Episode Transcript
[00:00:05] Welcome to Prompting Curiosity, a podcast for the AI curious. No coding background required. I'm your host, Dr. Shantae Cofield, also known as the Maestro, and I created this show to explore what these AI tools actually are. Really, though, are the files in the computer, how to use them, and what they might mean for how we think, work, create, and move through life. Whether you're skeptical, intrigued, or already experimenting, you're in the right place. All that I ask is that you stay curious. All right, let's get into it.
[00:00:38] Hello, hello, hello, my curious people, and welcome to episode 58 of Prompting Curiosity. I'm your grateful host, the Maestro, and today we are talking about Claude Skills. So these are not new at all. They've been a thing since last October. October 2025. But today, the day that I am recording it, not the day this drops, I'm recording it. Uh, I went for a walk with my guy, K. I've. I have spoken about him before in the podcast. Um, he is moving. He's taking his family. They're going abroad for a year, which is so awesome. Um, but we went on a walk today because he's leaving in, like, two weeks. And anytime, uh, I talk to him, I'm all inspired and I get excited about stuff. And I was like, I'm gonna give skills. We didn't even talk about skills, really.
[00:01:19] Uh, but I came back and I needed an episode topic, and I was like, ah, maybe skills and I'll try it and. And here we are.
[00:01:26] So I will say that, as per, always, the best way to truly understand this stuff and anything AI related and just anything in life, uh, is to try it out yourself. So I actually watched a few YouTube videos about Skillz when they first launched last October, and I kind of got it, but, like, couldn't really fully wrap my head around it. And funny, I think part of the reason that my walk with K inspired this is because I sent him the video, one of the YouTube videos that I watched, and I was trying to understand it, and then he, like, just dove all in and, like, did the most.
[00:01:57] Uh, but I watched it and I was like, okay, I don't really have a use case for this, so I'm gonna not worry about it right now. Um, but came up today and I was like, why don't I give a try? And I played with it for, like, 20 minutes before making this episode, and now I fully understand it. And it's not like, oh, my God, everyone should use it. But I'm like, okay, I understand it. And you, uh, Know, the point there being, if you want to understand this stuff, it is a matter of just when you feel ready, when you feel bored, uh, giving it a try, actually just diving in. So we're going to start off with the definition and this comes directly from the anthropic support page. So per Anthropic skills are folders of instructions, scripts and resources that Claude loads dynamically to improve performance on specialized tasks.
[00:02:43] Skills teach Claude how to complete specific tasks in a repeatable way.
[00:02:49] The best skills, they solve a specific repeatable task and are focused on one workflow rather than trying to do everything.
[00:02:59] Um, for my techies in here, the core, the core file that's inside of a markdown. Excuse me, the core file that's inside of the skills folder is a markdown file. So remember a marked off file, markdown file, simply a plain text file. It uses symbols for formatting.
[00:03:15] So if you like one headers or bold, it's going to be symbol, it's going to be a symbol as opposed to like doing all the HTML stuff. Um, and this is a preferred file format for LLM. That's light, it's small, it makes sense. So the translation of what I just said, a skill is simply instructions for a set way of doing something. That is it. The cool part here is that once a skill has been created, Claude will then load and use that skill automatically when it's relevant for the task that ah, you're performing. That's what it means by it loads dynamically. Not like it doesn't like always load. It's just like, oh, I see that this user is typing about this thing. This would be helpful for me to use the skill and then it calls it, right?
[00:03:58] So if you're performing a task that Claude should be using a particular skill for and it's not, you can also tell Claude and be like, use my brand guideline skill or whatever for this, for this thing, uh, to make this PowerPoint. But it should know, but we know should be breaking under not working. Uh, so in that case you just have to just tell it.
[00:04:20] So we're going to get techie for a seckie. Um, because to me one of the coolest parts about skills is how they work and that's called progressive disclosure. So it would be hugely inefficient and expensive. I recall using all these tokens if Claude read the entire skill folder, like in all the stuff that's in it, anytime it thought that it might be helpful to use a skill for a particular task. Ask right, if you're like, oh, if it was going through and you're having conversation, it's like, oh, I think it would be helpful for it to you. And it says Claude, right? Claude's like, I think it would be helpful to use this skill and read the whole thing only to be like, m. Nope, not helpful. And it would take a long time. I mean this is fast, but still, it's just. You hear how cumbersome that is. So instead what Claude does is it uses what's called progressive disclosure, which means that it scans the skill description first, which is limited to 200 characters and it only loads the full instructions full or a skill once it's decided that the skill is relevant to the task. So I think that's pretty cool. If you're like, I don't care, well then, sorry, it's the last 20 seconds you don't care about. But as for like, you know, all the different things you can do, all different types of skills. Claude does have a director directory of skills, like inside of Claude. Um, but. Or I should say Anthropic has a directory of skills, but there weren't that many in there when I looked. And to me, the best way to use these skills and the reason I made this episode is for you to make your own. Right. Custom skill.
[00:05:41] Um, also worth noting, for my techies in the crowd, you can go to GitHub and search through a ton of skills there. Anthropic did open source, uh, repo. It's Anthropics with an S. Anthropics forward slash Skills. Um, but as we're always be careful about what you download to your computer. Like I don't know if I'd be downloading them, but um, I would just make my own.
[00:06:05] So how do you make one? How do you make your own? So as per always just ask the robot. You could just ask Claude. Um, Claude actually has a skill. Creator skill. How meta. But um, honestly, all you need to do is tell Claude that you want you wanted to help you create a skill and then you're off to the races. That's literally I did today. I was like, hey, I was talking back and forth with it about like the episode and I was like, actually, you know what, I think it'd be helpful to actually make a skill. Let's just do that. How do I do that? And it was just like, all right, it. Here we go. And move off to the races. So a few things to note as you are making the skill as per always the better the input, the better the output. So I'm going to really strongly suggest that you go back and forth with Claude about what you want this skill to do. Like, what do you want it to actually be? And then test it as you go to see if it works the way you want. Right? It works fast. And you're like, hey, you know, I want, um, I'm going to share, like, the skill that I made. But, um, I kind of want to give you an example. If you're like, okay, this is like a branding skill, my brand guidelines. And so then you'd have it generate something and you would see, like, did it follow it? And if it didn't, then you'd be like, all right, we have to change this and, like, change the instructions for this. That's all.
[00:07:13] Uh, so, uh, that's the first thing to note. The second thing is that Claude will actually write and format the skill for you. Like, I have, you know, I had an episode on what, how to make a custom GPT. Or maybe it was called what the is a custom GPT. This is like my. My what the series.
[00:07:29] Um, I kind of.
[00:07:31] When I'm thinking about naming the episodes, I'm like, you know, I'm just going to, like, go with this. WTF? Like, WTF? What the is a custom GPT? What the is a.
[00:07:40] Uh, what the is MCP? That was episode 4051, I think.
[00:07:45] Um, but either way, Claude will actually write and format the skill for you. So that is one of the other reasons that it's nice to just use clause. You don't have to worry about how to type it and how to summarize it and what the description should be or what the title should be. I mean, obviously you can help, you know, and say, actually, I want you to name it this. I did that for mine. I was like, actually, I don't like that name. Let's change it. Um, but the actual formatting of everything, it does it for you so you know that it'll work correctly.
[00:08:10] Um, next thing to note it, Claude is supposed to be able to save your skills. Like, I made this skill while we were just in a regular chat, and it was like, all right, it's been saved. And then I, like, went into my skills library and I was like, but where is it? Um, so I had to actually, like, when you. When you ask it to make a skill, it makes like a. Like a. What is it called? It calls it like a artifact.
[00:08:35] And, like, it makes like a document, and you can read it and see it, and then it's in Like a little, ah, section of the chat and there's a button that says save. And so I did have to actually go and like manually click that, which is like, no, it's not a big deal. But, um, it is worth noting. So if you like go and do this thing and then you're like, I tried to use the skill and like, it didn't work. And it's like, because it may not even, it may not have saved, you have to actually save it yourself.
[00:08:58] Last thing to note, the name of the skill and the description are super important, right? The description is the only part that Claude reads before deciding whether to load the rest of the skill. So if it's a vague description or a bad description, then you just made up yourself, um, this, it kind of means like, the skill doesn't technically exist and so it probably won't fire. So again, this is why it's just helpful to have Claude write the description. I.
[00:09:22] So like I said, I made this, I made the skill inside of a chat with Claude. All of this can be done inside of a regular chat. Um, but there is another more formal way to create a skill. So your skills, we're going to go, if you want to go look at them and see if you have any already, because sometimes cloud will make them. Um, depending on how you've been using Claude, like, it can have some loaded already in there.
[00:09:43] Um, but, uh, your skills live in the settings section of your account and then from there you go to customize and at the top of your screen you'll see a button that says add.
[00:09:57] And when you click it, it is, it pulls up a drop down, it opens a drop down. There's four options, and the first one is create with Claude. Second one is right skill instructions, third one is upload a skill, and the fourth, which is pretty cool, is record your screen. So I did not try the record your screen option. I knew that it existed and so I was like, oh, this is where it is. Um, but that's pretty cool. And it definitely is something that's worth noting that it exists, right? There's so much that AI can do, uh, and that, you know, largely I'm scratching the surface just using. I'm never just like trying to use things, to use things. It's like, do I have a use case?
[00:10:33] Um, and so to me, it's nice to know about the things because then one day if you're like, actually, I have this problem and this would be helpful if I could do this to solve it, and suddenly you're like, oh, I could do a screen recording. So Claude sees how I do this and then I just have it replicate that and it'll do it on its own. So very cool feature. Have yet to try it. You know, at some point I will try it and then when I do, I'll make an episode.
[00:10:55] But, um, in that same section, right, when you go inside the settings and then go to customize and all your skills are in there, you can access your skills. That's where you can actually click on them. And if you want to edit them or anything like that, you can do that or you can just have Claude edit them via just a regular chat. Right. So, side note again, to me, the most amazing part about AI is the fact that we can use natural language, whether we're typing it or speaking it, to do all of this. Like, that's crazy to me. You just describe the thing you want and it will make it, it will build it. That is wild, right? Absolutely wild. So, uh, just again, to save the 11 billionth time, your best bet for truly understanding what the fuck I'm talking about and understanding what skills are, you know, and for getting ideas on how they could be helpful for you, it's to go through the process of building one yourself and then you're like, oh, I see, I understand all right. But I do want to give you an example. So I'm not just like, you know, go figure it out. I do want to give an example. I know it's helpful to hear examples. Um, so I'm going to share the one that I built, one of the ones that I built today, um, while I was doing, you know, quote unquote research for this episode, so I built a skill called client debrief and the descript. I'm going to give you the description and by, from the description, you'll understand what it is. So, and don't forget, Claude wrote this. Uh, the description says, helps process a client call after it's over, when the user is still stuck on something or didn't fully land where they wanted to during the call itself, pulls the client's transcripts, surfaces every distinct thread in what and what's going on for them, and offers a synthesis read of the pattern underneath, then stays in the conversation as a thinking partner. Use whenever the names of a specific client.
[00:12:43] Excuse me? Use whenever the user names a specific client and says they're still thinking about the call, didn't wrap it up the way they wanted or wants to work through what to do next with somebody. Eg, I'm stuck on what to do with X client or that call with that client didn't really land. Help me think it through.
[00:13:01] So I already know before I, you know, get ahead of you. If you are concerned about anonymizing things, then do that. And you can still use that client's name, right? You don't have to pull the transcript directly in. You can, can give them a different name, that's fine.
[00:13:15] Right?
[00:13:16] So I just want to put that out there, throw that out there. And if you have problems with that, issues with that, that's. You don't have to use it. Okay.
[00:13:24] Um, but that is the skill that I made today and there was a lot of back and forth that I did with Claude. Right. I was testing it and being like, I don't really like the output you're giving me, like, this is not helpful. And this wasn't the initial skill that I went in trying to build. I didn't really know what I wanted to build. And I was like, okay, maybe we can just do this with these transcripts. And then I was like, this is not fucking helpful actually, like I can't see how I'd use this again and again. And Cloud was like, yeah, you're right, not helpful. And I was like, thank you for always just saying yes to me. Um, but as per, always, like go back and forth with the thing iterate, try it out, test it and see. And they'd be like, nah, this is not it. Or like, actually this is helpful. Um, so that is what we landed on, that I was like, yeah, I'm not going to use this thing every day. But I could see the utility of it. And I actually, because I had a client in mind and I was like, oh, this is helpful for me and be able to, being able to go back and forth, there's context for this. It will pull the salient points without me having to highlight them because I know what the salient points are from every conversation that I have with a client is and are. And so it was nice to put a few transcripts in there and then be like, I know what the salient points are. I want to see what you're pulling, why you're pulling those. And then we're going to change that so that you're going to pull the things that I want and then we can have conversations and you know, whether you're highlighting patterns or identifying patterns or, you know, I'm using it as a thought partner, um, or I can just reference the transcript again, you know, that's super. Super helpful. Right. So, um, you can also make a skill that pulls exact verbiage from transcripts, and then you just use that.
[00:14:55] That verbiage to. For marketing. Marketing materials. That's probably, you know, that second one that is probably something that you're like, oh, I see how, like, there could be very big utility with that. So, you know, realistically, the. The sky is the limit when it comes to creating a skill, which I realized probably makes it harder to come up with a skill that would be helpful for you. Right. AI continues to be, you know, here's a solution. Go find a problem. Um, but if you have a task that you always want done in the same way, that is the foundation for a good use case for a skill.
[00:15:27] Hopefully you're sitting there as I'm saying this, and you're like, wait, that sounds like how you've taught us to use a project both in Claude and in ChatGPT. And you are correct. And honestly, to me, there is a lot of overlap. Right.
[00:15:43] The main differences, however, are, uh, that skills can be used anywhere in any conversation at any time. You can use it in cloud code, you can use it in coworking, whereas the instructions that you come up with for a project, they only apply to conversations in that project. Can't use them everywhere else.
[00:16:03] Next, skills are simply specialized procedures.
[00:16:08] Projects, on the other hand, they're folders, right? A project is a full. Is basically, it's okay, nice way to think about it, right? Is a full folder and they allow for. And they are designed for providing background knowledge that's always loaded with every chat. Right? So with a folder. Excuse me, with the folder with a project, you can upload all other things to it that it will reference each time, and it will have access to that each time.
[00:16:30] Yes. Right. Uh, you can create custom prompts inside of a project, but the main utility of having a project is that you've created this workspace that always has context about that topic. Right? So, for example, I have a project that is for prop. For prompting curiosity for this. For this podcast. This way I don't always have to explain the podcast or the audience or, like, what I'm doing. Each time I can just be like, hey, I want to make a new episode. Um, here's what I'm thinking, and we can go back and forth because it has all this context about what the prompting curiosity is. Right? Makes sense again, for the 11 billionth time, it will make way more sense when you actually try it out for yourself. All right, last part. Last here. How I Use AI this week. We're going to wrap it up there. Uh, each episode, I share a quick example. If you're new here, welcome. Uh, each episode, I share a quick example of how I used AI that week. This week, I used AI to transcribe my coaching calls, obviously. Hopefully, as I was going through, uh, talking about that skill, you were like, maybe like, hey, wait, transcripts from where? What are you doing? Uh, so I use AI. I use a tool called granola, um, and I absolutely love it. I've been using it for the past few weeks. Uh, maybe you've seen ads for it. You probably will after this episode, because they're always listening. Um. Um, I. I'm a big fan of Zoom. Um, and I use the Zoom AI all the time. And, you know, it produces the notes at the end. I really like them. I think it does a really good job. But granola is better because it's just right there. It's. It's going and taking notes as you are. It do. It's transcribing in real time. As you're on the call, you can take notes inside of it as well. Uh, and then right immediately afterwards, it produces an AI, um, call notes, if you will. There's different styles you can have it do. It's. And what it produces is so good. The way it organizes it. It pulls from across the conversation. It's not just, like, bullet points, you know, of the conversation. It pulls different things in so that it can make this. This outline is so, so, so good. So if you haven't heard of it, check it out. I have no affiliate link. Um, I'm on, like, the free plan. I don't know. We'll see if it, like, actually starts charging. I don't know, but it is definitely worth it. You do have to connect it to your calendar, which I didn't want to do at first. And then I was like, what the Am I trying? What am I hiding? I don't know if I can hide. Um, so I sucked it up and just did it. And it's worth it, right? The notes that it generates immediately after the call are so good. And again, like I said, it's. It's transcribing in real time. So you can scroll back and you can see this. You can see the transcription. You can scroll back if you just missed something, um, or if you just want to reference something verbatim, like maybe a client said something like, oh, that was really good. You can just go open the chat right there and see it and scroll back and get it. It's just, to me, is just so good. It is the difference between, you know, using Siri for just dictation and using, uh, Whisper Flow, which I did a whole entire episode on. I'll link that in the show notes, right? With Whisper Flow, it's just because it's AI like, it does hears you correctly and it. And it learns and it's. It's really good. Um, so you also don't have to, you know, have an annoying note taker. So like whatever that. The fireflies or whatever that thing was otter, like, you don't have to let anything into the zoom room. It just is taking notes as you go. Um, you don't have to sign in to Zoom afterwards, like get the transcript or get the notes. It's all just right there. It's super convenient, it's super sleek, and most importantly, the output is really, really good. So that's the main way that I use AI the second. The second piece, little like just to nerd out again, is that I used the granola MCP. So remember, it's episode 51. I'll link that one as well. What's an MCP? Right? What the is MCP? Uh, MCP is model context protocol. But also it is a thing and it is a, uh, a connection, right? So it can say it is.
[00:20:29] You can have m. An MCP server and you can also have this MCP as like, this is the protocol. But either way, it is how. In this case, it is how Claude connects with an outside tool, right? It's how AI connects with another tool. That's what an MCP is. Um, and so in this case, this is how Claude connected with Granola. So, you know, earlier I said that as part of the skill, it pulls the transcript and then it goes through, summarizes all that stuff.
[00:20:58] Summarizes, analyzes it. Um, and you have to kind of have conversations back and forth. How does it do that? Via this mcp, which is just like going into Claude and. And basically clicking a button and then signing into granola. And it's a connector, right? Um, but via that. I was unable when I was inside of Claude to just be like, hey, go and pull all of the transcripts from my recent calls and let's build this skill. As simple as that. And to me, that is really awesome. Like, copy and paste gets annoying, it gets tedious, it gets messed up. You know, that was the big impetus for me building out that podcast, um, automation, because I'm like copying from this thing over to that thing. And still it saves so much time and something like this saves so much time where it's able to go and get those transcripts. And all I have to say is, hey, can you get the transcripts, Reference these transcripts. And it's connected to um, granola and it's set to go. So I love it. Um, realistically, you know, this is a use, this is a reason to pay for granola. So this way you have like more backlogs of your, those transcripts if you want. Um, but.
[00:22:06] And then you can, you know, use, use Claude and talk, talk to Claude about analyzing them, pulling from them, referencing them, which is pretty cool. Um, so yeah, there's that you can set these things up to go automatically, but there's a little bit outside of the scope of this little section. Um, but suffice to say I, uh, you. These are the ways that I used AI this week. I think it was really cool. Uh, and I encourage you to, to give it a, Give it a shot, give it a try if it, if it tickles your fancy. But I'm looking at the time that my friends is all today. Hopefully you found this episode helpful. We nerded out a little bit. Uh, if you did find it helpful, consider leaving a rating or a review. I love them, I really do. So don't forget I also have a companion newsletter and bloggity blog, the Curious Companion that drops every Thursday. And it's basically the podcast. By basically I mean exactly the podcast episode in text form. I do put a little bit more in into the podcast episode, not gonna lie, um, like, you know, little tangents and things like that that I would put in here that I don't put into the newsletter. But if you prefer to read or you just want a written record of things, join the newsletter or check out the blog. You can head to prompting Curiosity.com forward/newsletter or forward slash blog. Or keep it simple and just check out the link in the shadows.
[00:23:24] As always, endlessly appreciative for every single one of you. Until we chat again next Thursday, stay curious.
