So, I'm going to
1:54:16copy this. This profile is right now
1:54:17with my corrections. You might go
1:54:19through it and give it some more
1:54:20corrections. Turn it into a claude skill
1:54:22that takes any draft I give it and
1:54:24rewrites it so it sounds like me. The
1:54:26skill should do these things. Should be
1:54:28named uh let's call it VIP brand voice.
1:54:32Before starting, I'm also going to give
1:54:34you my uh AI slop and band words list.
1:54:37I'm just going to tweak this prompt
1:54:38again. Ask for this
1:54:43next before starting to bake them as
1:54:47references
1:54:48into the skill. I'm just going to copy
1:54:51this build prompt now and we'll just
1:54:53wrap that so it's nice and tidy for
1:54:55everyone. Uh go back to Claude and I'm
1:54:57going to paste my build prompt in here
1:55:00to give it the outline of actually what
1:55:03we're going to do. And when you're
1:55:04ready, it's going to ask for the next uh
1:55:06bit. So, I've got my anti-AII slop,
1:55:09right? Like anti-AII writing
1:55:11essentially, which is built up. We
1:55:13analyzed and built this up over months
1:55:15and months and months of all the new
1:55:17things that AI comes out with. Um, you
1:55:20know, it does that rule of three where
1:55:21it does like boom boom boom between uh
1:55:24full stops. You know, it has m dashes.
1:55:26It's like in a XY Z world, all these
1:55:30sort of AI tells. Um, it's got all of
1:55:33that in there and sentence structures.
1:55:35We've also got a band word list as well.
1:55:37So, what you're going to do is you're
1:55:38going to copy this and take that and
1:55:40paste it into Claude. And then I'm just
1:55:42going to go back as well. I'm going to
1:55:44grab the add the band word list, which
1:55:47is all those words that we don't want
1:55:49written. It's also got banned phrases in
1:55:51there as well. Not just X, but Y. In
1:55:54today's blahy blah world, it's worth
1:55:56noting that all of this these horrible
1:55:58things you see it write. Um, and this,
1:56:01you know, furthermore, no one writes
1:56:02furthermore. So, I'm going to take that
1:56:04and paste that in there as well because
1:56:08it's going to bake those in as our
1:56:09reference files for it to reference and
1:56:11check things off. Now, what's
1:56:12interesting about it did ask me a couple
1:56:14questions when writing this, but what's
1:56:15interesting, obviously, it's done all
1:56:17the stuff we've learned so far. It's
1:56:18operated the skill creator. It's wrote
1:56:21up the skill. What's interesting now is
1:56:23the way we've structured the prompts as
1:56:24well. It's actually running test cases.
1:56:27So it's actually seeing how well the
1:56:28skill works and does it adhere to the
1:56:30rules we gave it to make sure yes we've
1:56:32got all the rules written down but the
1:56:34final output does it actually stay to
1:56:36our strict rules or not and then it's
1:56:39evaluating. So something that actually
1:56:41runs in your clawed account and the
1:56:42skill creator has it's able to actually
1:56:45run evals and evaluations and tests your
1:56:48skills. So, when you're building as
1:56:50well, and I hadn't mentioned this
1:56:51earlier because it goes super deep, but
1:56:53when you're building your skills, you
1:56:55can ask Claude to actually test them for
1:56:57you and evaluate them, and it's actually
1:56:59going to check that it does what it says
1:57:01on the tin. Like, it follows the process
1:57:02that you've given it um and give you a
1:57:05review of that test. You can see it's
1:57:06actually running six agents right now
1:57:09automatically, includ
1:57:12after it's packaged the skill to
1:57:14actually then say, "Hey, I'm going to
1:57:15run one with the skill on some test
1:57:18content. I'm going to run one without.
1:57:20I'm going to run a second one with the
1:57:21skill. I'm going to run a second one
1:57:23without. And the same for this as well.
1:57:25So, it's running three tests on a
1:57:27baseline of the skill of using the skill
1:57:29versus not using the skill and comparing
1:57:31the output and rating it against the
1:57:33criteria in the skill to give us our
1:57:35feedback. So, you can see the process
1:57:38that Claude is working through on the
1:57:39right hand side here, which is super
1:57:41interesting that a lot of people don't
1:57:42know it can actually do. And so this is
1:57:44going to help test and it's also going
1:57:46to help it feed back and make edits and
1:57:48update to the skill before we even get
1:57:50to testing it. So it's getting rid of
1:57:51some of the lowhanging fruit that's not
1:57:53quite working properly. So that when it
1:57:55comes back to us, we can run a test on
1:57:57it just for that more higher level. As I
1:57:59said at the start, these voice, this is
1:58:01a nondeterministic skill. So it's a bit
1:58:04more art. It doesn't have a structured
1:58:06templated output every time. It's a bit
1:58:08more of an art than a science to feel
1:58:09back and forth. And you'll probably
1:58:11constantly be updating your voice skill
1:58:13to be like, "Ah, I don't really like the
1:58:14way you phrase this all the time, like
1:58:16stop doing that or use this pattern or
1:58:18stop using this pattern." Um, as your
1:58:20brand evolves and upgrades over time or
1:58:23your personal brand voice as well. You
1:58:26can build them for either. You can build
1:58:28them for your company brand voice or you
1:58:29can build them as a personal brand
1:58:31voice. I actually have a personal
1:58:32copywriter brand voice for some of the
1:58:35marketing work to be honest. And also
1:58:36when we think about the context of how
1:58:38useful a brand voice skill can be if you
1:58:41have a team of people that are working
1:58:43or a marketing department um it's easy
1:58:46as one person to stay in control of this
1:58:47stuff. But across a across a company and
1:58:50every output in the company you want to
1:58:51make sure is in line with your brand
1:58:53voice. Then we can make sure even
1:58:55everything's being human written and
1:58:57people on keyboards etc. for emails and
1:59:01anything. We can make sure we're
1:59:02applying our company brand voice to
1:59:04everything that goes out. Making sure
1:59:06we're also catching any AI slot that's
1:59:08being generated in the company that's
1:59:10really obvious. We can also do it for
1:59:12personal brands. So, anyone who's
1:59:14building a personal brand with a team,
1:59:15we can easily get our ghost writers or
1:59:18copywriters or marketing experts or
1:59:20sales team to actually apply our
1:59:23personal brand voice over the top of it.
1:59:25So you don't necessarily have to do
1:59:26these huge edits to everything and
1:59:28rewrite them in your way or vibe because
1:59:30your ghost writer has this extra tool or
1:59:33this lever to pull in your company. So
1:59:36here we go. We've got our brand voice
1:59:38here which we can see the skill file.
1:59:40We've got the main set of instructions
1:59:42on the right hand side here. Of course
1:59:44hopefully that's less than a th000
1:59:46characters. Um it's got then the
1:59:48step-by-step process of how your brand
1:59:51voice should work. the one rule that
1:59:53outranks you. Change how it's written.
1:59:55Never change what it says. So don't go
1:59:57and completely change the argument or
1:59:59rewrite everything. We're not
2:00:01fact-checking easier. We're improving
2:00:03the text basically in brand voice, not
2:00:06just rewriting things to be completely
2:00:08different or restating things before you
2:00:11write. So before it does anything, it's
2:00:12going to check the voice profile. It's
2:00:14going to check the band words, anti-AI
2:00:16swap, and reference examples. It's going
2:00:18to pick the mode first. So it's got
2:00:20different modes it can run in. an
2:00:22editorial, so for internal guides and
2:00:25opinion pieces that it's going to run
2:00:27on. Brand recommendation for when we're
2:00:29talking about other brands, and then a
2:00:31partner brief if we've got a partner
2:00:33that's advertising on the website, for
2:00:35example. Then got its how to rewrite
2:00:38exactly like key things that I want it
2:00:40to rewrite and look for. It's got the
2:00:42hard rules, things it should never do
2:00:44that are top of mind. No m dashes, of
2:00:46course. Uh, no formal connectives, no
2:00:49paragraphs longer than five lines. I
2:00:52might even chop that down to two lines
2:00:53to be honest. And then references the
2:00:55band words and AI slot list. Frequency
2:00:58of words and certain phrases in there
2:01:00like never use hidden gem more than once
2:01:02in an article or per thousand words. And
2:01:04also report back what you changed. Then
2:01:06we've got the voice profile that we did.
2:01:08That was that first analysis that we
2:01:10did. British seasonaire idioms for
2:01:12example, grammar, British English, etc.
2:01:16You can see how deep this is. Sound and
2:01:18smell before sight. So, we really
2:01:20viscerally describe things. Um, not just
2:01:23like tell them what they look like, but
2:01:25actually what they sound what they smell
2:01:26using all the t the senses in our our in
2:01:29our writing. Uh, points the reader at
2:01:32the staff, not the tourists. Deflates
2:01:34the luxury he is selling on a site.
2:01:36Probably don't want these deflations in
2:01:38there on a site that's called.vip.
2:01:40Uh, and it's got some of the vibe in
2:01:42there and everything like that. So, I'm
2:01:45not actually perfectly happy with this
2:01:46skill. It's got examples of like a flat
2:01:50AI written version by like exact
2:01:52examples of my writing in there. It's
2:01:55got those band words and AI slots built
2:01:57in there. So, we can see it's not
2:01:58perfect. I'm not actually going to save
2:02:00the skill yet because reading through
2:02:01I'm like, I don't want that bit in
2:02:03there. So, I'm just going to say, hey,
2:02:04can we remove this section from our
2:02:06voice profile? I don't mean to deflate
2:02:09the luxury to be honest with you.
2:02:12Um, so I'm going to get it to remove
2:02:14that section as well and just uh tidy
2:02:17things up before we even think about
2:02:19testing it. We've got our brand voice
2:02:22review. So this is the eval that I was
2:02:24talking about. Remember I was talking
2:02:26about it will go through and test and it
2:02:28will run with the skill here. Run this
2:02:30through my brand voice and then it's
2:02:31given the output of what comes out
2:02:34versus also a benchmark of how it's
2:02:36actually performing. the outputs. We can
2:02:38scroll through and see the like
2:02:40difference between without a skill and
2:02:42with a skill. So, you can go through and
2:02:44see the difference for yourself. And
2:02:45then we've also got the the benchmark
2:02:47about like with the skill it passes 70
2:02:5097% of the time based on the rules how
2:02:52long it takes. Without the skill, it is
2:02:55much quicker, like 28.9 seconds. With
2:02:57the skill, it's a lot longer there.
2:03:00126.5 seconds. The token burn is much
2:03:03bigger with the skill, of course, cuz
2:03:04we've got extra work going on in there.
2:03:07Um, it's just got rid of my eval. Um, so
2:03:10you can see, yeah, again, we've got it
2:03:12broken down by each evaluation here on
2:03:15the pass rate with the skill was good,
2:03:1650% without, the time it took, etc. Um,
2:03:20and then we've got what it's actually
2:03:21doing when we're running the skill. It
2:03:23passed all of these checks. When we ran
2:03:25without the skill, you can see no long
2:03:28no sentence longer than 25 word. It
2:03:30failed that check. No M dashes rewritten
2:03:32in the copy. It failed that check. So
2:03:34you can go through and have a look at
2:03:36like where it's falling down. Sometimes
2:03:38no sentence longer than 25 words seem to
2:03:41be the main problem. Otherwise it looks
2:03:43like it's performing really well from a
2:03:44a data point of view. And then we've
2:03:46removed that and put a new rule in
2:03:50luxury is the product and don't undercut
2:03:51it cuz that was something that I was
2:03:53unaware I was falling into that trap
2:03:56when trying to write more authentically.
2:03:58But we can save that skill now. We've
2:04:00got our brand voice skill essentially
2:04:02ready to rock and roll. So, we can
2:04:04actually test it out. Let's go and just
2:04:06jump into a new chat mode here because
2:04:08I've got here this blog post, five
2:04:11reasons to ski in Corvel. You can see
2:04:14now, oh, where's Claude gone? We could
2:04:16say, we could post that in there and
2:04:18say, can you use my VIP brand voice
2:04:24to bring this article up to our brand
2:04:28voice, please? Now, we could also run
2:04:31obviously the SEO widgets that we built
2:04:34on here, but you know, we're not going
2:04:35to do that now. I just want to show you
2:04:37literally the brand voice running on top
2:04:39of this article. This example that I
2:04:42would use as a test because what I would
2:04:44do now is I would go and use it a bit
2:04:46and I would note down some things I
2:04:48didn't like. I would note down, hey, it
2:04:49does this, I don't like that. You keep
2:04:51doing this. go and make 10 or 20 notes
2:04:54and then come back to Claude and say,
2:04:55"Hey, these are all the things I spotted
2:04:57in my brand voice skill. Can we go and
2:05:00update it to make sure it's behaving
2:05:02like this and not like this?" That could
2:05:04be specific words. It could be patterns
2:05:06of writing. It could be all sorts of
2:05:07things that you're just like, "Oo, I
2:05:09don't like the way it does that."
2:05:10Because again, this is an art more than
2:05:12a science that you need to probably be
2:05:14working back and forth as you use it.
2:05:16But again, you can see uh using the
2:05:18voice skill and getting to work on that
2:05:20piece of content. Okay, so it's gone
2:05:22into its editorial mode. Remember, it
2:05:24had different modes that it ran there.
2:05:26It realized, hey, you have an article.
2:05:27We just need to to be in editorial mode
2:05:29for this article. It's given me through
2:05:31what it changed. Rewritten your opening
2:05:33pattern is read a question first pivot.
2:05:35Well, credential drops rewritten from
2:05:38the original to that. And we've got all
2:05:39the changes it's made in there. Hard
2:05:41rules I enforced. Remove the four m
2:05:43dashes. is the semicolons, exclamation
2:05:45marks were too many flagged for some
2:05:47calls in there. So, it's actually given
2:05:48me back some things that I I need to
2:05:50actually check on. But essentially, our
2:05:52article's come back and it is rewritten
2:05:54in our brand voice. It's beautiful. You
2:05:56see, it's not as uh as longer
2:05:58paragraphs. We've got some of the way
2:06:00that I'm framing my questions and and
2:06:02just bit more flow in the way I would
2:06:04write in there. Is it 100% amazing,
2:06:06perfect, as if I sat down and wrote this
2:06:08whole thing myself? Probably not. I
2:06:10would probably go and give something
2:06:11like this another clean pass over to
2:06:13give it some extra stuff and some extra
2:06:15story from me because it was just a
2:06:17blank article I gave it. I think that's
2:06:18the one thing it's really going to be
2:06:19missing now is that original story from
2:06:22me which would either come from a
2:06:24transcript or me sitting and doing some
2:06:25work to add some stuff in. Maybe I'd
2:06:27even ask Claude to interview me about
2:06:29each bit to get some personal story and
2:06:32some flow back in there direct from me.
2:06:34But you know, but be back in time for
2:06:36Last Lift Home is the sort of thing we
2:06:37would be writing in there. And that's
2:06:39essentially how your voice skill works,
2:06:41you know. So, let's dig in. Look at what
2:06:43actually makes it worth your week. Love
2:06:44that. Brilliant. We've come a long way.
2:06:46We've come a long way. And now we've
2:06:49built our voice skill.
2:06:52We're really coming to the end of this
2:06:54course. Now, there's a couple areas that
2:06:56I want to take you through before we get
2:06:58right to the end here. One is what
2:07:00skills should you be building for
2:07:02yourself? It's a good question. And
2:07:04there's so many different skills you
2:07:05could be building and a lot of people
2:07:07aren't really sure where to start. Yes,
2:07:09we built that first SEO department, your
2:07:11brand voice skill. Next, where do you
2:07:13go? Well, I always think start with the
2:07:15pain. What tasks do you actually help
2:07:17hate doing? What do you not like doing
2:07:19in your work, life, business that makes
2:07:21you go [sighs] slows you down on a
2:07:23Monday morning or, you know, leaves you
2:07:25feeling drained. That's a great place to
2:07:27start. So, I've actually got this prompt
2:07:28here that you can use to help you work
2:07:29out which skills are actually worth
2:07:31building for your work, not just the
2:07:33impressive fancy ones like a LinkedIn
2:07:35writer or something. It's going to
2:07:37interview you here. Um, and it's going
2:07:39to go through the process of
2:07:40interviewing you honestly pulling some
2:07:42things out and then giving you a report
2:07:44of what you should actually build and
2:07:46how to go about building them. Uh, so
2:07:48that's going to be tailored to you. If
2:07:50you take this prompt and run it in
2:07:51Claude, that'll um that'll take you
2:07:53through that process. I've also just
2:07:55whipped up 101 skill ideas kind of based
2:07:58on a lot of the skills that you saw in
2:08:00my account. Um, and I also ran an
2:08:02analysis across the internet. I was
2:08:04like, look, what are the biggest pains
2:08:06people have in their business that could
2:08:08be solved with Claude Skills? Go and
2:08:09look in Reddit, forum, social media to
2:08:11see what are people moaning about that
2:08:13we could actually build Claude skills to
2:08:14help them. Cash flow narrator, for
2:08:16example, helping people understand their
2:08:18cash cash flow for a month. A quote
2:08:20builder, an invoice generator. I have an
2:08:22invoice skill in my account where I just
2:08:24give it um information on the job I did.
2:08:27It might even know the job I did and it
2:08:29just pulls out a branded correct
2:08:31perfectly templated invoice for me.
2:08:33Super simple but super powerful. There's
2:08:35so many different options in here to to
2:08:37spark your ideas for what you should be
2:08:40building essentially. So this document
2:08:41again, you can get the link below which
2:08:43is in my AI recipes here. Click the top
2:08:46one there. The or might not always be
2:08:48top but complete flawed skills course
2:08:50guide. This is going to give you
2:08:51everything that we've gone through in
2:08:53this video today, including everything
2:08:55we just built that you need. It's got
2:08:57the the three ways of building skills.
2:08:59Remember, just wrapping a prompt,
2:09:00building a new workflow, using recorder
2:09:03skill. We've gone up the five levels, 1
2:09:052 3 4 five different levels. We've
2:09:08talked about the two different types of
2:09:09skill, a deterministic and a
2:09:11non-deterministic skill. We've built an
2:09:13orchestrator, a manager on top of them,
2:09:16and we've built a brand voice skill as
2:09:18well. So really now it's up to you is
2:09:21what are you going to build next? Go and
2:09:23use that prompt that I just gave you in
2:09:24the documentation
2:09:27here. Um and start building for
2:09:29yourself. Everything's here ready for
2:09:30you to go. This is the number one Claude
2:09:33skill that pretty much underpins
2:09:34everything I do. It links to all other
2:09:36skills and outputs I create. And the
2:09:38weird thing is I don't really see anyone
2:09:40else talking about this, showing this,
2:09:41building this skill that gives you an
2:09:43unique advantage over everyone else
2:09:45using Claude. and I'm going to show it
2:09:47to you today and then walk you through
2:09:49step by step how to build it. Now, you
2:09:53can also get everything I talk about in
2:09:54this video. There's a link in the
2:09:56description where you get my AI recipe
2:09:57vault. You can go to build your brand
2:09:59print skill. This is what we're talking
2:10:00today, a brand print skill that I've not
2:10:03seen anyone else use. And in here, we've
2:10:05got step-by-step recipes, including some
2:10:07of the all all of the prompts I'm going
2:10:09to use in this video. So, you can get
2:10:10that in the link below. But first, I
2:10:13want to show you how it works so we know
2:10:15what we're building towards without
2:10:17creating this brand print skill. You're
2:10:19going to get these generic outputs from
2:10:21Claude. We've all seen these before.
2:10:23They're just boring and they look like
2:10:25you've just printed this straight out of
2:10:27AI, which no one wants to see anymore.
2:10:30Whereas, when we actually build the
2:10:31brand print skill, what we get are
2:10:33beautifully branded outputs across
2:10:36everything you do, included. You can see
2:10:38this is a Word document. It's got my
2:10:39logo. It's got all my brand colors on in
2:10:41there. You might have seen some of my
2:10:43teaching dashboards in previous YouTube
2:10:46videos. My brand skill builds all of
2:10:48that as well. It can run across bulk
2:10:51producing invoices in the right
2:10:53template. It can even create my GEO
2:10:56audit mapping with all of the right
2:10:58colors in my brand as well as even
2:11:01building actual websites and apps that
2:11:03all carry my logo, my brand colors, even
2:11:06my font and feel for everything. The
2:11:08brand print skill pretty much links to
2:11:10every single output I make in Claude.
2:11:13And so, let me show you exactly how
2:11:16we're going to make this. Now, to start
2:11:18off, I want you to use the Claude
2:11:20desktop app cuz it's just going to give
2:11:21us way more power. And I want you to
2:11:23actually select the co-work setting in
2:11:26the middle here because this is going to
2:11:28allow Claude to actually run more
2:11:30autonomously and extract and build this
2:11:32skill, which is a three-step process.
2:11:34The next thing we need to do is we need
2:11:36to collect our brand assets. That means
2:11:38logos, fonts, your brand guidelines,
2:11:42imagery, any other assets you have. I
2:11:44like to put them in one folder on my
2:11:46desktop to allow Claude to access them.
2:11:48So, you can see here, I'm going to new
2:11:50project or folder. I'm going to go and
2:11:53add a folder here. I don't want to
2:11:55create a whole project because this is a
2:11:58one-time job that we're actually going
2:11:59to do. And then I'm going to go and
2:12:01select my brand assets folder on my
2:12:04desktop here and open that and allow
2:12:06Claude to access it. Next, I need to go
2:12:09and get my initial prompt. So, step one,
2:12:11get all your assets in one place in one
2:12:13folder. Step two, we need to run our
2:12:16brand audit. We can take this prompt
2:12:17here, which is in the AI recipe vault
2:12:19link I I told you about in the
2:12:21description. You're a brand systems
2:12:23analyst. I want you to audit my brand
2:12:24assets and produce a brand specification
2:12:27precise enough that another AI could
2:12:28apply my brand. It's actually going to
2:12:30go through all of the assets and extract
2:12:34everything we need in here to build our
2:12:36skill. So, we're going to take that
2:12:38prompt and we're going to place it in
2:12:40here. And actually, it says my assets
2:12:44are in the folder attached.
2:12:48And I'm going to run that. Now what's
2:12:50interesting while we let that work if I
2:12:52show you my actual skill which is this
2:12:55Elliot brand print skill what you'll
2:12:57start to see is actually it's extracted
2:13:00all of this information that I've just
2:13:02given it and also it's got reference
2:13:05files so it's also including all of the
2:13:07assets all of the font files all of the
2:13:10actual logo files and a reference to my
2:13:12brand guidelines so it's super powerful
2:13:14what we're doing here but the first step
2:13:15is running this audit now what you'll
2:13:17see on the right hand side here is the
2:13:20power that's going on here. It's listing
2:13:21the assets in there. It's extracting the
2:13:24color palettes, the ROS, the fonts, and
2:13:26actually using the brand guidelines to
2:13:27to map all of that typography hierarchy
2:13:30and the font file coverage, cataloging
2:13:33all the different variants and the way
2:13:34they should be used. So, it's not just
2:13:37throwing random colors and fonts and
2:13:39logos together, which is the real key
2:13:40here, which is why we need to run this
2:13:42step. If you wait a few minutes, you're
2:13:44going to get this output. You're going
2:13:45to get your brand audit and you're going
2:13:47to get an inventory of everything it's
2:13:49found in there, what they do. You can
2:13:50see all the fonts, logos, type come out,
2:13:53any flags that it needs that it's not
2:13:54sure on, the different colors that it's
2:13:56pulled from here. And you can see some
2:13:58of them are inferred, like we need to
2:14:00maybe go and check these to make sure
2:14:01they're 100% correct. But you can see
2:14:04it's pulled out those greens and those
2:14:05inks and cream colors that I was using
2:14:07on my brand. We've got all the safe
2:14:09pairings cuz I got brand guidelines
2:14:10about what goes with what. It's actually
2:14:13listing out all of that information as
2:14:15well as the typography, the Thunderbolt
2:14:17LC font, host hope, host Grotesque light
2:14:20font I use, as well as, you know,
2:14:22everything that's in there. It's got
2:14:24some missing things that we actually
2:14:25need to go and clarify. So, at the
2:14:27bottom, we need to actually go and
2:14:29answer all of these questions to
2:14:30actually confirm some of this
2:14:32information. Otherwise, it's just going
2:14:34to be a little bit off and it's going to
2:14:35be a bit frustrating. So, go through and
2:14:36answer all of these questions. Now, once
2:14:39you've gone through and answered those
2:14:40questions and corrected the output, it's
2:14:43time to build the skill, which we've got
2:14:45another super important prompt here.
2:14:47I'll just wrap it up there nicely so we
2:14:48can see it. We've got the next skill.
2:14:50The order is correct. Now, build me a
2:14:52claude skill called brand print that
2:14:54applies this brand to any output I ask
2:14:56for. You can see it's actually going to
2:14:58structure how the skill should be. So,
2:15:00it's got the the main overarching
2:15:02instructions of how this works, the main
2:15:05prompt here, the different assets that
2:15:07it's actually got and any references. If
2:15:09you have some references of like word
2:15:11documents, PowerPoints, presentations,
2:15:14etc., really good if you put your
2:15:15PowerPoint presentations in there, can
2:15:17pull those structure out from your brand
2:15:18assets as well. And then it's going to
2:15:20go and build the skill with all these
2:15:22rules, etc. in there. So, I'm going to
2:15:25copy this prompt and I'm going to go
2:15:27back to my chat. You'll see you should
2:15:28have your full brand specification V2
2:15:31here, which is super helpful anyway to
2:15:33have in your business. But I'm going to
2:15:34take that prompt, paste it in, and we're
2:15:36going to run that. This is going to take
2:15:38all of this analysis and audit and
2:15:40things it's pulled out, all these files,
2:15:42and build it into that Claude skill I
2:15:44showed you a second ago. Now, it takes a
2:15:46few minutes to actually run this
2:15:47exercise and build the skill for you.
2:15:49But once it's done, you'll see you
2:15:51should get this little brand print skill
2:15:53built. You can see when we open it up on
2:15:55the right hand side, it's got all of my
2:15:57fonts in there. It's got all of my logo
2:15:59icons. It's got references. Um, and it's
2:16:02got exactly how all of this should
2:16:04operate and how everything works. So,
2:16:06all we have to do is click save skill.
2:16:08We can save the skill and start testing
2:16:10it and using it, which is what we're
2:16:11going to do next. So, if I actually go
2:16:14back out, I'm going to go into my second
2:16:16brain here. This is pretty much where I
2:16:17do all of my work in Claude. It's got
2:16:19all of my context, all the information
2:16:21about my business, what I do, everything
2:16:23underneath, ready to rock and roll. So,
2:16:26this makes life super easy when I say
2:16:27something like, "Can you create a uh
2:16:31branded invoice for the king of England
2:16:35for joining my AI mentorship program?"
2:16:38If I just run that because obviously my
2:16:40second brain knows all about my business
2:16:42and, you know, blah blah blah. I don't
2:16:43have to give it all my address and
2:16:45everything or details about me. It
2:16:47should go and find the address for the
2:16:49king of England, etc. If it's thinking
2:16:50cleverly and build me my branded output.
2:16:54And I could do this for, as I said, web
2:16:56pages, Word documents, PDFs,
2:16:58spreadsheets, PowerPoints, anything I'm
2:17:00putting out. And here we go. Here it
2:17:02comes out. My logo, my colors, invoice.
2:17:05It's got all the billing details on
2:17:07there. Even has the nice uh invoice
2:17:09number that's coming out, which is some
2:17:11of the other settings I set up. But
2:17:12mainly what I want you to see is I just
2:17:14said, "Hey, create me an invoice. create
2:17:16me a branded invoice and it goes and
2:17:17auto applies my brand every time. Now,
2:17:20the other thing I want to say at this
2:17:22point is claude design. A lot of people
2:17:25will be like, well, why don't I just use
2:17:26claw design earlier? I can give it all
2:17:28my brand system in there and just have
2:17:31it work there. Well, this is actually
2:17:32claude design is slightly detached like
2:17:34it doesn't link in. I'm not working in
2:17:36there all every day through my work and
2:17:38creating outputs. I might move in here
2:17:40for very particular like slides for
2:17:43example. I I gave a presentation at AI
2:17:46for manufacturing. I gave a keynote
2:17:47speak. I wanted to create actual
2:17:49specific slides around that, not just
2:17:51fast outputs. So, I'm using claw design
2:17:54very intentionally when I have a serious
2:17:56design task that I need to focus on. But
2:17:58for everything else, I'm just using my
2:18:00brand print skill to get fast outputs
2:18:04when I'm working on all the boring stuff
2:18:07or even YouTube videos for my business.
2:18:09That's what I'm referring to. I built 43
2:18:12different Claude skills for marketing.
2:18:15And if they all disappeared tomorrow,
2:18:17there's five that I've rebuilt
2:18:19immediately that do the highest value
2:18:21work. They run this YouTube channel,
2:18:23help me grow my email list, and have
2:18:26even been reviving my dead LinkedIn.
2:18:29I've gone from post getting 500 odd
2:18:31impressions and no sort of comments to
2:18:34getting 7,000 impressions and plenty of
2:18:37comments and interaction engagement on
2:18:38my posts. And so we're going to start
2:18:40with this first skill. And I might as
2:18:42well use it on this actually dead
2:18:45LinkedIn post that didn't perform very
2:18:47well before I actually started using
2:18:49this first skill, which I call my
2:18:51director of psychology. Now, if you've
2:18:54studied anything about marketing and
2:18:56content, you'll know that psychology is
2:18:58an important part. You'll have heard
2:18:59about hooks, curiosity loops, and the
2:19:02zyic effect. And honestly, I went deep
2:19:04on this for probably hundreds of hours
2:19:06and ended up with information overwhelm
2:19:08and I never really translated any of
2:19:10this amazing information into actually
2:19:12implementing it in my content because I
2:19:14was just overwhelmed. And so what I did
2:19:15was I took the 15 main psychological
2:19:18principles that I wanted to use in my
2:19:20marketing and started applying them to
2:19:22my work through the use of Claude. And
2:19:24the amazing thing was I actually saw it
2:19:26working as I showed you. I started going
2:19:28from these crappy numbers to actually
2:19:30these great numbers even as just a test
2:19:32on LinkedIn and now I apply it across
2:19:34everything I do because I've baked it
2:19:37into this director of psychology skill.
2:19:40So all of those principles are used and
2:19:42overlaid on my work. Now it doesn't
2:19:44write the content for me and produce the
2:19:46scripts for me etc. I produce them and
2:19:48then I work with this skill to actually
2:19:51get our output dialed and introduce
2:19:53these psychological improvements to keep
2:19:56retention through my content. Now, what
2:19:58you'll get with this skill is actually
2:20:00some key changes. It's still working
2:20:02here. As you can see, you'll get some
2:20:05critical feedback about where you're
2:20:06losing people, where you're waffling,
2:20:08where you're not really actually
2:20:09drilling home what you should be doing.
2:20:10And the five changes or the key changes
2:20:12you need to actually make. I need to
2:20:14change my opening sentence, my hook. I
2:20:16need to make my count and my numbers
2:20:18work harder. Move some of the best
2:20:20lines. It finds some of the best lines
2:20:22that are hidden and moves them up and
2:20:23just tells me what to delete because
2:20:25it's just fluff and not adding anything.
2:20:27While this runs its full tear down, you
2:20:29can actually get all of these skills I'm
2:20:31talking about in this video today,
2:20:32including this first one in my AI recipe
2:20:34vault. Free link in the description.
2:20:36Find the five clawed marketing skills.
2:20:38Open this up and you can scroll down and
2:20:40you can actually rightclick download
2:20:42these skills to use for yourself. Back
2:20:44over here, our final tearown is actually
2:20:47done. And I actually love this because
2:20:49it gives me some interesting information
2:20:51about my my post or my script or
2:20:52whatever it is in terms of word length,
2:20:55word count, the readability level to
2:20:57make sure it's not just complete
2:20:59unintelligible
2:21:00corporate speak. Then we've got actually
2:21:03the first tow down. What's going wrong?
2:21:05Like where is the bad stuff that we need
2:21:07to delete or rewrite? What's not working
2:21:09in this post? I love this one. the only
2:21:11bit that the person can actually act on
2:21:12and it's the thinnest thing here. It's
2:21:14kind of tagged onto the end. You've made
2:21:16the whole argument delivered halfway and
2:21:18we're not really delivering anything
2:21:20interesting through this post. Now, it
2:21:22actually goes for the rewrite and the
2:21:24re-edit, not rewrite, re-edit. It gives
2:21:27us the initial input. It's cut that
2:21:29initial line. It's telling us why it's
2:21:31doing something. Plus, if you only do
2:21:33one thing, it gives you that one big
2:21:35change you can make. Now, I don't always
2:21:37use the visual output. Sometimes working
2:21:38just in the chat back and forth works
2:21:40really really well, but I use the tear
2:21:42down as well if I'm really struggling to
2:21:44visualize how to reorganize a post that
2:21:46I've written. Now, skill number two has
2:21:48helped me engineer some of the most
2:21:50viral videos and posts I've had on the
2:21:52internet. You can see here, this one
2:21:54actually, it was responsible for 348,000
2:21:57views and some decent revenue off this
2:21:59channel. It was also then a onetwo punch
2:22:02because the trend ended in another
2:22:04follow-up video with another over 100k
2:22:07views because it managed to pair my
2:22:09niche AI with something in popular
2:22:12culture to jump on a trend that no one
2:22:14else was spotting. And this is a skill
2:22:17called trend jack. And when you run the
2:22:19trend jack skill, it's going to ask you,
2:22:22you know, what mode like what channel
2:22:24are we looking for? Do we want to go
2:22:25extra depth, etc. But what it's actually
2:22:27doing, it's not just finding news. Like
2:22:29anyone can do deep research on news.
2:22:31It's actually using something I like to
2:22:33call the play call. And it's helping you
2:22:35find a topic and a shape and an angle
2:22:38here that is unique to anyone else. So
2:22:41anyone can talk about a piece of news or
2:22:43a topic, but we want to give it a
2:22:45particular angle and map that back to
2:22:47our content. And this is exactly what
2:22:48the trend jack skill does. It has a
2:22:51first run setup as well. So, if it's the
2:22:52first time you've ever run it, it's
2:22:54going to help you edit and improve the
2:22:56skill for your niche, for example. And
2:22:58now, what levels this up as well? If you
2:23:00can pair it with a connector like Ampify
2:23:02or something that helps it scrape even
2:23:04more of the web, it's going to give you
2:23:06even more detail, but even on its own
2:23:08with deep research and web search, it's
2:23:10going to do a really great job. Let that
2:23:12run for as long as it takes. It might
2:23:13take a few minutes cuz it's going deep.
2:23:15It's not just surface level. And it's
2:23:17going to give you your four
2:23:18opportunities ranked on the left hand
2:23:20side here, which is is really nice. But
2:23:22I really value on the right hand side
2:23:23here this full report that I get my full
2:23:26trend brief for the day. We've got the
2:23:28pulse just a quick update on what's
2:23:30going on across my industry. Claude
2:23:32Claude Co-work went to all paid plans on
2:23:35mobile and web which is a nice little
2:23:37upgrade for a lot of people. This one as
2:23:39well that some people might have missed.
2:23:40Claude can now send Gmail. They can send
2:23:43emails from Gmail. All little pieces of
2:23:46content that are perfect for me. It also
2:23:47gives me a little pulse across all of
2:23:49the other channels that I might be uh
2:23:51might be looking at and gives me some
2:23:53outstanding like what are the outliers
2:23:55here. You can see this video here
2:23:57actually is massive outlier in the
2:24:00industry which is an interesting angle
2:24:02rather than just you know your standard
2:24:04sort of tutorial walkthroughs. It's
2:24:07actually using AI and actually testing
2:24:09what happens when you let them unleashed
2:24:11which I really love that angle. Plus
2:24:13then we get our Tik Tok breakdown, our
2:24:15Instagram breakdown. This thing goes on
2:24:16and on before we get to the real key
2:24:18stuff, the video opportunities rank. And
2:24:21as you can see here, number one, recast
2:24:23the marketing skills video as a
2:24:25constraint experiment. So I could take
2:24:27this sort of video and repackage it,
2:24:29re-angle it, and say, "Hey, in my next
2:24:32video, I let my five top clawed
2:24:34marketing skills actually run unattended
2:24:36for a week. This is what happened." Now,
2:24:38the first two skills we talked about,
2:24:40they help increase quality. They help
2:24:42improve ideiation. The third one is more
2:24:45about volume. You see some of these big
2:24:46grazers like Alex Horoszi, Gary Vee,
2:24:49they're pumping out dozens of clips
2:24:51across their channels every single day
2:24:53and they preach that you need to be
2:24:55doing more, which is fine for those of
2:24:57you that have time or the funds to have
2:24:59a whole marketing team. But what if
2:25:01you're short on time and you don't have
2:25:02a big budget to hire other people? Well,
2:25:04this is where my Clipper Army skill
2:25:06really takes on a new level. You can
2:25:08give it a long form video. In my case,
2:25:11sometimes I get my phone and when I'm on
2:25:12a podcast, I have it filming right here
2:25:14or I just do long pieces to camera. I
2:25:17can give it that video and then it goes
2:25:20through and starts clipping out the best
2:25:22pieces of content. You can see here, for
2:25:24example, it takes one year of focus. You
2:25:26pick one thing and do so much work on it
2:25:30that you can't not succeed. And it
2:25:32doesn't just clip randomly. It actually
2:25:34goes through with some really clear
2:25:36rules. Never cut midword. Keep it
2:25:38chronological. Don't reorder some of my
2:25:40sentences. Never touch the source file.
2:25:42I don't want it editing the actual main
2:25:45file. I want it pulling out clips. Now,
2:25:47I've given it loads of editing rules,
2:25:49but the real big thing, the hot thing
2:25:51about this skill is it step two is
2:25:53finding the hot moments. It's looking
2:25:56through your video to find the strongest
2:25:58lines and the strongest moments that
2:26:00stand alone, that open on something
2:26:01important, that resolves a point, that
2:26:04strikes tension, has a contrarian take,
2:26:07is a story, or there's a lift in
2:26:09delivery and the tone that we're doing.
2:26:11It pulls them out, and then runs the
2:26:14tight edit that you just saw to deliver
2:26:16them for me straight into my folders.
2:26:19So, from here, all I have to do is
2:26:20publish the clip. Maybe I'll add
2:26:22captions within the actual Tik Tok or
2:26:25Instagram app or whatever it is. But
2:26:27these are ready to go. 15 new clips. You
2:26:29can see here it's actually running the
2:26:31process for us now. And it's depositing
2:26:33the clips into my folder here. So it's
2:26:35up to 16 now. We had 13 at the
2:26:38beginning. So it does take a little
2:26:40while to actually run. But to be honest,
2:26:42I've been running this process at the
2:26:43moment and it's not like it's tapping
2:26:45out my usage at the moment. It's not
2:26:47used burn through all my usage cuz this
2:26:49skill is really well defined. I'm not
2:26:50just telling Claude to go and edit the
2:26:52files. Actually has a strict process
2:26:54that it's following. So that's how
2:26:55you're going to generate more volume in
2:26:57your marketing. Now, one thing that
2:26:59people tend to miss is they post on
2:27:01social media, they post their content,
2:27:02but then they don't actually capture
2:27:04that attention. Whereas this next skill
2:27:06is going to help you do what I've done
2:27:08and actually build an email list that's
2:27:1029,000 approaching 30,000 subscribers,
2:27:13which is an absolute dream. I can't
2:27:15believe I've done that. even off a
2:27:17YouTube channel that itself still only
2:27:19has 43,000 subscribers. So, my
2:27:22conversion rate to my email list is
2:27:24incredible. That's really all down to my
2:27:27lead magnet skill. You can see in here
2:27:30though, it's going to help with lead
2:27:31magnet ideation and creation. Now, as
2:27:34you'll see in this video, I use
2:27:35something called the AI recipe vault to
2:27:38actually give people free things and
2:27:40they can also sign up to my email
2:27:42newsletter in the process. That's how
2:27:44this whole process works. But I wouldn't
2:27:45have come up with this idea without
2:27:47working with a skill like the lead
2:27:49magnet. Now, the real key to making this
2:27:52work is to run it on some sort of
2:27:53context for your business. If you just
2:27:55run it in a blank chat, it's not going
2:27:56to work very well. I'm actually running
2:27:58this on top of my deep context, and it's
2:28:00got information about what I do, my
2:28:03ideal customer profile, some of my hits
2:28:06in the past, some of my stories to
2:28:08actually build tailored ideas for my
2:28:11newsletter. And what we get out is not
2:28:13just a list of PDFs that we can write.
2:28:14No one wants a link, a PDF that they can
2:28:16just read through anymore. They want
2:28:18something more interactive, more useful,
2:28:19something they can actually do. It's
2:28:21either going to be some sort of
2:28:22calculator, uh, question diagnostic, or
2:28:25even something they can implement like
2:28:26the non-technical founders prompt vault
2:28:29here. I know prompt vaults go really
2:28:30well. I have my mega prompt chest, which
2:28:32people absolutely love. Well, we've got
2:28:33the 5day build.
2:28:36Uh, the 10day automation sprint. This is
2:28:38actually quite interesting here. I think
2:28:4010 days, 10 working clawed workflow
2:28:42automations. I can actually guide you
2:28:44through in an email series, video series
2:28:46or something. Essentially a sort of free
2:28:48build challenge where we build one
2:28:50automation a day like some of the stuff
2:28:52that I have running here. I mean, I'm
2:28:54sure people would love to see my guest
2:28:56radar build, for example, which searches
2:28:59for great podcasts and writes outreach
2:29:01pitches, etc. Showing people how to
2:29:03build this sort of stuff would probably
2:29:04go down really, really well. Let me know
2:29:06in the comments if you actually want
2:29:08something like that. I can certainly
2:29:09build the this 10day automation sprint
2:29:12where we walk through one automation
2:29:14build a day for your business. And and
2:29:16in fact, we actually had some risk notes
2:29:17at the bottom. It was like, it might be
2:29:19too good to give away. Maybe you should
2:29:21actually charge for this stuff. But hey,
2:29:23that's a bridge we can cross later
2:29:25because I want to get on to my final
2:29:27skill for the day. Now, we've improved
2:29:29the quality with psychological triggers.
2:29:32We've improved topic selection and
2:29:34trends. We've increased volume. We've
2:29:37generated leads to our email. What's
2:29:40next? This final one is how are we going
2:29:41to take these email newsletter
2:29:42subscribers, potentially turn them into
2:29:44paying customers or build a sense of
2:29:46community around your brand? Because
2:29:48it's called email sequences. And this is
2:29:51something I'm missing. If you signed up
2:29:53for my email newsletter, you'll know you
2:29:55get my free AI recipe vault, for
2:29:58example, and then you'll get a welcome
2:30:00email about me. And then it doesn't
2:30:01really do anything else. I don't really
2:30:03nurture or onboard people or give them
2:30:05more information about me or what they
2:30:08can expect. This has got everything
2:30:10built into it to help us. So, I'm going
2:30:12to say email sequences. Can you help me
2:30:15build out the rest of my welcome email
2:30:17series for people who have signed up to
2:30:19my AI recipe vault? So, we're going to
2:30:20run that and it's going to start to walk
2:30:22us through it. When you actually break
2:30:24down into this skill, you'll see it's
2:30:25got different sequence types it can run
2:30:27in here. It's got the welcome sequence,
2:30:29which is what we're going to do now.
2:30:30nurture, conversion, launch. If you're
2:30:33launching a new product or service,
2:30:35re-engagement, I've got a lot of cold
2:30:36subscribers on my email list, maybe I
2:30:38need to actually start thinking about
2:30:39how I can do a re-engagement to see if
2:30:42these people actually want to still
2:30:43receive emails from me or not. And post
2:30:45purchase. So, this is a really deep
2:30:47skill. You'll see here as well. What's
2:30:49the lead magnet? Well, we should have
2:30:50just built that in the last step. So,
2:30:52that should be an easy thing for me. I
2:30:53actually run this on top of my folder
2:30:55structure, which is my like second
2:30:56brain. It has a connector into kit as
2:30:59well, my email software. So, it can pull
2:31:02all of this information it needs to
2:31:03actually start outlining my welcome
2:31:05series. So, it's going to ask you some
2:31:07questions through this, like what are
2:31:09you trying to do? How many emails? Let's
2:31:11go with two more. And even for me, it
2:31:13could actually put them straight into
2:31:14Convert Kit, but we're just going to run
2:31:16them as markdown here as well because I
2:31:17actually want to outline and actually
2:31:19write a lot of these myself. What's cool
2:31:20is it's actually used my email
2:31:22connector. And by the way, you can use
2:31:23like the Mailchimp connector, etc. if
2:31:25you use a different email service to
2:31:27actually go into ConvertKit or Kit as
2:31:29it's now called and look at my welcome
2:31:31email sequence to actually analyze what
2:31:33is in there at the moment and then be
2:31:35able to build from that. I don't even
2:31:36have to copy and paste anything in which
2:31:38is super powerful because that's where
2:31:40connectors actually start automating
2:31:42some of this work for you and this
2:31:44analysis and this output. I'm actually
2:31:46going to do my next video is going to be
2:31:48my top marketing connectors hint one of
2:31:50them is going to be kit. So, keep an eye
2:31:52out for that video.

About this transcript

This page contains the full transcript of CLAUDE SKILLS FULL COURSE 2.5 HOURS (Beginner to Expert) by Eliot Prince, generated from the public captions YouTube serves with the video. The transcript has 31,595 words across 4,366 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.