The argument in one line.
Build a Claude operating system by stacking folders with contextual files (global instructions, memory, brand voice, workstations, clients) so Claude loads only the relevant context for each task and produces consistently high-quality outputs without repetitive back-and-forth.
Read if. Skip if.
READ IF YOU ARE…
- You're a solopreneur or small agency owner using Claude regularly who wants to systematize your prompts, brand voice, and project memory in one portable folder structure.
- A content creator, designer, or strategist who works across multiple clients and needs Claude to remember your brand guidelines, past decisions, and workflows without re-explaining them each session.
- You have the Claude desktop app installed and are comfortable with markdown files, but feel like you're starting from scratch with every conversation instead of building on previous work.
SKIP IF…
- You've already built a sophisticated Claude workflow using custom GPTs, vector databases, or other AI infrastructure — this teaches folder-based organization, not advanced integration.
- You primarily use Claude through the web interface and don't have access to the desktop app's file-linking capabilities that make this system work.
- You're looking for ways to automate complex business processes at scale — this is personal OS design, not enterprise workflow automation.
The full version, fast.
Claude works best when the right context loads into the right folder at the right moment, and a deliberate folder architecture beats raw prompting every time. Build a root OS folder containing a CLAUDE.md instruction file that defines global rules and a MEMORY.md that persists decisions across sessions, then layer in a brand-context folder holding three artifacts the model reads before producing anything: a voice profile, an ICP and positioning document, and a visual identity with design tokens. Nest workstations for distinct functions like content, finance, and ops, each with its own CLAUDE.md and memory, and nest client folders the same way so brand context never bleeds across accounts. Package repeatable processes as skills, route outputs to a projects folder, and use dispatch to trigger work from your phone.
Members feature
Chat with this breakdown.
Modern Creator members can chat with any breakdown — ask for the hook, quote a framework, find the exact transcript moment. Unlocks at T2: refer 3 friends + add your own API key.
Create a free account →Where the time goes.
00:00 – 01:11
01 · Cold open and promise
Most AI tools create more management than they save. Claude Code is different because you build around your own workflow. Promise: a system that knows your projects, voice, visual identity, and remembers previous work.
01:11 – 06:50
02 · Initial setup core files
Create LifeOS root folder with CLAUDE.md and MEMORY.md. Download starter templates. Connect Claude Cowork. Code view vs Cowork view walkthrough — code view is better for reading and editing markdown files.
04:12 – 06:50
03 · Memory system deep dive
CLAUDE.md tells Claude to open MEMORY.md silently every session, write to it on remember commands, maintain Last Updated timestamp. Behavior rules go in CLAUDE.md; project facts go in MEMORY.md.
06:50 – 17:05
04 · Brand context and starter skills
Three brand context files: voice-profile.md, positioning-icp.md, visual-identity.md. Three starter skills spin each up interactively. Demonstrated live: ICP asks 7 questions; voice profile extracts patterns from writing samples; design tokens scrapes colors from a URL.
17:05 – 25:35
05 · Workstations
Department subfolders (content, finance, ops) each with own CLAUDE.md and MEMORY.md. Parent rules inherit down. Routing map in parent CLAUDE.md. Key limitation: Cowork UI cannot walk up the parent folder tree; code view can.
25:35 – 28:25
06 · Multiple clients
Clients folder with brand_context, workstations, projects, MEMORY.md per client. Brand context is not inherited from parent. One command builds the entire structure.
28:25 – 30:07
07 · Remote dispatch
Claude mobile Dispatch tab sends tasks to desktop Cowork. Must specify folder context in the task text manually — no auto-injection.
30:07 – 31:51
08 · Closing and CTA
Recap three pillars: brand context, skills, project folders. Build skill documents for repeated tasks. MEMORY.md single-file has a ceiling at scale. CTA to Agenci Academy and next video on skill building.
Lines worth screenshotting.
- Claude works best when you inject the right context into the right folder at the right time — context placement is the core skill of effective Claude OS design.
- Using Claude Code out of the box without a folder-based operating system is like running a business without any file organization — technically possible, but wasteful.
- A CLAUDE.md file is the instruction manual loaded into every conversation; a MEMORY.md is where Claude writes persistent decisions — together they replace the need to re-explain your context every session.
- Separate department workstation folders — one for finance, one for ops, one per client — let you apply different rules and context to different types of work in the same system.
- Building your folder structure before you start working drastically improves daily output quality — the upfront investment pays back on every task.
- Mobile dispatch means your Claude OS is accessible from your phone without losing any of the context, memory, or brand instructions stored in the desktop folder.
- Brand context files — voice profile, ICP, design tokens — give Claude persistent knowledge of who you are and who you serve without repeating it in every prompt.
- Markdown files are both human-readable and optimized for LLMs — the same format that's easy for you to edit is also the format the model processes most reliably.
The folder IS the prompt.
Claude OS playbook
Every output quality problem with Claude is a context problem, and the fix is architecture, not better prompting.
- Start with two files: CLAUDE.md for rules, MEMORY.md for facts. Nothing else until the root folder feels limiting.
- Build brand_context/ from real writing samples, not invented — voice-profile.md extracted from your actual posts.
- Create workstations as subfolders for each domain so context stays scoped and does not bleed between departments.
- Wrap every repeated task into a skill document with a trigger, steps, and a defined output file location.
- Use code view when you need cross-folder context — Cowork UI cannot walk up the parent folder tree.
- Copy brand_context into client subfolders rather than relying on inheritance — it does not trickle down in the Cowork UI.
Terms worth knowing.
- CLAUDE.md
- A configuration file placed in a project folder that Claude Code reads automatically at session start, providing persistent instructions, project context, and behavioral rules without requiring the user to re-explain them each time.
- MEMORY.md
- A markdown file in a Claude OS setup that stores accumulated context — decisions made, projects completed, preferences noted — so Claude can reference past work across separate sessions.
- Claude OS
- A folder-based personal operating system built around Claude Code, using structured directories, instruction files, and context documents to give the AI persistent, role-specific knowledge about a user's work and brand.
- ICP (Ideal Client Profile)
- A detailed written description of the specific type of client a business is best suited to serve, including their industry, size, pain points, and goals, used to focus marketing and service delivery.
- Design tokens
- A set of standardized named values — colors, typography, spacing — that define a brand's visual identity and can be referenced consistently across design and development work.
- Workstation (Claude OS)
- A dedicated subfolder in a Claude operating system configured with context files specific to one functional area — such as finance, marketing, or client work — so Claude loads only the relevant rules when working in that space.
- Brand voice profile
- A documented description of a creator or company's tone, writing style, vocabulary, and communication principles, used to ensure AI-generated content sounds consistent and on-brand.
Things they pointed at.
14:35channelSkill builder video linked at end
Lines you could clip.
01:00
“Getting this structure right before you start building will drastically improve the quality of outputs you get day to day.”
14:35
“Skills are effectively process documents that help you repeatedly achieve a specific result given a certain process.”
29:09
“Injecting the right context at the right time, that is fundamentally what this is all about.”
30:40
“Out of the box, Claude is effectively a generalist. It can do everything pretty well, but actually your specific work processes are what is gonna take your outputs to the next level.”
Word for word.
metaphor
00:00For years, people have wanted an AI system that actually helps organise and execute on their projects both personal and for their work. But most tools end up creating more management than they actually save. And that's why Claude Code is actually different.
00:12Because now you can build a system around the way you work. One that knows your projects, your voice, your workflows and how you like to organise your life.
00:21And if you're still using ClawCode or Cowork out the box watching this video will massively improve the results you're gonna get. So we're gonna build out your own complete personal operating system from scratch using the desktop app. And by the end, you'll have a tool that repeatedly produces great results in your brand voice using your visual identity and always remembering your previous projects.
00:42It's gonna be portable for each one of your clients and also allow you to set up separate rules for each department like finance and ops if you need to. Getting this structure right before you start building will drastically improve the quality of outputs you get day to day.
00:56So let's get started building it. And by the way, as we go through, we'll be providing complete starter templates to make the setup incredibly easy for you.
01:05So if you haven't already, you're gonna go and download the Claude desktop app either for Mac or Windows. And then when you open that, you'll be on the latest version. So we can either work inside the Cowork view or the code view.
01:16So for the sake of this video, let's assume you're working inside Cowork. But what we actually need to do is go to our desktop and create a new folder structure and start adding in those files. So I want you just to go to your desktop, create a new folder, we'll call it LifeOS or WorkOS whatever you want to call it and we're gonna go and open that folder.
01:31Inside that we're gonna start with two files. First we have our Claude. Md file which is loaded into every single conversation.
01:38You can think of this as the instruction manual that's going to tell Claude Cowork how to behave. And if you've not come across a markdown or a dot md file before it's just a way of formatting text that distinguishes hierarchy of information.
01:51So think about headings with one hashtag, subheadings with two hashtags, bullet points as dashes, etcetera, etcetera.
01:58So it's what most of your files will look like, and it means we can actually just go into them and edit them really easily. It looks very human readable, but it's also optimized for the underlying models too. Then we'll have a memory file called memory dot m d.
02:10And as it sounds, we're gonna get Claude to write to this when we're making important decisions. So we're gonna tell it to do that from inside our claude dot m d, which as we've already discussed is always loaded into the conversation. Because I wanted to keep this nice and easy for you, I've created a base template for you to download for everything we're gonna run through today, and the link's down below in the description.
02:29So inside the resources folder, you can see a few different things, but we've got the Claude dot m d and the memory dot m d files, which we're gonna start with. So jump into one of those documents, then we're gonna need to download those as markdown.
02:42So we're gonna hit file, download, and then we're gonna hit markdown down here. And then we're gonna do exactly the same in the memory dot m d file, which is gonna store information about our active projects, about you, about the contacts you work with, and the key decisions we're gonna make.
02:56So we're gonna drag them across into our life o s folder. Now ultimately, what we're trying to do is make some simple changes to Claude out the box so that Claude knows us better and can get to a 90% version of a result that we've asked for without too much back and forth between it. That is simply the goal that we're heading for.
03:12So most things we're gonna end up using Cloud Code for are producing written content, doing some analysis and planning, or producing visual content. Now if we jump back into Cowork, we're gonna need to make sure that we're connected to our Life OS folder. So you're gonna click choose a different folder and then find the relevant folder and make sure that you're connected into that.
03:31Then when you run a test conversation, you can see that on the right hand side, we now have all of the folders that are in our underlying folder structure on the right hand side. So we have Claude dot m d that's appeared here, and later, the memory dot m d will appear as well.
03:46So we're gonna need to click into that Claude dot m d. It's gonna load up the folder instructions, and we can see all of the information in the markdown text formatting that's inside of our Claud dot m d session.
03:58Now that's not very readable. So, actually, a good way to view this is to go to the code toggle, start a new session in code, and do exactly the same. Connect up to the relevant folder down here.
04:09Start a test conversation. And on the top right here, you can open the files folder. Click on the Claude dot m d.
04:17Close the files view. And then if we expand that, you'll be able to see that Claude dot m d inside the code view is actually properly formatted markdown. So it views everything as we would structuring a proper document in markdown with all the title headers in bold, etcetera.
04:34So the code view is a much easier way to see and read and edit these files. So we're gonna run back and forth between the code and the co work view throughout the course just to get the best of both worlds there. Let's go through some of the rules that we've given it as a baseline.
04:48So we're effectively telling Claude, this is how I want you to work with me. So we've got how we work together. I want you to be conversational and clear.
04:56So just tell me the facts. I want you to be short and succinct. I want you to operate and give me your single best recommendation.
05:03Don't give me 10 alternatives. Just tell me what is the one recommendation you'll give me. I want it to flag when it doesn't know something.
05:09And also, we'll move on to this. Use my voice. Read my voice profile.
05:14Read my brand context anytime you actually produce assets that I want you to produce. Ask questions upfront on complex tasks. Don't just assume.
05:21And then this is important for the architecture later. I wanted to know which folder and file it's operating within. So if we're operating within a client file, we are only pulling that client context and not other clients context into it as well.
05:34And with our memory. Md file what we're effectively doing is telling it in Claude. Md to always open that memory.
05:41Md file and you can think of this as your short term important memory store, this file. It should read it. It should let that shape the responses but never mention that.
05:50And when I tell you something like remember this then it should always go ahead and actually save it to the memory dot m d file straight away. And importantly, I add anything to the memory dot m d, I'd like it to update the last updated so that we can maintain recent information only.
06:03Now ultimately, what we're trying to do is make some simple changes to Claude out of the box so that Claude knows us better and can get to a 90% version without too much of that back and forth. That is it. That is why we've introduced this Claude dot m d file.
06:15Why we've got the memory. Md file for short term recent memories. So when you distill it, most things we're gonna end up using Clawd code for are producing written content, doing some analysis and planning, or probably producing some visual content like and the likelihood is that will either be for us and for our own clients if we have any.
06:35So we need to give it the right resources to personalize it to you, and we can break this down into three files. And I've actually built some skills for you.
06:43We'll come to skills later that help you spin up these files for you really simply. But first, let's talk about what they are. So we have the idea of brand context, and this is how Cowork is gonna know your business.
06:55So we're effectively giving it three files or resources. First is the voice profile. So this is how you sound.
07:01Would people know that that content you've produced is you if it didn't actually have your face on it? That is what we're trying to achieve here with the voice profile. Then we've got the positioning and ideal customer profile or and this should be specific and pointed.
07:14So it's what you stand for, who your content is aimed at, and what your content is not or what it's against. And then finally, when we're building anything visual, we want to craft a visual identity. So think of this as your palette, your fonts, and also rules for why each one is actually there or when that specific accent color should actually appear in your assets.
07:32So if we jump back to the Google Drive and I've included templates inside here. You can open any of these and see the rough templates that make up these brand content file. But a better way to do it is just actually run through it and do this inside of Cowork using the starter skills zip.
07:45And this has got the three starter skills that are gonna spin out your brand context so that things start to sound like you and look like you created them too. So inside of co work, we're gonna go to customize. We're gonna hit skills.
07:56We're gonna add a skill here. And what we're gonna do is create a skill, upload a skill. So we've gone through and added all of the zip files as individual skills, and you should now be able to see the starter dash design tokens, ICP positioning, and voice profile.
08:09This one might be renamed, uh, to visual identity by the time you're watching this. And you can click into any of them and start reading about the skills.
08:16We have this preview window, and we have the voice profile builder. So it's gonna help us build a voice profile so that Claude writes like you, not like a robot. So if you've never come across skills before, it's probably the single most powerful concept that you will use inside Claude Cowork or Claude code, and they're effectively process documents that help you repeatedly achieve a specific result given a certain process.
08:37So for example, many people are gonna spin up voice profiles after watching this video. So I've created a skill so that we can do that in a consistent way so that every single outcome is a file, a markdown file that's in the folder brand context, and it's called voice profile.
08:52And we're basically doing that by giving it different steps. So step one is we're gonna ask you for writing examples. Step two, we're gonna extract voice patterns, and it gives it some rules for extracting those.
09:02Three, it's gonna actually come back and actually confirm that this sounds like you. And then step four, it's gonna write the file. So these are just starter skills that effectively help us create consistent voiceprofile.md documents, and you can do this for any of your tasks.
09:17Anything that you're gonna do more than once or twice, you should spin up as a specific skill, and all they are are process documents for doing a specific thing with additional context that we can reference. I've got plenty of other resources on skill building if you want to watch them too.
09:30So we've got these starter design tokens, the ICP positioning, and the voice profile. So to use any one of those and to test usage, there are two ways you can do it.
09:40We'll go back to the conversation window, and you can either invoke it as a slash command. So type in slash starter, and you'll see any of these. So let's say for voice profile, write me a LinkedIn post, or you could actually just say, write me something in my voice, and the description of the skill itself would tell Claude to automatically invoke that.
10:00So it says, let's build your ICP and positioning. I'll ask seven questions one at a time. Positioning first.
10:05What do you believe about your industry that most people in your space wouldn't say out loud? So most content is AI slop. We're looking for AI generated content that has a human touch about it.
10:15And you can see it uses this ask user questions feature to basically take our answers and convert this into a document eventually, which will be our ICP positioning dot n d. And you can take your own thoughts.
10:26You can take thoughts from others in the industry. You can go into the chat mode and you can get examples of things you should choose from. But this is just a starter point to actually generating these documents which will help all answers sound more like you.
10:38So what are you against? The advice given is based on shallow demos. A lot of the YouTube content about AI content is pretty shallow.
10:46It's not production ready systems. I'm trying to hit the angle of actually using this in a production business, and I do this on the daily with thousands of business owners inside my Agenci Academy too. So this is what we're all about.
10:58In one sentence, what differentiates about how you think or what you do compared to everyone else? I'm gonna say same as before.
11:05This is just a demo. Now it's moving on to our ideal customer. I'll say someone who can save time through plug systems.
11:12And you can also pick the option to help you think it through. It will come back with a few angles which will help guide me on specifically how to more information. So what specific problem drives them to find you?
11:23Which one of these best captures what your customer is actually losing? And I would say probably tried AI, got slop and can't ship. Most people try AI, they treat it like Chatuchi BT and actually end up getting pretty poor outputs.
11:36But building a system like this helps massively improve the quality of your outputs. What have they already tried that didn't work? Help me think it through.
11:42They've tried raw ChatGPT, AI courses, shallow demos. And you can see as we go through on the right, it's kind of got like all of the steps that we're going through in the skill has been put in this progress bar.
11:54So we've gathered positioning answers. We're going then gonna gather ICP answers. It's gonna synthesize and write that into brand context and then review with me or you and adjust.
12:04So this is a really clean interface in Cowork because you can see against the plan what is the progress that we've made against that plan as well as all the files and the context that we've used down here. And ultimately, Cowork just a UI layer built on top of the code interface at the bottom. And at the end, it spit out an ICP document.
12:20And you can see inside our LifeOS folder, we've got the brand context folder that's been created in ICP .md. So head back to Claude, and we can actually open the markdown documents, anything with m d, in preview mode by just clicking them on here.
12:33And we've effectively generated our ideal customer profile here. So here's what we stand for. Here's what we stand against.
12:39Who our content is for, what they care about, what they don't respond to. And any of this, we've obviously done just to, like, a shallow run through of this, but you can input a bunch of examples that will make this sound much closer to your brand.
12:51So now we're on the stage reviewing with Simon. Feels accurate. Now let's run the voice profile.
12:55Now here's where it gets really powerful. We built into this skill the ability to actually just take writing samples from you. So you could paste them from LinkedIn, or if you've already got a voice profile, you can paste it in here.
13:05You can actually connect this and get Claude Cowork to connect up to your Gmail to assess the last, you know, 20 emails that you sent. You can attach a link to, let's say, a blog post that you've written or a LinkedIn post, and it will go and effectively scrape the details of those posts.
13:20By providing those examples, it's gonna extract voice patterns across six dimensions. So you can see our progress has been updated here.
13:26So it says drop three to five real writing, LinkedIn posts, blog posts, emails, newsletters, whatever sounds most like you. And you might already have this tone of voice inside your chat projects. So I've got tone of voice examples.
13:38Gonna highlight all of that. Here's my tone of voice examples, and then I'm just gonna paste that directly. And it can now effectively start extracting voice patterns across the six dimensions that we've defined in the skill, confirm with sample sentences, and then it will write our voiceprofile.md document, which will then appear in the LifeOS folder.
13:55So you can see how you start building up a picture of this is who I'm targeting with the ICP and positioning. This is how I talk. This is my voice and my brand voice, and that sits in voice profile.
14:05And then your visual identity is gonna determine everything that you create, like slides, as well as the other brand context of the voice profile. And now it's gonna write that brand context slash voice profile. Now these are all starter templates.
14:17And inside our agentic operating system in our agentic academy, we have a full fifteen minute interview where it goes through and takes load samples of your voice profile to actually build something that is realistically, to a 95% version, exactly how you sound.
14:32And people have been really impressed with those results, but this will be a quick way to spin up an easy 80 version of what you sound like in your post. And it gives some examples there of how you sound, examples, etcetera.
14:45It's kept pretty elementary at this point like we mentioned. So I say, yes. That's good.
14:48Now my visual identities design tokens, please. And now it's starting to pull the context from the design token skill.
14:56It says, do you have a website I can pull your brand colors and fonts from? If you don't, then it's gonna guide you through a few questions to grab this manually. But we do and we're gonna go school.com/scrapes.
15:08And hopefully, it can pull some colors from that website. Now if we expand open this folder and zoom in a little, we're really starting to build out a folder structure underneath here.
15:17So we've got the skills that we had and can add our own skills there. We've now got the brand context with the ICP and positioning and the voice profile inside here. As well as how Claude should actually work, and we'll continue to run through that in a moment, and the memory file too.
15:31But it actually couldn't grab that information from school, it's either gonna grab it from there or I can give it a different website to grab that information from. And it's locked in a palette for us from a previous website.
15:41We're just gonna let it pick the fonts for us. But obviously if you have fonts then insert those there. You want it to look as much like you as possible or as much like your brand as possible.
15:52And we'll come to how you can do this for individual clients too, but we're just going through step by step on how to do it for your brand first. But the point still stands that effectively when we use certain skills that we create for ourself, it will leverage and we will tell it to reference certain design things.
16:08So say we are creating a skill that creates our LinkedIn carousels, and we want that to be consistent with our brand, then it's gonna read off for all the copy that it creates per slide, It's also gonna leverage the design tokens or the visual identity to work out the colors and the accents. There's an easy way to get to a sixty, seventy, 80% version.
16:28We can open the document in here. You can see it's got our colors in a table here. It's got our fonts.
16:33It's got our logos and assets if you've added them and a visual style as well. It talks about shadows, hierarchy buttons.
16:40So it's a basic design system that all of the things we produce from this point onwards that were visual will leverage. So you're really starting to build a picture here of your identity for your brand or your client's brand. We've run through the brand context, and effectively, we've taught co work how to know your business, your voice, your audience, and your visual identity.
17:00We've added a few of your first skills and used those to actually generate files inside your underlying architecture. And that's all we're doing here. We're effectively creating files that are loaded at the right time into Claude when it's needed.
17:13And all the skills that we set up are gonna be accessible across all of your projects, which brings us nicely to how to set up multiple workstations. So if you do have a finance team or you work on finance or you work on ops, you might wanna separate the context from each of those so the outputs for those specific tasks are even better.
17:32But before we move on to that, if you found it valuable so far, then do me a favor and hit the subscribe and like buttons below this video because they massively help me. Anyway, back to it with the routing map that we've added to claude.md for your different workstations.
17:47So you can consider a workstation as like a permanent area of your business that has its own rules. So we'll have its own claude.md document, which are rules that apply everywhere.
17:56That's what we're running through now. We have our memory. Md which is memories that it can leverage.
18:00We then have a brand context folder which contained information for our brand. And then now we add a workstations folder and within that we would have the different workstations with their own Claude.
18:11Md, their own brand context, their own memory, and their own resources or reference files.
18:17So what we're just starting to do is nest workstations underneath the parent folder so that when you're doing your monthly finances, it's basically using a different set of knowledge that you want Claude to work on versus when you're writing your LinkedIn content, for example.
18:31That's why we'd separate finance ops and content. And we pass it those rules by giving it its own claude.md, its own memory dot m d with specific decisions to that specific workstation that don't apply to all of your workstations.
18:45And it's important to know that everything we've already said in our parent claude.md are global rules that get passed down. So we don't need to repeat information inside the workstation called the m d.
18:54It will always read the parent and not replace that information. So it's only specific rules that we want to add inside this subfolder.
19:02So the way that this works is that we've got a routing map and Cowork is gonna use this table to decide which work station to load. So every time we set up a new work station, it's gonna be added into here. So say we want one for marketing, it's gonna add a new work station for marketing in this table.
19:18And what we've done in the global clause.md is talk about how we actually set up those work stations. So because this claude.md is loaded into the conversation, all I need to do is go into a new session, ask it to set up a workstation for me, and it will set up all of these things.
19:32A Claude dot m d, memory dot m d, a workstation name resources folder. So let's go back to co work and we don't need to go back to the folder structure because this will be created for us by Claude following our Claude dot m d document.
19:46So let's say set up a workstation for content, read through Claude dot m d to understand exactly what it's trying to set up, and it will then create our new progress tracker based on what it needs to do and start hitting those tasks and going through those tasks. And one thing I didn't mention was every workstation underneath the parent folder will also have access to all of the skills that we added into the global folder.
20:08So this will be able to access the brand context skill, the voice profile skill. It will have access in this subfolder. So this workstation handles our long form writing and our posts.
20:18And what I want you to understand here is that we're actually just creating more folders. It's a folder structure or folder architecture that supports injecting the right com the right context at the right time.
20:30That's how we make Claude perform better. We're just injecting the right information at the right time by building a solid architecture with parent inheritance and no overlaps.
20:40And it's created our structure. So we can actually click this button here to view the claude.md that is created for the content workstation.
20:47But we can see now that workstations has been created a content folder with its own Claude dot m d, memory dot m d, and resources dot m d. And we didn't have to create any of this. We can open any of these.
20:57You can see that we've got the content workstations memory, it's just a template, but it's been updated today. Now workstations don't have their own brand context, so they inherit the brand context that we have in the parent folder above it.
21:09Finance doesn't need its own voice profile, for example, but your other clients do, and we'll come to that in a bit. But for now if you wanted to work specifically within that workstation we would change the folder from LifeOS, choose a different folder and we'd actually go and the workstations folder and click into the content and actually work inside that content folder.
21:31So instead of working inside LifeOS, we're now working inside the content and we can say tell me what the from that specific folder, which we can also open through here, and it will understand that it's in the content work station specifically.
21:46That's how we work within the context of this. But it also has rules inside the Claude dot n d to read up the chain to the parent rules too.
21:55So this is your work station, the folder where all writing produced under your name gets drafted, edited, saved. And it's got a general workflow inside this claude.md. So remember claude.md gives us our how to instructions.
22:07So it's gonna confirm format and platform, check our voice profile is correct, and draft those and add them to a projects folder, which we'll come back to.
22:15And this means now we've got specific rules that can be applied to only the content workstation that we don't need to waste the context of other workstations for. For example, putting no em dashes substituting for commas, use commas, full stops, or rewrite the sentence. Every single thing we do doesn't need to have that rule, but our content production system does need to have that rule.
22:34And that's why it's important to have these workstations broken out. So let's go into our code view. We've got our brand for our root folder, our life OS folder.
22:42We've got our global instructions on our global memory. And now we have our workstations with content and ops, and we can set up multiple of those.
22:50And that brings us to the projects folder because we'll be producing a lot of content through each individual workstation. So we want an area to effectively store all of our outputs.
23:01So I'm gonna say create a LinkedIn post about Claude desktop updates. So it's starting to create the draft post, but I've just figured out a limitation of Claude Cowork in this method actually where code does act as something superior.
23:14So Cowork seems to not inherit the information for like brand context from the folders above it. So I told it that despite in the ClaudetteMD, it's saying actually to access that information, which is entirely possible in the code view.
23:28I said make sure to use the voice profile from the root folder. So it actually said there's no brand context slash voice profile inside the content folder, which if you remember is our subfolder. And it actually can't access it.
23:39So we either have to make a local copy in the content folder inside the workstations of the voice profile, I e go into the folder, copy the brand context folder, go into the workstations, go into the content, and actually paste in that brand context. Rename it to brand context and do it this way and therefore it has access.
23:57But as soon as you make changes to your voice profile, for example, at the parent directory level, gonna have to also make those changes down here. However, if you're actually operating inside the code view instead, and I just tried this out, you can just say create a LinkedIn post about Claude desktop updates. Make sure to use the voice profile from the root folder.
24:15It's actually able to go up the parent tree and understand the voice profile from the root folder despite the fact we are in the content folder, workstation folder when we're doing this. So it's ended up spitting out this result, which is our LinkedIn post that it's run.
24:30And then we'll say, great. Add it to our projects folder. It will inherit Clawd context, but it won't always inherit specific brand context folders that you spin up.
24:38And then you can see if you refresh the files screen up here that we've now got the projects. So inside that specific workstation, we've got the outputs of that workstation. Projects, content, and then dated, and then we've got our post dot n d, which we can actually open up inside here too.
24:54So we can see the post there, copy it directly from here, take it to LinkedIn, whatever, or connect it up to the next stage, which is actually scheduling it via LinkedIn. Or the alternative structure is actually instead of having that inside individual workstations, can have a global project folder and have the all outputs put into that global project folder.
25:14But like we said, you're gonna need to use code instead of co work to do that easily. But that's quite easy to amend in your Clorida MD, whatever structure you want, whatever structure suits your outputs there. Now everything we've covered so far is specific to your brand.
25:27But what happens if you actually have multiple clients? Well, if we reopen the Claude. M d, so exactly the same as a workstation, we would ask it to create a client, specific workstations which are nested inside those, or specific memory dot m d or claw dot m d files that sit inside that client themselves.
25:46And it's basically going to add the brand context, the memory, workstations, and project folders.
25:53So we'll go back and say add a new client called Acme Corp, and what we'll see again then is that client added. And we could have done this in Cowork. It would have been exactly the same process, and we're just setting up the infrastructure to manage multiple clients, workstations, and different brand contacts and projects within those.
26:10And it's now gonna add the clients folder and all of these subfolders within that. So we've not had to do a single bit of work. All of this is defined in the instructions that we give it inside the claude.md inside our parent folder.
26:21And you can see it's now clients, Acme Corp, brand contacts, and then you've got the individual files within that brand contacts folder.
26:29It should also have a Claude dot m d file inside there, so I'll make sure to update the overall Claude dot m d to to do that by the time you download this. And don't forget, they inherit everything that comes above it unless we specifically override it. So things like brand context will not be inherited.
26:45They will be taken from the folder that we are working from within. We need to make sure if we wanna work for a project or a workstation inside Acme Corp, then we are working inside that specific folder in our co work. So ultimately, what you've built today is a series of folders and context that feed into it.
27:03But we've set up the architecture for something much larger, which is being able to operate within multiple workstations for your brand and your voice to get the visual styles of your brand and to also operate within individual clients where they each have their own brand context.
27:18And all of our outputs can be put into one place and be produced at a certain level of quality that we're happy with. Now this folder structure is just scratching the surface on the quality level that you can achieve by injecting the right context at the right time, but that is fundamentally what this is all about.
27:35 and spinning up the right context. That's why we use things like skills, which are effectively bundles of context to do specific tasks.
27:44The Claude dot m d, which are global instructions on how short how Claude should operate for you. And the memory dot m d, where it's gonna actually pull things every single time we have a conversation with it and always remember those rules. Now we've not even touched on optimizing all those different things or taking skills off the shelf to do tasks like build your LinkedIn carousels, build your slides, build your Instagram posts, generate images in a certain style, and all of those different things you can do through packaging.
28:11Skills is something that we cover much more deeply inside our agentic academy, which you can find down in the link in the description below. Now a couple of things I wanna show you before we finish this course that will help you take this more mobile, which is the ability to actually operate this whole system with the underlying injection of the right context on the go.
28:30And that's using what's called dispatch. So inside the co op window, you go to dispatch. So then when you hit dispatch, you're basically able to, as it sounds, dispatch tasks from your phone to your computer.
28:40So on your phone, you'll make sure you have the Claude app downloaded. You can go to dispatch in the menu here.
28:47And then once you're there, you can actually insert tasks that you want it to do, like create me a LinkedIn post about core code versus co work. And then when I hit enter on that, you can see that this now appears on my laptop as well. So as long as my laptop or computer is running in the background, we're able to effectively dispatch tasks here.
29:05Now the one limitation of this is that actually it won't have full access immediately to all of the different files that you want it to. But you can see this is very different to this the voice style that I used earlier because actually this is just using ClawCode out the box and you have to specify which folder and which context to use if you're gonna use this dispatch function.
29:25Now I just wanna finish off by talking about the thing that is gonna make this so incredibly useful for you. So if you zoom out, we've got our personal brand context folder, which tells Claude how we like our outputs.
29:37We have our skills, which tell Claude exactly how to generate our specific business or personal or work outputs. And we have our project folders, which contain all of our specific contextualized results that have been output from Claude.
29:52But the one thing that I would build out from here is building out those skills processes. That is what's gonna take your outputs to the next level because . It can do everything pretty well, but actually your specific work processes, the way you consolidate your finances every month, all of that is very specific to the way that you work, and therefore, I'd highly recommend investing in building out those skill documents.
30:16We have a video specifically on how to build skill as well, which I'll link at the end of this video. But I just wanted to finish off by saying we what you've built today is a small snippet of a Claude operating system or what you might see called an Agencik operating system. And this system will do wonders for you over a period of time.
30:33We've covered things like memory, project outputs, workstation, clients, and all of that injects the context at the right time.
30:42But there are limitations to what we've built today. So storing memory in this way with one memory dot n d folder is good for a short period of time, but it becomes very data heavy and hard to recall information. So if you have lots of projects, lots of clients, or need to recall decisions or information you made, then we need to go and actually improve things like the memory.
31:02We need to build out all of the different process documents for our common tasks into skills like social media content creation, creating carousels, taking video transcript and repurpose them into blog posts, taking our meeting notes and converting those into action points. So all of these things, we've actually personally built out into our own agentic operating system using this structure, the underlying structure that we talked about today.
31:24And if you prefer not to actually just build all of this by yourself and get some plug and play systems straight out the box built using best practice, then you can just grab that out the box, and we continue to update it every single week with best practice. Because our aim is ultimately to save you a ton of time. So if you're interested in that, check out the link in the description below.
31:43And if you wanna see how to build out good skills and build those skills in a modular way into skill systems, then check out the next video.
The Hook
The bait, then the rug-pull.
Every AI productivity tutorial promises less back-and-forth and better outputs. Simon Scrapes delivers on that promise with a concrete folder architecture that turns Claude from a generalist chatbot into a personal operating system that knows your voice, your clients, and your workflows before you type a single prompt.
Named ideas worth stealing.
01:11model
Claude OS Folder Architecture
- CLAUDE.md global rules
- MEMORY.md short-term memory
- brand_context/ (voice, ICP, design)
- skills/ (process documents)
- workstations/ (department contexts)
- clients/ (per-client contexts)
- projects/ (outputs)
A nested folder structure where Claude reads context from the folder you are working inside, inheriting parent rules but not parent brand context. Inject the right context at the right time.
Steal forAny team or solo builder wanting Claude to produce consistent branded outputs without lengthy prompts every session.
06:50list
Three-File Brand Context
- voice-profile.md — how you sound
- positioning-icp.md — who you serve and what you stand for
- visual-identity.md — palette, fonts, asset rules
Three markdown files that tell Claude how you sound, who you serve, and how things should look.
Steal forAny Claude project where output quality and brand consistency matters.
08:50concept
Skills as Process Documents
A skill is a markdown file defining a trigger, step-by-step instructions, required inputs, and expected output file location. Claude executes it consistently every time. Anything done more than twice should be a skill.
Steal forRecurring content workflows: LinkedIn posts, carousels, client onboarding, weekly reporting.
04:12concept
MEMORY.md Decision Rule
Behavior or process rules go in CLAUDE.md. Facts about projects, people, or decisions go in MEMORY.md. Uncertain: ask Claude where it belongs and confirm. Every write to MEMORY.md updates the Last Updated timestamp.
Steal forAny Claude setup that needs persistence across sessions without external databases.
How they asked for the click.
30:07product
“If you prefer not to build all of this by yourself and get some plug and play systems straight out the box built using best practice, then you can just grab that out the box.”
Soft sell positioning the paid Agenci Academy community as the advanced done-for-you version of what was just built. Also plugs next video on skill building. Mid-video subscribe CTA at roughly 17:30.
Visual structure at a glance.
