The argument in one line.
Building a persistent AI co-writing system in Claude Code with structured context profiles and reusable skills produces consistent, branded output without one-off prompting.
Read if. Skip if.
READ IF YOU ARE…
- A newsletter creator or Substack writer with 6+ months of publishing history who wants to scale output without hiring help.
- A content professional (ghostwriter, SEO writer, copywriter) managing multiple client voices and needs a system to switch contexts without rewriting briefs.
- A founder or business owner shipping regular written content who's frustrated rewriting the same brand guidelines and voice notes for every prompt.
- A marketing team at a startup that needs to generate on-brand content variants at scale without each person recreating the same setup.
SKIP IF…
- You're brand new to Claude and haven't tried basic prompting yet — this assumes you already know what Claude can do and want to go deeper.
- You write primarily in fiction, narrative, or creative work where context engineering and JSON profiles don't map to your process.
- You've already built custom GPTs, n8n workflows, or Make.com automations for content — this video is foundational tooling, not advanced orchestration.
The full version, fast.
Most AI writing fails because models lack context, not intelligence � so swap one-off prompts for a persistent file-based system inside Claude Code that loads who you are, who you write for, and how you sound every session. The architecture is four parts: a CLAUDE.md system prompt that defines rules and routing, structured JSON context profiles (Voice DNA, ICP, Business Profile) generated through interview prompts, a knowledge folder holding your past content, and reusable Claude skills that package expertise like Substack notes or LinkedIn posts. Open the project in Cursor with the Claude Code extension, ask for content in plain language, and Claude autonomously reads context and invokes the right skill � producing on-voice, on-audience output without re-prompting.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Where the time goes.
00:00 – 01:14
01 · Cold open + promise
Opens on the thesis — Claude Code is not just for developers. States the promise: a full co-writing system, downloadable starter kit included. Lists 5 things covered in the class.
01:14 – 02:36
02 · Audience targeting
Who this is for: newsletter writers, Substack/Beehiiv creators, email marketers, social media, business owners & CEOs scaling their voice, ghostwriters, SEO, brands and teams.
02:36 – 05:38
03 · Why AI writing fails
The core problem framing: AI doesn't fail because it's not smart enough — it fails because it lacks context. Identifies 5 failure modes: generic output, inconsistent voice, heavy editing required, re-explaining from scratch each session, prompt roulette.
05:38 – 08:24
04 · From prompting to systems
Side-by-side comparison: Prompt Engineering (one-off, no memory, inconsistent) vs Context Engineering (persistent, predictable, voice-consistent). Formula: context + instructions = quality output.
08:24 – 09:44
05 · What is context engineering?
Defines the 4 pillars: structured information (organized JSON profiles), persistent files (context lives in the project), reusable instructions (skills written once), system prompts (rules that define how AI operates).
09:44 – 12:25
06 · Why Claude Code over Claude Desktop
Argues Claude Code is categorically different: file system access, bigger context window, reliable MCP connections, persistent CLAUDE.md, skill discovery, agent support. Claude Desktop cannot replicate this.
12:25 – 14:51
07 · Access and pricing
Recommends Cursor (AI-native editor) with Claude Code VS Code extension. Subscription vs API key — recommends subscription. Alex uses the $200/mo Max plan, runs Opus 4.5 all day, never hits limits.
14:51 – 22:43
08 · Folder architecture and CLAUDE.md deep dive
Walks through the full folder structure: .claude/skills/, context/, knowledge/, CLAUDE.md. Shows the actual CLAUDE.md template: system identity, context routing, workflow rules, skills guide. Gives away starter kit download for paid Substack members.
22:43 – 24:37
09 · Installing Claude Code in Cursor
Live demo of installing the Claude Code for VS Code extension in Cursor, opening it as a panel — shows it looks like a normal Claude chat, not a terminal.
24:37 – 26:52
10 · Context profiles: Voice DNA, ICP, Business Profile
The 3 core JSON context profiles that power the system. Explains Voice DNA as the most important — focuses on tone/personality, not just word patterns. Shows example JSON structure. Gives Substack members 3 creator prompts to generate their own profiles.
26:52 – 33:01
11 · Live demo: 20 Substack notes from a newsletter
Pastes newsletter into knowledge/drafts, asks Claude to write 20 Substack notes. Claude reads system instructions, reads Voice DNA + ICP, invokes the Substack notes skill autonomously, outputs 20 notes, then creates a folder and saves them — all without explicit prompting.
33:01 – 35:10
12 · Recap and offer pitch
Recaps the 4 components (concept, architecture, profiles, skills). Pitches paid Substack ($20/mo for starter kit + resources) and the Co-Writer System high-ticket training program waitlist.
Lines worth screenshotting.
- Context engineering beats prompt engineering because a system that knows who you are, who you write for, and how you sound produces consistent output across sessions — a prompt produces consistent output only for that session.
- The four root failures of AI writing are: generic output, no voice consistency, starting from scratch every session, and relying on prompts instead of a persistent system.
- A CLAUDE.md file is the system prompt for your entire co-writing environment — every agent, skill, and conversation inherits it automatically.
- JSON context profiles (Voice DNA, ICP, Business Profile) are the identity layer — they tell Claude who you are and who you write for without consuming conversation context.
- The .claude folder is the command center of any Claude Code system — agents, skills, settings, and CLAUDE.md all live here and compound over time.
- Claude Code generating 20 Substack notes from one newsletter draft without additional prompting demonstrates that a well-configured system can run autonomously on familiar tasks.
- Claude Code is not developer software — it is an operating system for knowledge work that scales writing output the same way IDEs scale coding output.
- A co-writing system that knows your voice eventually produces output that requires minimal editing — the marginal cost of each new piece approaches zero.
- Ghost writers, newsletter creators, SEO writers, and brand teams all benefit from the same architecture: persistent context + reusable skills + autonomous agents.
- The 1% of 1% positioning is real: using AI to write + adding context around that writing + building a system around the context creates a compounding advantage most competitors will never build.
- Skills that auto-activate remove the cognitive overhead of remembering which tool to use — the system selects the right skill the same way a human expert selects the right approach.
- Reexplaining yourself to Claude at the start of every session is a symptom of missing context profiles, not a permanent cost of using AI for writing.
The folder IS the product.
JoeFlow / MCN+ playbook
Alex is selling a folder structure with JSON files and markdown skills — and it converts. The same architecture powers JoeFlow's Sessions, Chef, and Batch.
- The 'From Prompting to Systems' comparison slide is a ready-made format for any JoeFlow explainer — adapt it to show Sessions cockpit vs one-shot Claude.ai.
- Voice DNA + ICP + Business Profile as a setup wizard is a concrete MCN+ onboarding feature — three JSON-generating interviews before the user ever writes a word.
- Alex's autonomous skill invocation demo (Claude finds + runs the skill without being told) is the exact moment Joe needs to capture for the Chef orchestrator demo reel.
- The 'you wouldn't hire a writer without context' analogy maps directly to JoeFlow: you wouldn't hire an agent without giving it your CLAUDE.md.
- Alex's two-tier offer ($20/mo entry + high-ticket training) is the same shape as JoeFlow standalone + MCN+ — confirm this structure is correct before changing pricing.
Terms worth knowing.
- context engineering
- The practice of building structured, persistent files of instructions and background knowledge that an AI always has access to — contrasted with prompt engineering, which rewrites instructions from scratch each session.
- Claude Code
- Anthropic's command-line AI assistant that runs locally and supports persistent project files, skills, and custom instructions, making it more powerful than the standard web chat interface.
- CLAUDE.md
- A Markdown configuration file placed in a Claude Code project that acts as permanent system instructions — the AI reads it at the start of every session to understand its role and rules.
- context profile
- A JSON or Markdown file containing structured background knowledge — such as brand voice, target audience, or business details — that Claude loads automatically to personalize its output.
- Voice DNA
- A context profile document that captures a creator's unique writing style, tone, vocabulary, and patterns so an AI can generate content that sounds like the creator rather than generic AI.
- ICP (Ideal Client Profile)
- A structured document describing the target audience — their demographics, pain points, goals, and language — used as a context profile so the AI always writes for the right reader.
- Claude skill
- A Markdown file that packages specialized instructions for a specific task, which Claude Code can invoke autonomously when a relevant request is made.
- Substack
- A publishing platform for email newsletters and long-form articles, used here as an example content format that the AI co-writing system generates automatically.
- content operations
- The repeatable systems and workflows a creator uses to consistently produce, publish, and distribute content at scale — as distinct from one-off writing sessions.
Lines you could clip.
05:38
“Most AI writing fails. And it's not because the AI isn't smart enough. It's because it lacks context.”
05:39
“You wouldn't hire a writer without showing them your past work, explaining your audience, and describing your style. So why would AI be any different?”
03:58
“If you stop using this traditional sense of prompting your way through everything and start actually building a system, now you're in the 1% of the 1% of AI users when it comes to writing operations.”
08:15
“The formula really is context plus instructions equals quality output. Better input, better output every time. Simple as that.”
28:59
“You don't even have to prompt anymore. You just do the system setup once and you never have to prompt again.”
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
metaphoranalogystory
00:00In today's Claude masterclass, I'm gonna show you how you can build your first AI co writing system inside of Claude code, and I'm gonna give you every single tool, every resource that you need as a downloadable to do that and to get started. So as I said, master class Claude code for writing, how to build a co writing system that scales.
00:17And this is super important because, you know, I think, like, 99% of people out there who use AI when they hear Claude code, they just see that that second part of code, and they think it's technical. It's for developers, and this is just simply not the case anymore.
00:28And if you're not using Cloud Code for, you know, everything outside of development for content operations and writing operations, you're totally missing out. It's, like, a million times more powerful than regular Cloud AI, whether that's the web app or or the the website or the desktop app.
00:42And you're gonna see that once you start using Cloud Code for this type of work, it's gonna unlock things you they don't even know were possible. I use it every single day now. It it's, like, the backbone of all of my writing operations.
00:51So that's what we're gonna do today, give you the the the building blocks you need to start doing this. Alright? So Cloud Code for writing.
00:57Now I broke down five main points of what we're gonna learn today in this in this masterclass. The first is the concept, why context engineering beats prompt engineering. Then we're gonna look at the setup, how to actually access and set up Cloud Code.
01:07It's it's easier than it might sound. The architecture, we're gonna look at the folders and the files and how they connect. Then we're gonna look at the components, such as the Claude MD file, which is gonna be like your system instructions, context profiles, which I'm gonna explain, and Claude skills, which I'm gonna give you.
01:22And then you're just gonna see me building this from start to finish so you can follow along as well. Now who is this for? Like I mentioned, Claude code is not just for developers.
01:30So if you ever had that thought, get that out of your head right now. This is for anyone. Anyone who needs to to scale up or improve their writing operations and wants to use AI as a co writer.
01:41And this can be newsletter creators. This can be anybody on Substack, Beehive, email marketers. We can do social media creators, business owners and CEOs, founders who might need to scale their voice, professional writers.
01:52This is a big one. If you are a ghost writer or a tech writer or a a a content writer for a company, SEO, anything like that, you need to be using Cloud Code.
02:01Like, you absolutely need to be using Cloud Code, and you're gonna make a ton of money. You're gonna scale up your your your writing and improve the quality of it like you did not even know was possible. And then lastly, brands and teams.
02:11If you are a a marketing team, a team at a startup, anything, you really should be comfortable using this to, you know, scale up your your brand's content and target the right ICPs, etcetera.
02:22So that's who this is for, and that's just a small list. This is really anybody who who needs to, you know, write with AI in in any capacity.
02:30So if you take a look here, you can see this is kinda what we're gonna build here. I'm gonna show you all inside now after this. But what you're looking at here is Cursor, the tool I use, the coding tool I use with Claude code inside of it.
02:42And on the left, you can see the folder that I'm working in, which is basically my writing system folder. In the middle here, you can see actual Claude installed through an extension on Cursor, which I'll show you how to do, which is super simple. And here on the right is the markdown preview of what it is I'm working on.
02:55So in the screenshot here, this is actually me working in Claude code, setting up this master class that you're watching right now. So, you know, that that's it.
03:04It's all connecting in one way or another. Alright? Now before I show you exactly how to set this up, I just wanna cover a couple, like, really important points that I I just I talk about all the time, but it's just so important because there's so much bad AI stuff out there, bad AI advice.
03:19There's a lot of bullshit, a lot of just bad prompt libraries, bad just all around bad AI use and advice, and we need to break that.
03:28And that's why I talk about it all the time because, honestly, with the amount of people, like alright. If you think about it this way.
03:35There's already you know, if you're using AI in any capacity, you're already ahead of a lot of people who don't use AI. Right? And then if you're using AI to write, you're even ahead of more people now.
03:44Right? And if you are using AI to with context at the same time, now you're getting ahead even more.
03:52And among all of those those people, you know, probably 99% of them are just using prompts. So if you stop using this traditional sense of, like, prompting your way through everything and start actually building a system, now you're, like, in the 1% of the 1% of AI users when it comes to writing operations, and you're gonna stand out, and you're gonna go way ahead of them.
04:10Because, like, me on Substack, on Twitter, or X, or YouTube, or wherever it is I am, especially on, like, Substack and in more written content platforms, there are so many people still sharing, like, crazy amounts of prompts and stuff that that are just so, like, so, like, a year ago. You know?
04:25Like, it's shit I was seeing a year ago that was useful, maybe two years ago, and now it's just, like, getting old. Like, you should not be doing that. But alright.
04:32So most AI writing fails, and it's not because the AI isn't smart enough. It's because it lacks context. If you're using a simple prompt for everything, that's why your AI is gonna fail, not because it's it's not capable.
04:41Yeah. A lot of people talk a lot of shit about how AI can't do this, can't do that. That's not they just don't know how to use it correctly.
04:46So if any of these sound familiar, you definitely need to be building a system with Cloud Code. If your output sounds generic like everyone else using, you know, Chatbook Tea, if you don't have any consistency between sessions, different voice every time, if you struggle with that, if you often have to edit, you know, heavily or even edit at all, like, you have to do any serious amount of editing that takes up a lot of time, you shouldn't be doing that if you're looking for scaling up your production.
05:09If you're starting from scratch, always reexplaining who you are, what you're doing. This is a serious problem. You should not be doing that.
05:15And if you're playing on a, say, prompt relay, hoping this prompt works better than the last, throw this prompt or that prompt, like, please just stop. You need to move past that. Alright?
05:24The real issue is the AI doesn't know who you are, who you write for, or how you sound. Like, that's three of the key issues among many, many, many. But that's three of the fundamental issues that you cannot achieve without setting up a proper system, which Cloud Code helps us do.
05:38So this little quote here, I kinda quoted myself. That's kinda that's kinda it's kinda bad, but I quoted myself here, you wouldn't hire a writer without showing them your past work, explaining your audience, and describing your style.
05:48So why would AI be any different? Right? So, you know, that's that's applicable to anybody.
05:52If you're a founder or CEO and you hire a ghostwriter, you know, you would obviously need to share samples of your work, your voice, your audience, describe all of that to them. Same thing if you're writing for a team.
06:01If you're you're a company or you're doing marketing or anything like that, you would not be able to hire on new team members, new content writers, new whatever without explaining everything to them. So you cannot do you cannot expect AI to just magically know all this stuff.
06:14Alright? So what we're doing here is moving from prompting to systems.
06:19Alright? So prompt engineering, you know, this is a very simple you know, this obviously isn't a prompt here, but just for you to have an idea, it might look like something like this.
06:27Write a LinkedIn post about productivity, then you might have a you know, sometimes prompts can be a page, two, three pages long. You know, there there are some prompts that are really good. I'm not, like, a 100% against prompts.
06:35I use them. I give them away. I sell them.
06:37But you need to know where a prompt is is should be used and where it shouldn't be used. So if you say, you know, right here, write a LinkedIn post about productivity. Write this in a a more casual tone or a more business professional tone and do this and do that, you're just gonna get generic output.
06:50You're gonna get output, and guess what? You belong to that 99% who are publishing shit online that nobody likes to read, and it's not valuable, and it's just getting really annoying to read that as well. I'm sure you know it's getting very annoying to scroll on your feeds and see this over and over and over again.
07:04I don't even care anymore at this point. If I see something that I know is AI generated, I don't I don't mind that at all. That doesn't there's no that doesn't bother me.
07:10Right? Like, you can even look at some of my writing and say, oh, that was generated AI. I use AI writing.
07:14Like my whole thing is teaching you how to use AI writing. That's not the point. The point is it just provides a value, and it's just like this this words this just words that mean nothing, and and that's what we don't want.
07:24So the problem's not writing with AI. Problem's not even people noticing something's written by AI. The problem is just not providing value.
07:30Well, context engineering changes that. Right? Because context engineering, what we can do is load up things like our voice, our ICPs, our ideal client profiles, our skills, and then the output now is no longer based off of a prompt.
07:42Instead, it's all of this together. So it's writing in our unique voice, targeted to our unique audiences, using our unique skills, and that is where you start to provide value even when using AI for writing.
07:51Okay? So is also what we're gonna be building here. The difference, prompt engineering, one off prompts.
07:56You hope for good output. You don't have memory between chats, you know, with some caveats there, like using projects and stuff. You often have to re re explain everything or at least a lot of things, and you get inconsistent results.
08:06Well, context engineering, which is enabled by building something out in, Cloud Code, we're gonna get a persistent system, predictable output, memory built in, context always available, and a consistent voice.
08:17Okay? And the formula really is . Better input, better output every time.
08:23Simple as that. Now what is context engineering? Context engineering can mean many things to many different people depending on what it you're working on.
08:30You know? So somebody who is might be a developer, might have a different definition of what context engineering is to somebody like me who uses AI for writing. But for me and for us who are concerned with writing operations and scaling up with AI or co writing with AI, context engineering to us is structured information.
08:45Alright? We wanna give a lot of organized profiles in context, not just long markdown files or just, like, not we're not dumping everything into AI. We're being very selective in pointing it into the right areas, and also making sure that when it gets to those areas, the information is structured in the most optimized way for whatever it is we're doing in that at that time.
09:05The second is persistent files. We want context that lives in our project. We want context that is always accessible, always available to the AI.
09:13We don't wanna be surfing around for it, looking every time we're trying to, you know, write something. The third is reusable instructions. These have been made so much easier with the invention of, like, Claude's skills and things that we're gonna talk about here.
09:26Our system instructions, we need skills and instructions that we write one time and we use it forever. Or, hey, we use it forever.
09:33We don't need to write a prompt or copy and paste a prompt every time. We give the AI access to the skill that is just reused all the time. And then lastly, system prompts, the rules that define how our AI operates is super important as well.
09:45What is ClaudeCode? ClaudeCode now is what's gonna enable us to build this system. And ClaudeCode is Anthropic's official agentic coding tool, but it's secretly perfect for writing systems.
09:54So do this now before everybody learns this, you know, and and everybody starts doing it. It's perfect for writing systems. This is what it looks like here inside of, for example, Cursor.
10:03You can you know, a lot of people think it looks like that terminal. You see a lot of people use Cloud Code, and you see the little the little Cloud Code robot y terminal type stuff on people's computer. But if you just install install an extension here in cursor, it just looks like a regular Cloud Chat.
10:15It doesn't even look like coding. You forget that you're even in a coding environment. And why do we wanna use Cloud Code for writing?
10:22Well, first, is it has file system access. It can read and write our files in the project we're working in. So we're just gonna open a folder, and now Cloud Code can can literally run wild in this folder and do what we need it to do.
10:33And, you know, a lot of people, like, just before jumping on this this masterclass, I saw somebody answer one of my Substack notes saying, oh, you don't like, they were kinda saying, you don't need to use Cloud Code for everything. You could just use Cloud Desktop with all of its integrations. Because Cloud Desktop, like the app you can just download on your desktop, is not Cloud Code.
10:48We're talking just Claude desktop where you would be working in projects and stuff like that. They have extensions. You can hook up to MCPs.
10:54You can hook up to stuff like this, but don't get fooled. It's not the same thing as Claude code. Number one, the MCPs I find are way more unreliable on Claude desktop.
11:01It does not work smoothly like it works on Cloud Code being, like, your connections, MCP, which is your connections to other tools. It does not work easily, but not just that.
11:11There's a big difference between having Cloud Cloud in a folder, easily able to read, like, entire archives of our content, organize stuff, write stuff for us. It is so much more efficient if we're working with Cloud Code inside of a folder.
11:23Also, the context window is way bigger. You're not gonna run into problems like you do on Cloud Desktop. So believe me, I would not be telling you to write in Cloud Code and build all this in Cloud Code if you could just get the same quality output in Cloud Desktop.
11:35There would be no reason for me to do this. So you can believe, and my promise to you is that the moment that it becomes this easy to build somewhere else, I will bring that to you. But right now, Cloud Code is the best place to do it.
11:45Secondly, persistent context. We have a Cloud system file that can load automatically with every session. So every time Claude fires off and does something inside of your folder, it reads the system instructions first.
11:56We have skill discovery. We can upload in our folder Claude skills and allow Claude to autonomously invoke those skills as an agent whenever it wants or needs to. We don't need to copy and paste all over the place.
12:08We can work directly in our content.
12:12What happened here? And lastly, project based. We could build out different projects with different systems.
12:18So I have a folder for myself. I have a folder for each one of my clients. I have I have folders which act as different projects.
12:24Okay? Alright. So now we're gonna get into it.
12:27Let's let's start building this out now. Enough of the the the preliminaries. Now we build this out.
12:32How to access Cloud Code? Like I said, it is a you can access it in various different ways, but I use Cursor, okay, an AI native code editor with built in Cloud Code support.
12:44But, really, you can just access it in your terminal, integrate it with an editor, which I'm gonna do. You can even access it online, which don't do because it doesn't work the same way that we're gonna be doing here. So I highly recommend just downloading Cursor.
12:53It's free. And once you're in Cursor, you can install a Cloud Code extension, which I'm gonna show you right now, and it's gonna make it look exactly exactly like, you know, a regular Claude.
13:04Okay? Alright. So as for access, you can access it any of those ways.
13:07We're gonna be using cursor. And then what do you need? Okay?
13:09So you can access it two ways. You can either use Claude code with your Cloud subscription, your regular Cloud subscription, which I recommend, and I'll tell you why in a minute.
13:18Or you can use an API key and fire off with API costs. Okay? Now the problem with the API key is depending on how you use it, you know, if you're not cautious, this can add up quickly.
13:28You can, like, blow through API API costs, and it's just I just don't like it. I use the a b API for other things, but when it comes to this, building my cowriting system, I much rather just connect it to my Clod subscription.
13:40And me, I use the highest max subscription on Clod, which is 200 a month. Now this is not necessary if you don't write a lot or if you don't do a lot, but I'm imagining that if you're watching this video, if you're interested in building a system, it's probably because your career is somewhat built around writing operations, or you want it to be built around writing operations.
13:58And if that is the case, I do recommend just going for the highest plan you can afford. Because, look, at $200 a month, which is what I have, this gives me access to Opus 4.5, the the highest model they have.
14:09I literally I legit run Opus 4.5 all day long. I run you know, I write eight to ten hours a day every day. I write for myself.
14:19I run my entire content operation solo. I have multiple clients that I build that I run these systems for, and I never real I never hit my limit of collectors. I paid $200 a month, but I can't even begin to tell you the the return on investment I've gotten from that $200 a month.
14:32It's not even funny. I I couldn't even put a number on it. Okay?
14:35So that's what I say. But they they have other ones. You can they have the $20, the $100, $200.
14:39So, you know, you're just gonna have to make that decision about, you know, what it is. You could start off low, see how it goes. But that's what I use.
14:44And when you first log in to Cloud Code, on cursor, whatever, it's gonna ask you if you wanna use your subscription or your API key and just follow its instructions. It's super easy. Okay?
14:52What we're gonna build, we're gonna build a couple things here.
14:56We're gonna build a Claude MD system instructions. We're gonna build a context folder for who you are, who your brand is, a knowledge folder, and our Claude skills. Alright?
15:05And this is what the folder structure is gonna look like. I'm just explaining all of this for you before we get building. But you're gonna see this is what the folder structure's gonna look like.
15:13We're gonna have Claude. We're gonna have its skills. We're gonna have context knowledge in claude.md.
15:17This dot Claude folder is gonna be a hidden folder, but but we'll get into that. Alright.
15:22Enough of this, actually. Let let's just go. What you're gonna do to get started right away is for the paid members of my subsection, you're gonna have all of these resources available to you.
15:33And so it's super easy for you to literally just download this file I'm gonna give you and open it up in cursor. If you're not a paid member, you're gonna have to build it all from scratch.
15:41But if you're a paid member of my Substack or if you wanna come over and join us, you're gonna get access to this ClodCode starter kit that I built for you. Okay? And what you're gonna do here is you're simply just gonna download the zip file, and it's going to put this folder right on your computer.
15:54Alright? You can read through this Notion page here. It's gonna give you all the information that we're gonna talk about right here.
15:58But you're gonna download this. Alright? And what this is gonna download for you is right here is a Claude code starter kit.
16:04Alright? This is literally all you need to get started. So I just bypassed all of that extra work you would have to do manually building each one of these these folders, and now you can just use this.
16:14So you can rename it to whatever. I'm just gonna keep it as Quadco SorterKate. You can name it as, you know, mine's like Alex co writing system.
16:19You can name it whatever you want. And what you're gonna do is go into cursor, and this is what cursor looks like after you download it.
16:25And you're gonna click open a project, and we're gonna go to our downloads, you're gonna open this new folder I just gave you. And when you open this new folder that I just gave you right here in cursor, you are gonna see that I built out everything for you already.
16:38All these files I just told you about. We have the dot Claude hidden folder with my Claude skills inside of it. We have the context folder where you're gonna have your three context profiles we're gonna talk about.
16:48You're gonna have a knowledge folder that I just have some placeholders here for you to put, like, content drafts notes, but you can reorganize these folders, however makes sense for you. Then we have our collab.md system instructions, which I'm gonna show you here.
16:59Okay? And here, this is what cursor looks like.
17:03We have on the left our folder. In the center, we have where we will have our chats, our terminal, etcetera. And on the right, this is their little agent, which I actually don't really use, so I'm just gonna close this right now.
17:12And this is how we are going to start this out here. Okay? Now back to our little presentation.
17:18The dot Claude folder is a hidden folder that has all of your all of Claude's important files inside of it, and included in these files are your Claude skills. So Claude skills are package expertise that you can package inside of a folder, format it a certain way, and Claude can use that for really anything.
17:34Right? So if you take a look at the screenshot here, this is what my personal system looks like, and these are some of the skills I have in my Cloud folder. I have things like business profile creator, how to guides, LinkedIn posts, LinkedIn profile optimizers, sales email sequences, thought leadership newsletters.
17:48I have so many skills, like, many. I have over a 100 right now loaded up throughout for different for different things for Claude to do.
17:55And you could see that you also have a agents folder here, which is we're not even gonna touch the topic of agents today. That's for, like, the higher training program stuff we're gonna talk about later. The most but you can eventually put agents inside of here as well, which you cannot do with cloud desktop.
18:07So back to that point about somebody saying you could do everything in cloud desktop. Well, you can't put agents in here like we're gonna do, you know, in the future.
18:14But this is these are all the skills here. You're gonna have a couple to begin with. And the context folder is going to be where you put your context profile.
18:22So if you know me by now, you know that I always talk about how important it is to make context profiles in JSON documents. So gonna start with three here, a voice DNA, an ICP, and a business profile. I won't explain this again because I explained it so many times.
18:35Basically, what this is, a context profile is gonna tell the AI, you know, in a very optimized way, in a very optimized format, just context about your brand or whoever or whatever it is you wanted to know about. So for us, I have three fundamental context profiles that every single one of you should use.
18:49Doesn't matter what industry you're in or what you're doing, you should probably be using these three, and I'm gonna give you prompts to create these three. We need a voice DNA, the most important of all.
18:58This is what makes your AI sound different than every other AI as best as you possibly can, and you're gonna iterate on that, you know, over time forever until you can really get the AI to sound how you want it to sound. In ICP, who you're writing for, who you're targeting, who your audience is, and a business profile, just a general look at what your business is so the AI can just have a quick understanding if it ever needs so.
19:17So that's what we're gonna do with the context. The knowledge folder, which I showed you, where I put some placeholders there, like drafts and notes, you can organize this however you want, but this is exactly what it sounds like. It's your it's clogged AI brain.
19:28Okay? You can put your entire archive of newsletters, your company's content, blogs, notes, voice notes, anything in there.
19:35That's like a whole another thing we could talk about later, but I'm gonna leave that up to you to just set it up however you want depending on whatever it is you do. And then we got the claude.md file, which is a markdown file that sits in your project, which I'm gonna show you is the system prompt.
19:50And Claude reads it automatically at every session. This matters because you don't need to paste instructions every time. It automatically loads it.
19:58We have persistent rules, context routing. We're gonna tell it how it it should work, the skills it should use, and we're gonna read through that. Okay?
20:06Inside of your Claude dot MD file, you're gonna have a couple different things here, like your system identity, where it can find your different profiles, your context, workflow rules, and available skills.
20:19And here, we don't we don't even need to look at this because I can show you exactly what it looks like. So take a break for a second and think about everything I just told you.
20:28So I showed you all the different files, and now you can see what it looks like actually built out here. We have our dot cod file with our skills, as I told you. This is what a skill looks like.
20:36So this one, for example, is for Substack notes. Tells it how to write, for example, a Substack note. We have our context.
20:44So you would have like your business profile context in a JSON. You would have your voice DNA.
20:49Alright? We have knowledge where you would put all your different content, and then we have our system our system instructions here. Okay?
20:56I'll open this in, like, even preview so you could read it even better. But this is what our MD file looks like. So these are our system instructions for Claude.
21:03This one I'm gonna give you in the starting kit tells Claude that you are my AI co writer. Your role is to help me create high quality content that sounds like me and resonates with my target audience. Alright?
21:14Now as I mentioned before, Claude is gonna read this every time. This is the most important thing that you have in this entire system because this is the instructions for how Claude works in your entire system.
21:22So you can always update this, always make it, you know, iterate on it explain your system. But for us, we're gonna tell it it's our co writer.
21:29We're gonna break down the system architecture for it. We're gonna tell it that we have context profiles about who I am, who I write for, what I offered, where they're stored. We have the three context profiles of voice DNA and ICP, a business profile.
21:40We're gonna tell it that it has skills available to it, which are reusable instructions. These are stored in its dot cloud slash skills folder. These are packaged expertise.
21:49So instead of explaining how to write something every time, these skills contain the instructions once and cloud can use it forever. Then I give instructions on how to use this system. Before writing anything, Claude should read the relevant context profiles that are in our context, check for applicable skills in its skills folder, and then reference any past content in knowledge when relevant.
22:08Alright? And you can change this third one here. If you don't have content, you can take that out.
22:12If you wanted to not reference content unless you tell it to, you could change that. For example, reference past content in knowledge when relevant and explicitly told to do so. You could put that in there.
22:22Right? Then we're gonna give it a writing workflow when I ask you to write something, identify which context profiles matter for this task, read those to understand voice, audience, and business context, check if there's a skill for the content type, and if there's a skill, follow its instructions, and then this will produce content that matches my voice and serves my audience.
22:38Give it a guide of what our context profiles are. Okay? We give it a skills guide of everything we have in there.
22:44We tell it where everything's at. We tell it what I expect, and we give it a quick reference guide. And this is just a template.
22:49You can adjust this based off of how you wanna use it, and you can yourself ask Claude to adjust it whenever you need to. Alright? So that's what our our our settings look like.
22:58So now we need to put Claude code in here. And to do this on cursory, if you come here, you could see that there's extensions. And you're gonna wanna download this this install this Claude code for Versus code extension on cursor because this is just a really easier they're like a smoother way of interacting.
23:12So you're gonna install it and enable it. And once you do that, if you open up a new file here I see other people sometimes have, like, a Claude button that they can click.
23:20I don't know if I'm I don't have it or if I did something wrong, but here, I'm gonna click Claude code open. And then you can see Claude code opens right here in my cursor. So it looks like Claude.
23:29It just looks a little different, but this is Claud Code here. And this is access to it's running inside of our folders here. Right?
23:34So, for example, you know, we could tell it various we could we could tell it, like, oh, write a write a sub stack note or write help me write a newsletter.
23:44You can it's not like you could even be flexible with it. You can ask it for its advice. You can ask it for help, for brainstorming, or you could just target and tell it, like, do this specific thing, and and we'll see how all that works.
23:55But we need to finish setting up the system first. Okay? To finish setting up the system, we need the context profiles.
24:01So as I showed you, we have the clauda m d file, which is the system instructions. We have the skills already in there, which I gave you, two free skills, a sub site note generator, and a profile optimizer for your social media. And as time goes on, I'm gonna be dropping more and more of those.
24:15But we gave it those skills, and now the last part that's really missing are the context profiles, and these are what power everything. And like I said, they're structured JSON files that give Claude deep knowledge about you, your audience, and your business. We have the core three we're gonna do, the VoiceDNA, the ICP, and the business profile.
24:32Voice DNA. A voice DNA captures your tones, phrases, patterns, and personality in writing, but I created this technique for I created my own technique for making these voice DNAs in a very specific way because what I found was that oftentimes AI, when I gave it a voice DNA, it just wanted to repeat the same type of words or phrases I would use.
24:49So I worked really hard and still am working on iterating this. I I always am working on it. On making it not focus so much on my words, but more on my, like, tone and my personality and who I am as a person and what I sound like so that we can kinda break that if she's using the same phrases all the time.
25:02So this is an example of what it might look like. You know, we have, like, voice DNA, the core essence of who it is, an emotional palette, communication style. Then we have the ICP, who you write for, the problems, language, goals, objections.
25:14And you can have a a good idea of what that looks like here. We're gonna tell it who it is we're writing for, the demographics, the experience, etcetera, and then business profile, just what we offer, positioning differentiators, and and it's all formatted in JSON for the AI to understand very easily.
25:29So how can you create these now to to get your system up and running? You're gonna use the creator prompts that I give you. You're gonna run them in Cloud and answer the questions, whatever it is it tells you.
25:37You're gonna provide writing examples for your voice DNA. You're gonna get structured content, and then you're gonna save them. So here, what you're gonna do is come to this starter kit here that I gave you in the beginning where you could download the folder.
25:49And when you come down to the context section, you're gonna see that I also included three prompts that you can use to build your context profiles. The first is a voice DNA creator prompt.
26:00You're literally just gonna attach this prompt in Claude. You can use regular Claude. You can go on the regular website, the desk.
26:04It doesn't matter. You're just gonna drop this prompt in Claude, and you're gonna attach writing samples or, you know, you're gonna write if you don't have any. And you're gonna attach these samples and run that, and you're gonna get a voice DNA.
26:14The second one is ICP. You're gonna run this prompt. Same idea.
26:17Open a new chat. Run this prompt. It's gonna interview you.
26:20It's gonna ask you questions about your brand. You're gonna answer all these questions, and when it's done, it's gonna output a profile for you. And the last one is a business profile prompt, which is going to the same idea.
26:31It's gonna interview you, ask you some questions about your business, you're gonna answer it, and it's gonna format this as a context profile. So once you get those, you're gonna go back to cursor, and you are going to upload them into your context.
26:44So I went ahead and uploaded my personal ones that I I just did up here so we can practice with them. So you're gonna see here I have my voice DNA. You can see how big it is.
26:53This is my own, Alex's voice DNA. I have my ICP, and I have my business profile. So we have that uploaded.
26:59Okay? Now the skills, like we mentioned before, we have two in here that I gave you already. You can find these.
27:04There's repositories of skills. You can build your own. I I offer some in my community in Substack.
27:09In the training program that I'm gonna show you later, which takes us even to another, you know, level, you're gonna get all my agents, all my cloud skills that I use in my daily work. But these are your skills. These are two that you can just start with.
27:19This is what it looks like. So a skill is specifically built for Claude. It's called a Claude skill, and it's a folder that looks like this.
27:27And it comes with this skill dot m d, the skill markdown file, and it has to be formatted in a certain way. So you can see that this one's for sub site notes, so it generates high performing sub site notes. Alright?
27:35And this is what it looks like here. And these can be invoked, which is what I'm gonna show you at at any point. Alright?
27:41And then for knowledge, you would put your content like we like we mentioned. So let's just practice a little bit here so you can just get an idea of what this might look like. I'm gonna use voice dictation now, and I'm just gonna ask Claude, tell me what you understand about this system, your instructions, and my context.
27:57And that's it. It's just like talking to a regular Claude, but now it's Claude code, and it has access to everything here. Right?
28:02So I read through everything, and it knows now that based on its instructions, it's a co writer. Its job is to produce content that sounds like me. It knows its system structure.
28:11It knows its workflow. It knows its expectations and whatnot.
28:16If you could explain my voice DNA, how would you explain it? Go look at the the file. Let's just see if it can access the file so you can see.
28:25And here you can see that this is like a little bit of a different stuff you're gonna notice in Cloud Code. You can see what it's doing. So here it's reading my voice DNA.
28:32It's it read it. Here's how it's gonna explain it. You're a 28 year old founder who sounds like a friend showing something cool, not a guru selling transformation.
28:39You figure out how to use AI without creating garbage content, and you share what works with genuine excitement, not manufactured hype. Right? So that's just a a cool way of of seeing what it understands about you.
28:48Now if you another really cool thing you might not notice that you you probably didn't even know you could do in Cloud Code is this will save here. You can access your chats, and you can open up multiple. So we're gonna click this little plus on the top right, and now I can open multiple Cloud Code chats.
29:00So I can be working on so many different things at once. Right? And that that comes in handy.
29:07So now what we can do basically is let's let's try a full a full writing like operation here. We're gonna open up Cloud Code again.
29:16We're gonna open up a new Cloud Code. And this time, let's go let's go take a look at my substack.
29:27And let's say we'll go with this article I have here.
29:33I'm gonna copy it. This entire one I did on creating infographics with with Notebook LM the other day.
29:40And you know what I wanna do? I wanna put this in my my let's say I wanna put this in my drafts here. Okay?
29:45I'm gonna open a new file here, and I'm gonna say newsletter one.
29:54This is just a basic example. I'm gonna paste it in there. Alright.
29:57So we can either pretend here, I'm gonna save it, that you're building out your content inside of here so you have it saved or you just dropped it in, whatever. So let's say, you know, I'm working in the system. I just finished a newsletter.
30:08You know, I was talking to Claude. I was having help me. Hey.
30:09Help me write this newsletter. You know, here's a transcript I did of a video. Can you write the newsletter and my voice DNA, etcetera, etcetera?
30:16And now we have it saved into our drafts. So how this system can literally work is we can just tell it, hey. Make a sub make me 20 Substack notes based off of my latest newsletter and save them somewhere.
30:29That's all we gotta tell it. You don't even have to tell save it somewhere. You just tell it, hey.
30:32Write me these notes. You don't have to prompt it. You don't have to do anything.
30:34And what it's gonna do is it's gonna invoke its skill to write the notes. It's gonna look at its context profiles, and it's gonna do it all for you autonomously. So let's just sell it right here.
30:43Write me 20 Substack notes based off of my latest newsletter in your knowledge base, and we're gonna tell it that. You're gonna see it's thinking here.
30:52We're gonna open it here. The user wants me to write 20 It's gonna be a little hard to see here, but 20 sub stack notes based on their latest newsletter.
30:59First, I should read the context profiles to understand their voice and audience, then I should check the newsletter draft they have open. Then I'm gonna use the sub stack note skill if it's available. So it reads everything and then invoke the skill.
31:08You can see right here the skill that it's using is this subsect note skill I showed you. It invoked it itself, and you have to give it permission every time it does stuff like this. So do you wanna proceed with this skill?
31:18Yes. Now it's using the cloud skill to write the subsect notes. You see how you don't have to prompt anymore?
31:23You just do the system setup once and you never have to prompt again. So it's gonna read the skill, and then it's going to output the final product.
31:31So here's what it's doing. It's reading the skill, and now it just created 20 substack notes for me like that.
31:37No prompting. Just telling it to create me the skill based off of my newsletter. Alright?
31:42Just gonna keep doing it, keep doing it. And you have to you don't even like, you shouldn't really even be scared about, like, you know, adjusting this folder. You can adjust it however you want.
31:51You can tell Claude to create folders for you, to delete stuff, to clean things up. So for example alright. Look.
31:57It created these 20 Substack notes for me. Right? Now I can literally tell it these are great.
32:02Create a Substack notes folder in my knowledge base and put all of these inside of it.
32:09And now Claude has access to this project. It can create folders by itself. So you're gonna see.
32:14I'm gonna close the thinking here just to get it out the way. But it's working here. It's going to going to create a folder all by itself.
32:23Okay. So after literally, like, a second, you're gonna see it's gonna ask for permission, allow it to write these notes. I'm gonna give it permission.
32:29It's going to create it. I'm gonna come over here into my knowledge, and now you can see there's a new folder here for substack notes.
32:35And you're gonna click, and there's a file here of all my substack notes that it just did. So, you know, like, can literally work with Claude as your partner, as your assistant to to organize things and whatnot. And here, if you don't like to look at things in native like markdown, you can come to the top right, you can open previews, and you can see, you know, you can see the the preview in markdown.
32:55So, yeah, that that's how you do that. That's how you set it up. That is a very simple way of setting up a writing system.
33:02I'm gonna run through this to make sure I didn't miss anything here. We talked about the skills, how the skills work. So how the skills work, you know, like I mentioned, very important.
33:10Claude will discover these skills automatically. Like, could tell it like, I could have told it, like, hey, invoke your sub sec notes skill to write it, but that's the beauty of it. You don't need to.
33:17It's an agent. It it knows how to autonomously invoke those. Alright?
33:21And do it all by yourself. Skills you can create, these are like skills I have. You can create substack notes, LinkedIn posts, email sequences, how to guides, sales pages, Twitter threads, newsletters, anything else.
33:31Anything you can think of, if you can write it, you can skill it. And then just for you to really just visualize this one more time, because I know sometimes it could be difficult visualizing that inside of a folder.
33:40So I tried to do many slides just visualizing this. You ask Claude, write a sub stack note. Claude's MD system instructions is gonna define the rules.
33:47It's gonna understand his task. It's gonna look at context profiles to understand your your your context, your business. It's gonna look at what skills it has available to write it, invoke that if it has it.
33:56And then in the end, you're gonna have output that sounds like you or your brand, that's targeted at your audience, that is using a expertise skill to do so. Alright?
34:04And your new workflow becomes as simple as that. Open your project, Claude loads the MD file automatically, ask for content, like write a LinkedIn post about this, write a sub stack note about this, help me do a newsletter.
34:14And I'm showing you very simple, you know, use cases. Listen. You can expand this out.
34:18We're talking I do this for all my newsletters, for all my content operations, for SEO, like, really complex SEO work I do inside of here. If you're somebody who does technical papers, white papers, I don't even know, governments government papers, consulting, anything like that.
34:32Like, this it's all it's all possible. Claude is gonna read the context.
34:38It's gonna use its skills, and you're gonna get the output. And that's it.
34:43You can keep building this now forever. Keep building on top of it and whatnot. Alright?
34:46So just to wrap this this class up, this master class, what we covered, the concept, context engineering beats prompt engineering. You need to build a system, not one off prompts.
34:55The architecture, we have the Claude MD file, the context profiles, and the skills. These all equal your co writing system.
35:02The profiles give us consistent outputs, tell the AI who we are, what we do, and the skills give it packaged expertise. Okay? Your starter kit is here.
35:12For my substack members, for my paid members, your starter kit is here. You can download this file. You can open this folder right in in on your computer and just kick it off right away.
35:20There's no need for you to start writing everything from scratch to start experimenting. Gonna get access to all of this, create your context profiles. It's it's really simple.
35:29You know, the thing I can tell you is I can sit here all day and teach you about it and talk about it, but there is nothing more valuable than just getting this damn folder, opening it up in collide code, and just start messing around with it. And it doesn't matter. Just test it out.
35:39If the first if your first whole system comes out and it's shit, it's not working like you want it to, which I highly doubt that's gonna happen, but if it happens, delete it. Download the folder again. Start again.
35:47Keep trying it out. That's the most valuable way you can learn. And like I said, now as for my Substack, if you're not familiar with it, we have the free tier where you're getting, like, this video master class plus some limited content.
36:00The paid Substack, it's only $20 a month, and you get access to all of this, the tools I'm talking about, the resources I'm talking about here, all of my weekly guides, my how to guides, um, on Substack. And then coming down right here really soon, the doors are closed right now, but it's the co writer system, which is my high ticket training program, which I've already ran version one, but now version two is in the works based off of building your entire writing system inside of Cloud Code.
36:21Think of everything you just learned right now, but scaled up 20 times, and me handing you all of my personal operations. Like, I'm gonna give you I'm gonna give you everything here.
36:30My actual MD file, which is multiple pages long. I'm gonna give you all the agents, which we haven't even talked about today, but you can create agents inside of Cloud Code, actual agents that I have five or six running for me right now that I'm gonna give you the exact same agents for you to put in yours.
36:45Every single one of my skills, my entire cloud skill library, which is, you know, numerous skills, context profile skills to build instead of using a prompt, you can use a skill to do it.
36:55Step by step builds along with me in exclusive community access to other co writers who are working building these systems. Right? So check that out.
37:04I'm gonna drop the wait list below as well. Make sure to sign up for that. That's coming real soon.
37:07I'm working on recording all of it, finishing it all up. It is the AI co writing system. Sign up for that.
37:13Share comments on this. Share it with anybody who you think needs to be building a writing system. Drop any comments below.
37:19Let me know any questions you might have. Answer any newsletter. You know where to find me.
The Hook
The bait, then the rug-pull.
Ninety-nine percent of people who use AI see the word 'code' in Claude Code and stop reading. Alex McFarland's thesis is that this is the most expensive mistake a writer, founder, or content operator can make right now — and his 37-minute masterclass is the corrective.
Named ideas worth stealing.
08:24list
Context Engineering (4 Pillars)
- Structured information
- Persistent files
- Reusable instructions
- System prompts
The 4-part framework that makes AI writing output predictable and voice-consistent, as opposed to prompt engineering which produces one-off, context-free outputs.
Steal forJoeFlow onboarding — this is the exact argument for why Sessions + CLAUDE.md beats just using Claude.ai
14:51model
Co-Writing System Folder Architecture
- .claude/skills/
- context/ (voice-dna.json, icp.json, business-profile.json)
- knowledge/
- CLAUDE.md
The four-folder project structure that gives Claude persistent identity, audience awareness, and packaged expertise across every session.
Steal forJoeFlow project templates — offer this folder structure as a downloadable starting point for power users
24:37list
The 3 Context Profiles
- Voice DNA (tone, personality, communication style)
- ICP (who you write for, their pain, language, goals)
- Business Profile (what you offer, positioning, differentiators)
Three structured JSON files that give Claude deep knowledge about creator identity. Voice DNA is the most important — Alex emphasizes it should capture personality and tone, not just vocabulary.
Steal forMCN+ onboarding — these 3 profiles as a setup wizard would make any AI-assisted content feature dramatically better from session one
05:38model
Prompt Engineering vs Context Engineering
Side-by-side comparison. Prompt Engineering: one-off, no memory, re-explain everything, inconsistent voice. Context Engineering: persistent system, predictable output, memory built in, consistent voice. Formula: context + instructions = quality output.
Steal forJoeFlow marketing — this comparison slide is a template for explaining why session context > one-shot prompts
How they asked for the click.
VERBAL ASK
35:11product
“For my paid Substack members, your starter kit is here. You can download this file, you can open this folder right on your computer and just kick it off right away.”
Two-tier pitch: $20/mo paid Substack (starter kit + weekly guides) as entry, then high-ticket Co-Writer System training program waitlist. Soft sell — no countdown, no urgency. Relies on demonstrated value from the live demo.
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
OTHER LINKSAlso linked in the description.
Visual structure at a glance.
