§ 01 · The Hook
The bait, then the rug-pull.
The promise is buried in the second sentence: not more coding tricks, but a different category of use. Riley Brown opens with a flattering claim his audience already believes, then pivots to the thing they do not know yet -- that Claude Code can run as a general-purpose agent with persistent, reusable skills.
What the video promised.
stated at 00:22 "We are gonna talk about how to set up Claude skills, how to use Claude code as a general agent, and how to set up your skill.md file." delivered at 24:55
Where the time goes.
00:00 – 01:13
01 · Cold open
Bold assertion that Claude Code is the best, then introduces Claude skills as the most important thing to study right now.
01:13 – 02:48
02 · What is a coding agent?
Lovable demo: types make a landing page about pizza, explains sandbox VMs, positions Claude Code as the superior local equivalent.
02:48 – 05:14
03 · Claude Code on your computer
Replicates Lovable in an empty terminal folder. Generates neo-brutalism pizza page. Runs on localhost:8080.
05:14 – 07:32
04 · Claude as general agent
Switches to Obsidian notes folder. Types do what is in my queue -- Claude finds the right file unprompted and writes a research plan.
07:32 – 10:12
05 · Skills intro: folder structure
Opens Cursor, creates .claude/skills/summarize/SKILL.md. Explains the description field is the trigger for when Claude invokes the skill.
10:12 – 13:50
06 · X Post skill from annotations
Drags annotated Twitter screenshots in as references. Claude analyzes patterns and generates skill. Red text equals annotations not post content.
13:50 – 19:57
07 · Typefully API integration
Adds Typefully API key via key.txt. Claude extends the skill. Tests live -- drafts appear on Riley Brown and Vibe Coding Explained accounts.
19:57 – 24:55
08 · Fresh context test
Resets Claude context. Fresh prompt fires the xpost skill automatically (green dot). Generated posts include GitHub link unprompted.
24:55 – 27:20
09 · Conclusion + folder diagram
Visual diagram of .claude/skills/skill-1/SKILL.md. Teases Part 3 agent workspace. Links to Luma webinar.
Visual structure at a glance.
Named ideas worth stealing.
13:09
concept
Skills as SOPs
Skills are standard operating procedures for agents. The description field is the trigger that tells Claude when to invoke the skill. High-quality description = reliable invocation.
Steal for Framing any repeatable workflow Joe does as a skill he builds once and reuses forever
00:00
model
Progressive revelation ladder
- Lovable (familiar)
- Terminal + local Claude Code
- Notes-based general agent
- Cursor + skills (the unlock)
Teach the familiar before the unfamiliar. Each rung lands the next one. By the time skills appear the audience is already sold.
Steal for Any tutorial video Joe makes about tools his audience has not tried yet
13:56
model
Annotated examples to skill loop
- Collect real examples
- Annotate them (why you made each choice)
- Dump into references folder
- Ask Claude to extract the skill
- Claude writes the SOP
You do not write skills manually. You give Claude annotated examples and it writes the skill for you. The annotation is the instruction.
Steal for Building any brand-voice, posting-style, or workflow skill for JoeFlow morning batch, Rewriter templates, or Chef prompts
Lines you could clip.
13:09
"All skills are, the way I think about them, they are just SOPs or standard operating procedures for agents."
18:37
"I turned Claude into my personal marketing team. Social media copy on demand, email sequences in seconds, brand voice locked in permanently. Zero prompting every time."
19:59
"Claude Code just became a general agent, not a chatbot -- an agent that does your workflows."
How they spent the runtime.
Hook length22s
Info densitymedium
Filler15%
Things they pointed at.
How they asked for the click.
25:56
link
"In the description down below this video, you will find a link to a webinar -- a big livestream where we are gonna create a really cool agent interface with a ton of different skills."
Soft close, no hard sell. Points to Luma event. Reasonable for a free educational video.
Word for word.
HOOK opening / re-engagementCTA the pitch analogy
00:00HOOKYou probably already know that ClaudeCode is by far the best coding agent in the world. It can create entire apps from a simple prompt, both front end and back end. It can blow your mind in terms of the design of that app. But what you might not know is you can actually use Claude as a general agent. And Anthropic released something really interesting a few weeks ago. They released what is called Claude skills,
00:25which is just a special type of markdown file which allows you to surround Claude code with relevant skills. So if you wished that you had an AI agent that had relevant skills, I believe that this is probably the most important thing that you could be studying right now. And so in this video, we're gonna talk about how to set up Claude skills. We're also gonna talk about how to use Claude code as a general agent, and we're gonna talk about how you need to set up your skill dot m d file and what folders, and I'm gonna show you how to just use AI to create these skills.
00:59Let's talk about building a general agent with relevant skills for your business, whether you're a content creator, business owner, aspiring entrepreneur. We're going to build an agent with relevant skills to you. And we're gonna use that agent. Let's dive in. Okay. So we're gonna start off by talking about what is a coding agent. You may have heard of Lovable, and this video is not about Lovable. We're not even gonna use Lovable. I just wanted to make a point real quick. So many people are talking about vibe coding. You can use tools like Lovable, Replit,
01:27VibeCode. These are coding agents that you can use to build apps. And whenever I go to Lovable and type in, uh, make a landing landing page about pizza.
01:41Right now, what it's doing is it's spinning up a sandbox, which is basically a virtual computer, and it's going to run an AI agent or an AI coding agent on the side. Now it's gonna generate a bunch of files
01:55within a directory or code base and those files are going to make an app. And so this is an AI coding agent at work building an app. Okay. So it's done. The lovable coding agent took my prompt, make a landing page about pizza. It generated all of the files. This coding agent decided what files it needed to generate. It generated them and then it probably reviewed those files
02:22and then it ran the program in a virtual computer somewhere so I can see it on my, uh, browser because this is within a web browser. And so this is an example of an agent made specifically for creating simple applications. A very narrow agent experience where it's made specifically for creating certain types of apps with certain stacks. And we could do this same thing on our computer. Right? We could do the same thing on our computer. If we were to go to finder,
02:53we can just go straight into our downloads. Right? We can go straight into our downloads and we could go new folder. We can go like this. We can say using Claude code.
03:04Right? So we don't even need any IDE or something like cursor or lovable. I can open a terminal window. I can open a new terminal at this folder and we can just run Claude. So now we're running Claude code in this empty folder called using Claude code which is an empty folder. And I can just say, please build a beautiful, amazing landing page
03:29for my pizza shop, Riley's Pizza Shop. Make it in the style of neo brutalism mixed with a friendly pizza energy. So just like Lovable, we can run this. And Claude Code in this empty folder is going to create the application right here. Okay. Check this out. It just generated an index dot HTML file and there we go. It created a file the same way that Lovable
03:57is creating all the different files and putting them in a folder in some virtual machine. It's just a virtual computer. We're doing this on our computer. It's doing the same exact thing Lovable is doing. And the point I wanna make here, you can just use these coding models on your computer and coding files, right, this is an HTML file and this is a CSS file. This is not the only
04:21type of file you can create with Claude Code. You can create any type of file. So it is done and if we just copy this link right here and we paste in localhost eighty eighty, we can run this on our computer. This is running locally on our computer where Lovable in Lovable's case, it just created a virtual machine and it ran it on that computer and it showed you what it created
04:46on the browser. Now we're running this fully on our computer and we didn't use any like software in between. We generated the code files with an AI coding agent on our computer. It used an AI model from Anthropic servers to create those files. So that's where it costs money. But we didn't use any other software other than getting Quad to generate all the files and we're running it on our MacBook Pro.
05:11HOOKAnd this is the website that it created. And you'll notice that in my opinion, I've tested every single one of these coding agents, just raw Claude code is the best in the world. So Replit and Lovable, they have their own coding agents. Claude code out of the box is better than all of them. And so what I want to explain now
05:32HOOKis how you can use this as a general agent. So this right here is Obsidian and this is just a basic notes app where every file instead of the files being an HTML file or a JavaScript file or a CSS file which ultimately created this
05:51HOOKapplication that we can use, instead of it being one of those types of files, these are all markdown files. So the way that I use general agents is I actually use Claude code at the same location as my
06:06HOOKnotes. Like this is my agent workspace. This is kinda how I think about this. This is a note book and Riley Brown uses this as his personal note book so that he can be a better CMO and co founder of the Vibe Code app. And I basically state all of my goals. Why am I stating this? I I know my goals. Why am I writing this? Because I don't use this by myself. I use this with my general agent which is Claude Co. And what I can do here is I can actually click on this
06:34or I can right click on this and hit reveal in finder. So here we can see Riley content creation. That's basically what I called this file and you'll see here all of these different files here,
06:48um, like this is just a wrapper of these files that are showing up on my computer. Right? This is just in my computer and so the same way that I ran code in a new folder, I can run Claude code in this folder. So I can reveal this in Finder. What I'm gonna do is I'm gonna right click services,
07:07new terminal at folder. Alright. New terminal at folder. Now, you can see here that it is running. So we're in this folder right here, which is the same folder where all of this is located, and now we can run Claude. And so now we have Claude code running
07:24in the folder that has all of my notes and so I can use it with Claude code. So I could go into tasks, right, and and this is how I keep track of my tasks. I could go into queue, let's say these are both done, right, this is these are the things that I wanna do, and so in my queue, one thing that I do wanna do, and I can just drag this in here, one thing that I do wanna do is I want to explain
07:48Claude skills to my audience and make a video. Right? This is one thing I really wanna do. Um, I want to, um, share the file structure, um, with of Claude skills, and then I want to take my notebook and examine what Claude skills would be useful for me to create. And I'm I'm, uh, do this onto this page. Don't create or edit any other files. Just make a plan and do it. Okay. So we just created this markdown file of this thing that I wanna do. The cool thing is I can just ask Claude Code to do it. I can just say, do what's in my queue. Right? That's all I need to say. I can, yo, do what's in my queue. And because the first thing that this agent's gonna do is analyze the read file,
08:31it's going to see how it's organized. Right? It's gonna know that there's tasks, it's gonna know what the queue is, draft tasks, so it's just gonna know to check the queue and because Cloud Code is such a powerful agent and it's really good at looking through the code base, in this case, which is just a bunch of my notes, it's gonna find the right location. It's really smart. And it's gonna find this page and it's just gonna do what I want it to do. And we'll just wait for Cloud Code to do that. And you'll see here that we never said to look at this file. It doesn't know that I have this file open. It just found what's in my queue, and you can see here it's examining notebook for skill ideas.
09:09It found the right file, and now it's creating the relevant things. Allow it to edit this session. Boom. Look at that. So it just added all of this. It added all of this to this document. It told us what skills are. Right? And it told us that
09:28from the Anthropic skills repo, it it basically did research. It's found the Claude skills repo and now it is telling us that you can create within the Claude dot m d file, you can create a skill by creating a folder with the skill name and inside that folder,
09:48it needs to have a Claude dot or a skill dot m d file and then you can put optional, uh, resources within that folder. So you can even put assets, references, etcetera. And I'll get to that in just a second. The point I wanted to make is the same way Lovable was a coding
10:05specific example of using an agent, this is a notes specific example for using an agent. So I promise you, we are we're working our way up to creating Claude's skills. I just wanted to illustrate the point. I wanted to talk about what a domain specific agent was like the lovable agent that can create apps. Then I showed you a note taking example where you can use Claude code as a general agent. Now we're gonna use it in a wide open world within Cursor.
10:33And we can do literally whatever we want. I like to use Cursor because it's just the best in my opinion for quickly using Claude code and it's ironic because I use we can just say Claude code skills. So what we're gonna be creating Claude skills in this project, I'll just name it that.
10:54And now it is opening up this directory in an IDE. So Claude cursor is just an IDE like Versus Code where you can view all the files. And so the way that I vibe code is I'm just gonna come in here and the first thing that I'm gonna do is hit command j. We're gonna open up the terminal and we're gonna type Claude. Then we're gonna press this right here to open up the side panel.
11:18And so now we have Claude code on the left or on the right controlling this directory which is Claude code skills. And what we're gonna do here is I'm gonna say please take a look at this direct or at this repo, and I'm gonna paste this GitHub,
11:38uh, to the Anthropic skills right here. Please create one simple skill in the right format so that you
11:49will use this skill when you need to. Okay. So right now it's creating an example skill. What it's going to do is it's going to create a dot claud file. It's going to create a dot claud file. As you can see here, it's created this dot Claude file.
12:07It's also going to create a skills folder as you'll see in just a second. Okay. There you go. It created a folder called dot Claude, and then it created a skills folder within that, and then it has other folders beneath it. So it created this summarized folder with a Skill MD. That's all you need to create a skill. You just need to put in the dot claud folder because Claude code knows to look in the dot claud folder for certain things like sub agents and now skills. We're not talking about sub agents in this video, we're just talking about skills. So we created the skills folder. And within this folder, right, we see this skill dot m d. And every single skill dot m d file should be formatted with the name
12:49and description. Right? It should have a name and description. You can see here that this one that I've created is called brand voice, which is skill number one, and then there is a name and description. And so the name is just a human friendly title. It can be anything.
13:05HOOKBut the crucial part is this description because it tells Claude Code when to use this skill. And so we're reaching a point where you can get agents to do things with a simple markdown file. But what you need to do is you need to make sure that you format it correctly so Claude Code knows when to follow your instructions. Because basically, all
13:26HOOKskills are, the way that I think about them is they are just SOPs or standard operating procedures, SOPs for agents. And so, right, these are instructions on what to do in certain scenarios. And if you put a high quality description, it'll know what instruction manual or SOP
13:47HOOKto use when it needs to do a specific task. And so I don't really wanna use this summarize skill. I actually wanna create a skill that's much more relevant to me. Okay. So I just wanna show you what I've been doing. I've been annotating. I've been annotating my Twitter posts because I can give these to Claude.
14:06I can put these in Claude's skills so the agent knows how I post. And so I can say, short video attached
14:14to this photo. This is a Twitter account that I run, by the way, that helps me grow VibeCode. And I can say,
14:22senior Vibe Coder is rage bait
14:28and works. Uh, don't do this often. Right? And I just put this directly on the image and it helps me kind of annotate my images here. And I'm just gonna hit I'm just gonna exit out. Let's save. And we can save all these. And so what we can do here is we can just take all of these files.
14:47We can drag them in here. And what I'm gonna do is I'm just going to put a new skill here. And what we could do is we could just drag these images in here. Now we're gonna go to Cloud Code. I want you to replace this,
15:01uh, summarize skill. And because this is a folder, right, I can just type at summarize. Right? I can say replace this summarize skill, this skill. The whole summarize skill should be replaced
15:17with a Twitter post x post skill. This skill should allow the agent to create tweets in my style and use the images that are in the at new in the new skill here folder as a reference. So please create that skill now. So we're basically going to tell Claude Code to analyze these images. It's gonna look for patterns in how I speak.
15:44And actually, one thing I can mention, I just hit the backspace. Just know the red text in these images
15:53are annotations. So that's not part of the post. These are my annotations to you on why I did certain things. So take that into consideration when constructing this skill. Right? So I annotated why I did what I did. So this gives some extra context to Claude code, and it's going to packages it package it as a skill. And so in theory, whenever I ask for a Twitter post to Claude code in the future,
16:13uh, within this repo, it should be able to do it in my style because I've given it relevant context. Okay. So there you go. It created this skill dot m d file, and all it did was analyze all of those images. It converted them into these, uh, markdown formats, and it converted it into a skill. What it didn't do is I didn't specify this. Uh, that was just a temporary folder. Please remove that folder, but add those images
16:42into that skill folder references because this doesn't need to be here. This can just go into the skill and right? So right now we have two skills. We're gonna add another skill in just a second. So we have one skill
16:56and every skill needs a claw dot m d file. So it just created this, and now it has these references that it can use right here. What we can do if we want, references to posts, and we can select all of these and we can just like move them in here.
17:14So we have references and then the skill dot m d file. You need this skill dot m d file. Right? And so what we can do, I want you to create a markdown file in the root.
17:29And when I say in the root, I mean, right, there's the the folder that we opened up originally, is Claude code skills. This is the root folder. So it should put it below the dot claud folder. The dot claud folder is also in the root. The skills folder is in the dot claud folder, thus not in the root. Just wanted to be clear. Uh, in the root and come up
17:52with 10 ideas ideas for the Twitter copy based
18:00on my last video where I explained Claude skills. Should match.
18:10And then I can, uh, give it that GitHub repo link. I mentioned skills and explained
18:18why it's important for general use cases for like marketing
18:26and as general agent. Okay. So let's say I made this video because I'm making this video right now. What I would probably do is feed in the transcript of this video and then it would generate 10 options. So it would have more context of the video we're creating. But since I'm making the video right now, I can't give it the transcript.
18:44Um, but that's okay. It's going to create a markdown file in the root and it's going to use this skill. Okay. So it is done. Okay. You'll see that when you see this green dot and it shows x post right here, when you see this green dot and it has x post, that means it's using that skill. Right? It is using the x post skill
19:07and we're gonna allow it to create these files and boom, Twitter ideas. So here we go. Cloud Code just became a general agent, not a chatbot, an agent that does your workflows. Anthropic
19:21just released skills, reusable prompt templates, invoke with slash commands, marketing teams. Okay. So this is really good. Oh my. This shouldn't be possible. I turned Claude into my personal marketing team.
19:34Social media copy on demand, email sequences in seconds, brand voice locked in permanently. Zero prompting every time. How? Claude skills. One markdown file, infinite workflows. This is very good. So it wrote this in my voice using this skill. But it gets much cooler than this. So if we were to go to,
19:59um, let's go to Arc. So we're gonna go to typefully.com. And in Typefully, we can go to settings
20:07and we could go to API and we can hit new API key. And so this is going to be test key
20:18for me, create API key. I'm gonna delete after this video and then what I'm gonna do is I'm gonna click on this docs and this explains exactly how to use this API key. So what I'm trying to do here is I don't want it to just come up with ideas, I want them to draft them in this tool where I can post directly to social media. So in theory, I could have an agent
20:41come up with tweets and automatically draft them within Typefuly because that's what you can do here on this platform. Right? We could have it directly post this
20:53as a, um, a draft here. Right? It can schedule a new draft. And so that's exactly what I'm gonna tell it. So I want you to add to our x
21:06post skill. I'm using the software Typefuly, and I want you to instead of basically, of putting these in a document, I want you to actually directly schedule these as a draft on Typefuly. And so I am going to give it these docs to the API. Here are the docs to read. Here is my API key. It is in a text file. Right? So I don't want to expose it. What I'm gonna do is I'm just gonna go I'm gonna go key
21:28dot t x t. I'm gonna paste my key right here. So I pasted my key in this file, and I can just go at key dot t x t. It is here. Please add to the skill using Claude's skills,
21:43and I'm going to mention this link right here. Please make this skill such that it schedules it as a draft
21:53to this Typefulli account. It should not post or
22:00schedule. Okay. Now we're gonna wait for the skill called XPost to be,
22:07uh, messed with. Okay. So it is done. And here we can see that your accounts available is Riley Brown, Vybe Code, Vybe Code explained. Okay. So this works. Maybe. Maybe.
22:20Uh, please test this by drafting
22:26one of the items in the list that we generated in and then we can just go Claude
22:36skills Twitter ideas. Okay. Okay. It looks like it is done if we come back here. Okay. It created this. Right? It created this on Vibe Coding Explained. Okay. Now on the Riley Brown account, draft
22:53a post that says, hi, this is a test. Okay. So it's done. If we go back to ARC, we go to Riley. Here we go. We see, hi, this is a test. This was thirty four seconds ago.
23:10So that's pretty cool. We can have it draft post. I'm gonna just reset the context here. So we're opening up a new Claude instance to reset it. Right? We're resetting it, uh, so that there's no context in here and I'm gonna say,
23:26I just you created a Typefuly video that explains how I created a Clog Code skill in order to post on all my Twitter accounts via Typefuly. I want you to draft up three x posts
23:44and draft them on Typefuly on the Vibe Coding Explained account. Oops. I misspelled it. Typefuly is the app.
23:55Okay. So we interrupted it, but it was using the expo skill. It should go back to using the expo skill. Okay. Do I wanna proceed? Yes. Okay. So the drafts were created. Oh.
24:08It has a link associated with it? Wait. What happens if we just go to this link? Wow. Skills repo. It even included that repo. No way. You can build custom skills for Claude code, automate repetitive workflows, create one click commands, share skills across projects, zero coding required. Just dropped a full walk through how to do this. Wow. Okay. So it created these
24:32posts using the Typefly or the XPost skill. And in this skill dot m d file,
24:40it has access to this code. Right? You can it has this code that I can that the Claude code can use to post this on social media. And what I need to do now is I need to delete my API key because I probably exposed it. Alright. So there you go. We just created a general
24:58agent and we gave it a skill. And you know what? I think that's enough for you to get started. I don't wanna like go crazy and create, like, an agent swarm or an agent with a ton of different skills. But we just created a we used Claude within cursor to create an agent, and then we just created one simple skill. And this is enough for you to get started because as you saw, I just gave it the link which I'll put in the description to the Claude code skills. So Claude code can make your skills for you.
25:27And remember, you just create a you know, you have the root folder. Right? And then within your project, just have a dot claud folder, then you have another folder,
25:39uh, which is your skills folder. And then within that, you just have the skill name. So this is just skill, uh, skill dash one folder. And then in here, you have a, uh, a skill dot
25:55CTAm d file. And within that file, it will have a name and description. And Cloud Code will do that for you. Just make sure that when you create these skills, right, because you can create many of these. You can create skill one,
26:08CTAwhich we did, um, and we didn't create more than one skill. Right? You can create as many skills as you want. I wouldn't go overboard, but if you wanna create a general purpose agent and use Claude Code, this is how you can do it and you can create a bunch of different skills. And so in the description down below this video, you'll find a link to a webinar. I'm just gonna do like a big livestream
26:30CTAwhere we're gonna create a really cool agent interface with a ton of different skills. We're gonna do it live. It's gonna be like three to four hours long. And so if you wanna sign up for that, that's in the description. It's gonna be sick. But, uh, yeah, this is just part two of my general agent series. I made part one on Obsidian and I'm just gonna continue this because I'm becoming much more interested in creating
27:00skills with also apps. We're gonna create a little interface where we can alternate between using an app that we create and the agent and we can create things that both me and the agent can use. It's gonna be like an agent interface or an agent workspace. And that's what we're gonna talk about in the next video. Uh, so thank you guys for watching. Peace.
— full transcript
§ 05 · For Joe
Steal the progressive revelation ladder.
format to steal
Start with the tool your audience already trusts, then walk them up a ladder to the thing they did not know existed.
- Open with a bold claim your audience already agrees with, then pivot to the thing they do not know yet.
- Build the ladder: familiar tool to your tool to the unlock. Three rungs, not one jump.
- Use live demos with real mistakes visible -- exposed API keys and misspellings build more trust than a polished cut.
- Name the abstraction on screen. The phrase you want to stick is the one you make visible.
- The annotated-examples-to-skill pattern is a product demo format: show input, Claude working, output. Replicable by anyone.
- The description field framing (it tells Claude when to use the skill) is a clean mental model worth stealing for any Claude skills content Joe produces.
§ 05 · For You
What you can actually do with this.
for the builder watching
You do not need to write your workflows from scratch -- give Claude one good annotated example and let it write the SOP for you.
- Pick one repetitive thing you do weekly: writing social posts, summarizing meetings, drafting emails.
- Collect five to ten real examples of you doing it well and annotate them with why you made each choice.
- Dump them into a .claude/skills/name/references/ folder and ask Claude to build the skill from your examples.
- Test it by opening a fresh Claude session and asking it to do the task -- if the green skill dot appears, it worked.
- Once it works, you never have to write that prompt again.
