0:02 Claude just got even more powerful with
0:04 the recent Opus 4.6 update. And if
0:05 you're still starting from scratch with
0:07 every conversation, reexplaining
0:08 everything, pasting your formatting
0:10 rules in, pasting your brand guidelines
0:12 in, and then doing the same thing all
0:14 over again tomorrow, then you're leaving
0:16 serious productivity on the table. And
0:18 here's the thing, even after all that
0:19 effort, the outputs still aren't
0:21 consistent. Today, I'm going to show you
0:22 how to go from this super long,
0:25 exhausting prompt to this, all while
0:27 maintaining consistency and quality.
0:29 We're not just prompting anymore. We're
0:31 building a personal AI operating system.
0:32 And the feature that makes this possible
0:34 is called claude skills. And by the end
0:36 of this video, you'll know what claude
0:38 skills are, why they're such a big deal,
0:39 and you'll know how to build your own
0:41 skills, turning any workflow you're
0:43 already doing into a portable, reusable
0:45 system that works anywhere inside of
0:52 So, what actually is a Claude skill?
0:53 Here's the simplest way to think about
0:56 it. A skill is like an instruction
0:58 manual that tells Claude how you want a
1:00 specific task done, what steps to
1:01 follow, what tools to use, what
1:04 standards to meet, so it executes the
1:05 same way every single time
1:07 automatically. Think of it like a
1:09 playbook. A coach doesn't reexlain every
1:11 play before every game. The playbook
1:13 exists, the team knows it, and they just
1:15 run the right play when the moment calls
1:17 for it. That's exactly what skills do
1:18 for Claude. Now, technically, a skill is
1:21 a folder, and inside that folder is a
1:23 required file called skill.md. That's
1:25 where your instructions live. And as an
1:27 option, you can include scripts,
1:29 reference documents, and assets like
1:31 templates. Before skills existed, you
1:33 had two options. Either you repeated the
1:35 same detailed instructions in every new
1:38 chat, which is slow and inconsistent, or
1:39 you used a clawed projects with custom
1:42 instructions baked in. And projects are
1:43 great, but they have two limitations
1:45 that start to hurt as your workflow
1:47 grows. First, they're not portable. So
1:49 the instructions that you put inside a
1:51 project only lives inside that specific
1:53 project. So when you start a new chat in
1:55 a different project, you're starting
1:57 without that context. The second thing
1:59 is that they're not stackable. So you
2:01 might have one project for copywriting
2:03 and another for data analysis, but you
2:05 can't directly combine them into a
2:07 single shared workspace. They operate
2:09 side by side, but not as one unified
2:12 system. Skills solve both problems. You
2:14 take the instructions for how you want
2:16 something done, package it into a
2:18 folder, and then upload it into Claude.
2:20 Now, once activated, that skill becomes
2:22 portable across your account. You can
2:24 use it in regular chats, inside
2:26 projects, in cloud code, and in API
2:28 workflows you set up instead of being
2:30 locked to just one project. Now, there's
2:32 three categories of skills. You have
2:34 official skills built by Anthropic,
2:36 things like document creation, theme
2:38 styling, internal communications. Then
2:40 you have custom skills that you build
2:42 yourself. And you have community skills
2:44 shared by other users. But be careful
2:45 with these. I'll talk more about
2:47 security later. All right. So to get
2:49 started with actually using cloud
2:51 skills, the first thing we need to do is
2:53 we need to go to settings and then under
2:55 capabilities then you're going to see
2:58 something called code execution and file
2:59 creation. You need to toggle this
3:02 setting on. Once this is toggled on,
3:03 you're going to scroll down lower on the
3:05 page and you're going to see skills just
3:07 below memory. Now when you click on
3:09 example skills here, you're actually
3:11 going to see a list of anthropic created
3:13 skills and you can toggle on the skills
3:14 that you actually want. But here's a
3:16 quick tip. I recommend only having a
3:18 couple of these that you're actually
3:20 planning on using toggled on at one time
3:23 because otherwise Claude might actually
3:25 get confused and call a skill that you
3:27 don't want inside your chat. So, just be
3:28 mindful of that. Now, here's another
3:30 quick tip that's really important.
3:31 Skills are supposed to trigger
3:33 automatically when they're contextually
3:35 relevant, but honestly, they don't
3:37 always based on my experience. So, I
3:40 recommend going into your settings under
3:43 general and then under your profile and
3:45 adding a line that's telling Claude to
3:48 always consider relevant active skills
3:50 when responding. That one sentence makes
3:52 a huge difference in how consistently
3:54 the skills activate. Okay, so now that
3:56 you have the foundation, let's put this
3:58 into action with our first use case. One
3:59 of the fastest ways to see skills in
4:01 action is to take an official skill that
4:03 already works and extend it with your
4:06 own standards. No building from scratch,
4:08 no complex setup, just adding your brand
4:10 on top of what's already there. All
4:12 right, here's the scenario. Claude has a
4:14 built-in skill called theme factory, and
4:16 it applies pre-built visual themes to
4:18 anything Claude creates, like landing
4:21 pages, documents, or web components.
4:23 It's got about 10 themes built in, and
4:24 they look decent, but they're generic.
4:26 They don't look like your company. So,
4:28 let's fix that. So, let's say we have a
4:30 company called AuraNC Digital, a tech
4:33 forward marketing agency. I've got three
4:34 things ready when it comes to the look
4:36 of the brand. I have the company logo, a
4:38 brand color palette with hex codes, and
4:40 our font pairings. And that's all we
4:42 need to get started. Now, before we
4:44 actually create our first skill, I want
4:46 to make sure that under settings and
4:48 under skills, we have our skill creator
4:51 skill toggled on. Otherwise, this is not
4:52 going to work. So, just make sure you do
4:54 that. Now, I'm going to create a new
4:56 chat. And then in that chat, I'm going
4:58 to upload the word document that has our
5:01 logo, colors, fonts, etc. And then I'm
5:02 going to give Claude this prompt. Read
5:04 the theme factory skill. Then create a
5:06 new custom skill called Aura Sync theme
5:09 that extends it with Orura Synync
5:10 Digital's brand standards. And then I'm
5:12 going to give it the brand standards in
5:14 the in the word document. And then
5:16 basically say you know what the look of
5:17 this should actually include dark
5:21 background accent colors etc. And then I
5:23 say use the skill creator skill to
5:25 ensure proper formatting. Now watch
5:27 what's happening. Claude first reads the
5:29 theme factory skill to understand the
5:31 existing theme structure. Then it reads
5:33 the skill creator skill to make sure
5:35 that it's properly formatted and then it
5:37 analyzes our brand, the fonts, the
5:39 colors, the logo and builds a new skill
5:40 that layers the brand on top of the
5:42 existing theme system. And after it's
5:44 done working, here is what we actually
5:47 get. The skill.md file. Now, here's a
5:48 quick thing that we want to pay
5:50 attention to here is the description
5:52 field in the skill file. This is one of
5:53 the most important lines in your whole
5:55 skill because this is what Claude reads
5:57 when it decides when to trigger the
5:58 skill. So, you want to make sure it
6:00 includes what the skill does. And
6:02 specific phrases might be something like
6:05 Orura sync branding or use our brand
6:08 colors or apply our company theme. The
6:09 more specific your triggers, the more
6:11 reliably it's going to fire. All right.
6:12 So, now I'm going to download the zip
6:14 file. Then I'm going to go to settings
6:16 and under capabilities where it has all
6:18 my skills, I'm going to click on the
6:20 plus sign. And then I'm going to upload
6:22 the zip file as a skill under the skills
6:24 section and I'm going to toggle it on.
6:26 Awesome. So now we have our skill
6:28 created. Let's test it out. So inside a
6:30 new chat, I'm going to give Claude a
6:32 quick brief about our company and then
6:33 I'm going to ask it to build a landing
6:35 page for this company. I'm also going to
6:37 add a few more details about Aura Sync
6:39 so that it knows what type of
6:40 information to actually put on the
6:42 landing page. I'm not going to say
6:44 anything about styling because our skill
6:46 has all of that information. All right.
6:48 So, after working for about a minute or
6:50 so, let's see what it came up with. Wow,
6:52 I actually think that looks really good.
6:53 It has that dark background that I was
6:56 going for, the electric blue accents on
6:58 the buttons and some of the text. And it
7:01 looks very much on brand. The only weird
7:02 thing is that the logo kind of didn't
7:04 come out uh exactly how I wanted it to.
7:06 I put it on a transparent background and
7:07 it didn't quite get that right, but that
7:10 would be an easy fix. But this looks
7:11 like pretty close to what I was going
7:13 for. Notice that I didn't mention the
7:14 name of the skill anywhere in the
7:16 prompt. Claude triggered it
7:18 automatically because the task matched
7:20 the description of the skill. And here's
7:21 the thing, this same theme now applies
7:24 to everything I create for this company.
7:25 Watch. I'm going to ask Claude to draft
7:27 a one-page executive summary about
7:30 Orsync's Q1 performance. And when I do
7:32 that, here's what it came up with. You
7:33 can see that the report has the same
7:36 colors, the same fonts, that same logo.
7:38 So that one skill is applied everywhere
7:40 automatically. And because it's a skill
7:41 and not a project, I can close that
7:43 conversation and open up a completely
7:46 empty conversation. And the Orura Sync
7:48 theme is still there. It's accountwide.
7:50 So my brand is now part of my Claude
7:52 profile. And this approach works for any
7:54 official skill. You could extend the
7:56 brand guideline skill with your own
7:57 design system. The point is you don't
7:59 have to build from scratch. Start with
8:01 what already works and make it yours. So
8:03 you just saw how easy it is to take what
8:05 Claude already does well and customize
8:07 it to your brand. But skills are a layer
8:08 that sits on top of Claude's core
8:11 features, projects, artifacts,
8:12 dashboards, and if you haven't locked in
8:15 that foundation yet, skills won't land
8:16 the way they should. If you're looking
8:18 to become better at using Claude and you
8:20 want a road map to get you up to speed
8:21 quickly, then I suggest you take a look
8:23 at this free resource from HubSpot
8:25 called the complete guide to Claude AI.
8:27 This guide covers the entire cloud
8:28 foundation, setting up projects,
8:30 building artifacts, and interactive
8:32 dashboards, designing real workflows, so
8:34 that when you start adding skills on
8:36 top, everything else clicks into place.
8:38 Inside, you'll find project templates
8:40 that you can copy directly, like a full
8:42 content marketing setup with exact
8:44 commands and document lists, an
8:46 interactive marketing performance
8:48 dashboard, and more. It even includes
8:50 best practices and guidance on how to
8:52 implement Claude across your entire
8:54 team. My favorite section is the one on
8:56 using Claude as an executive assistant.
8:57 It gives you step-by-step instructions
8:59 on setting up automated morning
9:01 briefings, a content optimization
9:03 system, meeting follow-up, and task
9:06 tracking across projects. It's super
9:07 practical and helps save hours every
9:09 week. Download this guide using the link
9:10 in the description, and thanks to
9:12 HubSpot for sponsoring this video. All
9:14 right, so before we go deeper with our
9:16 next use case, here's a quick filter.
9:18 Not every workflow needs to be a skill.
9:20 So, when you're deciding whether or not
9:22 to create a skill, ask yourself three
9:24 questions. Number one is, how repeatable
9:26 is the task? If you only do it once,
9:28 just stick with a regular prompt. But if
9:30 you're doing the same thing weekly with
9:32 the same format, the same rules, create
9:34 a skill. The second question is, does
9:36 consistency matter? If the output needs
9:39 to hit a specific standard every time,
9:41 like a brand guardian skill that scores
9:42 content against your tone of voice, for
9:44 example, that's a skill. And number
9:47 three is, could somebody else run it? If
9:49 you could hand this to a new team member
9:50 and say just ask Claude and the
9:52 instructions are clear enough, it's
9:54 skill ready. And if you have two out of
9:56 three of those, build the skill. Now,
9:58 extending skills is the quick win. But
10:00 the real power comes when you build
10:02 skills from scratch that encode your own
10:04 workflows. And specifically, when you
10:06 encode what's already working so you can
10:08 repeat it on demand. Here's the
10:09 scenario. You have a piece of content
10:11 that performed really well. A YouTube
10:14 video with great retention, a blog post
10:16 that ranked, an email sequence that
10:18 converted. You know it worked, but you
10:20 probably can't explain exactly why in a
10:21 way that's repeatable. That's what we're
10:23 going to fix. I'm going to paste the
10:24 hooks from my three best performing
10:26 YouTube videos. These are the openings
10:28 that grabbed people. You know, the first
10:30 30 to 60 seconds that determines whether
10:32 or not someone stays or clicks away. And
10:33 I'm going to ask Claude to find the
10:35 patterns. So, I'm just giving it this
10:37 prompt right here. Not going to read out
10:39 the prompt. And after Claude thinks for
10:41 a bit, here's what it comes up with. So,
10:42 Claude identifies that there's a
10:44 consistent structure across all three
10:46 hooks. So there's an anchor where I meet
10:48 the viewer exactly where they are,
10:50 naming the tools or the situation
10:52 they're already using. Then there's a
10:54 diagnosis, so I validate what they're
10:55 already experiencing and surface the
10:57 real problems that they haven't
10:58 identified yet. Then there's a
11:00 credibility signal, so I share my own
11:02 experience and I build trust with that
11:04 person. And then there's a reframe,
11:06 which is the key move. I disrupt their
11:08 assumption about what the problem
11:10 actually is. And finally, a promise. I
11:13 stack multiple specific outcomes so that
11:14 they know exactly what they're going to
11:16 be getting from the video. Now, here's
11:18 the thing. I knew the hooks worked, but
11:19 I couldn't have written that five-part
11:21 sequence down if you asked me. It was
11:23 more of intuition, but that can't be
11:25 replicated. But now, Claude gave me a
11:27 framework to follow. Now, I'm going to
11:30 ask it to take this analysis and package
11:32 it as a reusable Claude skill called the
11:34 hook generator. This skill should accept
11:36 any topic as input and produce three
11:39 hook variations following this exact
11:40 structural pattern. All right. So after
11:42 running for a moment, here is the skill
11:44 that generated on the right hand side.
11:46 So you can see there's some really
11:48 detailed instructions here, including
11:50 when the skill gets triggered, the five
11:52 element framework that it broke down
11:55 when it was analyzing my hooks, etc. So
11:56 now on the top right hand corner of the
11:58 screen, I'm going to click on copy to
12:00 your skills, that button there, and now
12:02 it's going to be part of my skills
12:03 within my capabilities under my
12:05 settings. Okay, so now let's actually
12:07 test this hook generator. I'm going to
12:09 start a brand new chat and I'm going to
12:10 pick the most boring title I can think
12:12 of and I'm purposely not going to give
12:14 it a lot of instructions. I'm just going
12:16 to give it basically one sentence. Okay.
12:18 Generate three hooks for a video called
12:21 five tips for better time management. So
12:23 now the skill is going to do the heavy
12:26 lifting because the prompt is so basic.
12:27 All right. So let's see what it came up
12:29 with. Wow. So look at these. Each one of
12:31 these follows the exact same five
12:33 element structure. the anchor,
12:35 diagnosis, credibility, reframe, and
12:37 promise, but with completely different
12:39 angles. The first hook enters through
12:41 the productivity tool trap. So, you've
12:43 got the apps, you've got the systems,
12:44 you're still not getting to what
12:46 matters. And the reframe is time
12:48 management isn't about time, it's about
12:50 energy management. Pretty good. The
12:51 second hook is all about being
12:53 overwhelmed and being really busy. So,
12:55 your schedule's packed, you're
12:57 exhausted, and the reframe is that the
12:59 real skill isn't doing more, it's being
13:01 ruthless about doing less. I like that.
13:03 And the third hook enters through the
13:05 plans falling apart angle. So you plan
13:08 your week perfectly on Sunday and by
13:10 Tuesday it's fiction. And the reframe
13:12 here is stop building rigid plans for a
13:14 chaotic world. Build systems that absorb
13:16 the chaos. I think that's really good.
13:18 So those are three completely different
13:19 angles built on the same structure as
13:21 what worked for my top performing
13:23 videos. And each one of these sounded
13:25 like something you'd actually want to
13:26 watch because the structure underneath
13:29 is proven. This is the real unlock.
13:31 You're not just saving time, you're
13:32 taking the patterns from your best work
13:34 and actually making it explicit and
13:36 clear. You're training your AI on your
13:39 own success formula. And this method
13:41 works for any repeatable structure.
13:43 High-converting email sequences, reverse
13:45 engineer the framework and skill it
13:47 sales call scripts that close, reverse
13:49 engineer the flow and skill it. Add copy
13:51 that outperforms, reverse engineer the
13:53 hooks and angles and create a skill.
13:55 Report formats that your executives
13:57 love, reverse engineer the layout and
13:59 create a skill. Your best past work
14:01 becomes the template for all your future
14:03 work. That's the real power of building
14:05 skills from scratch. Now, let's combine
14:07 everything. This is where multiple
14:09 skills work together with Claude's
14:11 built-in capabilities and connectors,
14:13 and it starts to feel like a workflow
14:15 engine. Now, for our third use case,
14:17 here's the situation. I have a CSV file
14:20 with 50 leads, company names, industry,
14:22 employee count, revenue, engagement
14:24 scores, all these different data points.
14:26 I want Claude to analyze them, score
14:28 every single one against my ideal
14:30 customer profile, and then build a
14:32 presentation that I can bring to my next
14:34 Friday sales meeting. And because I'm
14:36 going to be doing this analysis over and
14:38 over again, I'm going to create a skill.
14:39 So, I'm going to give it this prompt
14:41 with scoring rules for our ideal
14:43 customer profile criteria. And this
14:45 skill is going to contain which
14:47 industries I prioritize, which company
14:49 size is my sweet spot, how I weigh the
14:51 engagement scores, which source channels
14:53 are signaling high intent, and what
14:56 automatically disqualifies a lead. And
14:57 all of that logic is going to be baked
14:59 into the skill so I never have to
15:01 explain it again. All right, so Claude
15:03 works for a minute or so, and here's the
15:04 skill that it comes up with on the
15:06 right. You can see this is super
15:07 detailed here, and it has everything
15:09 that we need to know about scoring our
15:12 leads against our ICP criteria. So
15:14 again, I'm going to go to the top right
15:15 hand corner of the screen and I'm going
15:17 to click on copy to your skills, that
15:19 button. And now our skill is added to
15:21 our capabilities within our settings.
15:22 All right. And now we're going to test
15:24 this out. So let's create a brand new
15:26 chat. And I'm going to upload my CSV
15:28 file with all my leads. And I'm going to
15:30 give it this prompt. Score these leads
15:32 against our ideal customer profile and
15:33 give me the full breakdown with their
15:35 tier assignments, key reasons, and
15:37 recommended actions for each lead. Then
15:39 summarize the overall pipeline. All
15:41 right. Now, here's what it came up with
15:42 when I gave it that prompt. All right.
15:44 Now, watch what's happening here. Claude
15:46 is actually reading the lead scorer
15:48 skill to understand my criteria. It
15:50 processed all 50 leads through the
15:52 scoring engine. And then it delivers a
15:54 full spreadsheet with two tabs in it.
15:55 Every lead scored with points,
15:57 breakdowns, and recommended actions in
16:00 one sheet and then a pipeline summary in
16:02 the other. We've got 25 hot leads, five
16:05 warm, five cold, 15 disqualified. And
16:07 each one of these has the key reason
16:09 here. So it actually provides the reason
16:11 for the score and the recommended action
16:13 that we should take for each of the
16:15 leads. So these are really actionable
16:17 insights that we can walk into our
16:19 meeting with wellprepared. Now here's
16:21 where this gets even more powerful. I
16:22 also have Gamma connected as a
16:24 connector. That's an AI presentation
16:26 tool. If you want to set this up for
16:28 yourself, you go to settings, then
16:30 connectors, and you'll find Gamma in
16:32 there. Now once it's connected, Claude
16:34 can actually send data directly to it.
16:35 So, I'm going to take these results and
16:37 turn them into a polished deck with one
16:39 more prompt. So, I'm going to say, "Now
16:41 that you've scored the leads against our
16:43 ICP, create a presentation with these
16:46 results in Gamma." So, Claude takes all
16:48 the scoring data and sends it to Gamma
16:50 through the connector. And in this demo,
16:52 it actually even picks a dark theme here
16:55 with cyan and teal accents that closely
16:57 match our OruraNC palette. And the way
16:58 it's coming up with that is it's reading
17:00 the styling skill that we created for
17:02 the Ourync brand within our first use
17:04 case. So, it's using that as a reference
17:06 point so that it feels on brand right
17:08 out of the gate. Pretty amazing. So,
17:10 after working for a couple minutes,
17:11 here's what it came up with. We have
17:13 slides with a scoring methodology. It
17:15 came up with 10 different slides. It has
17:18 a pipeline summary of the 50 leads and
17:20 the distribution of that. It has a slide
17:21 with our action plan on it for the sales
17:24 team and the key patterns and insights
17:25 from all this information. That's a
17:27 presentation I can open, edit if I
17:30 needed to, and actually export it as a
17:32 PDF or a PowerPoint and bring it to my
17:34 sales meeting. And that entire deck was
17:36 generated from one prompt. Pretty
17:38 amazing, right? Now, this is the key
17:40 concept of this example. Don't build one
17:42 massive skill that does everything.
17:45 Build small, focused skills that each do
17:46 one thing well, and let Claude
17:48 orchestrate them with your connectors.
17:50 The lead scorer skill handled the
17:52 analysis logic. Gamma handled the
17:54 presentation and it even got the style
17:55 inspiration from the first skill that we
17:57 created in this video. Three
17:59 capabilities working together from two
18:01 simple prompts and this scales to
18:03 whatever your business needs. You can
18:05 pull data from notion through the notion
18:06 connector, run it through a custom
18:08 analysis skill and output a formatted
18:10 report. The combinations are as specific
18:12 as your workflows. Now, if you were
18:14 doing this inside a project, you
18:15 wouldn't need to actually repeat the
18:17 workflow details in the project
18:19 instructions. Just tell Claude which
18:20 skills to use for which task. For
18:22 example, you could say, "For lead
18:24 analysis, use the lead scorer skill. For
18:27 presentations, send to Gamma." And then
18:29 just let Claude handle the rest. Now,
18:31 before you go building a dozen skills
18:33 tonight, here's some guardrails. I
18:34 touched on this at the beginning of the
18:36 video, but there are community skills
18:38 available on GitHub created by other
18:40 people. But downloading a public skill
18:42 from GitHub is like running code from a
18:44 stranger. So before activating any
18:46 community skill, open up the skill.md
18:48 file and read it. Look for hidden
18:51 instructions, references to external
18:53 endpoints, anything doing more than the
18:55 skill claims. Prompt injection is a real
18:58 risk. So audit before you activate. I
19:00 personally recommend just sticking with
19:02 anthropic skills or skills that you
19:04 create on your own, just like we did in
19:05 this video, where possible. The second
19:07 thing to keep in mind is something I
19:08 mentioned at the beginning of the video,
19:10 which is to have only a few skills
19:12 toggled on at one time. Here's another
19:14 reason why Claude uses something called
19:16 progressive disclosure. The description
19:19 field of every active skill sits in
19:21 Claude's system prompt at all times. The
19:23 full instructions only load when
19:25 triggered. But more active skills means
19:28 more context consumed before Claude even
19:30 starts your task. So just keep that in
19:31 mind. The next tip is to keep things
19:33 organized by naming your skills in a
19:35 consistent way. So start each custom
19:37 skill with your company name, for
19:39 example, or initials so that they stay
19:41 grouped together and then use short
19:44 hyphen style names like brand guideline
19:46 review instead of spaces. Now, here's a
19:48 quick debugging tip. If a skill isn't
19:50 triggering properly, you can ask Claude,
19:52 "When would you use the brand guardian
19:54 skill, for example?" It'll quote the
19:56 description back to you, and if what it
19:57 reads doesn't match how you're actually
19:59 using the skill, then you'll know
20:01 exactly what you need to fix. And my
20:04 last tip is that skills are static. So
20:06 they don't evolve with conversations. So
20:08 you need to set reminders regularly to
20:10 actually go in whether it's every
20:12 quarter or every month to actually make
20:14 sure that the skills are still valid.
20:16 Remember to review and update these
20:18 regularly. Skills give Claude your
20:20 methods. Connectors give Claude your
20:22 tools. Combine both and you're not just
20:24 automating tasks, you're building agents
20:26 that execute your business strategy
20:28 across your entire tool stack. So give
20:30 this a try this week. Use one of the
20:32 techniques we covered here for your own
20:34 task and let me know how it goes in the
20:36 comments. If you enjoy this video, then
20:37 please give it a thumbs up. It really
20:39 helps the channel. And if you want to
20:40 learn more about how to use AI to level
20:42 up your work and your life, then click
