Big Idea

The argument in one line.

A working agentic operating system is not the dashboard you look at, it's four layers built bottom-up: skills that turn repeated prompts into reusable files, memory organized as router files instead of a flat folder, routines that schedule those skills, and app connectors that link the agent to everything else.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…

You already use Claude Code regularly and want a structured way to organize skills, memory, routines, and app connections instead of prompting from scratch each time.
You're building or considering an AI consulting offer and want a packaged 'command center' concept to sell or demo to clients.
Your Claude Code workspace has grown large and messy and retrieval or context has started to slow down.
You want to move from manually running Claude Code sessions to scheduled, unattended automations.

SKIP IF…

You're brand new to Claude Code and haven't written a skill.md file yet, since this assumes you already know what a skill is.
You have no interest in cloud- or VPS-hosted agents or paying for an always-on service.

TL;DR

The full version, fast.

The video argues a slick agentic dashboard is only 20-30% of what makes a Claude Code setup productive; the real value sits in four layers the creator calls the ARMS framework: Applications, Routines, Memory, and Skills, learned bottom-up. Skills start as prebuilt or hand-written skill.md files and mature into 'thin skill, rich reference' packages triggerable headlessly with claude -p. Memory moves from a flat folder of files (one workspace hit roughly 60,000 files) to router files, a root CLAUDE.md pointing to department-level docs, and eventually a visual graph of how everything connects. Routines move from local Claude Code schedules that only run while a computer is on to an always-on cloud agent kept in sync via Syncthing, with a VPS-hosted Claude Code positioned as the likely next step. Apps progress from built-in connectors to Claude-assisted connector search to building bespoke connectors and full custom apps.

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 →

Chapters

Where the time goes.

00:00 – 00:43

01 · Intro

Cold open: Claude 5 models are more capable, but most people's agent setups haven't caught up. Promises a four-part framework and introduces the creator, Jay.

00:43 – 03:54

02 · My Agentic OS

Live tour of the creator's dashboard (RUBRIC): micro apps, calendar, email summary, skills deck, routines/scheduled tasks, an artifacts ring for past client work, and a link into a visual second-brain graph. Notes the dashboard doubles as a packageable service for AI consulting clients, with a mid-roll pitch for the RoboNuggets paid community.

03:54 – 05:09

03 · The ARMS framework

Argues the dashboard is only 20-30% of the value; the rest is workspace organization. Introduces the ARMS framework (Applications, Routines, Memory, Skills) as a pyramid learned bottom-up, and offers a free 9-page PDF guide.

05:09 – 10:35

04 · Part 1: Skills

Three levels of skills: Level 1 prebuilt or self-written skill.md files (accessed via the desktop app's skill directory, e.g. Anthropic's skill-creator); Level 2 'thin skill, rich reference' packages where skill.md routes to brand books, templates and other files; Level 3 headless triggering of skills via claude -p, wired into the dashboard's skills deck.

10:35 – 14:34

05 · Part 2: Memory

Three levels of memory: Level 1 a flat filedump folder (one workspace grew to ~60,000 files); Level 2 router files, a root CLAUDE.md pointing to department-level files like CONTENT.md; Level 3 a visual second-brain graph for navigating and searching the workspace.

14:34 – 18:33

06 · Part 3: Routines

Three levels of routines: Level 1 local routines in the Claude Code desktop app that only run while the PC is on (demoed with a daily YouTube-to-newsletter routine); Level 2 an always-on cloud agent (Hermes) kept in sync with the local workspace via Syncthing; Level 3 (coming soon) running Claude Code itself on a VPS so no separate sync tool is needed.

18:33 – 21:06

07 · Part 4: Apps

Three levels of app connections: Level 1 built-in connectors via the desktop app; Level 2 a 'search connectors' skill that finds official MCP servers or community CLIs/APIs (demoed on Adobe Premiere Pro); Level 3 building custom connectors (CLI Printing Press tool) and full custom apps, including the dashboard itself.

21:06 – 21:38

08 · Wrap up

Recaps the ARMS framework end to end and repeats the call to grab the free PDF guide from the description.

Atomic Insights

Lines worth screenshotting.

A visual agentic dashboard captures only 20-30% of the value of the system behind it; the other 70-80% lives in how the workspace and context are organized underneath.
The rule for turning a task into a skill: the moment you find yourself prompting Claude for the same task twice, make it a skill.
A skill file is not just a markdown file. The most powerful skills pair a thin skill.md router with rich reference files like brand books, templates, or code snippets.
Headless skill execution via 'claude -p' sends a one-shot prompt with a chosen model and effort level, letting a skill run outside a chat session and get wired into a dashboard or internal app.
One creator's workspace grew to roughly 60,000 files before he built a visual second-brain system, and the size was slowing down both retrieval and how fast his plan usage burned.
A CLAUDE.md file works best as a router, not a table of contents: it should point to department-level files like CONTENT.md rather than list every file the agent might need.
Local routines set up inside the Claude Code desktop app only run while the computer hosting them stays on.
An always-on cloud agent needs its own dedicated computer to keep firing scheduled routines even when the local machine is off.
Syncthing, a free open-source file-sync tool, is used to mirror skills and memory files between a local Claude Code workspace and a separate cloud-hosted agent machine.
Running Claude Code directly on a VPS is framed as the likely next step for routines, since it would remove the need for a separate file-sync tool between two machines.
The fastest way to find a connector for an app is to have Claude search for it: check for an official MCP server first, then community-made CLIs or APIs.
When no connector exists for a tool, a CLI-based 'printing press' pattern can generate a bespoke connector, demonstrated for MyFitnessPal and a community platform with no existing agentic connector.
The entire visual dashboard shown in the video is itself just one example of a custom application built on top of the same skills-and-memory foundation.

Takeaway

Organize Claude Code in four layers before you scale it.

WHAT TO LEARN

A capable agent setup isn't a prettier dashboard, it's skills, memory, routines, and app connections each built up in stages, starting from the bottom.

01Intro

A capable agent setup isn't a prettier dashboard, it's skills, memory, routines, and app connections each built up in stages, starting from the bottom.

04Part 1: Skills

Turn a task into a skill the second time you catch yourself prompting for it, don't wait until it's a recurring headache.
A skill folder that pairs a thin skill.md with rich reference files (templates, brand guides, code) produces far better output than one giant markdown file.

05Part 2: Memory

A flat folder of files degrades an agent's retrieval speed and burns plan usage faster as it grows; one workspace hit roughly 60,000 files before this became a real problem.
A root context file works best as a router pointing to department-level files, not as a single document trying to hold everything.

06Part 3: Routines

Local, in-app scheduled routines stop working the moment the host computer is off, so unattended automation needs either an always-on cloud agent or a persistently-running server.

07Part 4: Apps

Before browsing a connectors directory by hand, have the agent search for an official MCP server or a community CLI/API first, it's faster and finds options a manual search misses.
When no connector exists for a tool you rely on, a CLI-based connector-generation pattern can build one instead of waiting for an official integration.

Glossary

Terms worth knowing.

ARMS framework: A four-part model for organizing a Claude Code based agentic operating system: Applications, Routines, Memory, and Skills, best learned bottom-up starting with Skills.
skill.md: The markdown file inside a Claude Code skill folder that gives the agent instructions on how to execute that skill; in richer skills it routes to additional reference files.
claude -p: A Claude Code feature that runs a one-shot prompt headlessly, without opening an interactive chat session, using a chosen model and effort level.
Router file: A markdown file (like a root CLAUDE.md or a department-level file) that points an agent to where relevant files live, instead of listing or containing everything itself.
Second brain: A visual, graph-style view of a workspace's files and how they connect, used to navigate and search a large Claude Code workspace faster than a traditional file explorer.
Hermes: A third-party always-on cloud agent service the creator uses to run scheduled Claude Code routines continuously, independent of whether his own computer is on.
Syncthing: Free, open-source, peer-to-peer file synchronization software used here to mirror a local Claude Code workspace's skills and memory files to a cloud-hosted agent machine.
MCP (Model Context Protocol): One of the three common formats (alongside CLIs and APIs) that community or official connectors use to link an AI agent to an external application.
Connector: Official or community-built integration (as an MCP server, CLI, or API) that lets an AI agent read from or act on a specific external application.

Resources

Things they pointed at.

00:43productRUBRIC (Agentic OS command center product) ↗

05:56toolAnthropic skill-creator (prebuilt Claude Code skill)

16:58toolHermes (always-on cloud agent service)

17:17toolSyncthing ↗

19:30toolAdobe Premiere Pro MCP (community GitHub connector)

20:00toolCLI Printing Press (connector-building tool)

21:27linkRoboNuggets free classroom / Agentic OS PDF guide (code R66) ↗

03:33productRoboNuggets Claude Code Masterclass community ↗

Quotables

Lines you could clip.

00:00

“Claude has evolved with today's generation of Claude five models being a lot more powerful than everything that came before. But the way most people set up their agents and their operating systems have not caught up.”

clean cold-open thesis, no setup needed→ TikTok hook↗ Tweet quote

03:54

“This visual interface captures only around 20 to 30% of the value of this agentic operating system because the remaining 70% of the value really lies on what's underneath.”

reframes a flashy dashboard as the smaller half of the value→ IG reel cold open↗ Tweet quote

14:34

“A routine is just a prompt that Claude sends to itself at that given time that you set.”

compresses an automation concept into one plain-English line→ newsletter pull-quote↗ Tweet quote

16:58

“The reason why Hermes is so powerful is because it is twenty four seven. It's always on.”

punchy value-prop line for always-on agents→ TikTok hook↗ Tweet quote

The Script

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.

metaphoranalogy

Claude has evolved with today's generation of Claude five models being a lot more powerful than everything that came before. But the way most people set up their agents and their operating systems have not caught up. So today, I'll teach you this different framework of setting up your Agenic OS so you can get the full power from these new models to make your systems faster, have your setup cost less, and ultimately be more productive than you ever thought possible.

I'll also break this down in four simple parts so that by the end of it, you'll be using agents better than 99% of people. And if you're new, my name is Jay.

I spent over a decade working with brands you may know, have been in AI since my master's in data science. Now I'm running an AI business in one of the largest AI communities globally. Let's dive into it.

So this is my AgenTic OS or more specifically, the visual command center for my AgenTic operating system. And from this one view, I can access summary information for the applications that I use daily, like events in my calendar and time zones that I care about, a quick summary of my emails, including the messages that Claude is flagging as needing my attention.

There's also quick links here to some of the micro applications that I created myself and I use on the daily, which I'll talk more about in a bit. And I also have custom widgets here like this one for YouTube because, obviously, I do a lot of content. And here, I also have a view of my routines and scheduled tasks and which ones are going to fire on what time.

And for some specific skills where it makes sense to trigger them from this dashboard, I also have this skills deck where I can adjust the effort level as well as the model to be used for this specific run and also run it straight from this dashboard. And because these are widgets and Cloud Code is actually really good in creating these for me, you can freely adjust the size as well as the placement of these widgets and even create new ones depending on what you need.

And lastly, because of the way that I use Cloud Code where I just consistently ask it to create artifacts for me, I just had it make me this artifacts ring where I can just easily find the assets and artifacts that it made for me in the past. So for example, if I'm looking for artifacts for a client called Stropro, then I can just search for that and I'm able to also open this specific HTML file that it created for me back in the August 5.

And lastly, and very importantly, here in the center, if I click on that, that will just give me access to my own second brain system, which is important at least in the work that I do because it's crucial for me to visualize these systems so that I can explain them better and so that I can visually show the skills and other files in my workspace, which is important in my work.

Now I'll show more of that second brain in a bit, but really this whole AgenTek OS dashboard, it's great especially if you're a visually motivated person like myself. And this also works well if you're serving clients and if you're into AI consulting because creating something like this for a client is also a service that you can package and sell.

To give a quick example, this one's a design mock up for a financial services firm here in Australia. Here's another one for a company called Bito Green. And the point being is that if you have the foundations in place for your own personal AgenTek OS, then with today's really powerful AI models, then it's really quite easy to customize it to whichever client that you are serving.

And by the way, if you want to learn how to build and sell AI systems that businesses actually pay for, then that's pretty much all we do over at the RoboNuggets community, where not only do you get access to the Cloud Living master class, which we update every week and takes you from zero to mastery with the latest on AI, but you also get access to our agents as a service course, which walks you through how to actually get paid for all these AI skills that you are learning.

You also get to be part of a genuinely great community of AI builders. In fact, you can see just some of the recent wins our members are getting from the program right here. So if you want to start earning from AI, then check that just in the pinned comment below.

Now back to the video. Now having a dashboard like this is great. It's visually appealing, and you get to see all aspects of your work and your business, which is pretty much why I use it as my homepage now on the daily.

But I'd say this visual interface captures only around 20 to 30% of the value of this agentic operating system because the remaining 70% of the value really lies on what's underneath. Because a big part of this operating system is how you've organized your context and your workspace so that your system and your AI agent, in this case, Claude Code, works for you and not against you.

And there's many ways to organize your own operating system, but at least how I think of it personally and how I teach it in our community is via what I call the ARMS framework. It's very simple to remember because it's sort of like giving Cloud, which is your AI agent employee, its own ARMS, its own workspace. And the core idea here is if you figure out how to best set up these four aspects of your OS, then you'll be way ahead of 99% of other Agenic AI users.

And those four elements of the ARMS framework would be your applications that you use, the routines or scheduled tasks that you run, your memory system, as well as the skills that you have your agent use. And in my view, the best way to learn this is actually from bottom up. So for most users of these AgenTik AI systems, first, you learn about skills and then you set up your own memory system.

Then once you're confident there, that's only when you rise up and you can actually schedule your own routines or even create your own applications or create connectors for the applications that you're using. And so what I'll do for the rest of this lesson is I'll just go through this from bottom up. And for each of them, I'm going to share three levels of how you can use them so that you can just freely skip ahead to the ones that you don't know yet.

But given the depth of what I plan to cover in this video, for sure you'll pick up a few nuggets along the way that you probably haven't known yet. And by the end of it, if you watch the whole thing, then you'll be able to level up your own AgenTic skills to the point that you can also build out an AgenTic OS similar to what I have here but customized for your setup.

And also just to make it easy for you, I've also published this nine page PDF guide where if you read through that or just send it to your cloud code, then your agent will be able to guide you on how to set up this operating system as well. So you can just grab that in the description below. So now let's start with the first element which is skills.

Now you most likely have encountered skills before because they're basically just shortcuts to your SOPs, to your standard operating procedures. And the basic principle here is when you find yourself prompting Claude for the same task twice, then it's probably good to make it into a skill.

Now in the very first level, if you're just getting started, it's more than likely that the first skills that you access are the ones that are prebuilt by Ontropic. And the way you would have accessed those is in the cloud desktop app under customize. You'll have a section here for skills where you can browse the ones that Ontropic gives to you.

But from here, you can see that one of the more popular skills from Ontropic is this skill creator skill. And the reason why that is is that generally, I do advise people to create their own skills as well because each of our work is really custom to us. And so when you get to the habit of creating your own skills, the faster that you can get more refined and better results from Claude.

And you can find inspirations for skills that you can create everywhere. For example, a few days ago, I found this post on x, which looks to be a really good tip just to make your computer run a bit faster. So what I did a few days ago is to just paste that whole tweet and then just invoke the skill creator skill.

And if you send this, that will just let Cloud Code create that skill for you so that you can test it out, which by the way, at least for my workspace, this has been quite effective. So if you've been having trouble with Cloud Code or even Codecs clogging up your systems and making it slower, then this might be something that you wanna try out as well.

Now once you've created or tested out a few skills yourself, then you start to realize that there's a second level to this. Because contrary to what some people might believe, a skill file is actually not just the markdown file. Because some of the most powerful skills that you can add to your arsenal actually have rich references that they can pull from.

And just to make that a bit clearer with an example, if I go ahead and find that cleanup skill that we made earlier, this one's a pretty thin and light skill because it only has this one skill dot m d as you can see. And if we go ahead and open that, really, a skill dot m d is just a markdown text file, and this just provides instructions to your cloud code on how to execute this specific command.

But let's say if I find a more complex skill like this one for Robo, then you can see that this skill that I invoke with slash Robo actually has multiple files connected to it. And if I just open the folder of that so you can see better, then you can see in this folder that there is this skill file, which is the markdown file that gives instructions to Claude code on how to use this skill.

If we go back here, we can find that as well. And from this skill dot m d, you can see that it's essentially functioning as a router to these other reference files that are also from within that slash Robo skill folder. And the reason why that's important for this skill specifically is because this is actually the design system that I've been using for a lot of our videos and a lot of our company assets.

And so if I open this brand book HTML, you can see this provides us some guidance on the robo style. So it has guidance on the fonts.

It has guidance on the color palettes. And so having visual references for your skills like this works really well, especially for skills that are meant for design. And so for some cases where you want the skill to do more complex tasks, what you can actually do is to enrich it and not be limited to just one skill.md to house all of your skills as files.

So for example, this PDF guide that I was talking about, the reason why this is well designed and already knows our brand is because the way that I actually create this is using that really good slash robo skill. So to make something like this, what I do is just give a simple prompt like make a PDF guide with slash robo on how to set up an agentic OS, and then I just answer a few questions that Claude has so that it is aligned with my intention.

And because that slash robo skill is already so well defined with a lot of visual artifacts and references, I actually get a well designed PDF guide like this in just one or two prompts. And if you need a quick starter prompt just to let Claude find your thick skills and actually make them into ones with richer references instead of stuffing everything into the skill dot m d, then you can use this prompt to screenshot it to kick start that process.

And when you have those, now you can get to level three, which in some instances is actually useful to trigger skills even outside of your chat sessions. To give one use case, that cleanup skill that I created before, I actually added it to this skills deck so that I can trigger it from this page instead of having to open up Cloud Code in another terminal or chat session and typing out slash cleanup there.

Because most of the time, I actually run this skill whenever I see my device slowing down. And so when that's done, what it essentially does is also provide an output or a report similar to this one where if I open that, it just provides me with this summary of the results of that skills run. And the way that you run any of your skills headlessly, meaning without having to open up a chess session, is through this Claude feature called Claude hyphen p.

And without getting too technical, what that essentially does is spin up a quick session where it sends a one shot prompt to Claude using the model and effort level that you want. And you can see for this specific example, the only thing that was sent to Claude is this slash cleanup command. And so this becomes useful when you want to integrate your skills into dashboards like these or even as part of internal applications, let's say, that you want to create for your team or your company.

And like with most of the stuff that I'll teach you today, the great thing about the tools that we're using now like Cloud Code or Codecs is that you actually don't need to learn how to use any extra technical tooling in order to make this happen. As long as you're aware of this feature like Cloud hyphen p, then what you can do is copy a prompt like this, which you can just screenshot and send to your Cloud Code, and that can get you started with integrating any of your skills into your own operating system.

Now let's go to the next layer, which is memory. And it's important because the more that you use Cloud Code, the more context or files you have that you create. And at the very first level, what you have really is a workspace with a bunch of files.

So for example, for my case, the folder in my computer that I set as my agentic operating system or workspace is this folder called Robo. And you can see there's a lot of files and folders already in here. Now when you're just getting started and you only have a few files, it'll mostly be okay for you experience wise.

But the problem starts to arise when your context and your files build up so much that it's actually making it harder for your agent to find things. So for example, for my case personally, when I built out this second brain system and I pointed it to that robo folder, it's only then that I found out that I actually have something like 60,000 files already in that folder.

And so you can imagine it's probably difficult for the agent to navigate through that, which would have a direct impact on number one, how fast you can retrieve information from your workspace. And number two, how quickly you run out of your usage in your plan. And so as soon as you start to see some slowdown in your systems in retrieving memory, I advise people to move to level two, which essentially is just organizing that workspace that is optimized for an agent.

And I say optimized for an agent because if you remember, if you're a millennial or a Gen X, when Windows or Mac operating systems first came to be, you probably had a period in your life when you wanted to organize things neatly into folders and to properly rename the files and folders so that you, as the person, can easily navigate through that workspace and actually find the things that you need.

In this new paradigm where your agents are the ones operating on your files, you don't really need to pay as much attention to the names and the navigability of your files from within this more traditional File Explorer view. Because with these agentic operating systems, at the minimum, what you should have in your workspace are what I like to call as router files.

And that's best illustrated here in this second brain system where the center of it is your claud dot m d. And that by itself is a router. And in my claud dot m d, that just provides claud context on the different departments that I work with so that when I work on content, it knows that it just operates within this set of files.

When I work on my community, let's say, it knows that these files are the ones that are relevant and so on and so forth. And then for each of these departments, I also have a dedicated router file. So for example, if I search for the content dot m d file in here, what you'll see inside that markdown file is just a list of skills as well as reference files so that when I'm looking for something content related, they'll be able to look at this list and immediately navigate through my files and find the stuff that I want.

And so instead of focusing on organizing your workspace in terms of folders and files, which is probably second nature for a lot of us who grew up with this older, more traditional operating systems like Windows or Mac. But remember, since these agents like Cloud Code can parse through files at lightning speed, it's always much better to just give them these router files so that your agent can find what you need in the fastest way possible with the least amount of steps.

And again, like I mentioned, you can start this setup too just by talking to your agent. What you can do is just copy this prompt and give it to your Cloud Code, and it'll get you started with setting up those router files. And then once you're ready to upgrade your memory system, then you can move on to level three, which is having your visual second brain system, which is the one that I was showing you earlier.

And the reason why this is important for a lot of people is that it allows you to see how all of your files and folders connect, and also it allows you to search for things faster. And you've already seen me do this through this video when I'm showing my skills or how my claw.md connects to different departments or how these different files connect.

That's actually really valuable for a lot of people who are more visual and can understand things better when you show it to them in this format versus having to take them through, again, that traditional file explorer view, which is not really the most engaging format. The other thing is if, let's say, you want to go and search for the cleanup skill in here, usually with file explorer, it is much longer to find.

But earlier, as I illustrated, if I look for that cleanup skill from my OS or my second brain system, then I can immediately find that and I can immediately show it and preview it. Now let's go and talk about routines. And to me, is the third layer because once you've mastered skills and memory, that's only really when you get the confidence to actually let your agent do tasks even without you monitoring it.

And that's the essence of routines because routines are essentially just scheduled tasks. And at the first level, the way that you do this is really simple because it comes out of the box with Claude code. And in the Claude desktop app, if you head to routines here on the left sidebar, you'll be able to draft routines in here just by talking to Cloud in natural language.

And you can see I have a couple here. The one that I use a lot is this YouTube to Substack daily, in which if you open that, you can see exactly when that repeats, which for this one is every day at 8AM. And in a nutshell, a routine is just a prompt that Claude sends to itself at that given time that you set.

So for this specific routine, what it does is it just turns any new video that we have on the channel and it drafts that into a newsletter post in my tone of voice. And when that runs in the background, that also puts the artifact here in my AgenTek OS. So let's say this previous video that I had called the six new rules of Cloud Code.

When it's time for me to review that, I just open it, and it provides me with several options and drafts that I can just iterate and review. And because this uses a custom skill as well for that routine, then I already have something like 70%, 80 confidence that this is within my tone of voice, and I only need to apply some minor edits to it for it to get to production.

But if you notice here, here in my routines, I only have a few in my Cloud Code desktop app. And the reason why that is is because at level one, even though they're really simple to set up, the key limitation with these local routines is that they only run while your computer is on. And so a lot of my routines are actually in level two where I have those scheduled tasks running in the cloud so that even if my computer is off, I have the assurance that those scheduled tasks will run even without me.

And there's many solutions to this now. OpenCloud probably popularized it before. Grockbot is also pretty new, although it's paywalled to a pretty high price point at the moment.

So the one that I personally have set up and use is Hermes. And I have several other tutorials on Hermes on this channel. Or if you're part of the community, you can also just go through this Hermes agent masterclass to use it in the best way possible.

But in a nutshell, the reason why Hermes is so powerful is because it is twenty four seven. It's always on. And the reason why it's twenty four seven and always on is because for most people who are using Hermes, they give it its own computer.

Like for myself, I give it a computer in the cloud. And so that's why if you look at my routines firing board in here, most of the scheduled tasks that I have personally are already loaded in my Hermes agent.

But a key aspect of this that most people miss is that if your Hermes agent has its own computer, then how can it access all of the skills and all of the context that you have built up with Cloud Code? And there's many ways to do this, but for me personally, and I think for a lot of people who are just getting started with this, what you can use is a tool called SyncThing.

And SyncThing is this really good and free open source software that basically what it does is literally just sync things between your computers. And so if you point it to your workspace that Cloud Code works on and also install it in the computer that your Hermes agent uses, then you'll be able to sync the files that you want, including all of the skills and memory files that you want to share with Hermes.

And as usual, all you need to get started is just one single prompt, and you can use this one if you're interested to try it out. Now when it comes to the next stage of routines, this is actually something that I think is coming soon by default for Cloud, and I actually know some users who are tinkering a lot with this type of setup where they get a VPS, a virtual private server, essentially a computer in the cloud, and install Cloud Code into that.

So all of the files and all of the context that their Cloud Code builds with them lives in that space. And so in that format, you get the best of both worlds. Right?

Because you get routines that don't die out and actually run twenty four seven, and you're operating through just one agentic platform, which is Cloud Code in this case, but you can also use Codecs. And you don't have to use an external tool like SyncThing in order to sync files between your two different setups.

So it's highly possible that OpenAI and Anthropic will offer something like this in the future. But obviously, because of file storage and security concerns, that's probably something that you can expect sometime in the near future instead of now. Now let's go to the final element, which is applications.

Because if you're trying to do any real work with your agents, then you need to be able to connect to your apps. And at the very first level, the way that you connect to these applications is probably through the Cloud Code desktop app. From here, if you go to customize and under the connectors section, here you can find and browse the different applications that you can connect to.

But in my view, this is actually not the most efficient way to connect to these applications. Because if you go to the second level, you can actually just have Cloud Code connect to these applications itself or search for what connectors exist instead of you having to do it for Cloud. At least for me personally, what I use to find connectors is this skill that I have called search connectors.

And what it basically does is search the web for the official connectors, if there's any, or if there's none. Sometimes there's community made connectors that are in the form of CLIs or command line interfaces or APIs or MCP.

So, basically, these are the three usual formats that people use to connect their AI agents to their applications. To give an example, I use search connectors for Adobe Premiere here and look to check if there's an official one from Adobe. It also look for anything community made, and by the end of it, it provides a recommendation, which right now is this open source repo, which is available on GitHub.

So if I want to use this connector, then I'll just continue this session and ask Claude Code to scan it if it's safe and to set it up so that we can start using it. But the great thing about these AI agents is that if you take it to the next level, what you can actually do is to build your own connectors and to build your own applications.

Now there's many ways to build your own connectors, but the one that I use personally is the CLI printing press from Matt VH, who is the cofounder of Lyft. And I actually made a separate video on this if you want to check it out. But essentially, what you can do with this is to create connectors for applications that don't have them.

So in this video, I created one for MyFitnessPal as well as for Skull because no AgenTic connector really exists for those platforms. And, of course, nothing stops you from creating your own applications as well.

Like, for instance, this whole dashboard is one type of application. But apart from that, you can see I also built out these micro apps that I use almost of the daily. I have this app where all of the generations for images and videos that you see on this channel and in our business actually end up in this masonry grid so that it's much easier to preview.

I, of course, have that second brain system, which I've shown already quite a lot. And those ExcaliDraw illustrations that you're seeing, I actually have Claude build out this landing pad where I can just copy in those artifacts that it creates for me and just use them depending on the visual that I'm trying to communicate.

And so whenever it makes sense, I encourage you to try out creating your own applications yourself because then you will realize just how powerful these agentic platforms can be. So there, that is the ARMS framework in full. And I hope that was useful for you to craft your own perfect agentic operating system.

And if you want the full guide along with all of the prompts that I shared here so that you can start creating something like this for yourself or for a client as well, then feel free to just grab this PDF and send it to your cloud code, which you can find down in the description. That's it for this one. Thanks for watching till the end as usual, and I'll see you all next time.

Thanks.

The Hook

The bait, then the rug-pull.

Most people's Claude Code setup hasn't caught up to what the newest models can actually do. This breakdown walks through a live command-center dashboard, then unpacks the four-layer ARMS framework underneath it, skills, memory, routines, and apps, each with three levels from beginner to always-on cloud automation.

Frameworks

Named ideas worth stealing.

03:54acronym

The ARMS Framework

Applications
Routines
Memory
Skills

A four-layer model for building a Claude Code based agentic operating system, learned bottom-up starting with Skills, then Memory, then Routines, then Applications.

Steal forStructuring how to teach or document any AI-agent workspace, from a solo Claude Code setup to a productized offer for clients.

05:09model

Three-Level Maturity Model (per ARMS layer)

Level 1: use what's built-in
Level 2: organize/enrich it yourself
Level 3: make it visual, headless, or always-on

The same three-stage progression is applied to each of the four ARMS layers, from a beginner default to a fully custom, automated setup.

Steal forAny onboarding sequence or course structure that needs to serve both beginners and advanced users of the same tool.

CTA Breakdown

How they asked for the click.

VERBAL ASK

21:06link

“If you want the full guide along with all of the prompts that I shared here... grab this PDF and send it to your cloud code, which you can find down in the description.”

Soft CTA repeated at both the framework intro and the outro, pointing to a free PDF guide in the description; a separate paid community/masterclass pitch is dropped mid-video during the dashboard tour.

MENTIONED ON CAMERA

00:43productRUBRIC (Agentic OS command center product) ↗

17:17toolSyncthing ↗

21:27linkRoboNuggets free classroom / Agentic OS PDF guide (code R66) ↗

03:33productRoboNuggets Claude Code Masterclass community ↗

FROM THE DESCRIPTION

PRIMARY CTAWhere the creator wants you to go next.

AFFILIATECommission earned if you click.

OTHER LINKSAlso linked in the description.