WEBVTT 00:00:00.000 --> 00:00:03.359 Welcome to the deep dive beat. I am really looking 00:00:03.359 --> 00:00:06.639 forward to this one. Oh me too because Well, 00:00:06.639 --> 00:00:09.060 let's set the stage here. You think you're a 00:00:09.060 --> 00:00:12.339 Claude power user, right? You have a notion folder. 00:00:12.800 --> 00:00:17.640 It is packed with 30 amazing prompts But here 00:00:17.640 --> 00:00:20.899 is the harsh truth copy pasting them is costing 00:00:20.899 --> 00:00:24.539 you time It's costing you money and it wastes 00:00:24.539 --> 00:00:27.239 computing power. It is wildly inefficient. So 00:00:27.239 --> 00:00:30.379 today we are exploring Claude skills It's a totally 00:00:30.379 --> 00:00:32.899 new paradigm. The mission today is simple. We 00:00:32.899 --> 00:00:35.299 want to stop teaching the AI our preferences 00:00:35.299 --> 00:00:37.810 from zero. Every single time. Right. We will 00:00:37.810 --> 00:00:40.450 walk through eight practical skills today. They 00:00:40.450 --> 00:00:43.409 transform repetitive tasks into automated, reusable 00:00:43.409 --> 00:00:45.829 workflows. It completely changes how you interact 00:00:45.829 --> 00:00:48.469 with AI. But, you know, before we build specific 00:00:48.469 --> 00:00:50.210 skills, we need to understand the core concept. 00:00:50.250 --> 00:00:52.090 We have to look at what a skill actually is. 00:00:52.130 --> 00:00:54.270 Yeah. And the meta tool used to build them. Exactly. 00:00:54.350 --> 00:00:56.329 Let's explain the problem first. Long -pasted 00:00:56.329 --> 00:00:58.929 prompts eat up your tokens. Right. And tokens 00:00:58.929 --> 00:01:01.250 are basically the tiny pieces of words an AI 00:01:01.250 --> 00:01:04.170 reads and processes. That's right. Heavy usage 00:01:04.170 --> 00:01:07.129 limits vanish incredibly fast when you do this. 00:01:07.909 --> 00:01:11.969 The AI has to read your entire setup every single 00:01:11.969 --> 00:01:14.310 time. It's exhausting. It really is. It wastes 00:01:14.310 --> 00:01:17.409 so much computing power just establishing a baseline. 00:01:17.969 --> 00:01:21.909 So a Claude skill is different. It is a packaged 00:01:21.909 --> 00:01:24.849 workflow. Think of it like teaching a chef your 00:01:24.849 --> 00:01:27.989 favorite recipe once. Oh, I like that analogy. 00:01:28.090 --> 00:01:29.950 Yeah, instead of explaining what a kitchen is 00:01:29.950 --> 00:01:32.170 every time you want a sandwich. Right. And behind 00:01:32.170 --> 00:01:35.170 the scenes, these are stored as skill .md files, 00:01:35.650 --> 00:01:38.689 basically a hidden text file saving your exact 00:01:38.689 --> 00:01:41.329 workflow steps. Exactly. They don't clog up your 00:01:41.329 --> 00:01:43.209 context window. They only load when you need 00:01:43.209 --> 00:01:45.290 them. You just call them using a slash command 00:01:45.290 --> 00:01:46.890 and the skill name. You know, I have to make 00:01:46.890 --> 00:01:49.629 a vulnerable admission here. Beat, I still wrestle 00:01:49.629 --> 00:01:51.530 with this myself. I literally just copy paste 00:01:51.530 --> 00:01:53.870 my favorite massive prompts from Notion every 00:01:53.870 --> 00:01:57.019 morning. Oh, we all do it. It's a hard habit 00:01:57.019 --> 00:01:59.780 to break. It is. And that is exactly why we are 00:01:59.780 --> 00:02:02.060 starting with skill number one, the skill creator. 00:02:02.239 --> 00:02:05.180 The tool that builds the tools. Right. With the 00:02:05.180 --> 00:02:07.159 skill creator, you don't input a short prompt. 00:02:07.700 --> 00:02:10.759 You describe your full workflow, start to finish. 00:02:11.259 --> 00:02:13.259 Let's look at the YouTube brand scout example. 00:02:13.680 --> 00:02:15.819 That was a fascinating use case from our sources. 00:02:15.960 --> 00:02:18.409 Yeah. Finding brand sponsors on YouTube channels 00:02:18.409 --> 00:02:21.590 right now. You tell the skill creator your exact 00:02:21.590 --> 00:02:24.069 step -by -step process. You want Claude to scan 00:02:24.069 --> 00:02:27.009 a transcript. You want it to identify brands. 00:02:27.610 --> 00:02:29.229 Then you want it to format them into a table. 00:02:29.569 --> 00:02:32.150 And you need specific recommendations, like strong 00:02:32.150 --> 00:02:34.710 fit or skip. Wait, so if my description is vague, 00:02:34.930 --> 00:02:37.150 how does Claude know what I actually want the 00:02:37.150 --> 00:02:40.129 skill to do? That is the genius part. Claude 00:02:40.129 --> 00:02:43.150 proactively asks follow -up questions to lock 00:02:43.150 --> 00:02:45.770 down the exact workflow criteria before saving. 00:02:46.169 --> 00:02:48.990 It forces you to clarify. Then you test it with 00:02:48.990 --> 00:02:51.930 real inputs. So Claude asks follow -up questions 00:02:51.930 --> 00:02:54.849 to nail down the exact workflow steps. Exactly. 00:02:54.930 --> 00:02:57.650 It interviews you. That makes sense. Now that 00:02:57.650 --> 00:03:00.430 we know how to build skills, let's pivot. Let's 00:03:00.430 --> 00:03:03.349 apply this to a very common daily task, creating 00:03:03.349 --> 00:03:05.830 and refining written content. Yeah, writing is 00:03:05.830 --> 00:03:07.710 where most people spend their limits. This brings 00:03:07.710 --> 00:03:11.150 us to skill number two. Hook Forge. People spend 00:03:11.150 --> 00:03:15.710 hours agonizing over opening lines for articles, 00:03:16.110 --> 00:03:18.370 for social media. Oh, just staring at a blinking 00:03:18.370 --> 00:03:21.490 cursor. It's painful. Hook Forge automates that 00:03:21.490 --> 00:03:24.129 friction. You just provide the topic, the audience, 00:03:24.229 --> 00:03:26.930 and the content summary. The skill then generates 00:03:26.930 --> 00:03:29.349 multiple hook variations, and it doesn't just 00:03:29.349 --> 00:03:31.669 give you the text. It provides the angle and 00:03:31.669 --> 00:03:34.229 explains why it works. It teaches you the psychology. 00:03:34.430 --> 00:03:36.460 Right. But once you have that draft, you hit 00:03:36.460 --> 00:03:39.039 another wall. The draft sounds robotic. Yes. 00:03:39.240 --> 00:03:41.979 Which brings us to skill number three, the humanizer. 00:03:43.000 --> 00:03:46.099 AI content has a very stiff flavor. It relies 00:03:46.099 --> 00:03:48.840 on statistical averages. Exactly. It picks predictable 00:03:48.840 --> 00:03:52.039 words. The humanizer edits AI -generated content 00:03:52.039 --> 00:03:54.500 to improve flow. It increases sentence variety. 00:03:54.639 --> 00:03:56.759 It removes those predictable AI transitions. 00:03:56.879 --> 00:04:00.319 Like Delvin. Furthermore. Yes, it hunts those 00:04:00.319 --> 00:04:03.639 down, but it's careful. With Humanizer, it preserves 00:04:03.639 --> 00:04:05.639 the author's voice instead of overwriting it. 00:04:05.979 --> 00:04:08.300 With Humanizer, is there a risk that it completely 00:04:08.300 --> 00:04:10.819 erases my personal writing style to make it sound 00:04:10.819 --> 00:04:14.419 human? You would think so, but the specific instruction 00:04:14.419 --> 00:04:18.800 in this skill is guarded. It preserves your original 00:04:18.800 --> 00:04:22.209 meaning. It acts like a polish, not a rewrite. 00:04:22.490 --> 00:04:24.870 Got it. It preserves your unique voice while 00:04:24.870 --> 00:04:28.310 removing those robotic AI clichés. Yeah. It makes 00:04:28.310 --> 00:04:31.209 the AI invisible. Sometimes the words are perfectly 00:04:31.209 --> 00:04:35.029 human, though, but the format is wrong. Or the 00:04:35.029 --> 00:04:37.730 complexity is entirely wrong for your audience. 00:04:38.329 --> 00:04:40.449 Beat. Let's explore reshaping information. See, 00:04:40.449 --> 00:04:42.810 this is a huge bottleneck. You have the research, 00:04:43.089 --> 00:04:45.370 but you need visual structure. Enter skill number 00:04:45.370 --> 00:04:47.769 four, the infographic builder. Right. Imagine 00:04:47.769 --> 00:04:51.329 a dense 20 -page report. This skill extracts 00:04:51.329 --> 00:04:53.689 the key ideas. It removes the fluff. It creates 00:04:53.689 --> 00:04:55.930 a scannable outline. With a title and headings. 00:04:56.069 --> 00:04:58.870 Yes. And even visual suggestions. It is totally 00:04:58.870 --> 00:05:00.970 ready for design. It bridges the gap between 00:05:00.970 --> 00:05:03.990 text and design. But sometimes format isn't the 00:05:03.990 --> 00:05:06.149 issue. Sometimes it's the sheer complexity of 00:05:06.149 --> 00:05:08.290 the language. Claude loves to over -explain. 00:05:08.370 --> 00:05:10.670 It really does. It gives you every nuance. That 00:05:10.670 --> 00:05:13.649 is why we have skill number five, Caveman. I 00:05:13.649 --> 00:05:16.410 absolutely love that name. It's perfect. This 00:05:16.410 --> 00:05:19.250 workflow forces Claude to use simple language. 00:05:19.850 --> 00:05:23.509 Fewer words. Absolutely zero jargon. It is amazing 00:05:23.509 --> 00:05:26.170 for token reduction. And for explaining complex 00:05:26.170 --> 00:05:28.189 things to beginners. But let me ask you this. 00:05:28.889 --> 00:05:32.250 If the caveman skill cuts out so much text, aren't 00:05:32.250 --> 00:05:35.110 we losing the actual factual value of the original 00:05:35.110 --> 00:05:38.689 content? It seems risky, but no. Caveman targets 00:05:38.689 --> 00:05:41.290 jargon and filler words. It targets those long 00:05:41.290 --> 00:05:44.189 introductions. The core information remains totally 00:05:44.189 --> 00:05:46.370 intact. Right. It cuts the fluff and jargon, 00:05:46.730 --> 00:05:49.790 but keeps the core message entirely intact. Exactly. 00:05:49.930 --> 00:05:51.949 It just removes the cognitive friction. We've 00:05:51.949 --> 00:05:53.810 completely stripped the fluff. Before we talk 00:05:53.810 --> 00:05:55.790 about scaling this exactness, let's take a quick 00:05:55.790 --> 00:06:01.410 pause. Sponsor. And we are back? We have reshaped 00:06:01.410 --> 00:06:03.850 text. But what if we need to scale content across 00:06:03.850 --> 00:06:06.810 platforms or apply this precision to software 00:06:06.810 --> 00:06:09.850 engineering? Scaling and coding require extreme 00:06:09.850 --> 00:06:13.009 precision. Let's look at skill number six. Expand 00:06:13.009 --> 00:06:15.610 and contract. This is a style -neutral skill. 00:06:15.790 --> 00:06:17.430 Completely neutral. You can tell it to expand 00:06:17.430 --> 00:06:20.170 a short thought with context and examples. Or 00:06:20.170 --> 00:06:22.269 you can contract a massive report into a tight 00:06:22.269 --> 00:06:24.730 summary. Without losing the original tone. Right. 00:06:24.790 --> 00:06:26.730 It's like an accordion. You turn one idea into 00:06:26.730 --> 00:06:30.430 a tweet, then a newsletter, then an essay. The 00:06:30.430 --> 00:06:32.769 tone stays locked. That is incredibly useful. 00:06:33.449 --> 00:06:35.339 But... Let's transition to pure logic. Let's 00:06:35.339 --> 00:06:37.500 talk about code. Still, number seven, the Carpathy 00:06:37.500 --> 00:06:39.839 guidelines. This is vital for developers. Yeah, 00:06:40.000 --> 00:06:42.639 inspired by Andrei's Carpathy. This workflow 00:06:42.639 --> 00:06:45.540 forces Claude to understand existing code before 00:06:45.540 --> 00:06:48.560 making edits. That is a massive shift. Normally, 00:06:48.660 --> 00:06:51.199 an AI just overwrites things. Exactly. It acts 00:06:51.199 --> 00:06:53.519 like a careless junior dev. This skill forces 00:06:53.519 --> 00:06:57.439 it to prefer small, focused changes. It has to 00:06:57.439 --> 00:06:59.720 explain decisions. And highlight side effects 00:06:59.720 --> 00:07:03.509 and risks. Beat. Whoa. Two secs silence. Imagine 00:07:03.509 --> 00:07:06.350 an AI instantly understanding the context of 00:07:06.350 --> 00:07:09.269 an entire massive code base just to safely inject 00:07:09.269 --> 00:07:12.230 one tiny perfect line of code. It is mind -blowing 00:07:12.230 --> 00:07:14.110 to watch. It stops you from breaking production. 00:07:14.360 --> 00:07:16.620 Is the Carpathy skill mostly for writing brand 00:07:16.620 --> 00:07:18.439 new apps from scratch, or does it have another 00:07:18.439 --> 00:07:21.160 purpose? It can start from scratch, but it is 00:07:21.160 --> 00:07:23.439 actually most valuable for reviewing, fixing, 00:07:23.680 --> 00:07:26.480 and modifying existing complex project code bases. 00:07:26.899 --> 00:07:30.519 Ah, it actually shines when reviewing and safely 00:07:30.519 --> 00:07:34.019 modifying existing complex project code bases. 00:07:34.139 --> 00:07:36.899 Yeah, it's a critical safety net. So, if we can 00:07:36.899 --> 00:07:40.069 trust Claude to evaluate code base risks. We 00:07:40.069 --> 00:07:42.610 can use it to evaluate real -world risks. That 00:07:42.610 --> 00:07:45.129 sets up our final workflow, skill number eight. 00:07:45.290 --> 00:07:48.389 decision framer. Moving way beyond a simple pros 00:07:48.389 --> 00:07:51.129 and cons list. Right. It evaluates complex choices. 00:07:51.470 --> 00:07:53.870 Let's use the sources example. You are choosing 00:07:53.870 --> 00:07:56.769 between launching a paid newsletter, an online 00:07:56.769 --> 00:07:59.430 course, or a membership community. Those are 00:07:59.430 --> 00:08:01.970 tough, compounding choices. Very tough. Instead 00:08:01.970 --> 00:08:04.430 of recommending just one, the skill surfaces 00:08:04.430 --> 00:08:07.449 your assumptions. It highlights risks. It maps 00:08:07.449 --> 00:08:09.930 out the consequences of each trade -off. Exactly. 00:08:10.350 --> 00:08:13.050 A community needs daily moderation. A course 00:08:13.050 --> 00:08:15.620 is heavy up front. The framer provides a structured 00:08:15.620 --> 00:08:18.899 framework when intuition isn't enough. So I plug 00:08:18.899 --> 00:08:21.139 in my problem with decision framer, just make 00:08:21.139 --> 00:08:23.660 the final choice for me. Not at all. The AI's 00:08:23.660 --> 00:08:26.259 job isn't to choose for you. It highlights the 00:08:26.259 --> 00:08:28.759 consequences of each path, so your human judgment 00:08:28.759 --> 00:08:30.759 is better informed. No, it highlights trade -offs 00:08:30.759 --> 00:08:32.679 and consequences, so you make a highly informed 00:08:32.679 --> 00:08:35.259 choice yourself. Precisely. It sharpens your 00:08:35.259 --> 00:08:37.639 strategic thinking. We have covered a lot today. 00:08:38.120 --> 00:08:40.950 Let's synthesize the big idea here. The overarching 00:08:40.950 --> 00:08:44.070 critical insight is that a usable Claude skill 00:08:44.070 --> 00:08:47.269 starts from a repeated workflow. Yeah. Not from 00:08:47.269 --> 00:08:49.850 a clever prompt. Right. It is not about generating 00:08:49.850 --> 00:08:52.610 more prompts. It's about building a library of 00:08:52.610 --> 00:08:54.970 reusable systems, systems that perfectly match 00:08:54.970 --> 00:08:57.690 how you work. You are creating automated habits. 00:08:57.889 --> 00:09:00.629 It frees you from the tedious setup phase entirely. 00:09:00.889 --> 00:09:02.769 I want to leave you with a provocative thought. 00:09:03.629 --> 00:09:06.629 Beat. If AI can now standardize and automate 00:09:06.629 --> 00:09:09.230 all of our repetitive workflows and frameworks, 00:09:10.029 --> 00:09:13.830 What uniquely human, messy, creative skills should 00:09:13.830 --> 00:09:15.970 you be focusing your newly freed up time on? 00:09:16.190 --> 00:09:18.309 That is the question of the decade. I highly 00:09:18.309 --> 00:09:20.309 encourage you to take action on this. Pick just 00:09:20.309 --> 00:09:22.789 one repeated task you do every week. Turn it 00:09:22.789 --> 00:09:25.149 into a skill today. Test it out. See how it changes 00:09:25.149 --> 00:09:27.490 your workflow. Start building your library. Thanks 00:09:27.490 --> 00:09:29.889 for joining us in this deep dive by UTRO Music.