The argument in one line.
Claude Code becomes a business operator, not just a coding assistant, when you wrap it in a persistent CLAUDE.md, reusable skills, guardrail hooks, external MCP tools, and agent teams that you then deploy to the cloud to run without you.
Read if. Skip if.
READ IF YOU ARE…
- A complete beginner who has never coded but wants to build and deploy real automations with Claude Code from scratch.
- An aspiring or early-stage AI-automation-agency operator who wants a repeatable system to deliver workflows to paying clients.
- A creator or solo builder who wants to automate content research, packaging, and analytics instead of paying for a stack of SaaS tools.
- Someone who already uses Claude Code casually and wants the structure behind skills, hooks, MCP, sub-agents, and agent teams.
SKIP IF…
- You are a senior engineer who already ships production agent systems and wants deep internals rather than a beginner walkthrough.
- You want a vendor-neutral comparison of coding agents; this is Claude Code specific and agency-monetization framed.
- You need exact, current pricing and model versions; the numbers shown are the presenter's live-session figures and drift.
The full version, fast.
This masterclass turns Claude Code from a chat-in-your-editor tool into a system that runs a business. It moves through five modules: foundations (install, pricing, permission modes, and the auto-loading CLAUDE.md instruction file), customization (skills as reusable SKILL.md prompts, slash commands, lifecycle hooks stored in settings.json, and MCP servers that grant new abilities like browser and database access), prompting and context management (split prompts, plan mode, one conversation per task), and agentic workflows (sub-agents that delegate like employees, agent teams that share a task list and talk to each other, eight practical hacks, and cloud deployment via Modal, Railway, Render, or Replit). It ends with two live builds: a four-stage YouTube content pipeline and a premium Ferrari landing page shipped GitHub-to-Vercel in minutes. The throughline is that guardrails and structure, not raw prompting, are what make agent output reliable enough to sell.
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 – 01:43
01 · Intro
The five-module map and the no-fluff promise; resources live in his Skool templates vault.
01:43 – 06:11
02 · Pricing & Hosting
You pay Anthropic not Claude Code; Pro vs Max plans; VS Code, Cursor, Antigravity, terminal, and desktop as hosts.
06:11 – 19:55
03 · Installation & Extensions
Install the verified extension, log in via subscription, tour the interface, permission modes, effort, and built-in slash commands.
19:55 – 32:05
04 · The CLAUDE.md File
The auto-loading instruction manual; four pillars; write-first vs /init; the three-layer architecture.
32:05 – 1:13:26
05 · Skills & Slash Commands
Skills as SKILL.md prompts, marketplaces, building a lead-gen skill with skill-creator, commands, and skill evals.
1:13:26 – 1:24:17
06 · Hooks
Lifecycle automations in settings.json: finish-notification, block .env edits, and a file-edit audit log.
1:24:17 – 1:37:24
07 · MCP
Plugins that grant new abilities; Chrome DevTools screenshots and a Supabase vector table; the token cost of MCPs.
1:37:24 – 1:57:51
08 · Prompting & Context Management
Split prompts, plan mode, one conversation per task, CLAUDE.md and skills as reusable prompts.
1:57:51 – 2:15:43
09 · Sub-Agents
Delegation like hiring employees; building an email-writer agent; triggering, tool restrictions, and agent memory.
2:15:43 – 2:33:02
10 · Agent Teams
Teammates that share a task list and talk to each other; a four-agent sales pipeline; the cost of parallelization.
2:33:02 – 2:51:55
11 · 8 Hacks for Agentic Workflows
Folder architecture, skills, spec-to-code, permission modes, parallel sub-agents, context management, and Pixel Agents.
2:51:55 – 3:11:41
12 · Deployment
Running workflows 24/7 via Modal, Railway, Render, and Replit; the shared copy-error-to-Claude debug loop.
3:11:41 – 3:32:20
13 · Build #1: YouTube Content Pipeline
His real four-stage system: ideation, packaging, video assets, and post-video CTR analytics.
3:32:20 – 3:48:20
14 · Build #2: $10,000 Ferrari Website
Plan, execute, review, deploy: AI image to animated video to a premium landing page shipped to Vercel.
Lines worth screenshotting.
- The Claude free tier gives you no Claude Code access at all; you have to be on a paid Claude plan to use it.
- Using the Anthropic API console instead of a Claude subscription quietly costs far more once Claude Code starts burning tokens.
- CLAUDE.md auto-loads every session, so a good one means you never re-explain your project, conventions, or folder structure again.
- Consistency beats peak quality with agents: an average-but-consistent output is more useful than a brilliant-but-unpredictable one.
- A skill is just a prompt in a SKILL.md file, and that same file works across Claude, Cursor, Codex, Gemini CLI, and Copilot.
- Claude reads only a skill's name and description first, so a good description is what makes it pick the right skill without wasting tokens.
- Skill evals prove the point numerically: the presenter's title skill scored 100% with the skill versus 50% without it.
- Hooks are org-level guardrails in settings.json, like a PreToolUse hook that blocks Claude from ever editing your .env secrets.
- Two MCP servers can eat over 10,000 tokens of context, so vet MCPs by GitHub stars and convert proven ones into cheaper skills.
- Splitting one seven-task mega-prompt into reviewed task-by-task prompts means you fix task four without redoing the whole build.
- The more tokens you pile into one conversation, the dumber Claude gets, so start a fresh conversation per unrelated task.
- Agent teams differ from sub-agents because teammates can talk to each other, which is why you use them only for genuinely dependent work like front-end plus back-end plus tests.
- Chained agent steps compound error: 0.9 accuracy over five steps is only about 59% success, which is why frameworks and guardrails matter.
- Deployment just means renting someone else's always-on computer so your workflow runs at 3am without your laptop open.
- Every deploy platform shares the same debug loop: copy the red error from the logs, paste it into Claude Code, let it self-heal, redeploy.
- A $10k-style Ferrari landing page went from AI-generated image to animated video to a live Vercel site in roughly two minutes of deploy time.
Structure and guardrails are what make agents sellable.
WHAT TO LEARN
Raw prompting is not the skill; the skill is wrapping Claude Code in persistent context, reusable instructions, safety rails, external tools, and delegation so output is reliable enough to ship and deploy.
02Pricing & Hosting
- Pay for a Claude subscription rather than API access, because Claude Code silently burns tokens and the API bill compounds fast.
- Pick VS Code or Antigravity as a beginner; they are effectively the same host, and you rarely use their built-in AI anyway.
04The CLAUDE.md File
- Write a CLAUDE.md covering product, tech stack, rules, and folder structure so the agent never makes you re-explain your project.
- Value a consistent average output over a brilliant unpredictable one, because reliability is what you can build and sell on.
05Skills & Slash Commands
- Treat every skill as just a prompt in a SKILL.md file, and lean on its name and description because that is what Claude reads to pick it.
- Run skill evals to measure improvement; the same title skill scored 100 percent with it versus 50 percent without it.
06Hooks
- Use hooks in settings.json as guardrails, for example blocking any edit to your .env so secrets can never be deleted or overwritten.
- Add a PostToolUse hook that logs every file edit with a timestamp so you always have an audit trail of what changed.
07MCP
- Add MCP servers sparingly because two can cost over ten thousand tokens, and convert the ones you rely on into cheaper skills.
- Vet an unfamiliar MCP by its GitHub stars before trusting it with access to your browser or database.
08Prompting & Context Management
- Split a big build into reviewed task-by-task prompts so you fix one milestone without redoing the whole thing.
- Start a fresh conversation per unrelated task, because a bloated context window measurably degrades the agent.
09Sub-Agents
- Delegate to sub-agents like employees, each with a scoped tool-access tier from read-only to full access.
- Give sub-agents an agent-memory folder so they remember feedback and improve across sessions.
10Agent Teams
- Reserve agent teams for interdependent work like front-end plus back-end plus tests, where teammates need to talk to each other.
- Weigh the cost, since parallel agent teams can run far more expensively than a set of sub-agents for the same result.
118 Hacks
- Structure folders with the three-layer architecture, because chained steps compound error to about 59 percent over five steps.
- Manage context deliberately with compact, clear, and context commands to keep long sessions from degrading.
12Deployment
- Deploy to Modal, Railway, Render, or Replit to run workflows 24/7 without your laptop open.
- Debug every platform the same way: copy the red error from the logs into Claude Code and let it self-heal, then redeploy.
13Build #1: YouTube Pipeline
- Systematize creative work into staged skills the way the pipeline splits ideation, packaging, video, and analytics.
- Track views-to-clicks CTR per video so you make more of what converts and less of what does not.
14Build #2: $10k Website
- Ship client products end to end by generating assets with AI, building with front-end skills, and deploying GitHub to Vercel in minutes.
- Follow plan, execute, review, deploy so a premium build stays structured instead of turning into endless reprompting.
Terms worth knowing.
- CLAUDE.md
- A markdown instruction file Claude Code reads automatically at the start of every session, holding the project description, tech stack, rules, and folder structure so the agent has context before doing any work.
- Skill
- A reusable instruction manual (a SKILL.md prompt, optionally with reference docs and scripts) that Claude reads before a task so it follows a consistent step-by-step process instead of improvising.
- Slash command
- A shortcut typed as /name that triggers a specific skill or built-in action directly, instead of describing the task in a full prompt.
- Hook
- An automatic action that runs at a specific lifecycle moment (before a tool runs, after code is written, when Claude stops, or on notification), configured in settings.json and applied to every session.
- MCP (Model Context Protocol)
- A plugin standard that gives Claude new abilities by connecting it to external tools and services like a browser, Supabase, or Gmail through an MCP server running in the background.
- Sub-agent
- A specialized helper agent with its own instructions, tools, and model that the main Claude delegates a task to and that reports its result back; sub-agents share context but cannot talk to each other.
- Agent team
- A group of agents run by a team lead that share a task list and can communicate with one another, each with its own context window, used for larger interdependent projects.
- Plan mode
- A permission mode where Claude drafts a reviewable step-by-step plan before executing, letting you add comments and approve before any changes are made.
- Bypass permissions
- A YOLO permission mode where Claude executes everything without asking for approval, best reserved for simple, well-defined tasks you are watching.
- Context window
- The fixed token budget of a single conversation; as it fills with messages, files, and tool calls, the agent degrades, which is why compacting or resetting matters.
- Three-layer agent architecture
- A CLAUDE.md folder pattern that separates directives (what to do), orchestration (decisions and error handling), and execution (the actual code and API calls) to make agent workflows reliable.
- Deployment
- Moving a workflow off your own machine onto an always-on rented server so it runs on a schedule 24/7 without your computer being on.
Things they pointed at.
51:40toolAnthropic skill-creator skill
2:40:00toolPixel Agents (visual workspace extension)
3:35:00toolKie API (image + video generation)
3:36:40toolKling 3.0 (AI video model)
Lines you could clip.
00:20
“No fluff, no jumping around, just a clear practical roadmap that takes you from zero to building real systems.”
40:00
“A skill is just a prompt. Don't make it more complex than it is.”
1:45:00
“The more tokens you use in a session, the more stupid Claude Code becomes.”
1:40:00
“A minute of planning saves you ten minutes of building.”
2:34:10
“Point-nine to the power of five is a fifty-nine percent success rate, which is simply not good enough when we're building agentic workflows.”
2:53:00
“Deployment means putting your workflow in someone else's computer, so it's always on and it runs twenty-four seven.”
1:25:50
“Skills are instructions, hooks are guardrails, and MCP are new abilities. They work together, not against each other.”
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.
00:00Welcome to the most comprehensive ClawCode masterclass where I take you from a real beginner in ClawCode to someone who's actually able to build agentic workflows and full systems for yourself or for other businesses. Now I use ClawCode pretty much every single day to run my 6 figure AI business, and I'm also teaching it to over 10,000 people in my community.
00:18And so I've put together this masterclass in the exact order that I wish I had when I had started out. And as always, no fluff, no jumping around, just a clear practical road map that takes you from zero to building real systems. Now I've divided this master class into five different modules.
00:33So here's exactly what you're getting. Module one is foundations. This is where we install ClawCode, go through how it actually works, permissions, setting up your claud dot m d, your claud directory, so ClawCode understands your projects before you type a word.
00:44Module two is customization. This is where we build custom skills, slash commands, hooks that automate actions for you, and MCP, which connects claw code to every external tool in your stack. Module three is prompt engineering.
00:56And here, I'm gonna be breaking down how to talk to claw code so you get better outputs every single time. Module four is agentic workflows. This is where we talk about sub agents, agent teams, and eight hacks I've discovered after building hundreds of workflows, and at the end, deployment.
01:10So everything runs twenty four seven without you. And module five is the build. No more theory.
01:14You're gonna watch me build real projects from scratch using everything that we covered so you can see exactly how it all comes together. Now I'm also gonna be leaving time stamps right below in this video so you can stop the video and then you can come back to it and start wherever you left off. And as always, if you want all the resources and templates of this video, then you can check them out in the second link down below.
01:33In my free private school community, you can go to the classroom section and you can go to the templates vault. Now with that said, let's dive in.
01:43Alright. Here we have ClawCode. First thing is pricing.
01:46How much does this actually cost? Now bear in mind that we are not gonna pay ClawCode. We're gonna pay Claude, and through playing Claude, we get access to ClawCode.
01:55We get access to ClawCode work and we get access to Claude itself. Okay? Now Claude has three different tiers.
02:01It has a free tier, and by the way, if you have the free tier, you do not have access to ClawCode. So ClawCode, you have to pay for it. Then we have the pro tier, which is good for everyday productivity, which is $17 a month, and it gives you access to 45 messages for five hours.
02:14That is the limit on usage that you can use ClawCode. And this is the best tier that I can give beginners starting out unless you're creating some really complex applications or you have a lot of automations running in your business.
02:24And then we have the max plan. So let me put myself here. And the max plan gives you five x of the usage of the pro plan, which would be about 225 messages for five hours, and you will pay around a $100 a month.
02:36Now I personally have the max plan, so I'm personally paying $100 per month on Clockhode just because I use it every single day for my business and I automate a lot of stuff that I have day to day. Now you also have the option to pay 20 x.
02:49Right? So instead of five x, it will be 20 x of what the pro plan gives you, and you will also be paying double the money, which means that from a $100 a month, you'll pay $200 a month. So as a beginner, I definitely recommend sticking to the cheapest plan, which is a pro plan for now.
03:02And if you see yourself exceeding the limit or exceeding the token cost or usage, then you might wanna upgrade to the max plan. Alright. So next thing is hosting.
03:11What does that look like? How do we set it up? And which applications do we use to actually use ClockCode?
03:16So first thing is using Versus Code. Versus Code is what we call an IDE, very fancy way of saying a place where you can look at all your folders that you have with that ClawCode is using and you can also install ClawCode and you also have access to the AI of Versus Code itself and it's completely free to to download.
03:33Right? Now similarly, we have Cursor. So Cursor is a very popular one for coders just because Cursor itself has their own AI and their own AI is very very good when it comes to coding and doing all the things that like building applications and so on.
03:46So if you want the power of Cursor plus ClockCode, meaning if you're a bit more on the techy side and you're a bit more of a developer, then I recommend using Cursor. The only limitation of Cursor is the limit. So it's gonna get a bit expensive, especially when you have to pay for Cursor's AI but also for the Clockwork AI.
04:01And I can tell you from experience that most times, we just use Clockcode inside these platforms. We never use the AI in these platforms at all.
04:09But I usually go Versus Code because I don't use the AI in Versus Code anyways, so to me, it doesn't really matter. Then we have anti gravity, which is the exact same as Versus Code. Anti gravity is one of the popular tools that came out recently, which is part of Google's suite.
04:22And anti gravity is the exact same as Versus Code and it also has its own AI that we can use. And I honestly see no difference between anti gravity and VSCode unless you wanna use the Gemini models inside of anti gravity to build your own interfaces and make them look pretty. Right?
04:36So either one works fine. Then we have the terminal, which is where we are gonna be using clock code inside of our computer without having to go to anti gravity or Versus Code. The only limitation with the terminal is that it doesn't look too pretty.
04:47It doesn't look very visual. It's sort of like code y and it looks a bit complex when you're starting out. So as a beginner, I do not recommend this.
04:53And then finally, we have Clawd. So Claude desktop also has the ability to use Claude code. The only limitation here is that you will not be able to see your files and, um, it won't be as structured as anti gravity or Versus code.
05:06Now for today's video, we're gonna focus on Versus Code and anti gravity because most of the videos in this course are either using Versus Code or either using anti gravity. But they're very, similar, so I'll show you exactly how to download ClockCode in each one of these platforms.
05:19So let's start with Versus Code. All you have to do is go to code.visualstudio.com, and you will get to this page right here, the open source AI code editor.
05:26You can press download from macOS if you're macOS, or you can press download here, and you'll be able to go on this page, and you can choose your device right here.
05:35I'm just gonna press Mac here. As you can see now, it's downloading the application itself. We have this little pop up come up.
05:41It says happy coding. We can drag this across. Now it's gonna download the actual application in our computer.
05:46Now once you have it opened up, you will get on a page that looks like this. So this is Visual Studio Code, Versus Code in the name. You'll have different buttons here, and you'll have the main interface right here.
05:55So this is gonna be the main place where we're gonna be using Cloud Code and actually building different things, whether it's automations or applications. Now in order to download Cloud Code, all we have to do is go down to the extensions, and I'll walk you through what each one means here, so don't worry. Extensions.
06:08You wanna look for Cloud Code, and then you'll see the one from Anthropic. Make sure it's the one from Anthropic and not this one here or not this one here.
06:17This is the verified one. So once you press the extension, you will get on a page like this. You can see Clog code for Versus Code, and you can simply press install.
06:25At the time of this recording, it has 10,263,533 people, which is crazy, and it's the version two point one point one zero seven. It was published ten months ago, so very, very new.
06:35Now I can press this button right here, which is for us to be able to open a ClawCode, and when we get on this page, I can simply say hello. Now bear in mind that if you don't see this page right here to say hello and so on, it's because it's gonna ask you to log in.
06:49Now as you saw from me, I can say hello. How are you? Because I already logged in to Claude inside here.
06:54But if you haven't logged in to Claude yet, then you most likely see this page that looks like this, which says that Claude can be used with your subscription or used with your API usage. Right?
07:03So you can either pay it through the cloud subscription or you can use the API usage. Now bear in mind that if you use the API usage, which is the Anthropic console, you will end up paying a lot more in the long run just because the usage is gonna increase by a lot, especially when you use cloud code, and you won't even realize it.
07:19That's the thing. So you can just press claud.ai subscription.
07:22You can press open. It'll take you to this page here. You can press authorize.
07:26It says build something great. That's all good. Now I can go back, and you'll be able to see that now I can actually chat to ClawCode itself.
07:33I can press x, and now we have the full interface. So this right here will be the page where we actually use Clockcode and we can run up different conversations with Clockcode itself. On the left hand side, we have folders.
07:43Now bear in mind that anything that you build in Clockcode and I mean anything will be stored inside a folder. So when I'm building a website, when I'm building an application, so where is the HTML stored? Where is the document stored?
07:54Where are all the images stored? Right? They're all stored inside a folder inside your computer and this is great for us because you always have a place to store your exact files.
08:03Right? And so on the left hand side, we can press open folder. Let's say I add a new folder under desktop.
08:08I can press new folder. I can name it example and I can press create.
08:13I can press open. You can press yes. I trust the authors.
08:16Now you won't be able to see ClawCode anymore. So to see it again, you can press this button right here, ClawCode, and you can press x again. And now you're back in this interface.
08:24So now we have the folder set up. We have everything here. We can start adding files.
08:28So file. We can start adding folders. So folder.
08:32And that's how we start to add what we call the Cloud directory. So the directory of Cloud. So where are all the files stored?
08:38I can press delete and we're back to zero. Okay.
08:41So everything that we build or the specific project that we're working on will be stored under the example folder right here. Now to show you how that looks like on your computer, if I add a file here and I say file dot m d, say hello. Okay.
08:54And then I press command s or you can go to file save. Inside this folder called example, we have the file dot m d. And so now I can go to file, I can close the folder, and I can go back, I can open the folder again, go to desktop, I can open the folder again, press open, and now we're back exactly where we started.
09:11And so this way, you can have different folders for different projects, and you can always come back to clock code and start from zero. Now there are a lot of different things in the interface itself, so let's go through them step by step. So the first thing is the conversation history.
09:23So if you wanna switch between conversations the same way that you would do instead of Claude, you can go here to the clock itself and you can see all the past conversations that you've had or you can go to the web tab right here. And so now I can go to, let's say, write a LinkedIn post, which is a conversation that I had before, and we can always review the conversation that we had in case we missed anything in case we wanna keep going with the conversation itself.
09:43Then we have this right here, which is just the same thing as just pressing this button. It spins up a new Cloud instance, which means a new conversation. And then down here is where everything's at.
09:52So on the bottom side, we can see plus. In case you wanna add folders or context or browse the web, you can simply do it here. We typically do this if you wanna add more context to Clockcode.
10:01Then we have the filter actions. So there are lot of different things that you can do inside of Clockcode. One of them is using slash commands.
10:08So you can put slash something. So slash there's a lot of different things you can do. We can put slash clear conversation.
10:13We can put slash compact. We can do slash everything. And we'll go through exactly what a lot of these slash commands are later in the video, so don't worry.
10:21And then one of the most important things here is actually the permission mode. So if I press here edit automatically, this is the permission mode that we're gonna be using ClawCode in.
10:30Right? Now the permission mode might look different in six months and it looked different six months ago. Right?
10:35But essentially what it is, it's the way that you're telling ClawCode to act. Right? So if I put ask before edits, then anything that it does, it will ask for approval before making each edit.
10:45If I put edit automatically, and now Claude will edit your text or the whole file automatically. If I put here plan mode as a permission, then what it will do is that it will create a plan for the actual thing. Then we have bypass permissions, which is where Claude will not ask for approval before running the potential dangerous commands.
11:00Right? And then right here, we also have the new feature which is called effort. So if you want a very high effort for the task itself, then you can simply tell Cloud Code that you want higher effort.
11:09If it's lower, then you can put low. Now I typically have this on effort medium just because I wanna keep it balanced. Now the default permission here is actually edit automatically, and so the main ones are really ask before edits, plan mode, and bypass permissions.
11:21So let me show you an example. So let's say I tell Cloud Code to make a new file inside our folder or project and name it LinkedIn post and then you can add a LinkedIn post there about life with a happy tone targeting people who are business owners in The US.
11:36It's a very strange task but essentially it has to edit our folder, it has to edit something and it has to include the actual text itself. So from edit automatically, I can actually go to plan mode because I wanted to plan before it does anything.
11:48And here I can just press go. So now what happens is that Cloud Code is given the task. It will then think about what it's doing.
11:54It will create a plan of action. As you can see here, it says, I'll create a quick plan for this and then we can execute it. And then the plan is something that we can review right here, context, steps, verification.
12:04So the plan is to create a LinkedIn post file. The user wants a new file in their project folder containing a LinkedIn post about life written in a happy tone and targeting US business owners. It's a very strange request but we can press yes auto accept and now what happens is that it creates a file on our folder right here called LinkedIn post that I can open up and here we have the LinkedIn post.
12:24Now obviously this is just an example to show you how plan mode works. Right? Now let's say I wanted to use bypass permissions.
12:30Bear in mind, this is not something that you see as default, so you will not have access to this as soon as you get to Versus Code. What you have to do here is go to the settings here and you can look for allow dangerously.
12:42There we go. You can turn this on right here, which means that you're not allowing bypass permissions mode recommended only for sandboxes with no Internet access. So that makes no sense, but just make sure to turn this on.
12:52And all this mode does is that it will not ask you for approval for anything that it does. That's why it's called dangerously skip permissions because it's dangerous if you don't know what you're doing. It's dangerous if you give it access to different softwares and files and you're not there to actually see what it does.
13:08Right? So let's say I wanted to use bypass permissions and let's say I give it the exact same prompt right here and let's say I said Instagram post, right, what you will see now is that it will not give us a plan, it will just straight away make the file itself without us having to do anything.
13:23As you can see, it made the file here called Instagram post, life as a business owner that we can use straight away. Right? So on the one hand, this made a plan.
13:29It asked us for approval. On the other hand, it didn't ask us for anything and admitted itself. Now the way that I see it is that I usually go to plan mode whenever I want to create a plan for a specific project that I'm doing and then once plan mode is done, if it's something that's a bit more complex that needs my input, I put it to ask before edits because it will ask me before it does anything.
13:48But if it's just a very easy task, then we can put bypass permissions because I don't wanna be there approving everything all the time. Right? Just understand what it is that you're building and what Clockcode has access to.
13:58Now like we mentioned, the other way is using a software called Google anti gravity. Google anti gravity is known for making websites, for making nice interfaces. You might have seen it on YouTube.
14:07Right? But it's also a way for us to use Clockcode. And so I can just go here to antigravity.google.
14:12You can press download, and right here, you'll be able to see your own computer. I can press download for Apple silicon. You'll then have this page right here similar to the Versus Code.
14:21We can drag this across. Now it's downloading anti gravity to our applications. And once you get to anti gravity, you will get on a page that looks like this.
14:27So very similar to the one that we had at Versus Code. On the right hand side, you have the IDE's AI. So anti gravity's AI, which is the agent here, which is similar to the one that we had in Versus Code.
14:37On the left hand side, you have the folders, so the same thing as Versus Code right here. And then we have the extensions. We can go to Clock Code.
14:43You can simply install if you haven't installed it. I already did. And once you have it installed, you can press this button right here, and you can start working on it.
14:51And the same way that we're using the example folder in Versus Code, we can go here, desktop, example, and we can open it up inside of anti gravity. We open it up.
15:00We can press yes. I trust the authors, and you'll be able to use the exact same folder that we're using in Versus Code. We will use it in anti gravity as well.
15:07So that's why I say they're pretty much the same thing. And one more thing when it comes to setting it up, you can also use your terminal. So if I go to terminal, new terminal here, you can simply press Claude, and you can say, yes.
15:18I trust this folder, and you'll be able to use Claude in terminal as well. So if you say hello, you can see that it says hello.
15:24How are you? Now here we can see the model that we're using, how many context tokens we have, what plan you're on, the email that we're using, and then the organization, and then the actual folder example as well.
15:34Now as a complete beginner, I don't recommend using the terminal for now. Just use the actual interface here. It'll be much easier, especially if you're just starting out, uh, with the platform itself because you don't wanna overwhelm yourself too much.
15:44Alright. So now you have ClawCode installed inside of anti gravity and instead of Versus Code. If you're just using ClawCode, I honestly wouldn't see a benefit of using one over the other, so just choose one and stick to it and keep it very simple for yourself.
15:56But if you're someone who's building something a bit more sophisticated and you want antigravity's Gemini model to help you with this, then I recommend just choosing antigravity cause you have the upper hand itself that you can use later on. Alright.
16:07So last thing to cover here is built in slash commands. So we mentioned this before that we have the little box right here, which gives us the ability to do actions without us having to give it a prompt.
16:17So here, if I put slash, I can see that I can call any different action if I wanted to. And you can see we have context, we have model, we have customize, and we have slash commands.
16:27So these slash commands are things, they're actions that we can do automatically just by calling them specifically. As you can see, we have batch, Cloud API, compact, context, cost, debug, extra usage, head dump, and we have a lot more here that we can use specifically. I'm gonna go through a few of them just so you get the idea of how it works.
16:45So let's say I had a conversation with Cloud Code. And I said, hello. I said, how are you?
16:50Now there are cases where the conversation is very, long and you wanna start from scratch without having to go to a new conversation here. So what you can do here is you can press slash clear conversation. So slash clear conversation, you can go here, and now you start from zero again.
17:04So now we have a new conversation, and the next slash command is called context. So if I put slash context and I press this one here, what Cloud Code will give me is a breakdown of all the context and tokens that I'm using inside my account.
17:16So we have a million of tokens that we're able to use. So far, we used 17,900 tokens, which is about 2%, and this is just by saying hello.
17:26How are you? Right? So think of every conversation as a bucket.
17:29Right? The more words you add to the conversation, the more water goes inside the bucket and it comes to a point where it can't take any more water and so you have to go to a new conversation. Right?
17:39And this is what the percentage is. How much water is inside the actual bucket? How many tokens are we using for the conversation itself?
17:45Because a token is just words. And also, it tells us the category of usage by type. So we have the system prompt, which is a prompt that we can't really see here, but it's taking up 6,200 tokens, which is about 0.6%.
17:58And then we have other things like system tools, we have skills, we have messages, free space, and buffer as well. Right?
18:05It's fine if you don't understand these terms right now because we're gonna cover them later down in the course. Then we also have MCP tools. So I downloaded MCPs inside my ClockCode, which is one of the modules that we'll go through later.
18:15And each MCP takes about x amount of tokens. So one seventy tokens is used by the Gmail authenticate.
18:22Authenticating the Google Calendar takes one seventy eight, and so the more MCPs you add, the more tokens. And then we have skills. So skills are essentially instructions that Clockcode uses to do a specific task and each skill requires different tokens.
18:35Right? And so here you can see a breakdown of all the tokens that you're using, all the usage.
18:40Now apart from clear conversation and context, we also have ones like compact, which is where you wanna compact the conversation so you can keep going in the same kind of context without having to go to a new conversation.
18:54We do that because every conversation has x amount of usage. So So let's say right now we have a conversation that takes 90% of its usage. So if you add a slash compact, it will take that 90%, it will break it down, it will summarize the conversation so that we can still have maybe another 60 or 70% left.
19:11Right? So we can keep working under the same thread without having to have a new conversation with ClawCode. Then we have cost as well to see how much it's costing you, debug to debug, and then we have others as well, which we will cover later down the line.
19:25Alright. So by this point, you understand how Clockcode works and how permissions are the things that keep you in control. But right now, every single time that you start a new session with Clockcode, it starts from zero.
19:35Cause it doesn't remember anything about your project, which is why in the next lesson, I'm gonna be breaking down arguably one of the most important files in ClawCode, which is the ClawCode dot m d g file. Then I'm also gonna break down the ClawCode directory. So every single time that you start a new session, Clawd code understands your project completely without you having to type a single word.
19:54Let's dive in. Alright. So what is the clawd dot m d file?
19:57I'm sure you've probably heard of it. You've seen it somewhere, but essentially what it is, it's the product instruction manual for Clawd code. So essentially ClockCode, whenever we're working with a workspace, it needs to have some sort of memory and the memory is where all the files are, what the product workspace actually is, what is our objective, who we are, and all that stuff.
20:15Right? And so that's exactly what the ClockCode. Md file actually is.
20:18Now the claud. Md file loads automatically which means that when we work with a new workspace, it loads automatically in a way that before it starts doing anything, it always looks at the claud.
20:27Md file to get an understanding of the whole layout of the workspace before it dives in and does any single task. The same way that you would have an employee coming into a company, you give him memory and every single day that he comes back to the company to work, he already has the memory pre trained. So in this case, whenever we have a new conversation with Cloud Code, we are saying, hey, before you do anything, just look at the Cloud.
20:49Md file, understand our workspace, and then execute the task that I want you to do. Now the actual Cloud. Md file is divided into four different kind of steps or or pillars.
20:59Right? The first one is product description. So this is what this app does.
21:03It's not necessarily just an app. It's just saying this is what this workspace is. Right?
21:07For example, I have a YouTube workspace where I manage all my YouTube stuff. So this workspace is specifically designed for YouTube ideation, content creation, x y z.
21:17Right? Then we have the tech stack. This is looking at what tools and softwares we use.
21:21Then we have the rules and guardrails, which is what Cloud must and must not do. And rules and guardrails can be anywhere from the text they should use, the colors they should use, the fonts, the step by step process they should go through to then create something.
21:33Right? And these are all their different rules. It's kinda general, but you can have, like, any rules that you can think of, you can put it there.
21:38And then finally, we have folder structure where everything lives. Now this is one of the most important because as you will see when we make the actual file, there is a structure of folders that we add inside our workspace. The reason that is is because we wanna make things organized and Claude knows exactly where to store all the different files.
21:55So we'll have a file for the scripts, we have a file for the output which is the things that it gives us, we have a file for the ideation, maybe we have a file for all the passwords and keys, API keys that we use. Right? And so in having things organized, it's much easier for ClawCode to do things because it's not scattered everywhere.
22:12That's what it is. Now what ClawCode gains by having this file? Well, the first thing is that it understands your project better and every single time that you speak to ClawCode, you don't have to repeat all the things.
22:21Right? It already knows. The second thing is that it follows your conventions.
22:24This can be naming conventions, it can be patterns, it can be a structure that you use. Then like I mentioned, it looks at the rules. So every single time this happens, do x y z.
22:32Right? And it won't break what actually matters to you. And then finally, it's consistent.
22:36That is probably the most important thing because it's much better to have something that's average but consistent than something that's really really high quality but inconsistent because with AI specifically, it's very good at doing things. Right?
22:49Like it's nuts. It's crazy. But you also wanna have strict guardrails.
22:54Right, strict prompts that make it so that it's not inconsistent every single time. It actually gives you the output that you want every single time when you are doing that specific thing. So there's two different types of ways that we can set it up.
23:05The first way is using a three layer architectural structure. Sounds complex. It's very, very easy actually.
23:11And the second way is running the slash in it. Now the two different ways represent two different scenarios. The first scenario is method one, right, where we write the Claude RMD first and then we tell Claude to initialize the project and this is only when we have a new project, when we're starting from zero.
23:28Then the second method is when you already have an existing project. So let's say you're building apps or maybe a landing page or maybe some agents, right, you haven't done the Claude.
23:37Md yet, so now what we do is we take all the folders, all the files, Cloud will understand your whole workspace and from there, we tell it create a Cloud. Md file. So one is creating Cloud.
23:47Md file from scratch and then creating the folders and the other one is you already have the folders, now we create the claud dot m d. Alright. Method one is writing claud dot m d from scratch.
23:56So here we have the claud code project. I made a new project called claud and this is where we will make a file right here just pressing this button and name it claud dot m d And inside here and by the way, the reason why it's m d is because it's markdown.
24:10Markdown is just the way that we form a text. Right? Because this is title one, this is title two, two hashtags, then we have boldings which is hello, right, and so on.
24:21So with that said, we have the Cloud. Md file that we made. Now we have to actually write the instructions and then initialize the project.
24:28Okay? Now one thing I wanna show you as well is if I go to a new chat, I'm looking here cause it does mention make a cloud dot m d file for instructions. Claude will read this every single time.
24:38So even Claude code incentivizes you. It tells you, hey, make this file because it reads this every single time. Okay.
24:44So here we have the prompt that I'm gonna give ClawCode in the claud dot m d file. I'm gonna copy this. I'll bring it back.
24:52I'll go here and then paste it. So what does this prompt tell ClawCode? So you operate within a three layer architecture that separates responsibilities to maximize reliability.
25:02That's fancy names. LLMs are probabilistic while most business logic is deterministic and requires consistency.
25:09This system solves that problem. This means that LLMs are inconsistent, a business logic is consistent and so we want to solve for that inconsistency with this exact file, right?
25:19And then we give it the three different layers. So layer one is directives. We're basically saying anything that has to deal with what to do goes in the directives folder.
25:28Anything that deals with decisions goes and follows this framework right here and then everything to do with doing the work, so the execution is this folder right here and we're also making a dot EMV file where we store API tokens, passwords, and so on.
25:43Okay? And then we have other principles as well. I can save it first of all so it doesn't run.
25:48And I can say, can you initialize this project based on the claud dot m d file? Claud dot m d. I'm gonna press go.
25:56I wanna make this bypass permissions. It's reading claud dot m d. Now it's looking at all the different things that are inside, and as you can see, it's starting to make the folders.
26:04It's making the folders which are dot e m v executions directives dot t m p, which is what we call the ClawCode directory. The ClawCode directory is essentially where all the different files are stored, and the ClawCode directory usually looks like this.
26:18Right? It's with these lines. So we have ClawCode's folder, which is this main folder, and inside the ClawCode folder, we have ClawCode.
26:24Md. We have dot emv, which is this one here. So add your API keys, OpenAI, Google API, passwords that we're gonna use.
26:31Execution, we have directives, and we have dot tmp, which is empty.
26:36So let's say I go to a new tab and I can write this down. I can say build me a landing page for JEM Solutions, an AI automation agency that helps businesses to automate their workflows using AI agents, which have a hero section, what we do, testimonial section, and footer. I'm gonna press go.
26:51I'm gonna make this plan mode so it plans the landing page before it does anything. As you can see, it's looking at the cloud dot m d file.
26:59It didn't find any brand guidelines, which is fine. Now it's creating a plan. Alright.
27:02It just made the plan. We have the context, the tech stack, which is essentially what we're gonna use, the structure, and then we have decision design, hero section.
27:10I can just press yes auto accept. I can make this bypass permissions because that means that it doesn't ask me to approve things all the time.
27:18And the whole point of showing you this landing page is so that you see how the different files of the landing page are stored in the different folders here. Alright. It just made the landing page.
27:27I can open it up as expected. It's pretty decent, to be honest.
27:34Very simple. And you can see here that it made another folder called front end, which has all the different components of the landing page but it also made a MD file for the directives which is what to do. Right?
27:44Directive, build landing page, objective, tech stack, page structure, service section, testimonials, footer, file map, and so on. Right?
27:51And so this is showing you that all the different things like we saw here of what to do, SOPs, the objectives, the inputs, scripts, outputs, edge cases are all gonna be stored under the directives.
28:04So now you got to see exactly how we made the folders from scratch by adding the claw dot m d file first, but it's also method two which is using slash in it. So let's say I have a workspace called YouTube and YouTube has a lot of folders. It has context.
28:17It has executions, outputs, all that stuff. It has dot e MV, which are all the different passwords and so on, but the one thing that is missing is the cloud dot m d file. And so in this case, we wouldn't use the three layer architecture because the folders are already made.
28:31We already have the folders, right, which could have been done manually by me or someone else. And so in this case, to create the cloud. Md file, all I have to do is go to slash in it and now what it's doing is that it's gonna look at the whole organization.
28:45It's looking at all the different files that it has and it says, now I have a good understanding of the workspace. Let me create the cloud. Md file.
28:51So now what we've done is that we have the cloud. Md file which has the memory, it has the context of the whole workspace. So whenever we're making an automation, whenever we're making something else, when we start cloud from scratch like this, it basically has the understanding.
29:05And to show you another example, if let's say I go back to the one that I had before, let me delete the Cloud. Md. Let's say you had these done manually.
29:12I would go here. I would save this first of all. As you can see here, the first thing it's telling me is to tell Cloud Code to remember what you've told it using Cloud.
29:20Md. Well, in this case, I can press slash in it again, bypass permissions.
29:25What it will do is that it will look at all the different files that are here, which in this case is to build the landing page. As you can see now, we made the Cloud. Md file, which is the product overview, which is landing page for JM Solutions and AI automation agency.
29:37It provides guidance to ClawCode. So the architecture will be actually used to build things, the front end structure, the commands, the design system, and the important notes.
29:46Right? So that's a snippet of memory that ClawCode will now use whenever it makes something new. But the last thing I wanted to show you is a more advanced version of a ClawCode.
29:54Md file because under my YouTube workspace, I have commands. I have skills. Right?
29:58All these things which allow me to do much more with Cloud Code than you would just by having the default features. And so in this case, we can see that the Cloud. Md file, same thing here, provides guidance.
30:09What this is is a Cloud Workspace template, a structure environment for working with Cloud Code as an agent assistant across sessions, run slash prime at each start of the session to load a central context. So if you're running from scratch, all you have to do is go here and put slash prime and what this will do is that it will look at the whole structure, it will look at the cloud.
30:27Md file, it will look at everything to have an understanding, context that it needs to be able to tell me who I am, the workspace structure, and everything that it needs to know about me specifically.
30:37We have the workspace structure. Again, this is the Cloud Code directory. When you hear Cloud Code directly, that's all this means.
30:42So we have Cloud. Md. We have shell alias.
30:44Md. We have dot e m v here. We have n c p dot JSON, which is this file right here.
30:49The Cloud commands and skills. So Cloud commands and skills. Then we have references, execution outputs, references, execution output.
30:56Then we have commands. So telling me all the commands that I have. Then we have the YouTube content pipeline, which are all the different skills, the thumbnails, analytics and monitoring, skill system, additional skills, critical instructions, and notes at the end as well.
31:08And that's the exact Clot. Md file that I use every single time and the one that I always have sitting there as a way for Clotr code to understand who I am and to have context before I ask it anything. Quick one before we get back to the video.
31:20If you're looking to work with me one to one to be able to start and scale your own AI automation agency in a 90 one on one mentorship program, then check out the first thing down below. With that being said, let's get into it.
31:34Hey. Congrats on finishing module one. You've now got the foundations locked in.
31:39You have ClawCode running. You understand how it works. And most importantly, it knows your project inside and out.
31:44But as always, you've pretty much tapped into 5% of what ClawCode can actually do, which is why in the next module, we're gonna be taking ClawCode from a smart assistant all the way to building a fully customized system that's built around the way that you work. And it all starts with something that most people overlook, which is skills and slash commands.
32:04Let's dive in. Alright. So we have Claude Codes brain, we have skills which makes super Claude because Claude on skills is like Claude on steroids because it can do so much more than just the normal interface.
32:15What are skills? So skills are instruction manuals that Claude reads before doing a task. So the same way they use a recipe card, which are instructions to make something, in this case, a chocolate cake or it can be a burger, can be pizza, whatever it is, which is a list of instructions that show you how you can do that thing, Claude has ClawCode skills, which are instructions to handle PDFs, to create websites, to create presentations, and to do all those things.
32:38So that when we have a task that you give to ClawCode, let's say you say, let's make a website, and Claude, which is the agent, goes to the skills and says okay which skill do I have to use? In this case it will use the website skill and it will use the website skill to give us a perfect result.
32:52Now why is this important? Because without skills Claude guesses and the result is inconsistent. The reason why the result is inconsistent is because it doesn't have a proper way of doing things if it doesn't have skills.
33:02Right? Because skills are step by step instructions and so if it doesn't have step by step instructions, it does its own thing however it wants to do it in its own way And the way that AI does things can change over time.
33:13Right? And so what it does now might be different than what it does in thirty minutes. Right?
33:16Now with scales, Claude follows instructions, step by step instructions.
33:20And so with that said, because it has a clear prompt that it has to go through, it has to look at, the margin of error will decrease massive, which means that it will give us a much better output. Now that you know exactly what Clockwork skills are on a general level and why they're important, where do you actually store them?
33:34So I'm here in my folder called the YouTube workspace, which is where I manage most of my YouTube stuff, strategy, thumbnails, and ideation, titles, hooks, bodies, all that stuff. No.
33:42This is not scripted, so don't worry. But, essentially, if you see here, inside this folder, there are a series of folders.
33:49So dot Claude, brand assets, context, execution, outputs, references, scripts. By the way, if this is the first time that you've ever been introduced to ClawCode, check out this video up here because it will give you the overview of what most of these are and why they're, you know, that way. But if you already know that and if you already have the fundamentals, then this will make more sense.
34:04We will have a folder called the dot Claude folder. Now the dot Claude folder, you'll see a dot behind it because you actually not be able to see it in your folders. I know that doesn't make sense yet, but I'll show you.
34:12But, essentially, we have this folder called dot Claude, and inside the folder of dot Claude, we have another folder called skills. So that's where we store skills. So now if I go to a new page and I name this general, I can have the folder.
34:24I can press this right here, and I can put dot clawd. Boom. And now inside the doc clawd folder, I can then go here and put new folder and put skills.
34:33Okay? So now when we start adding skills, we add it to the folder itself. Now why do we do this?
34:37We do this because we wanna be organized. Right? And so Claude knows that every single time that it needs to look at a skill, it always goes to the skills itself, and it goes to all of them step by step.
34:47Now if you notice, the DocClaw folder also has commands. And so you can also go here and you can put new folder and put commands.
34:56So as you can see now, we have inside the DocClaw folder, have commands and we have skills. Obviously, nothing inside, but we'll add them later. For the sake of this video, I'm only gonna be talking about mainly skills, but that's essentially where we store Clockcode skills.
35:07Alright. So right here, if you've ever seen something that looks like this, which we call the, uh, folder tree instead of Clockcode, this is how the folders are stored. So dot Clawd is a folder.
35:15Command's another folder. Skills is another folder. Inside each folder, there's obviously documents here.
35:20So this contains only two documents, which we'll go through. And skills contains different folders, and each folder is a skill. There's a lot of folders and skills that I've set so far, but, essentially, every skill that we add under skills folder is a folder in itself because a folder consists of documents.
35:33If that didn't make too much sense, I hope this will. So here we have Doc Cloud skills. Instead of skills, let's say we have the LinkedIn writer.
35:38So let's say we wanna make a skill that writes LinkedIn posts. That's it. Inside each skill, we have three different things.
35:45Okay? We first have a skill. Md which is a process step by step instructions that we tell Cloud Code like, here's how the skill is done, and then we have two more folders.
35:54Okay? So each skill which is a folder has two more folders. The first folder is called references.
35:58Now just so you know, these two, they're not always gonna be there. Like sometimes you need reference docs or examples or templates, sometimes you don't. Right?
36:06So this is like if or or you need them, but in the case that you do need them, we'll have references and this includes documents of examples, of templates, or reference docs, anything that you need there. On this side, we have scripts. Scripts is the Python codes, which is again optional in case you need them, but essentially, if you're running some sort of automation, the code represents the automation itself.
36:22K. So this is the instructions. Hey.
36:24Here's how you do the task. References is like here's supporting documents that you would need to make the task happen, and the scripts will be like the Python codes, which you can think of like the different n a 10 nodes, the different steps of the automation, which are code. Right?
36:35So that's exactly how it's stored. And then going back to my previous example here of my YouTube workspace, which is where I store everything for my, uh, kind of YouTube projects. If I go to doc load, this is the folder that contains these skills folder as well.
36:46It also contains this right here, agent memory, agents, and commands, which I'll walk you through later on, but skills is the one that we have to worry And if I press skills, you can see that it has these many folders, which are all skills themselves. We have CloudMD generator. We have comment analysis report.
36:58We have competitor analysis, Excalidar diagram. We have a set of different skills that do different things. And if I press on one of them, let's say comment analysis report, I will be able to see the folder of references, the folder of scripts, and then the skill dot m d which is a document that outlines everything.
37:13Okay? And so skill dot m d, references and scripts are all folders inside the main folder of LinkedIn writer, inside the Doc Claude skills folder. So they're all folders.
37:21Right? One folder here, one folder here, one folder here, one folder here, one folder here, and then we have this code. M d.
37:27Okay? But that's the structure of different folders inside ClawCode. And this is pretty much the same kind of structure as a competitor analysis.
37:33If I go here, I also have references and and scripts. This also has the JSON file, which is just something that we use to store Google tokens, but some of them only have the skilled at MD. And so that's why I say that this is mandatory.
37:45You always need to have the skilled at MD. The references and scripts, don't always need to have because this right here for example, we don't need any reference files right? We just need a SkilledMD which has a bunch of instructions as to how to actually make the SkilledMD diagram.
37:56So here I have five different what we call marketplaces so the same way that you go to Amazon to get a product, we go to this marketplace to get a skill.
38:05The only difference is that these are free which means that we don't have to pay for the skill itself. At least as far as I know, I'm sure there's skills out there that you have to pay for but as far as these ones, you don't have to pay for them. Now the sites here, I'm gonna leave the links all down below so you check them out, but I have AITMPL, is a cloud code templates, which is the one that I personally use pretty much most of the time.
38:21Then we have this right here, which is also a pretty decent one. Then we have this, bunch of skills as well.
38:27Then we have this here. We have this as well that contains different skills if I go here. There we go.
38:34Now it might seem a bit intimidating at first when you look at this just because it does look very, very techy. I mean, all of this didn't really make much sense to me when I started, but it's actually very, very simple once you break this down. So let's say I go to the Cloud Code templates.
38:47Okay? And like we mentioned, skills are just prebuilt instructions. It's a prompt.
38:51Right? It does a specific thing. So let's say I wanted to create a website.
38:54Okay. If I wanted to create a website, I would definitely use the front end design. Why?
38:58It's because the front end design creates distinctive production grade front end interfaces with high design quality. Use this skill when the user asked to build web components, pages, artifacts, posters, or applications, dashboards, React components, HTML, CSS, layouts, or when styling, beautifying any web UI. It generates creative polished code and UI design that avoids generic AI aesthetics.
39:17Okay? So this is great when you're building a website, when you're building a UI, which is a front end, the thing that you see inside of software, and so we use this here. Now the skill has different components.
39:25It has the name right here. It shows you the amount of downloads that a skill has, and typically the higher the downloads, the better the skill is. In front end design, it's one of the most used skill there is out there, which is great.
39:35We have the install command, and what this is is a command where you can copy this and you can simply just paste it into ClawCode, and I'll show you how to do that, and it will automatically install the skill without you having to give it any instructions. Okay? Then we have the description of the Escher's TL.
39:48We have the license which is the just the document to I'm pretty sure it's to not use copyright, like to not copyright the the actual skill so you don't copy it and then put it on your own marketplace. Right? It's it's going for this person.
39:59And then we have the preview which is a preview of the actual skill and what it is and what it does. And then most importantly we have the code and the code really is a skill dot m d document. So when we spoke about this skill dot md here, all it is it's this is this prompt right here and so every skill is just a prompt right don't make it more complex than it is because that's all it is Just a prompt that tells Clockwork what to do.
40:20Now if I go here, just to show you how it looks like on a different website. Content ideas. Let's do content.
40:24Content creation. Content creation. Let's do a random one here.
40:27And here, we can see that we have the name of the skill. We have the description. We have the command, which we spoke about before, and then we have the file.
40:34So in comparison to here, where we can we are able to actually copy the prompt and paste it to to Cloud Code or do whatever. I mean, usually, you just use a command.
40:43It's the easiest thing to do. Just paste the command, and it will do everything for you. Here, you also have the command, but you have an option to download the documents.
40:50Okay? Because this right here will be a document for the skill. Okay?
40:53We're just a skill dot empty document. And that's it. And instead of downloads, you'll have stars.
40:56So 57,000 stars tells me that it's pretty good and that a lot of people have downloaded it, which tells me positive things about the skill. If I go to here, I can see that I have let's just file search. Let's see what this shows us.
41:09One click install. One click copy. So we can copy this.
41:13File explorer with the skill that I'm d. This is the full prompt. So they're all like I mean, the websites look a bit different, but they all have the exact same thing.
41:20Right? They have the description. They have the the skill itself.
41:23The most important thing is you understanding what skill do I need to be able to do what I want to do and then going into these marketplaces and looking at the skill that you need to then be able to copy and then paste it. I'm gonna copy this and I'm gonna go to Cloud Code in a new version here and I'm gonna paste this here.
41:38Okay and I'm gonna go and run. Now it's asking me if I can do this. I'm gonna press yes and actually turn this to bypass permissions.
41:46As you can see now, it made a folder called doc log. If I go inside, I can see that it made a folder called skills, and inside skills, it made a folder called frontend design. And inside frontend design, it had this skill dot m d document which is a full prompt that we saw here.
42:00The exact thing. And so like I showed you before, I made the folder manually but you can also have Claude make the exact same folder automatically for you. Now let's say I want to use a different skill.
42:10So let's say I went here and I said, okay. Let me use the senior architect. Okay?
42:14I go here. I copy this and then I simply just paste it here. Actually, me go to a new window and then I can go and press done.
42:22I can go to bypass permissions so it doesn't ask me every time to approve. Now you'll be able to see a new skill being added here. There we go.
42:29Senior architect. It's as easy as this. But senior architect in comparison to front end design has references, it has scripts, and then it has a skill that I need.
42:37So sometimes it needs references, sometimes it needs scripts to be able to fulfill the actual skill, but sometimes it doesn't. Right? Sometimes you just simply have a prompt, is just instructions of what it needs to do.
42:45Now bear in mind that the skill.md document is just a list of instructions and it's a standard across all the different LLMs. And so whether you're using Claude, Cursor, Codex, Gemini CLI, or Copilot, or any other of these 20 different tools, you will be able to apply this exact document to all of them.
43:03And that's why it's a very, important document, and it also makes skills way more valuable than people think. I'm gonna go here and delete this skill right here because I'm not gonna need it but next I want to show you how to build a website and how Claude builds a website before skills and then after skills. Okay so I'm also going to delete this front end design actually I'm going to delete the whole thing so Claude folder.
43:22I'm gonna go here, and I'm gonna say, can you make a website for my agency, JM Solutions? It's an AI automation agency that helps marketing agencies to build delivery systems and helps them to double their client delivery in thirty days.
43:35I'm not gonna put too much effort into this. I wanna keep everything equal and just show you the output. Right?
43:39I'm also gonna say the color scheme is gonna be dark blue, uh, and, yeah, give me the landing page in a link that I can access on my web. I'm gonna go here on plan mode because I wanted to kind of plan first before it gives me the actual website, and then we can go bypass permissions to to see the full thing. Alright.
43:54So after a few minutes, we got the plan here, which is a plan on the landing page, dark blue, context approach, color, sections. So we press yes and auto accept, and then we go to bypass permissions. So I don't have to do this every single time.
44:07Bypass permissions, that's it. Alright. I just made the landing page.
44:09I'm gonna copy this. I'm gonna paste it here. And here we have the landing page itself.
44:14W client delivery in thirty days. So as you can see, this is the design style that it made, k, with the actual cal.com link and everything, okay, which is not bad, which is not bad. Now let's say I go back and I delete this.
44:27Then k. And I give it the exact same prompt, but now I'm gonna use the skill front end design, copy the command, and then paste it here.
44:36And now we're gonna download the skill and then do everything again and show you the end result. K. So I just made the skill dot cloud folder skills front end design, and we have the skill.md, which is a prompt.
44:45Cool. Now I can go here. I'm gonna paste the exact same prompt.
44:48The only thing I am gonna say is can you use a different local host, uh, number to give me the link? Just because it gave me the link of local host 8080, and I wanna show you the difference between this and that.
44:58So I can press this. I can go to plan mode. It'll make the plan and then we go directly to build the website.
45:03K. You just made a plan. I can press yes auto accept and then go to bypass permissions.
45:07Alright. Now we made us a new link so I can copy this as well. I can go here.
45:10You can see this is the old website. I can paste the new link and we'll have the new website here, which is much nicer.
45:19It has numbers as well. It has testimonials in case we need any. And we also have the option to book a call as well.
45:24So no obligation, no pushy sales, just a clear roadmap for your agency. So as you can see with the skill right here, it just made the website a lot better because of the prompt that we gave it here. And you can apply this thing to pretty much anything that you do with these prebuilt kind of skills that you can see on the marketplace.
45:38And now that you know exactly how to go to a marketplace and use someone else's skills, in the next section, we're gonna look at how you can build your own skills step by step. I'm gonna be building my own, show you the full process so eventually you can do the exact same. Now to be able for us to make the skill, you need to understand the foundation of how a skill is actually made.
45:55Right? Now you know that a skill here has different folders. So for example, comment analysis report has this folder, this folder, and this document.
46:03Okay? And this document is by far the most important document because it has the instructions because not all the skills have the reference and scripts, but all of them do have an MD document. Okay?
46:11Now let's go here to this one for example and let me actually pull out another one as well. Let me actually go here, put this on the right so you can see the same kind of structure here. So first thing that we have with a skill.
46:21Md document is the name and description. So the name and description are used because when you tell Claude, hey, I want to make an Excalidar diagram or I want to fetch my competitors comments, right, what it does is that the first thing, it looks through all the skills that we have here which are gonna be like 12 only then does it look at the skill dot m d it looks like this first part name and description and then it chooses which skill it's meant to use.
46:45The reason why it does that is because if you don't have the small snippet of name description, it will go through all the files of all the skills and it will consume so many tokens and it will cost you a lot more. And so here we have a preview saying, hey, Claude, if you wanna use a skill, there's the description and there's the name.
47:01If you mention something similar to this, let's use that. It also increases the accuracy of which skill it uses. So first things first, name, descriptions.
47:09We have the name of the skill, Excalidraw diagram, then we have the description to use when someone asks to draw a diagram, make an Excalidraw diagram, or build an editable diagram default for all diagram requests. Okay? The whole file here is done in marked on formatting because that's the way the AI knows how to put the text in hierarchy.
47:25So heading two, heading one, heading three. Right? So it knows which one is more important.
47:28Right? So for this, it's a heading two. This is a heading three, and it goes through step by step what the skill actually is.
47:33K. So step one, understand the request. Step two, research if needed.
47:37Step three, plan the layout. Step four is generate elements. Step five, save and deliver.
47:42And then we go through step six, handle and feedback, critical rules, design principles, layout systems, and then color systems, topography scales, element schema, and then we have the JSON file, which is the type of design that Excalidar, which is a diagram that we use to make different diagrams, makes the diagrams itself.
47:59Okay? So that's the skill right here. On the other hand, let's say I go here.
48:02Comment analysis report, which is a report that I personally made, right, that I actually made step by step. We have the name description, then we have comment analysis report, prerequisites for the skill itself.
48:12This is what it does. We need this to start, so let's grab this first. Step one, fetch comments.
48:16So what it does is that inside the skill, it goes through the scripts, fetch comments dot p y. So scripts, fetch comments dot p y, which is the first, you can say, automation of the whole system. Fetch the top 200 comments by relevance.
48:27Look at these channels right here. Step two, analyze and write a report. Step three, generate PDF, and then it goes through error handling and partial runs and all that stuff in case it breaks.
48:35Okay? And so that's exactly how it's made. Now here I have a empty folder instead of Flockcode, which is general, which is the one that I used before.
48:42I'm gonna go here and create a DocClawd folder. Inside DocClawd folder, I'm gonna create another folder here called skills.
48:49And then I'm also gonna create another folder instead of docclot, which is called commands, which is a new thing that I'll walk you through. Okay?
48:56So commands will be the thing that allows me to call the skills itself. So for example, if I wanna call comment analysis report, I don't have to type comment analysis report. I can do slash comment analysis report, and this is a slash command.
49:08So I don't have to type the name. It it automatically goes there. It automatically calls the skill itself, which is a good thing to have when we do our skills.
49:14So inside skills now, I'm gonna create a skill. That's a new folder inside. I'm gonna name this YouTube title generation.
49:23Boom. And inside of YouTube title generation, I'm gonna create a file which is called the skill dot m d.
49:30And the skill dot m d will be the list of instructions of that specific skill. So the first thing I'm gonna do, I'm gonna paste this here, which I already wrote before, and this will be the name and the description of the skill itself. So we're making a YouTube title generator.
49:40We generate YouTube video title ideas used when asked to brainstorm titles, come up with video names, or generate title options for a YouTube video. And then now here, I can paste this quick sentence saying generate 10 title ideas for the given topic.
49:52Titles should be under 60 characters, create curiosity, and be relevant to a tech AI YouTube audience. And I can also say here's a few title frameworks that work well.
50:05And I can paste this here, which are some title frameworks that we use for YouTube videos. So for example, I failed that topic until I learned this. I never achieved or I never achieved goal until I learned this.
50:15Learn the skill and never unwanted action again. If you action like this, you're gonna be negative situation forever. Because it is all of huddle frameworks.
50:22And so this right here will be the skill that I'm gonna use. I'm going to press command s to save or you can go here file save. I'm gonna test it.
50:29Okay. So I'm gonna go here and say I want you to help me with title generation. I wanna create titles for my new video which is about how I broke past the six figure mark with my AI agency step by step.
50:42I'm gonna press go and now what it will do is it will think it will look at the YouTube title generation scale because that's a skill that it needs to use, and now it will give me 10 title ideas for my video, k, which is great. And so what did we do? We created the skill manually, and now we're testing.
50:57So we're testing that this actually works. And as you can see, it understood which skill to use.
51:02Of course, we only have one right now. And by the way, just because it has YouTube title generation, it doesn't mean that it's gonna use that every single time. And so we said, hey.
51:08I wanna help with titles. I wanna create a title for my new video, and then it gave me 10 title ideas, which is the exact thing that I told you to do in that skill. I never hit a 100 k until I learned this skill.
51:16Give me fifty minutes, I'll show you how to build a 6 figure AI agency. The smartest strategy to hit a 100 k with my AI in 2026. These are pretty good, uh, titles.
51:23It's not fun, but it'll get your agency to a 100 k way faster. I like that. And so now you saw me actually build a skill step by step.
51:29So now that you know how to make it manually, there's actually a trick that we use when making skills, and that's using the skill skill creator. So there's a skill that actually makes skills, which makes the whole process much easier. Now why does it do this?
51:39It's because I'm gonna press go. So you can actually find a skill under manage plugins and you'll be able to see skill creator.
51:47So you can find it here. I can actually just install it.
51:50I can look at skill creator and you can install it here. K.
51:53So now it's installed in your MacBook. Now it it's asking you, do you want it available for all your projects?
51:59Do you want it available just for collaboration or just for you? Only for this exact title or for this exact project. I like to have this skill installed across all my projects, which means that first of all I won't be able to see it here but it will be installed in a way where it's always available to you and you don't have to reference it or do anything because it's automatically gonna be used whenever you're making a skill.
52:19And so here I'll say install for you available in all your projects and now whenever I'm making a skill it will use this skill to make that skill. Now where does the skill creator come from? So the skill creator is something that Anthropic actually came out and made which is why we use it all the time and it's a skill that allows us to be able to use whenever we're creating skills.
52:37So as you can see, skill creator, a skill for creating new skills and iteratively improving them. At a high level, the process of creating a skill goes like this. Decide what you want the skill to do and roughly how it should do it.
52:47Write a draft of the skill. Create a few test prompts and run Claude with access to the skill on them. Help the user evaluate the results both on a quality level but also on a quantitative level.
52:55Rewrite the skill based on feedback, repeat until you're satisfied, and expand the test and try again on a larger scale. And now it goes through the whole process to make the skill which is much more efficient and even a higher quality rather than us making it manually.
53:07So we're gonna make a lead generation system skill. Okay? And this lead generation system skill will start by the input, which is niche, location, and amount of results.
53:22And that will be me just giving it to Cloud, like, on a chat basis, which is fine. And here we can use Apify Google Maps scraper.
53:30By the way, you don't know what Apify is, it's like the Amazon for scrapers. You go inside, you look at one scraper, which allows us to be able to go in Google Maps, extract the results based on what we want. So in this case, we want, let's say, roofing companies in North Carolina.
53:44It will scrape all of them, obviously, with the amount of results, and then it will go actually filter leads by website exist.
53:53And then it will go inside the website, and then it will give me a personalized icebreaker that I can use for LinkedIn.
54:09Okay? So this is an automation. Right?
54:12Something that you would build in an antenna, we're gonna build it here on ClawCode. And we can simply call it every single time that we want. So the first step I'm gonna do is explain to ClawCode roughly what I want.
54:20Okay? So I'm gonna say this. Hey.
54:22So the task of today is to make a lead generation system. The first step is me giving you a niche that I wanna target, the location of that niche, and the amount of results that I want. The second step is you have to go to Apify and use the Google Maps scraper to be able to do this.
54:35Then when you get the list of leads, you have to filter by which leads have a website. The ones that don't have a website, just discard them. Then the ones who have a website, just go inside the website, scrape the website.
54:45And then finally, based on a personalized template that I'll give you, you have to make a personalized icebreaker that I can use to connect people for LinkedIn.
54:53The output will be a Google Sheet with company name, location, website, phone number, and it's fine if they don't have a phone number, and then icebreaker.
55:09Okay? That's it. Ask me any questions if you have any and let's go from there.
55:13Okay. That's the prompt. Now bear in mind, I didn't set up Apify.
55:16I didn't set up anything so we're starting from scratch. Right? I'm gonna go to plan mode so we can plan things out and as you can see, I didn't tell it to create the skill yet because I'm going through everything step by step first and then I tell it at the end, hey.
55:27Now that you've done the whole thing step by step, you know what went right, you know what went wrong, just make the skill. Okay? I'm gonna press go.
55:33So how do you want to interact with the Google Maps API? Do you already have an API key or service set? I'm gonna say Apify.
55:39How should the output Google Sheet be created? Google Sheets API. Let's say we do a CSV file to make things easier.
55:45Text stack. What language framework do you want this built in? Python scripts.
55:48Yeah. I definitely do not wanna be building an anything workflow or else that would not be the point of the video. Submit answers.
55:53And if you're wondering whether something like this would be built in an attend versus Clockode, I would definitely build this in Clockode because it's much easier as well, and we don't have to pretty much worry about much. Now it's asking me, do you already have an API key for Apify?
56:05So I don't. So I can actually go find it. Apify.com.
56:09And by the way, if this is the first time that you hear about Apify, I would definitely recommend checking out this video up here, which goes through the full tutorial on how it works. I'm saying that because it is one of the most used softwares that I use when building automations. So here I am on the dashboard.
56:22I can go down to settings. I can go to API integration, and I can copy my key.
56:27And by the way, you get $5 of free credits when you apply or when you actually get into Apify. I have $29 because that's what I paid for, but you have the free version which is $5. And if you're anything like me and you wanna rinse the software, just make five different emails and then go through them.
56:41So for the LinkedIn icebreaker template, do you wanna provide it now, or should I build the system first and you plug in later? Let me give it to now. So instead of doing LinkedIn icebreaker, why don't we do this?
56:50Since we're gonna get their their number as well. Right? And by the way, can filter by people who have a number, like a phone number itself.
56:56I want you to find from the website so still do the exact thing that I asked you to, but I want you to find from the website the value proposition of the business, what makes them unique, and gave me a case study of a client that they've helped, if you can find it. And this is part of testing, working with ClawCode, and so on, which is good.
57:13For scraping websites to personalize icebreakers, how deep should the scraping go? Homepage only, homepage and about, homepage about services.
57:19I'll go homepage about services as well because I wanna know pretty much everything about it.
57:24It'll be a bit more expensive, sure, but that's fine. I can submit answers, and now we're gonna go ahead. Now it will remake the plan because I told it to change a specific thing about, uh, the process.
57:34So now it's saying that the input will be niche location number of results, the Apify Google Maps scraper to find businesses, filter keep only leads with both a website and a phone number, cool, scrape each website, home, about, and services, Extract value proposition, what makes them unique, and case study of clients who helped, if findable, and up it will be a CSV file.
57:50Great. Perfect. Because now it's asking me as well for the Anthropic API key to analyze websites.
57:55So I can go to the console here, which is the anthropicplatform.cloud.com. I can log in.
58:04I can go down to API keys. I can make a key. Name it Claude Claude code.
58:11Copy the key, and paste it here. By the way, the reason why I'm pasting it into others is because if I press, yes, I have one, I then have to wait a minute for it to understand that it told it, yes, I have one, and it will ask me.
58:21I just might as well just give it the API key itself. K. Cool.
58:24The plan is ready. Here's the summary, and it has everything here. So as you can see, we have the lead generation system, implementation plan, context, pipeline overview, files to create, implementation details, step one, Google Maps scraping, set to its filter, website scraping, cloud API analysis, CSV output, error handling dependencies verification.
58:41Great. Amazing. I can press yes auto accept, and now I can go to bypass permissions, and it will start building it out step by step.
58:50As you can see now, it's starting to make different files. Right? And it's not creating its scale.
58:54It's just creating the actual file itself because we didn't tell it yet to create the scale. It said it created the files. Great.
58:59Want me to run the test? Yes. Run the test with roofing companies.
59:05And I don't know why I'm very adamant on roofing companies in North Carolina. I've never been there in North Carolina, and I want and I want 100 results.
59:14Let's just go 50. Actually, 40 to make things faster. Actually, let's do 30 because I'm not trying to wait all day.
59:19So there we go. Run the test. Now it will start working on it.
59:22K. So now it's running the actual lead generation automation.
59:25If I go to Apify, I can see that it just scraped 34 results. Right? It just scraped it, and now it's gonna use those results to then go and scrape each website out those results, obviously, assuming that they have the website and the phone number.
59:36Alright. So I just finished running. We have the leads dot CSV document, which is this one right here.
59:41I can actually command a copy. I can go to a new Google Sheet right here. I can go here.
59:46I can paste this, and then I can go to data to text into columns. There we go. And now we can have different things here.
59:54So I can make this both a little bit of formatting here and there, make this white. Sorry, guys. I have to do this every single Time.
1:00:00Sorry. I also can't work with Google Sheets. And now we have the company name, location, the website, or the phone number, value proposition.
1:00:08So they do comprehensive roofing, gutter solutions with a family owned approach that prioritizes customer satisfaction and personal attention. They offer financing options.
1:00:15Okay. Cool. Unique differentiation.
1:00:17So this company is led by John Davis with thirty years plus of industry experience. They maintain an a plus BBB rating and don't consider a job complete until the customers are a 100% satisfied. Great.
1:00:26Now as you can see for case studies, interesting, most of them didn't have case studies which is fine. And so this right here is just an extra thing that we can do.
1:00:34I'm gonna leave it here, right, because some of them do. So we have a client testimonial from Tara Mansi in details on how Brian completed multiple home improvements project inside a building. And so the reason why I asked this is because let's say we're cold calling the business we can say something like this we can commend the business on having a case study and we can reference the name of the person who gave the case study right which is always good because personalization over anything is great and so that way there is list right, that we have here, which is great.
1:00:57What I can do now is this. I can say, okay. Great.
1:01:00Now what I want you to do here is the whole process that I run you through, I want you to create a scale. Alright. So now in terms of improvements, I want to do a few things.
1:01:07For example, location sometimes gets mixed up with company name. Website sometimes get mixed up with the location. Phone number, and this is pretty much all shifted for no reason.
1:01:16So I can go here and I can say, there was one problem. Like, example, Guardian Roofing of North Carolina. The location is LC, the website, and then the value proposition is the phone number and so on.
1:01:27Right? And so we are shifting everything by one.
1:01:30So I wanna make sure that when you give me the CSV, you check that all the columns that we have are clearly placed. Right?
1:01:36There are a lot of them that don't have a location, which is fine, so we can leave it there and you put things in the right place, but there are some of them as well that have different things in different places. So I want you to fix that and then give it back to me. And this is part of iterating the actual skill itself.
1:01:49Alright. So now it told us that it updated the leads dot CSV. I can go here.
1:01:52I can copy this and go back. I can make a new sheet, and I can paste it here. I can go to data, split text to columns.
1:02:00And as you can see here, now it fixed the area that we had before, which is perfect. There we go.
1:02:06Now with all this said, we can go to the next step and actually create this skill that we need. I'm in a new tab, and I can say, okay. Now that you generated the lead generation kind of workflow, I want you to create a skill so that when I call the skill, it basically runs through the whole process that we just went through.
1:02:22K? So let me know if you have any questions, but create a skill called lead generation system and then add it in the in the folder itself.
1:02:29I can press go. I can put plan mode, but essentially now it will use the skill of skill creator to then be able to create the skill. K.
1:02:35So now I could press yes and then I can go from plan mode to bypass permissions. Alright. So now it's finished creating the skill.
1:02:41I can go here to lead generation system and I can see that I have a document called skill dot m d and this contains the full instruction manual for that specific skill. So lead generation system, description, instructions, and we go on and on.
1:02:52The only thing they didn't do is that it didn't make a folder here called script and references because this right here and this right here, they're both references, even this, and then this right here is a script. So what I can say here is inside the skill folder for the lead generation system, can you also add a references and scripts folder to store the appropriate files?
1:03:12So the lead generation dot p y needs to be in a scripts folder, uh, and in the rest as well, like, based on what you know and, obviously, your definition of scripts and references, just include the files there to make things organized, and I can press go. As you can see now, I just made two different folders, references and scripts.
1:03:26And inside scripts, we can see that we have the lead generation.py, and then inside references, we can see that it have EMV requirements text and sample output CSV, which is great. One thing I wanna do though is I wanna delete these ones.
1:03:36So I can say, okay. Now that you added those inside the references and scripts folders, can you delete the other ones that we're not gonna use? As you can see, it deleted both of them, and it's also asking me, hey.
1:03:45Do you also wanna delete dot EMV and leads dot CSV in the output files? Yes. Delete them.
1:03:51There we go. So now we have a references folder. We have a scripts folder for the automation.
1:03:54In this case, the Python file, which doesn't make any sense to anyone here. It's just a file that runs the automation. Right?
1:04:00Because behind every automation in 9010, make.com, it's all Python. It's also JSON.
1:04:05Because as you may know, all the platforms in 9010 or make.com, behind the automations, it's all code. And so this is the code itself that runs the actual system and as you can see we start to add more and more skills with more and more folders. Now there's one more thing that I want to do and it's this which has to do with commands.
1:04:19Now I have a skill called lead generation system I want to create a command so that I can put like slash lead generation and then it calls that specific skill automatically Can you make that command? I can go bypass permissions, and I should be able to see a command being made here. Alright.
1:04:32So now I've finished making the command. As you can see, we have the leadgeneration.md here.
1:04:36So if I go to a new chat, I can press slash lead generation, and then that will activate the lead generation system that we have in the skills itself.
1:04:46There you go. And now it's gonna ask me, hey. What niche do I want?
1:04:49Location, number of leads, and output file. I was gonna say roofing companies again.
1:04:52MetSpas in London, and I want to find 30, and I want to name the file Medspa London 30.
1:05:07Boom. And I can go to bypass permissions. Okay?
1:05:11So that right there is the full process of creating a skill step by step and actually testing it and making a command so that whenever I go to clock code, I can simply just press slash the skill name and it will automatically call the skill itself. Alright. As we can see here, we have the result, MedSpa's in London.
1:05:26I can make a new sheet, data, and then we have the full list right here. And we just call this simply by pressing slash command of that specific skill.
1:05:35So now you know exactly how to build your own ClawCode skills. But what if I told you that your ClawCode skills could actually improve themselves twenty four seven overnight without you having to do anything? This is where we get to self improving skills.
1:05:46Like we saw earlier, the first way of actually self improving your own ClawCode skills is to do it manually. So we gave it the first basic version. It made this right here.
1:05:53Then we said, okay. This right here is a bit messed up. This right here is a bit messed up as well.
1:05:58Can we change this? And then it came and did this as well. And so what it did is it made it better.
1:06:02It improved itself. And this is great because it allows us to make a near perfect score or the output for our specific skills. But if you wanna take your skills to the next level and actually measure how much it improved by, well, this is where we get to the skill evaluation.
1:06:14So the first step of skill evaluation is making sure that you have the skill creator installed. So you need to have this installed and this isn't something that we spoke about in one of the first steps. Then once we have the skill involved, we then have to tell Claude to evaluate our skill.
1:06:26And so here I can say create an eval for my lead generation system skill. I'm just gonna do the YouTube title generation just to keep it more simple. So for my YouTube YouTube title generation skill, I can press go.
1:06:37Make sure it's bypass permissions. As you can see here, it created two eval files. One is eval dot JSON and one is trigger eval dot JSON.
1:06:43Basically, the whole point of eval is to make our skills self improving and to make the skills itself improve themselves over time. The first thing it made is three test use cases. So if I go here to the skill itself, it made a folder called eval, and we have two.
1:06:57So the first thing is test cases. So it uses specific inputs to run the skill. So let's say we're using the prompt of ClawCode beginner video.
1:07:03What it tests is framework usage, character limit, curiosity variety. Then it uses Versus code versus cursor versus WinSock comparison. So basically, these are three video ideas, and this is what it tests for.
1:07:12Right? Each eval has five to six expectations checking. Exact count, 10 titles.
1:07:17Character limit, 60. Framework adherence, topic relevance, curiosity, engagement, and variety. Whilst on this side, we have 20 trigger tests.
1:07:2310 should be triggered, 10 should not. Should trigger is give me YouTube title ideas, brainstorm video titles, help me name my video, etcetera.
1:07:29Should not trigger is writing scripts about YouTube, video script, thumbnails, SEO analysis, blog titles, channel growth advice, adjacent but distant tasks. And so now what it'll do is it will test the actual skill based on this criteria, and it improves if it finds that this output isn't good enough. So it will test all these three use cases, and it will test all these 20 triggers.
1:07:47With this said, I can now say run the eval from my YouTube title generation skill, and I can go press run. What this will do now is it will go and run this too and then give us a kind of a report. Alright.
1:07:57Perfect. Now it's finished with the evaluation. It's telling us that with the skill, it did a 100% correct.
1:08:01Without the skill, 33%. With the skill, 100%, and then average would be 50% here, and average would be a 100% here. And it also gave us full document here with the prompt, with the skill, the outputs that we got, four more grades.
1:08:14So six out of six passed, zero failed, which is perfect. Then we also have the benchmark comparison here, which gives us all the data of what the output looks like with the skill and what the output looks like without the skill.
1:08:26Okay? And so here we have the pass rate with the skill, without the skill. Obviously, here we have with the skill, a 100%.
1:08:31Without the skill, 50% plus minus 17. So as you can see, with the skill, it's a perfect score of a 100. And so what it did here is that it evaluated one, two, and three so that we have the ClawCode beginner, which is the first topic that it used, the first test use case.
1:08:43With the skill without the skill, you can see all the numbers here. Assertion, so all the different things that we needed inside that specific, uh, title. With the skill, it got everything right.
1:08:51Without the skill, only two. And it did that for all the three topics. So topic number one, topic number two, and topic number three.
1:08:57And then we have the analysis notes. So with the skill, achieves 100% pass rate across all three evals versus a 50% average without the skill. The biggest skill advantage is in the format compliance.
1:09:07Exact count, 10 titles, and character limit. Baseline consistency overproduced and exceeds length limits. And so all this is doing is it is testing three different use cases for our skill, and it's telling us, hey.
1:09:16Out of 10 times that we run this, how many times fail? How close is it to the desired output that we ask it to do? And so you can always have this evaluation for every single skill that you do.
1:09:23We'll be able to test it. Now, of course, YouTube title generation is a very easy system, but if we were to do something with the lead generation system, it would work pretty well just because there's a lot of moving components in the actual skill itself. And so at this point, you can tell it to improve the skill, run the eval again, and see the numbers and how it's improved.
1:09:38And so at this point now, we have a 100% accuracy rate, but let's say we had 80%. So what we would do in this case, if in case it has 80% or anywhere less than a 100, I can tell it, hey. It's missing x y and z.
1:09:48Can you improve the skill? It then improves the skill. It tells us, It's done.
1:09:50And then we ask it to give us the evaluation again. And this is a whole repetitive loop that it does to make sure that these skills are self improving. One more thing I wanted to go through when it comes to skills is the ability to call them using slash skill.
1:10:03Right? So in this case, in the dot claw folder, we have commands and we have skills. I've gone through skills, but commands are the things that we use to call that specific skill.
1:10:11Right? So for example, let's say I wanted to run YouTube title generation. Instead of saying, hey, can you run YouTube title generation?
1:10:19I would say slash YouTube title generation, and it will simply run automatically, right, just because we've added the command, which allows us to call it directly. Same way that if I wanted to run the views to clicks, I would go slash views to clicks analysis, and it will run the skill itself.
1:10:34And this is great for us because we simply don't have to manually write the skill itself. We can just put slash, call the skill, and you're good to go.
1:10:42And if I go to slash, I can see down to the commands, I can go down to the skills right here and I can see all my skills that I can call automatically and it will simply trigger that specific automation.
1:10:53And the way that we did that is let's say we have ExcaliDrawDiagram and we have YouTube title generation. If I wanted to call this one here, I would have to put slash ExcaliDrawDiagram and I simply don't have the command there.
1:11:04So what needs to happen because of that is I would have to explicitly say, hey, can you run my skill called the Excali draw diagram? Right?
1:11:15Which is quite inefficient, especially when you have 20 to 30 skills that you wanna browse through and just pick one automatically. Right? And so in this case, to turn that skill into a command, I can say, hey, can you turn my two skills into two commands and then so that I can call them using slash the name of the skill and then add the file inside the commands folder under the DocClot folder.
1:11:37Alright. Now I can press plan mode because I wanted to make a plan first. Alright.
1:11:41So here we have the plan itself, turn skills into slash commands. We have two skills, and we're going to create the commands as well. I can just press yes and auto accept.
1:11:49You can see here that it's gonna create a file called under the commands, under here. Same thing with this, YouTube title generation and Excalidraw diagram. I can press yes, auto accept.
1:11:59I can go to bypass permissions, and I can let it do its thing right here. Alright.
1:12:03So after a few minutes, it made the two different commands, which are two different files. So we have this one here, which says when someone calls slash Excalidraw diagram, use the Excalidraw diagram skill to handle the request. Same thing with the YouTube title generation.
1:12:15So when someone calls slash YouTube title generation, use the skill right here to handle request. That way, if I go here and I can put slash YouTube title generation, it'll run the skill itself right here.
1:12:26As you can see, it's looking at all the different references, then it should ask me what video do I want to make a title for. As you can see now, it's asking me for the video topic, angle, number of titles, And then if I go to a new conversation, I can do the exact same with Excalidraw. So I can put slash Excalidraw diagram, and it will start running the Excalidraw skill.
1:12:44And now it's gonna ask me what concept or system do I want visualized and what level of detail. And the next level would be obviously having 10 to 15 different skills with 10 to 15 different commands so that we're able to call the skills directly without us having to type the skills off or have a conversation with ClockCode.
1:13:01We can simply just filter by the skills. You can go here, and we can look at all the skills that we have and call them directly right here.
1:13:09Alright. So now you understand how to build custom skills and slash commands that make ClawCode do what you need in one line. But what if sometimes you want ClawCode to do things automatically without you even having to ask?
1:13:22Well, that's exactly what hooks are for, and that's what we're covering in the next video. Let's dive in. Alright.
1:13:27Now the one feature that I'm talking about is, of course, clockwork hooks. If you never heard of clockwork hooks, they are a bit more on the technical side, but they're actually very, very easy and very powerful when you understand how to use them. So what they are is automatic actions.
1:13:40They run at specific moments, like every time x happens automatically do y and it's almost like setting rules that your assistant always follows. Now the only difference between Clockwork hooks and Clockwork skills is that hooks are things that Clockwork uses regardless of whether you call them or not.
1:13:57They're on the organization level. Right? So you set a rule to every single time Clockcode runs, it always uses these hooks.
1:14:04Okay? And they're automatic things that I'll walk you through what they what they actually are. For example, let's say you give Clockcode a task.
1:14:10You ask Clockcode. Clockcode does the work and it uses these hooks, right, whenever it does the work. For example, we have pre tool use, so we have hooks that we have to use before the tool runs, post tool use, which are hooks that we have to use after the code is written, stop when ClawCode finishes, or notification when Clockwork needs you.
1:14:28So these are the four different types of hooks that you can have, uh, and they're just examples of what it could look like. Right? So you have a hook that runs every single time the workflow finishes or Clockwork starts or when it stops or when it needs notifications.
1:14:40And they're almost like triggers, what starts the actual hook itself. Now I'm here on the YouTube workspace and if I go to my Docloft folder, which is where the hooks are gonna be stored, there's a file called settings dot JSON.
1:14:52If I press in settings dot JSON, you'll be able to see the full code. Now the code itself is actually very, very easy to understand. Right?
1:15:00Now don't worry about the code too much because I'll explain exactly what what it is, but essentially, it's stored in a file called settings dot JSON. And so here I am on a new project. I can go to new folder here.
1:15:10I can name it docclaud and by the way docclaud is also the folder that we use to store skills, store commands, to store different things of ClawCode. And inside docclaud, we want to add a new file called settings.
1:15:23Json. Now a few things. First of all, that whenever we're making a hook, a hook is in sort of like automation or some sort of plug in, we are all making it under the same exact file.
1:15:34So this is not like a skill where you would need to add a new, uh, file for every single hook. We use settings dot JSON as the main file to store all the different hooks that we have across our organization.
1:15:45Okay? Second thing is that when we are, uh, installing hooks in settings dot JSON, you will be able to use this hook across different projects. This will make more sense when I show you, but essentially understand those two rules.
1:15:57First thing, you will be able to write all the hooks here. And then second thing, just as if you were able to to write all the skills in one file without you having to make different folders, but in this case, skills are different. And then you need to understand that this is on a organization level, not on a project level.
1:16:11With this said, let's go to here and the way to actually use hooks is you can either press slash hooks, which you will be able to only to do using your terminal as you can see right here. Terminal.
1:16:24Let me go here to terminal, new terminal, and I can press Claude. So here we are on a terminal view using Claude, and I can press slash hooks. View hook configuration for tool events, I can go enter and here we have zero hooks configured.
1:16:37So we have zero hooks that are configured right now in our organization. Okay? And here we have a read only menu to add or modify hooks.
1:16:45Okay? And so in the easiest way, what this is is you being able to say okay I want to run some sort of notification before the tool execution and so these are all triggers right when I say trigger I mean this is a trigger before the tool execution this is a trigger after the tool executes This is a trigger post tool use failure so when the tool fails.
1:17:05So let's say you want send yourself an email whenever a tool fails, you use that. Permission denied. After auto mode classifier denies a tool call then you use the hook Then you can make a hook there.
1:17:16And so you have different options that you can work with whenever you have these kind of statuses inside of flaw code. But let's say I go to notification, press enter, it says we currently have no hooks configured.
1:17:30To add a hook, you either need to edit settings dot JSON directly or you need to ask Claude. And now let's go to the normal view and let's talk to Claude.
1:17:39Say, okay. Hey. I want to make a hook.
1:17:43So here it's asking me what kind of hook do you wanna do. Here are some common hook events, and I have running before the tool is called, run after tool completes, run when Cloud sends a notification, run when Cloud finishes a response, run when sub agents finishes.
1:17:55Let's do this. Let's run a hook where when Cloud finishes a response, it sends me notification. K.
1:18:01So that's gonna be the hook that we're gonna make. So when Claude finishes doing something, it actually gives me notification. I can press yes.
1:18:08As you can see now, it's running the hook. And by the way, when you see something like this always in close code, it just means that this is the previous version. This is the updated version.
1:18:16And it's saying, do you agree that we can change this exact file? And I can say yes. And now it's gonna add the hook instead of settings dot JSON here, and it says that the hook is live.
1:18:26Every time I finish a response, you'll get a Mac OS notification saying that Claude has finished responding with the title ClawCode. And I can review or edit at any time via hooks.
1:18:34So in theory now, what we made is a hook that alerts me every single time ClawCode finishes doing something so that I'm able to then do whatever I want, work on different things simultaneously, and I'll know when I have to go back to clock code to see the output. So here I can say, what's the capital of Italy? And I will get a notification when it finishes running.
1:18:51Now this is the exact notification that I just got. The reason why I'm showing you here is because I'm using my my monitor so it doesn't actually show you the notification on the monitor as well. But essentially what would happen is that you would be talking to ClawCode at the same time you can be doing something else and you will know whenever you have to go back to ClawCode because of this notification right here.
1:19:11And there there is a hook. Right? And this will run every single time that you talk to ClawCode regardless of whether you have all these documents in your project, it will always run.
1:19:19Right? These are rules that will always run regardless of x y and z. So in this case, when ClawCode finishes, then we get a notification.
1:19:26Alright. So another example here could be that let's say we had a dot EMV document. So this dot EMV document is full of API keys for OpenAI, Anthropic, Stripe, all these things.
1:19:36And by the way, they're fake so don't try to copy them. Fake API keys. Let me just put API keys.
1:19:41None of these are real and what I want is that whenever I talk to ClawCode, maybe I could be doing a complete custom code, it's full stack project or maybe I'm doing a full website or maybe I'm building some some automations. Right? Ideally, we wouldn't want ClawCode to impact the dot EMV document because once this is impacted, let's say it deletes everything by mistake, we have to reconfigure everything.
1:20:01And all the automations break because they're all reliant on these passwords, right, to access softwares for the automations. So what we can do here is I can say, I want to make a hook. So in this case, we want to tell Cloud Code, hey.
1:20:15Don't edit the dot EMV at all. Right? And so in this case, it will be a pre to hook.
1:20:20So we would say, yes. So this is a pre to hook. Essentially, what we want is I want the Talcata code to not touch the dot EMV document.
1:20:29So don't ever touch the dot EMV document regardless of what it is that we're doing. So if you have to edit the dot EMV, then you can't do it because you have to stop yourself because it's sensitive information. So now we're making another hook.
1:20:38I can press yes. Let me go to bypass permissions so I can let it create the own hook.
1:20:44There we go. So now it added the free tool use onto our settings of JSON. So if I go here, I can see now that I have the stop hook, which means that whenever the clock code stops, it gives us a notification.
1:20:55So we get a notification here when Claude has finished responding. And then we also have the right edit one, which is the one that allows us to block EMV.
1:21:05And then we have the pretool use hook, which is the one that we use to be able to not let ClawCode touch our dot EMV document because, of course, it has all the passwords. So now here I can say create a dot EMV file with API key test one two three, and you can see that it says I can't do that. Your settings dot JSON has a hook that blocks me from writing to the dot EMV files since they contain sensitive information.
1:21:27You can create it yourself by running echo this this this. Right? And so this is extremely important.
1:21:32Right? I wish I knew this sooner because there are many many times where I told ClawCode to do something and it deleted something from here or it added something from here which messed up my whole project and so in this case what we're saying is whatever happens don't edit this Right? What I would recommend is you add all the API keys that you have, all the passwords that you have that ClawCode needs to have access to whenever it needs to run a specific automation.
1:21:55And then once you have those, put this hook here, make sure that now Clockcode is not able to edit the dot EMV to be able to make sure that you're safe whenever you're building the project. And last example, let's say I am building a project for a client. At the end of every session, I want to know exactly what files Clockcode touched and when, which is an audit trail.
1:22:13And so what I want is a log of all the kind of edits of different files that it made with the timestamps of when it actually touched them. So here I can say, add a post tool use hook because post tool is gonna be the trigger. After ClawCode finishes, it logs every file ClawCode edits or writes to a file called clawd slash log dot t x t with a timestamp, and we can press go.
1:22:32And this is great because a lot of the times that I'm building with ClawCode, I am making stuff, and then I'm forgetting which files are edited. And sometimes things just get deleted, and so this allows me to know exactly what got edited and what got deleted at what timestamp. Alright.
1:22:47So it just finished running. It said the post tool Hughes hook is now configured. I can see here that I have the write edit type command and I have the status login file changed to.
1:22:57All of this makes no sense to me. It should log everything to a claud slash log dot t x t document. Okay?
1:23:03So now if I go here and say, can you make a new markdown format document and name it YouTube title generation? As you can see now, it's getting the task but it's also updating the cloud/log.txt.
1:23:16If I go here, I can see that seventeen forty four zero nine, we wrote YouTube title generation.md. Let's say I wanted to say, can you also make another document or another five documents naming them, uh, hook number one, hook number two, three four five.
1:23:31And so essentially, it's whatever change you make, it will log it here. So as you can see, it made five different documents. If I go here, I can see that we now have five different logs.
1:23:39So one two three four five. It'll be hook one, hook two, hook three, hook four, hook five. That way you can have a log of all the edits that you've made so you can always come back to them in case it went wrong in some way.
1:23:52Alright. So now Clogcode can actually run custom commands and run actions on its own, but there's still one limitation, and that is that it can only work with what's inside your machine.
1:24:01What if you want Clogcode to talk to your browser, talk to your databases, talk to your APIs, or any external tools? Well, as of right now, it can't really do that without this one thing, which is called model context protocol, and it stands for MCP.
1:24:15And it completely changes what's possible. Let me show you. Alright.
1:24:17So the best way to explain MCP is they're just plug ins. Right?
1:24:22Essentially, other people build tools and you just plug them in. So you have Cloud Code, which talks to the MCP server, which is running in the background, and the MCP server will be the one controlling the external apps.
1:24:32So your browser, your databases, maybe it's Notion, maybe it's Supabase, maybe it's your Gmail. Right? Whatever that may be.
1:24:38And it makes it so much faster for us to be able to use the softwares because it is trained on how it should talk to the API documentation. And the API documentation is essentially the documentation which shows how you can actually automate something with the software. So you can install the MCPs.
1:24:52They run as servers. Right? And Cloud gets new tools and talks to real apps.
1:24:56So you basically don't have to configure things yourself. And so how are they different from skills and hooks? Well, skills are just instructions.
1:25:02We say do this task like this. Right? We have step by step instructions that CloudKit uses to fulfill a certain task.
1:25:08Now I made a full video on skills. You can check it out up here. Then we have hooks, which are guardrails.
1:25:12So this is giving it rules as to saying, don't touch this file. It's sort like the security guard. Right?
1:25:17Which says, hey. Before this happens, make sure this is in place. After this happens, make sure that is in place.
1:25:22Different rules. I've also made a full video on hooks. You can check it up here.
1:25:25And then we have the MCP. So the MCP gives us new abilities. Right?
1:25:29So we say now can use crow. Now you're able to connect to Notion and you know exactly what you should do. Right?
1:25:34And this is a new pair of hands that we give to ClawCode, which is why I say that it completely changed the way that I use ClawCode because it allows me to build things that previously either taken a lot more time to set up and things that previously were not So, again, skills are instructions, hook or guardrails, and MCP are new abilities.
1:25:49They work together not against each other, so they're different things. Right? But when you combine all those three things together, that's when ClawCode starts to become really, really, really powerful.
1:25:57Alright. So here I am on my main YouTube workspace, which controls my main YouTube content generation system. And inside this workspace, you can see that I have a file called dot m c p dot JSON.
1:26:07So dot m c p dot JSON is gonna be the file that we use to add the MCPs. As you can see here, if I zoom out a little bit, here we have an MCP server called Chrome Dev Tools. And so what does this do?
1:26:18This allows us to be able to use the web and access different things outside of ClawCode. Right? This doesn't mean scrape a website.
1:26:24This means going inside a website and taking a screenshot, for example. Now if you look at this, you're like, what the hell am I looking at? That's okay because you don't really have to understand the code.
1:26:33You just have to be able to talk to ClockCode to ask it what you want. Right? As you saw, to set up the MCP, we have to make a file.
1:26:39But before we do that, let me show you why we need it in the first place. Hey. Can you go to apple.com?
1:26:43Can you take a screenshot of the main page, and can you then give it to me here? So in theory, this should be able to then go to Apple, take a screenshot of the page, and then bring it back here as an image. As you see though, it says, I can't take screenshots of website.
1:26:56I don't have a browser automation or screen capture tools available. I can only fetch the page content, read screenshot files that already exist on your machine. So this is why we use MCP.
1:27:04And so what I can do here is I can go to file. I can write dot MCP dot JSON. And inside here, I'm gonna paste the code that we had before.
1:27:13File and save. And by the way, I'll show you how to get this code as well, so don't worry. Once we have this here, let me ask you the exact same question in a new tab, and you'll see how now it's gonna go to Apple, take a screenshot, and then bring it back.
1:27:25So you can see here it's saying, let me navigate to apple.com and take a screenshot for you. I can say yes. Let me do bypass permissions.
1:27:31It doesn't ask me. It said it successfully navigated to apple.com. It took a screenshot, and now we have the screenshot of the page, which is this one here.
1:27:38K? And so just in a matter of seconds, we were able to unlock a new feature in ClockCode just by adding this m c p dot JSON file here. You can start to see the different applications that come into this.
1:27:48This is a very, very simple example. One of the most used examples is Chrome DevTools, which is the ability to go to websites, take screenshots, and access the web, but we also have other ones as well.
1:27:56And by the way, just so you know, it actually opened up a new Chrome tab here, which is how it was able to take the screenshot and then bring it back to the workspace itself. Alright. So to find the snippets of code that I just showed you, which allow Cloud Code to have new features, plugins, we have to go to the mcpservers.org.
1:28:11Now if you're aware of skills, which is the ability for us to add instructions to Cloud Code and do different things, skills also have something similar. So if you go to skills.smp.com, we have what we call a workspace full of skills that we can copy and paste into Clockcode.
1:28:25So this right here is Amazon for MCP servers. And so as you can see, we have the Chrome Dev Tool, MCP, which is, uh, an option that we can use. We have Firecrawl.
1:28:33We have Superbase. We have Playwright. We have Xcode build MTP, and we have a lot more as well.
1:28:37Right? Cross official, search, web scraping, communication, productivity, and so on.
1:28:42Right? So let's say I go to featured. Let's say we go to the Chrome Dev tool that we had before.
1:28:47If I go here, I'll be able to see the name, the description as well. So this is the tool that allows your coding assistant, such as Gemini, Cloudcursor, Copilot, control and inspect a live Chrome browser.
1:28:58So it allows to control Chrome and go to a new browser and do whatever it is that you wanna do. It looks at the key features, the disclaimers, usage statistics, update checks, requirements, and we have the code itself.
1:29:10So here, we can copy this code, and that's the code that you'll be able to paste right here. Go here. You can paste it, and then make sure you press command s.
1:29:19Then it pushes everything back. That's why it kinda changed the structure. But it's the exact same code.
1:29:23Now bear in mind with MCPs, not all of them are vetted, which means that they weren't completely verified. So use your judgment as to what it is that you can actually be using. If I see that this is in the featured MCPs, then it's probably something that I am gonna be able to use because it's trusted.
1:29:37But let's say I go to cloud service, right, and find a random one, let's say AKS MCP, I don't really know whether this is good unless I look at the stars here, the GitHub stars. The higher the stars, the better it is.
1:29:48If it has two stars, then I'm most probably not gonna be able to use this because I don't know the impact it will have on my claw code and the things that it does. And if you ever copy some code and you have no clue what it actually means, you can simply just say, what is the m c p dot json that I have right now? What does it do?
1:30:01It tells you, and then I also told it to give it to me in one sentence and make it very, simple to understand. And here it says that it lets me control your Chrome browser, clicking, typing, taking screenshots, and inspecting things so I can help you test and debug websites directly from this chat. Now one use case for this specifically is, let's say you are making a landing page, and let's say you were going to look at the top landing pages from five different websites, and you wanted to take inspiration as to the top header was that on that landing page specifically.
1:30:26So here we have Apollo, we have MailSSO, we have Email List Verify, which are all different SaaS, so softwares that verify different emails or give you some data. Right?
1:30:35And let's say I wanted to make a new tool and make a landing page similar to these ones. And so here then I can say, can you go to each of these websites? Can you take a screenshot of the landing page and then bring it back and make a new folder under our workspace, name it screenshots of landing pages, and leave the pictures there.
1:30:50And I can press go. I can put this as bypass permissions. Okay.
1:30:53So now it's going inside apollo.io. It's going to mail.so, email list verify.
1:30:58The next one is zero bounce. Then we have verifalia.com. And then finally, have hunter.io.
1:31:03As you can see, it just finished everything. It made the folder of screenshots of landing pages. And here we have Apollo, email list verify, Hunter, Meldo SO, Scrubberyfalia, and Xerobounce.
1:31:13Right? And so all of these could be used then when you're making your landing page as inspiration. You can say, hey.
1:31:18Look at these images of landing pages, kind of emulate their style and make me a new landing page and call it x y z name. So another one that I use often is the one of Superbase. So Superbase is a software that allows us to be able to store data.
1:31:31It's sort of like a Google Sheets on steroids. Right? It allows us to add data to a table that we can then fetch using its API.
1:31:37So you're throwing a question. All the information is stored in Superbase, and Superbase gives you the answer. So here, what I can do is I can just configure this here.
1:31:44So I can copy this, which, again, usually is in the bottom. Right? Here, I can see is the name.
1:31:49We can see the demo of this, of how it works. Then it's giving me the actual setup. What I would have to do here is I would have to go and copy this.
1:31:58You can also look at the setup documentation here, which goes through everything that you have to do, but it's pretty much the same as the one we had here. So I can go here. And by the way, you don't actually have to make a new file.
1:32:07You can use the exact same one. Now we have this, and then I can say, can you add this to the MCP?
1:32:12I can give it the new MCP of super base. When I press go, put bypass permissions, it will now add the MCP to our new MCP dot JSON, which will be here.
1:32:22As you can see, it's making the change. I can say yes. And now we have the Chrome DevTools.
1:32:26We have superbase, and we have more and more and more and more as we start to add new things to our file. K.
1:32:31Let's say I wanted to have this document right here, which is a fake SOP for an ecommerce store. I wanted to add it to my super base so that we can store all of this in a table and then we can then fetch and can ask questions about the thing itself like a chatbot. I can go here.
1:32:44I can download as a PDF, and I can upload it right here. And then I can say, hey. I uploaded the PDF of the ecommerce store SOP and knowledge base.
1:32:53Can you please now make a vector database instead of super base, and can you add all the information there, vectorize it, and then, yeah, make a new table and call it SOP. I'm gonna make this plan mode, so plans. What it's now gonna do is it's then going to use the MCP of Superbase because it has to access the software itself.
1:33:10As you can see, it's reading the MCP dot JSON file because that's where we have the actual MCP itself. Now it's giving me a plan for creating the local vector database in Superbase. Sure.
1:33:20Looks all good. I understand half of it, but not even half of it, but it's saying all the things that I wanted to do. I can press yes, auto accept, and then I can put Papa's permissions to let it run on its own.
1:33:30So it's gonna take the PDF. It's we call it chunk it, so it's gonna chunk, meaning that it's gonna take the actual document. It's gonna split it up into different parts and then put it all into itself.
1:33:40And the VACI database, by the way, is just a a Google Sheet which stores all the information. So I just pulled up this page right here where I have to authorize the MCP connection. So MCP client connection, authorize API access.
1:33:52I can say, yes. Choose your organization. Authorize cloud code.
1:33:56Authentication successful. I said done. You farm a project, which takes two minutes to set up, make an account, and then pretty much make an organization, and that's it.
1:34:03And now it's going to create the table in the Supabase account. As you can see, it made a table called SOPs, and we should be able to see the information starting to come in. And as you can see now, we have the information that we had on the Google Doc.
1:34:15We have it here in the, uh, Supabase itself. Right? And the way that this works again is that we have each text, which is an embedding.
1:34:22Embedding is just a number. And when we say query the vector database, when we ask questions, it matches the numbers of the questions or the letters that we that we actually give it to the numbers here, and it gives us the answer. Right?
1:34:33But, essentially, what it did is that it used the MCP of Superbase itself to then send all the information here without us having to set up anything apart from that MCP code itself. Now although MCPs are great, there is one fallback, which is the cost. And so if I go here to slash context, I will now be able to see how many tokens the MCPs actually take.
1:34:54So here I can see that I'm using CloudOPUS 4.6. I'm using 15.6 k out of 1,000,000 tokens, so 2%, and only two MCPs use 10,800 tokens, which is quite a big amount considering the fact that we only added two.
1:35:07And you can see here that the EMCP tools, right, so each tool, which means that each type of action that it can do is requires tokens to run. And so we have click, which is one sixty three, close page, one twelve, drag, one fifty five, emulate, and so on. Right?
1:35:21And it keeps on going and going. So one important thing is that the more MCPs you add, the more money you have to spend on tokens. So make sure to not use 15 to 20 MCPs because they will take up a lot of your tokens and context.
1:35:34Just use the top two to three that I showed you and make those your main priority ones whenever you're using ClockCode. And so with the fact that you are spending a lot of tokens with the MCPs itself, usually, what we do is that we install the MCPs here.
1:35:46Right? So let's say we install Chrome Dev Tool or let's say we install the Supabase. We check if it works, which means that you test it, see if it did the right thing, and then you turn that, uh, MCP itself into a skill because skills require less tokens.
1:35:57And so with the healthy setup, you typically have a system prompt, which is the prompt of the actual AI itself, ClawCode. Then you have built in tools, and then you have MCP tools. You wanna have a few, and then you have your conversation, which means it's a lot of room.
1:36:09Because the more you speak to ClawCode, the more the tokens or what we call the context window increases, which means that you can't be having conversation with ClawCode forever because the tokens, the space of the conversation ends. Right?
1:36:21And so you wanna have as much space as possible for the conversation instead of having the system prompt, building tools, and having 20 different MCP servers, which take up so much tokens and so much space and context where to the point where you can actually have a proper conversation with Clockcode before the tokens runs out.
1:36:36You wanna be as efficient as possible with your tokens that you are using in ClawCode. And by the way, if you want all the resources that I'm gonna be talking about in this video plus being part of a community of over 9,000 people who are actively building with AI, then check out the second link down below.
1:36:54Alright. So congrats on finishing module two. At this point, you have fully customized Clockcode.
1:37:00You have custom skills, you have automated hooks, and it's connected to every single tool through the use of MCP. So now you have a pretty powerful setup. But if there's one thing I learned after working with Clockcode for hundreds of hours is that it's not so much about your setup or your tools, but it's about how you speak to Clockcode, which is the prompting.
1:37:17And that's exactly what the next module is all about, which is the prompting strategies that will 10 x the quality of everything that you do with Clockcode. Let's dive in. Okay.
1:37:24So the first hack is called splitting prompts. Now one of the biggest mistakes that I see when people use Clockcode is them adding huge and humongous prompts with a ton of information and a ton of tasks all in the same prompt and then expecting ClawCode to give them the up for the day one. An example of this could be them saying, build me a CRM with off database API dashboard emails and deploy it, which are basically seven different tasks, right, or four different tasks within the same kind of major thing that they wanna build.
1:37:48And there's a few problems with this. The first one being that there's a lot of context or tokens used in that same session that you're using Cloud Code. The second thing is that the output is not gonna be the best just because you're feeding it and you're telling it 20 different things at once.
1:38:00The same way that you tell an employee, hey. Can you build me a full software with this, this, this, and that without giving the person any pure structure? And the issue here, even when I started, was that I thought ClawCode would build anything.
1:38:09Right? It could build anything and anything if I give it one prompt and that's it. But it's much better to split prompts and take a more structured approach.
1:38:15So instead of putting this prompt right here with seven different tasks, we split the tasks up. Right? And we talk the ClawCode task by task.
1:38:21So we look at this, and we say, okay. The first task is setting up the project. Then it will set up the project, and then we review whether it did a good job or not.
1:38:28If it did a good job, then we go to the next thing. If it didn't do a good job, then we go back and fix it. Then we go to task two, which is knowledge base, and then we review if it worked.
1:38:35If not, then we look back. And we do this for the other tasks as well. And the reason this works a lot better is because when we have to fix task four, we don't have to fix the whole thing because we reviewed each one and it worked very, well.
1:38:45And so we go task by task, making sure that everything works until the end. Now is it gonna take more time to prompt ClawCode at the start? Yes.
1:38:51But, actually, you spend more time fixing this after it didn't do a good job rather than setting things up properly from from the start. So to make this more practical, I can give you, an actual prompt, uh, that I built so you can see the bad approach and then the good approach that we take. So on the one hand, we have a prompt that says, build me a full customer support chatbot with Next.
1:39:06Js front end, a knowledge base that uses embeddings, escalation logic that routes tickets to Slack, an admin dashboard analytics, user authentication with OAuth, and deploy the whole thing onto Vercel. Use Tailwind for styling and make it responsive. Now I'm not gonna run this for the sake of not having to wait three hours, but this right here, it will overwhelm Clockcode with so much context and so many things to do that the output, don't get me wrong, it will be okay, but it won't be as good as if you split things up step by step.
1:39:31Right? And so what we could do here is we have the first prompt, which would be this, let's say.
1:39:37Prompt one. Set up the next JS project with the TypeScript and Tailwind, create the basic folder structure with components, and we give it the first part. So when prompt one is finished, it did a good job.
1:39:45Then we go to prompt two right here. Then once prompt two is finished, so this is added in knowledge base, it goes to prompt three. And then we keep going until prompt four and prompt five.
1:39:54Right? So prompt four will be this. So all we did here is took this prompt right here, which has Next.
1:39:59Js front end and knowledge base, escalation logic. It has admin dashboard. It has OAuth.
1:40:03It has deploying. Right? So Next.
1:40:05Js front end will go here. Then we had knowledge base will go here. Then we had escalation logic will go here.
1:40:10Then we had dashboard, which will go here. And then finally, would have another one which is deploying the whole thing. And so you can see that it's it's kind of like a spider web.
1:40:17And the reason why we do this is because you take a more structured approach. Right? And so I'd walk you through how you review it and so on, but essentially, you would run this.
1:40:23Right? And don't worry about all these terms because they don't mean much to me either, but it's more the structure that we have to understand. Right?
1:40:28And so you have something simple with seven different tasks. You split them up into different prompt one, prompt two, until prompt seven. You review them individually to make sure that when you get the output, it's gonna be correct.
1:40:37You don't have to re prompt Claude all over again and fix it and debug it and spend two, three hours asking you why I did the wrong thing. The next hack is using plan mode. Now I know plan mode is not something revolutionary, but it's a big misconception as to what it actually does and when it's important.
1:40:49So whenever we have a new task in Clockcode, we ask ourselves whether it's hard or not hard. If it's hard, so if it requires different steps, then we use plan mode, then we review the plan, and then we approve it, and then we execute the actual thing to give us the output that we Right? An example of this could be an API with five endpoints or when you're building a full workflow, and I'll show you exactly what that looks like.
1:41:05Because it has different steps to it, and you want it to plan accordingly, take the time to show it to think through how it's gonna build it before it builds it. On the other hand, if you have something very easy like fixing a typo or changing some sort of color in a button that it made in a software, then you can just execute directly without needing the plan mode because there's not much planning to it.
1:41:20It's just a very, very easy task. Alright. So here I'm on my main hub in Claude code, which is called the YouTube workspace.
1:41:25Um, I automate pretty much 30% of everything that I do on a day to day basis with, like, YouTube. I would even go as far as saying like 40%. Now here under dot Claude, and by the way if you're new to Claude, if you have no idea what I'm talking about, if this looks very confusing to you, watch this video up here, it will give you a much better idea.
1:41:39But if you understand skills, which I'll walk you through in the last hack as well, Excali draw diagram. So that skill right there allows me to be able to make these diagrams automatically. To break down the topic for you guys to understand, I typically break it down in a visual way but also make it practical.
1:41:52So you see how I'm showing you the diagram plus also how you can use the actual concept here. And so in this case, let's say I wanted to make a video about antigravity versus clock code. By the way, if you want the video, let me know in the comments down below if you're interested.
1:42:03If not, also let me know. But if I wanted to do a video on antigravity versus claw code, well, I would want to basically, first of all, explain what antigravity is and how claw code is different from antigravity.
1:42:14And so I would want to make maybe three diagrams, I'd say, using Excalidraw, which is this platform right here, which allows me to explain my concept, the differentiation between anti gravity and claw code. Okay.
1:42:23So I told it to research anti gravity. I told it to research claw code. I wanted to look for the differences.
1:42:28I wanted to make three visuals and then give me back the visuals. Okay? So I passed like this as a plan.
1:42:32Right? And so instead of going bypass permissions, is basically where we go YOLO mode, we say, hey, take care of it yourself. Don't ask me for any help.
1:42:39Don't make any plan. Just go after it. Right?
1:42:41We can go and use plan mode to make a plan first before it gives us the output. K.
1:42:47So now I'm gonna press run, and you know it's in plan mode when it's blue right here. Bypass permissions is gonna be red. Now it's saying that it's gonna research both products in parallel to understand what we're working with, then plan the diagrams.
1:42:57K? So that's how it plans through. That's how it thinks through actually planning the the whole concept here.
1:43:00K. There you go. So after a few minutes, it made the plan, which is Google's anti gravity versus claw code three visual comparison diagrams.
1:43:06It gives me the context. So Michele wants to create x y z. Research summary, the so summary that I made.
1:43:11Three key differentiators. So interface, agent representation, verification, feedback, and deliverables, which is the three things that we want. Right?
1:43:17With color system, the layout, diagram one, diagram two, diagram three, diagram four, and the implementation steps, and the files verification. One more reason why I do also like this in specific is because when I use plan mode, I can also add comments. So I can say interface paradigm.
1:43:30Paradigm is a very strange word. Make it beginner friendly. I know it's a very stupid comment, but let's say you add the comment.
1:43:35When we press approve plan, then we will approve the plan but it will also take the comments that we add in the plan as context to change to whatever it is that we told it to change. But also I get to see exactly what it came up with because if I didn't like interface paradigm or if I didn't like agent orchestration I could say hey Can you find something else?
1:43:50Right? And we can go back and forth. Now let's say I I like the plan.
1:43:53Everything's good. Now I can press send feedback and keep planning. So send feedback because we we added one comment and if you add more comments so if you say here, let's say orchestration, you add turn this to organization.
1:44:02Turn this word. So we're just playing around with the names. I can go here and I can send feedback and keep planning.
1:44:09Or, actually, now that I thought of it, color system, I know anti gravity is not purple, so we can do use dark blue. Cloud code will be not blue.
1:44:19It will be, uh, use orange.
1:44:22That's it. Now I can go back. It will say we added four comments.
1:44:25It will take this four comments, and we can send feedback and keep planning. Okay? So now it will take the feedback, it will apply the feedback, and then give us back another plan with the updated feedback.
1:44:33There's a lot of feedback there, but you get the point here. It's basically crafting the whole plan before it gives it to us. Alright.
1:44:38So here we have the new updated plan. If I go back here, we can see that we have dark blue, orange, we have interface style, and we have agent organization. So these things changed.
1:44:45It updated the actual plan. And now I can say yes and auto accept. And now instead of making it edit automatically, I can now put bypass permissions because I know that the plan is good.
1:44:54I know that it's smart enough to be able to do whatever I wanted to do, and now it's just about doing it. Alright. So you just finished making the three diagrams.
1:44:59Now let's look at the output. So I can we look at the first one. I can go here.
1:45:03I can import scenes. So anti gravity, I believe this is an interface. Let's open it up.
1:45:08There we go. Nuts. Crazy.
1:45:10Insane. So here we have the actual diagram that I made. Google anti gravity and clock code.
1:45:14This is the diagram itself to show the the differentiation in the interface. The next one is agent organization, which is this one right here, which is on the left hand side is anti gravity. On the right hand side is clock code showing the difference.
1:45:25And finally, we have the agent verification, which is this one right here. So the verification and how they both work. Now the output was significantly better because we took the time to plan things ahead.
1:45:34On the other side, you could have just used bypass permissions, and it would have given you something, and I've tried it before. And the output just wouldn't be as good as you'd want it to be. So if a task requires more thinking, if it requires a plan, just make sure to use plan mode every single time.
1:45:47But the difference between a kitchen sink versus a clean session is the amount of tokens that it uses. So if you didn't know, a clock with session, right, one session means one kind of conversation here figure here I just press this it will compact it. I'll explain what that means but essentially each conversation has x amount of tokens.
1:46:02So if I go here and I put context, we get the context usage Right? So we get how many tokens have we used so far.
1:46:08So we have a limit of 200,000 tokens, and we have used about 83,900 tokens in one session. So the whole session that we just did, it used about 83,900, which is about 42% of each one.
1:46:19And it also gives us the actual, like, percentage broken down into, like, what actually takes the most amount of tokens. So the system message is 3.8. The system tools is 9.3.
1:46:27The MCP is 5.7. The system tools is 10.8 and so on. Right?
1:46:30And so you have all the tools broken down and you can see everything here. My point is that your prompts are better if you separate them in different sessions. Right?
1:46:37So instead of fixing a bug, making a new feature, refactoring something, or writing a doc all in the same session, right, even though they have nothing to do with each other, you would create a session for each one. And when I say session, I mean a new session here. So you press plus.
1:46:48This is a session. Conversation. You can call it conversation really, not a session.
1:46:51So conversation one, conversation two. Because the more tokens you use in a session, the more stupid the clock code becomes. The more it starts to think less and the more it starts to get overwhelmed with information.
1:47:00And so split them up in different conversations, and then you can start over. Now I call this prompt hack because sometimes you think that the problem is your prompt, but, really, it's the session tokens that you're using. So right here, we basically made three diagrams for a video that I'm planning to make.
1:47:12Now we're using about 42% of the actual usage. If I keep going with this conversation, we we could do that. Right?
1:47:17But the level of intelligence of Cloud Code will decrease the more you increase the tokens in that conversation. So if I wanted to deploy another workflow, let's say, analysis.
1:47:26It goes to my competitors. It goes through all their comments, see what people want, and it gives them back a report. Is it better for me to keep going on the session, or is it better for me to go to a new session and start from there?
1:47:35Well, obviously, it's better to go to a new session and start from there because you start from scratch. And if I go here and I put context, then this will now give me the context, and we're only using about 9% token usage.
1:47:44So now it has a fresh mind, and it can start doing the task that you wanted to do without being overwhelmed with all the information from the previous conversation that we had, did which something completely different. The next hack is using Claude. Md.
1:47:53So let's say you wanted to make an app, some sort of software app. Okay? And you wanted the app to use TypeScript.
1:47:58You wanted all the words to follow CamelCase. You wanted to use Tailwind. You wanted to use Supabase, and you wanted to use an app router, not pages.
1:48:06Now don't worry if you're not really understanding what all these words mean, but understand that these are all rules. Right? And so these rules, you can either tell Cloud Code every single time to do all these things or you can add them to something that we call a claud.md document where you basically build a one time prompt and it will auto load that prompt every single time in Claude so it has context, and we don't have to repeat everything over and over again.
1:48:25So Claude already knows. You just ask. So if you had to make a PDF document and we wanted the color to be blue, we wanted the the letters to be small, we wanted the, I guess, the the font to be italic font.
1:48:35Right? We can either tell it individually every single time, or we can add it to a prompt like this, and then it will use that every single time without us having to repeat ourselves. Now lucky for you, I made a skill in ClawCode called the Claw dot m d generator, and this includes a pull prompt on how to make the best Claude dot document, plus it gives you the example structure of one of them, for example, here.
1:48:53Right? So all you have to do is download the skill, which you can find in the second link down below. And here we can say, hey.
1:48:58I wanna create a claude.md prompt for my project, for my app that I'm building. I just want you to ask me any questions of what you need from me to be able to make this document, so we can press go.
1:49:06So now what it's gonna do is it's gonna look at the skill of all the requirements that it needs to be able to make the Claude. Md document, and then we're gonna start working on it. So as you can see here, it says, I have found the Claude.
1:49:14Md generator skill. Now let's walk through the process. So now it's gonna ask me where you should save the Claude dot m d file.
1:49:19I can press, uh, recommended. It's fine. And then tell me about your project.
1:49:22So what is the product name? What does it do? And who is it for?
1:49:24I can press other, and I can start talking to it. So let's say here I say, okay. So I wanna build a website.
1:49:28So this is for a website, and the website needs to be for my company, JM Solutions. We're an AI automation agency that helps marketing agencies to be able to build their back end delivery systems. And, uh, yeah, that's what that's what the actual thing is for.
1:49:41So you give context, and then the text tag we'll use is, I'm not too sure. Whatever you think is the best for build a website, let's just do that. It can submit the answer, and now it will start thinking about what to put in the cloud.
1:49:51Document. Okay. So it's not jam solutions.
1:49:52It's jam. But now when it makes the cloud. Md document, it will be an overview of what it is that we want to build.
1:49:58For your jam solutions website, what kind of pages features do you want? Landing plus services, keep it simple.
1:50:05Keep it simple. Actually, for conventions actually, let's keep it simple.
1:50:09And rules is gonna be any hard rules cloud must always or never do in this project. I have specific rules. There you go.
1:50:14And now it's most likely gonna ask me for rules. Because all these rules, I don't wanna repeat myself every single time I use Claude. I want Claude to use the Claude and the document when it's actually building the thing that I wanted to build.
1:50:25What are your hard rules? Mobile versus Zion. No inlay styles.
1:50:28There we go. Submit answers. And also other.
1:50:31My company name is JM Solutions, not JAM. Because here we have a plan to create the claud.md document.
1:50:38We have the context, project with the James Solutions, AI automation agency website, helping marketing agencies to build back end delivery systems. The tech stack is gonna be this, pages, conventions, file structure, rules, actions, verification, and we're good to go.
1:50:51Yes. And auto accept. Alright.
1:50:53So here it made the cloud dot md document. If I open it up, I can see that we have a full prompt, which is the one that we had before. So these are information, basically a summary of what it is that we wanna build.
1:51:03And this is great for us because every single time that we have to go out of Cloud Code and come back in, we don't have to to ClawCode what we want. We can just say it knows that every single time that we start talking to it, it will always look at the clawcode.md document. It will get context, and then only it will respond to us.
1:51:17Sort of like an employee having context of what it is that it has to do every day. I mean, imagine you give the AI the task to build a website. You give it all the different features and what it needs to do.
1:51:24Then you go and close ClawCode. You come back to it later, and then you tell it, hey. Keep going.
1:51:29And then it's gonna ask you, keep going with what? What it is you want me to build? And you have to re explain everything again.
1:51:33That's what this prompt is for. The last prompt hack is using skills. So we call it reusable prompts.
1:51:38So initially, we had the manual way, which is same effort every time. So we had session one where we write a full prompt. Session two, we write another full prompt, and we keep going session three, session four, and session five.
1:51:47And each session will take about five minutes, and we take about twenty five minutes that we wasted writing every single prompt manually. The second way is using skills. So we write the skill once, and we use it forever.
1:51:57So if you remember before, I showed you how I made a Excalidraw diagram using what we call skills. The skills are preset instructions that we gave ClawCode to perform a specific thing. And so all we have to do is create a skill one time, which will take about five to ten minutes, and we also have a skill to create the skill, which is which is awesome.
1:52:13And so whenever you wanna do a specific thing, you can always create a skill and then call the skill whenever you want to do that specific thing whenever you want. So And let's say we had the Excalidar diagram thing. So inside here, we have a prompt.
1:52:23This is a prompt that explains how to make the Excalidar diagrams. So this is the full prompt.
1:52:28Now I could either copy and paste this prompt every single time into ClawCode into a new session, or I could always just save it here under the skills, and it could always read this when it needs to make the actual thing. Now this is great for us because it saves me so much time whenever I wanna make any of these, to be honest.
1:52:44So for example, if I go back here and I go to the command, I can see here that I can call the CloudMD generator. I can call the comment analysis.
1:52:50I can call the comment analysis report, competitor analysis, which are all skills which perform different things. Now I've covered skills step by step in my Clawcode tutorial. So if this is the first time that you hear about this, just check out this video up here.
1:53:01It'll give you much more context. But if you already know what skills are, then let's actually build one step by step and show you how it works. Let's say I wanted to go on YouTube every day and check the amount of views that I have for the videos.
1:53:10Now I could either prompt Koloko to do this every day or whenever I want to. And by the way, before we build the skills, you can find all of them at aitmpl.com. These are preset skills where we have the skill creator, which is a skill that makes skills.
1:53:23We have the senior front end. We have the UI design. We have all these different skills that do different things.
1:53:27But sometimes I wanna make my own skill, which is my own kind of task. And so here I can say, hey. I want to be able to make a skill and the skill needs to be a skill that goes on the Internet every time and it finds the top news or top features that came out about ClawCode.
1:53:42Then it goes on YouTube and it checks the highest viewed video on ClawCode in the moment, and then it gives me content ideas that I could be using for my channel for ClawCode. Make a plan for it first, and then we can, uh, go ahead and make a skill. So this is the initial thing that we give it.
1:53:56I can just press go, and now it will start making the plan for the skill that it needs to do. Alright. So as you can see here, it made the plan for the skill that it's gonna make, and it's gonna be called ClawCode content ideas.
1:54:04So what it does is that it search the Internet for latest ClawCode news features. It searches YouTube for the highest viewed recent ClawCode videos. It then generates 10 to 15 ranked content ideas across reference against Michaelis content pillars, and then it outputs a markdown report plus style PDF to outputs.
1:54:18So now it's gonna be making all the skills for that specific thing, and then we can see phase one, phase two, phase three, phase four, phase five, phase six, and that's it. So from a single prompt, it made this full plan of execution that we can now simply press yes, no, or accept, and I'm gonna press bypass permissions so it does everything on its own.
1:54:36So as you can see here, it's going through the to dos, tasks that it needs to make, and then it's adding the folder called Cloud Code Content Ideas, which is a folder where it's gonna store our scale that we can always call when we want to to use it. Alright. So as you can see here, if finished making the scale, if I go to Cloud Code Content Ideas, have the references.
1:54:52So this will be the report in PDF that it needs to give us, uh, which is the output. Then the scripts, which is the actual Python scripts. It's kind of the the automation that it needs to execute.
1:55:01Generate PDF and search YouTube videos, and then it has the skill dot m d, which is the main document that explains to it exactly what we want to do. So with that said, I can go here, and I can say slash Claude content ideas.
1:55:16So it also made a command, and the command allows us to be able to to just find it easily. So I can press Claude code content ideas. I can put this in plant mode, and now it will start looking at all the different skills here, and it will go through each skill dot m d right here, and it will know that it needs to use the Clog code content ideas because it's associated with that command in specific.
1:55:34And so what it will do here is it will start calling the Clog code content ideas, and it will start executing everything that it needs to do. It says I have all the context needed. Let me write the plan and exit plan mode so we can execute.
1:55:43Alright. It made the plan of what it needs to do, all the steps, the key files, verification. We can go and press yes and auto accept, and we can put this on bypass permissions.
1:55:52So the idea behind this is that whenever now we have to go back to Cloud Code, instead of prompting it again saying, hey. Can you go on the web? Can Can you search this?
1:55:58Can you search that? We can simply just press slash and call the Cloud Code content ideas, and it will execute that thing that we made previously, which, again, will take us ten minutes at the start, but it will take us three seconds to be able to set this up or to call the tool, uh, whenever we have to run it. Alright.
1:56:12So now it gave us the top five content ideas. It finished running the actual thing, agent teams for agency operations, Cloud Code HTTP hooks, why don't delete my Cloud. M d, skills build a business command center, and Cloud Code for non coders.
1:56:25And if I go to outputs, I can see that it also made, uh, let's go to here. It also made a report in PDF format. And the PDF looks like this.
1:56:32So we have news and updates, the loop command which came out in Cloud Code, the cloud API skill, the effort levels in UltraThink, auto memory, security, agent teams, enterprise AI agent ROI, and we have the top YouTube videos latest by thirty days. So next drive is crushing it. Clockwork.
1:56:47There we go. Dan Martell, Sierra Legend. Brian Cassell, and we have the other ones as well.
1:56:52And it keeps on going and going and going, and we have the patterns observed, content ideas, and so on. So it took us, like, ten minutes to be able to make this whole skill, but now it's gonna save us a lot more time when we have to rerun it again. Because now when I go to a new Claude session, I can simply go and press the Claude code content ideas.
1:57:07And when I press this, it runs it over and over again. So instead of running the prompt over and over again, we have one time, and we save ourselves an eternity every time we have to run the whole thing.
1:57:19Alright. Congrats on finishing module three, which is the prompting guide. Now you know how to actually communicate with ClockCode to make it do things that you want it to do in a very, very effective way.
1:57:29And that matters because every single thing that we're gonna cover from now on requires that skill. So now we're gonna be moving into agentic workflows, which is sub agents, agent teams, workflow hacks, and the deployment side. And this is where you get Clockcode to stop acting like a normal coding agent, but start becoming the system that runs the entire operations for you.
1:57:47And this is the highest leverage thing that you can build with ClawCode right now. So let's dive in. K.
1:57:51So think of ClawCode as the business owner. Right now, what's happening is that you're asking ClawCode to do everything research, write, test, organize, right? And that's like running a one person business forever and so you can think of sub agents like hiring different employees.
1:58:06Each one has their own brain, each one has their own job description, their own tools and each one reports back to the main boss which is Claude in this case. Right? And so what happens here is that you tell Claude a task, you give it a task of what it has to do, and then Claude delegates which means that it gives the work to the right employee, to the right agent in this case, And the employee works in their own space and then hands back a clear summary to the actual agent which then gives it back to you.
1:58:31So you can think of this as a research assistant, a copywriter, and a QA employee. And so these are three different employees which are able for us to split up the tasks across different people so that each one is specialized in their own thing, and then they all report back to the main agent here. So now that you know exactly what it is, how do you actually create sub agents?
1:58:48So there's the first way which is using ClawCode the second way, which is doing it manually, and we're gonna look at both ways. Okay. So here I am on Cloud Code.
1:58:55I made a new folder called sub agents, and for us to be able to create the sub agent, I can type slash agents. Okay?
1:59:02Now if you're on this view right here, you will then see a symbol which is this one right here. This symbolizes that to run this exact command to mix up agents, you have to run it through your terminal. And to run it through your terminal, got it here, you can go up terminal, You can press new terminal, and then you will be able to see the terminal here.
1:59:20And I can press Claude to initiate Claude to actually make Claude work, and then I'll have this screen right here. Let me actually go maxed out, and then I have this page right here. So now we're gonna be working with this page specifically to make the actual sub agent.
1:59:33Okay? Now that we're here, we can press slash agents even though the spelling's wrong. We can press enter, and now it's telling us, hey, agents.
1:59:40Do you wanna create a new agent? I can press enter again, which means yes. It's asking me the location.
1:59:44So project or personal, I can do project. That's fine.
1:59:48So now it's asking me how I want to create the agent. In this case, I'm gonna do generated with Claude, and now I have to explain what it is that I want. For the sake of just showing you how it actually creates the agent, I wanna keep it very simple.
1:59:58I want a sub agent that Can review or that can take an idea of a of a YouTube video, and then it can spin up 10 different YouTube ideas, give me the 10 ideas plus hooks that I can use for that specific video.
2:00:11Now I'm using just a very simple content creation kind of use case, but at the end of the video, I'll also show you three different use cases where you can actually apply this inside a real business. Right? Now that we have this here, can press enter, and now it's gonna generate the agent from description.
2:00:24And we should be able to see the files and the folder being created here, which is basically how it stores the actual agent itself. Okay. So now it's asking me to select the tools, and I think of this as like the level of accessibility that you give to the agent itself.
2:00:36Like, for example, do you want to give it access to all the tools? Read only, uh, edit tools, execution tools, other tools. I'm gonna press continue because I can get pretty much access to everything.
2:00:46It's gonna ask me to choose the model that we're gonna use to be able to run a sub agent, and I'll walk you through, like, the model and which one should you choose in the first place. I'm gonna go for Sonnet because it says balanced performance best for most agents at Cisco. And then it's asking me for a color, so let's make this nice and pink.
2:01:02Then it's asking me where do I want to configure agent memory. Uh, keep it very, very simple. Use product scope because then it's gonna add it to your actual project.
2:01:09And then now you can see that it made a folder called doc log, and then I can press s to save. If I go here, I can see that it now made the agent memory, which doesn't really have much in it, and then it made the agent itself. So the agent itself, just so you know, each sub agent, all it is is a prompt.
2:01:26Right? The same way that you know skills, sub agents are just prompts as well.
2:01:30And the prompts is a list of instructions of what the agent should actually do. Okay. So now that we've made a sub agent step by step, let's go through the structure of the actual sub agent.
2:01:38K. I'm gonna press x here. I'm gonna expand this.
2:01:40And here you can see you have the name, description, model, color, memory, and then we have a whole description of the actual sub agent and what it needs to do. At this point, you have it ready, go to your chat window and then you can press file, you can press save or you can press command s and it's saved to your computer as well.
2:01:56And so what happens is that you'll start to have more and more agents which is more and more prompts, right, and so you can run agents simultaneously if you want. Alright, so now you know how to make it automatically using Cloud but just like everything, if you don't really understand the fundamentals of how it all works then if you're stuck on something or if Cloud doesn't do the right job, you won't know because you don't know what good looks like.
2:02:15So for us to create the sub agents, we first have to create a folder called the dot Claude folder. Now in case you haven't watched the skills video, check it out up here, but I use that exact folder itself to store skills as well. And so you have agents, you have skills, you have agent memory, you have commands, you have all these things in the same exact folder.
2:02:32And then we have to make a folder called agents, make a folder called agent memory, which is the memory of the actual agent, and then we have the actual markdown file, which is dot m d because it's the way that it's written like this with hashtags, with asterisks, uh, to organize the text, which is the instructions of the actual agent.
2:02:49So I deleted the previous sub agent that we had. First things first to make our own sub agent is make a folder called the dot Claude folder. And inside this folder, we can make another folder called agent slash memory.
2:03:01Boom. And then inside the dot Claude folder as well, we can put the agents. And so this looks like the exact something that we had before.
2:03:08And then inside the agent here, we can then press new file and call this name of the actual agent. So let's say we're doing email writer dot m d.
2:03:18K? And then you have the actual mark on file where you'll be actually writing these instructions. Now the instructions for a sub agent prompt are divided into different parts.
2:03:28We have the name. We have the description. We have the tools, the model, and then we have the actual instructions here.
2:03:33First things first, we have the name, which is email writer in this case. It's what you call this agent. It's how you will be able to refer to it later.
2:03:40K. So remember the name. Then we have the description.
2:03:42So in this case, write professional client emails based on the context you give it. You tell Cloud when to use this agent. Cloud reads this to decide if it's the right fit, and the description is the most important part because you tell Cloud when to use this agent.
2:03:54So when you tell her, hey. I wanna do x y z, Cloud will read this exact description to decide if this is the right agent for your request. Then we have tools.
2:04:02In this case, we just wrote read or write. The tools are the permissions. Right?
2:04:05So what this agent is allowed to do? Well, it's like onboarding someone into your own project management system but not giving them access to everything. Then we have the model, which is in this case Sonnet, which is the brain that we're gonna use.
2:04:15So which clock of brain do you wanna use for this exact use case? Now bear in mind that we have three different models. We have HAYKU, which is fast and cheap.
2:04:22We have Sonnet, which is why it's at balance recommended. Most agents use this. And then we have Opus, which is the most powerful.
2:04:27Bear in mind that the most powerful they are, the more costly they are. PayQuery is very, very fast, but it's only used for things like reading files, searching code, or quick lookups. So looking for different things that don't necessarily require a lot of brainpower.
2:04:39Then we have Sonnet, which is typically good for most things. It's medium cost, so it's cost effective, hence why it's we say balanced, but it's good for writing code, code reviews, and most of the daily tasks that you do out there, like writing emails or making content. And then we have Opus, which is the most powerful.
2:04:53You're rarely gonna use this, but you would use this if you're doing very hard debugging, if you're making some sort of full stack application and you need a agent that can go there and debug, meaning fix the code that you're writing, or if it needs really, really complex reasoning for it.
2:05:07Right? So, basically, the overall summary here is use Heiku for simple tasks, Sonnet for most work, and Opus only when you need the big brain. And once we're done with the model, then we have the instructions.
2:05:16So this is the rules and personality and guidelines of the actual sub agent. So for example, we have, uh, you are a professional email writer for an AI automation agency. When asked to write an email, you ask who it's for and the goal.
2:05:27You write a clear professional email, and you keep it under 150 words. Match the tone to the situation. Always end with a call to action.
2:05:34So this right here is the personality and step by step instructions. Think of this as an employee handbook. You tell it exactly how to do its job.
2:05:41Right? So these are the instructions that you give it. And so that right there is the structure of the dot m d file, which is the file where we're gonna give everything that we need, uh, to the sub agent, to Claude in this case, to know that, hey.
2:05:52This is the sub agent that we want to run. I'm just gonna paste the example that I'd already showed you. We have the name, description, tools, model, and then we have the instructions here as well.
2:06:00So now we can press command s to save it. We can go to Claude, and I can paste this one for here, which is write a follow-up email to my client, John, about the chatbot project. Now what it should do is it should look at the email writer MD, which is this right here.
2:06:13It will read through the instructions. It will look at what it has to do. In this case, the first thing it should ask me is ask who the email is for and what the goal is.
2:06:20In this case, we already told it who it's for, not entirely because it doesn't have the email. So it should ask us the email itself. And as you can see, it just made the email called email John follow-up dot m d where we have the two, John.
2:06:31And by the way, we could actually make it so that it can send the emails for us. But in this case, I just wanted to to write the email. We have the subject line checking in chatbot project and next steps.
2:06:40Hey, John. Hope you're doing well. I wanted to touch base on the chatbot project and to see how things are progressing on your end.
2:06:45We've made solid progress so far, I wanna make sure that we keep the momentum going. If there's any questions, blockers, or feedback you've been sitting on, now is a great time to surface them. I love to align on the next steps and make sure we're on on track with the timeline.
2:06:54Would you be available for a quick call this week or early next week? Happy to work around your schedule. Just let me know what works best for you.
2:06:59Looking forward to hearing from you. Best, Michele. Yeah.
2:07:01Very, very good email. And so as you can see, we just give it a simple prompt to write a follow-up email to my client, John, about the chatbot project. Then it knew that it had to use the agent email writer dot m d to fulfill the actual task.
2:07:12Now in the grand scheme of things, you will have the output. So some of them are outputs, right, which is this right here, which is the output of the task that we gave it. But sometimes it can get very, very messy.
2:07:20So what we usually do is we make a folder, which is not in the Doc Cloud folder, called outputs. And then we store this right here.
2:07:29Move. So that all the outputs go under outputs folder.
2:07:32It's all to keep things very, very structured. With that said, you got to see how I built it plus how I ran it step by step. Now in terms of triggering, which is, like, how do you start the agent itself, you saw me talking to the clogged agent and not really referencing the name of the actual agent itself, but that doesn't work too well when it comes to you having skills that are very, very similar.
2:07:49And so the main ways that we actually run the actual agent is three different ways. We had the natural language, which is similar to what I just did, but in this case, we ask it to use the code reviewer to check my changes, Use the name of the actual agent to do x y z to run the task. So Cloud figures out which agent to use, and then it runs a task.
2:08:07The second and the most common one is mentioning. So you're gonna use at, the name of the actual sub agent, to tell it to initiate x y z or to do whatever it has to do. Right?
2:08:16Create the task. Now this is guaranteed to trigger because you're referencing that specific agent only. Right?
2:08:22And then finally, we have session wide which is use one agent for everything. So use claud dash slash agent code reviewer. So my advice is to start with the first one, natural language because that is the easiest to start with and then move on to the second one mentioning.
2:08:34Right? The only reason for that is because most times you guys are not gonna have 35 different sub agents that you have to match. Right?
2:08:41And so you wanna keep it as simple as possible. So right here, to reference a sub agent, we can just say, can you make an email using the email writer agent to Jessica telling her we have dinner tomorrow? And then you put the actual name.
2:08:52Right? You can also not put the stash itself. It's just something that I like to do to keep it more precise.
2:08:57And then this right here will know that it needs to use the agent itself. It's gonna look through the agents here. Of course, there's only one.
2:09:04Right? And it drafted the email itself. The second way, like I mentioned, is you go to tag and you can see the different options.
2:09:10And you would use the email writer dot m d. Then you can say, can you make an email for James and Jack telling them thank you for the business meeting that we had before and tell them that the next steps is to have an onboarding call tomorrow 5PM EST to go through the next steps. And so because I was very specific and turned this off so you know that you're not reading it and not cheating, but we were exactly referencing the agent that we wanna use.
2:09:31And so there's no way that it will fail in using the agent because we give it the exact instructions to use that agent. And as you can see, it made the email itself, James and Jack, saying great meeting onboarding call tomorrow at five b s t, and then it made the whole thing. Right?
2:09:44Obviously, you could change this by putting your name. I didn't really give it the main folder, which is a claud dot m v folder just because, uh, I didn't wanna spend too much time on that, but you can give it more information as well. Now in case that you are using the terminal to run claud, you can simply go to slash agents again, and then you'll be able to see the email writer agent, which is this one right here.
2:10:00And you can simply press enter again. You can view, edit, delete agent. You can back agent, and you can do all sorts of things here.
2:10:06And, again, you can simply just give it natural language or you can press at email writer agent, and you can tell it whatever you want. Make an email to James tomorrow, and you can run this.
2:10:16And it pretty much does the exact same thing. I just wanna show both things because I know some people get honestly, some people, I mean, me at the start. I got a bit confused when it came to using the terminal versus using the normal Cloud chat and how those actually work.
2:10:26So as you can see now, one thing we didn't have before, I told Claude that if I don't tell her who it specifically I have to send the email to and what the goal of the email is, then it should ask me. And so here it's saying, who is James and what's the purpose of the email? James is a client and the purpose of the email is follow-up after a meeting.
2:10:40Whisper flow has changed my life, guys. You should definitely try it out. As you can see, have the full email, James, subject, and the body itself.
2:10:46Now one more thing that I spoke about before is restrictions and safety. Because the thing about sub agents is like they get really, really powerful because you can control what they can and cannot do. The same way that you wouldn't give an intern access to the whole company database, here's the exact same thing.
2:10:59And so we have the read only agent, we have the balance agent, and we have the full access agent. Now when you said read only, it can look at your files, it can search through the code, it can explore, uh, kinda what you have, but it cannot edit or write. Right?
2:11:11It can look but can't touch. It's perfect for research that you have to do without it actually doing something on your documents or changing anything in your computer. Then we have the balance agent.
2:11:19It can read search. It can also edit and run commands, but it cannot delete files. Right?
2:11:24It can build things but can't break things. It can only add but it can't remove, which is great because a lot of times when you have an agent that's out of hand that's going everywhere, it can delete things that later on you can't get back. Right?
2:11:34And so you wanna keep it normal and clear. And then we have the full access agent. It can read, edit, write.
2:11:39So full power only uses when you really trust the task you're doing. Typically, it's for tasks that are easy, that don't require a lot of brainpower, that are very, very structured. You have a step by step.
2:11:48It's done in this way, and it can be done pretty fast. So here you decide what each agent can do. The less access the safer it is, the more access, the more powerful, but the riskier it is for you as the person who's actually talking to the agents.
2:11:58And the way that you would change the actual permission is if I go here to the MD, you would have tools. Right? The tools is the permission.
2:12:04So you can literally think of this as if the word's permissions, but we use tools. Right? Now if you only want this to read, you can copy this.
2:12:13Read grep and glob. That's it. Now if you want a balanced, which is the most common, then you want this here.
2:12:19Read, edit, bash, grep, and glob. It's also gonna give you suggestions, but I'm gonna keep it normal for now. And then if you want the full access, you don't have to specify tools.
2:12:28Inherits everything. So if this is empty, then it will use any tool that it has access to. It's only if you specify the level of permission that you want the sub agent to have that it actually uses it per se.
2:12:38Now when it comes to sub agents, there's one issue, which is the issue of memory. Because every single time that you have a new session with ClawCode, so a new conversation, these sub agents start from scratch, which means that they don't have a previous memory of the conversation that you had before. Now as you can see here, we have a folder called agent memory, and this is where this comes into play because now we're able to actually have the sub agent remember the previous conversations, and we don't have to re explain things again.
2:13:02So all we have to do is go to the actual agent itself, the prompt, and then we can simply write below model and above this want it to include. But all you have to do here is put memory colon project. Okay?
2:13:12And you press save. So let me show you how it actually works here. So if I go to a new tab, of course, again, save the project so it's here.
2:13:18So I can go to a new session, and I can say email writer dot m d. That's fine. I can paste this right here which says make an email for John seeing that we have dinner tonight or tomorrow night at 5PM and make it really, really, really, really, really happy with an exclamation mark.
2:13:32Also, the tone to the memory. I can press go.
2:13:38And so now what it does is gonna launch the email, right, to actually make the email that we asked it. But it's also gonna save the tone preference in parallel so at the same time. Write the email and save the tone into the memory itself.
2:13:47So you can see now it's starting to work with memory dot m d. It's gonna save the tone preference to the memory itself, which is this folder right here. I can press accept.
2:13:54That's fine. Accept again. So as you can see here, it made the email dinner email dot m d with all the exclamation marks.
2:13:59It's on a voice. And then we have the agent memory. The agent memory has two different things.
2:14:03Has feedback tone matching dot m d, and it has memory dot m d, which is the main one. So feedback tone matching means this is the, uh, explanation of the actual tone of the email that we need to use so that we don't have to re explain to it again to use exclamation mark, to use x y and z.
2:14:18This list of tone for emails is gonna be used again and again when we're making new emails. Right? And then we have memory dot m d which has a memory index which means that every single time that we give feedback to the actual agent itself, it's gonna store the feedback here.
2:14:31And so we'll have more and more feedback so that it remembers the previous things and it gets better over time. Self improving sub agents. Now that we have the memory and the prompt itself, I can go here and I can say I can reference, uh, the email writer dot m d.
2:14:43Is it this? Yep. And I can say, can you send an email to Jack saying that we have football tomorrow at 5PM?
2:14:51Bring the boys over as well so all of them can have a good time and, uh, yeah, make it exciting. So as you can see now, as it's running, it's gonna read the memory dot m d file, and it's gonna do that every single time because it's with this specific agent itself. The more memories you have, so the more self improving the system will become over time.
2:15:06Right? And the more effective and the accurate the results are when it comes to doing the exact task. It's also looking at the feedback tone matching dot m d, and then we have the actual email itself with exclamation mark here, here, here, kinda looking at the matching here as well.
2:15:22Alright. So you can now spin up sub agents to handle specific tasks in parallel working at the same time. But what happens when you need multiple agents working together on a bigger project?
2:15:32That'd coordinating, sharing context, dividing the tasks. Right?
2:15:36That's exactly what agent teams is all about. And the way that I like to think about it is that agent teams is like sub agents on steroids. So let's dive in.
2:15:43Alright. So here I broke down Claw Chat versus sub agents versus agent teams, what they all are and how they're different. Alright.
2:15:49So the first one here is Claw Chat. This is you talking to Claw directly with no tools, no code, just a conversation, simple back and forth, and nothing in between. And typically, you use Clawd Chat only when you have a question or want some help writing something, just a normal conversation.
2:16:00Maybe you can write your emails. An example of this could be that you're building automation in, let's say, n a 10. You have an error.
2:16:06You drop the error message into Clawd Code and say, hey. Can you explain this error message? Or can you help me write my cover letter, brainstorm video ideas?
2:16:13Basically, the way that 99% of people are using AI right now, it's using Clockchat. Then we have sub agents, and sub agents just means that you have a boss, and the boss has three employees, but this is a bot, the employees cannot talk to each other. Okay?
2:16:25So if employee a is making something that then depends on employee b, well, that's a problem because they can't speak to each other and they can't communicate, right, when making a task together, which could decrease the quality of output and it just wouldn't work well. The results can only return to parent agent only and they have a shared context window so they have a shared context window which means that they all operate under the same window the same kind of tab in Cloud Code.
2:16:46And when do we use sub agents? Well sub agents are used when you're building code or you need Cloud to research explore or plan different things in parallel at the same time. This could be add authentication, write tests, right, where you have two different things which are not dependent on each other running at the same time or explore this code base for me, plan this feature then build it.
2:17:03And when you're building with sub agents you ideally want to have two different tasks or three or four or five whatever it is that you want, uh, but make sure that these tasks do not correlate with each other. So for example, let's say you have a YouTube transcript and you wanna repurpose that YouTube transcript into a LinkedIn post, a blog post, and maybe a Twitter post.
2:17:19Right? All these three things, you might think they're not dependent on each other but they are because the style of writing, the tone of voice has to be the same across the different content that you make. And so with sub agents, you're not able to have the LinkedIn post talk to the blog post, talk to the Twitter post to make sure that they all have the same exact tone of voice.
2:17:38And then we have the one and only agent teams. Okay? So the way the agent teams works is that you have a team lead.
2:17:44Do you give team leader tasks? Say, hey. Can you make this team of people that can do x, y, and z?
2:17:49Right? What this does is that it sponsor teams, so it creates a team and assigns tasks across the different teammates.
2:17:55And just like sub agents, it will start creating agent one, agent two, and agent three, and these agents all have the same shared task list, but the agents communicate with each other and each one has its own context. So it has its own context window and operates under its own kind of environment. Right?
2:18:10And so you have teammate a talking to teammate b, talking to teammate c, which can all see the same tasks that they have here, and the team leader is the one that kinda operates the whole the whole team. And this would be like an actual company. Right?
2:18:21All the employees are talking to each other, same departments, maybe different departments, but they can still speak to each other, correlate, and communicate with the head of department kinda in that in that role. Now you only use agent teams really when you need multiple cloud working on separate repos or separate kind of instances or when you have a big tasks.
2:18:36Right? Usually, it's development based, so it's more technical. For example, you can build a front end plus back end plus tests.
2:18:42Front end does depend on the back end, and it does depend on the tests. So you wanna have two different agents under the same team that are building the exact thing that you want, but bear in mind they can also communicate with each other. So the output will be much higher.
2:18:53Right? Because, obviously, the front end would depend on the back end working, and the test would depend on front end and back end working. So do you want them to communicate with each other at the same time?
2:19:01Or migrate five services at once, but also one cloud orchestrating the rest. Now in terms of the actual build itself, there is a repeat and improve process which is a four step cycle that we go through every single time when it comes to the communication, when it comes to speaking to the agent team or whatever it is that we wanna build.
2:19:14Right? The first thing is collect. So we wanna brainstorm ideas and it's gonna list every tasks and set clear goals just like a head of department would or just like a CEO would.
2:19:22Then we organize. So we rank back priorities, we set deadlines, and we block the time slots for the agents itself. Then we execute.
2:19:28So each agent will work through the tasks. We'll track the progress, and we'll have focus on what we have to do because each agent has its own task that it has to do. And then review, which is checking what's done, um, know what's left, and adjusting the plan in case we need to change.
2:19:40Alright? And then we repeat and improve, and we keep going on and on and on. And that's exactly how the team operates just like you would for a real company.
2:19:46And one more thing before we get to the build is that this is great because we have what we call parallelization advantage because a traditional workflow would have six tasks and let's say they all take one minute Right? So you have six minutes total.
2:19:58When you have an agent team, we have what we call parallel processing. So instead of each task being done individually at different times, here you'd have three different agents all working at the same time on their own tasks.
2:20:09Task one, task three, and task five will be done at the exact same time as well as task two, task four, and task six, which means that instead of spending six minutes or waiting six minutes, then you will wait two minutes because we have what we call parallel execution. And these are all fancy terms just to say that this basically saves us a lot of time.
2:20:23I'm speaking in in like minutes, but sometimes it can be hours that you would wait for a workflow to to work, and here you would literally get it within minutes, which is kind of mind blowing and I think that's where the future of agent workflows are gonna be. It's having one main agent which controls all the others and having them do the task for you without you having to be there.
2:20:41K. So here I am in anti gravity, our ID that we're gonna use. And by the way, anti gravity Versus code, basically the exact same.
2:20:47You can pretty much use Versus code as well, but I just switched onto Gravity because that's what I have here. First thing I'm gonna do here is press open folder. So I typically like to have a Claude code main folder and then have different folders inside.
2:20:57Just something that I do. And I can press new folder here and name this agent, so Claude agent teams, and I can press open.
2:21:05I already downloaded the extension, but you can go here to extensions. You can look for Claude code.
2:21:11Make sure it's the one by Anthropic, which should be the one here, Anthropic. And then you can install it, and then you'll be able to see a little button here which has ClawCode open, and you can press this button. X out of here, x out of here, and you go back to here.
2:21:23K. So here I am on the ClawCode documents. This is the official ClawCode documents which tells us how to basically start the team.
2:21:29I found this by going to code.claw.com/docs/en/agent/teams, and you'll see run agent teams right here. And if I scroll down, you can see when we use it, when we don't use it, how it works.
2:21:41And then here, it says to enable agent teams. So to enable agent teams, we have to make a file called settings dot JSON, and then we have to put this right here inside. I'm gonna go back here.
2:21:50I'm gonna press new file. I'm gonna name this setting is it setting or settings?
2:21:56Settings dot JSON. Settings dot JSON. And you'll see that this will now be turned into JSON because that's the data structure.
2:22:03And if I go back, I can copy this and then copy it back. And I can press file, save. What you could have also done is you could just let you copy this or you can just you go command a, copy the whole thing, and then you can paste it into Clock code and say, hey.
2:22:17Based on this documentation, can you please enable agent teams? And they will do it for you. But I like to be different, so there you have it.
2:22:23Okay. So settings. Json, we have this here.
2:22:25And now this is where we can prompt ClawCode to be able to make the agent. Alright. So I already have my prompt here that I'm gonna use.
2:22:31You can find them in the second link down below in my school community. So what this prompt says is create a multi agent team that turns sales discovery call transcript into useful sales assets. When I drop the transcript into the slash input, the system should generate the client summary, the product proposal, the deal updates, and follow-up emails.
2:22:47So we'll have four different agents. The first agent is call insights agent. So we analyze the transcript and extract the client company summary, core problem, pain points, desire.com's timeline, and the output will be a client summary.md, which will be a file, which will give us a summary.
2:23:00Then we have agent two, which will be a proposal builder agent. So this will generate a proposal. And then we have CRM agent, which will create a structured JSON CRM update, including company contact name problem and a bunch of other things.
2:23:12And here, we'll have the output which will be deal update JSON for the CRM, and then we have the follow-up email agent which will create a follow-up email. So, basically, what I'm telling the agent to do is I am gonna drop a sales discovery call transcript. So let's say I had a call with a client.
2:23:25I dropped the transcript of the call, so what we said on the call. And then the team will be consisted of four different agents, one, two, three, and four. So this agent will make the client summary document.
2:23:34This agent will make the proposal. This agent will make the CRM kind of update, and this agent will make a follow-up email. With that said, I'm gonna copy this, I'm gonna go back here and I'm actually not gonna be using this chat, but I'm gonna be using the terminal.
2:23:47Now I was a bit confused when I saw people using the terminal versus using the chat. The reason why we use the terminal here is because we can see more. So if I press Claude here, my cloud will open up.
2:23:56Like, this tells us everything. This shows all the settings. It shows all the small things.
2:23:59It might look a bit more complex, but it's very, very easy when you start using it and you you get the hang of it. I'm gonna press x. I'm gonna make this full screen, and we can use the terminal.
2:24:09By the way, can use the other one as well. It's just that I can't see all the features that I wanna show you, um, but it still works the exact same. I'm gonna go here, copy this, and then paste the prompt here.
2:24:16I'm gonna press enter, and now Cloud Code is gonna start working on this. Test tool loaded.
2:24:22We have this settings dot JSON, which means that we are enabling agent teams, and so it knows that it has to make a team of agents. It's now saying, hey. We have to create the input and output directory structure.
2:24:31I can just say yes. Proceed. I could also say yes and don't ask again, but I wanna show you all the different, uh, things that it asked me to do, which is fine.
2:24:38So so I can press yes. As you can see, it made the output and the input. The input is where we have the transcript, so we will be adding the files here, and the output is where we get the CRM, the emails, the insights, and the proposals as well for each transcript that we use.
2:24:51Now it's saying that it's gonna create a sample transcript and orchestration script in parallel, so it's also testing that it does work. Now it's asking whether we can make a transcript dot m d document to basically test a transcript of a call and make sure that all the agents work simultaneously.
2:25:04I can press yes. Is not writing the transcript dot m d, which I believe is gonna be here. Yeah.
2:25:08The input because that's the input. And then it's saying, do you want to make this edit to run pipeline dot sh? I can just press yes.
2:25:14So now it's asking me all the different small things that I typically would not wanna be asked because it's all code. Now I can go to yes and don't ask again. Okay.
2:25:22Now it's also making the folder called agents. So all the different agents that we have, the four different agents will all have a MD document. So MD just stands for markdown.
2:25:30It's the way that they basically put titles and subheadings and bullet points and so on to structure the information. And so each agent will have an empty document, and the empty document, it's simply just instructions of what the agent is. So then when we give it a transcript, it basically goes back to the empty documents and says, okay.
2:25:46What does this agent have to do? What about this agent? What about that agent?
2:25:49And the last agent as well. And it will start running all of them at the same time. Okay.
2:25:52There we go. So this is the structure. We have dot cloud slash agent, then we have agent one.
2:25:56Let me just take this out here. We have agent two. We have agent three, and then we have agent four to make the proposals.
2:26:03Because I'm pulling back up the terminal. I can actually just, uh, clear this conversation. Clear.
2:26:10I can start from zero. So now what we'll do is we'll actually test it. So we have everything done.
2:26:14We have the Cloud Agents set up. We have the input and the output folders. We have the script and then we have the the thing that allows us to be able to make the team.
2:26:21So I can just go back here, I can then go to this example discovery call. There's an example like example input that I would give the actual agent and I can say, hey, can you use this discovery call transcript to be able to activate the team and yeah.
2:26:36Activate the whole team agent team in Cloud Code and use this transcript as an input. Not a question. Paste it here, and now we press run.
2:26:43So this basically pasted the whole the whole call that we have. It's a short call, so nothing crazy, uh, but it's enough for it to be able to make all the things that wanted to make. So it's saying I can see the full agent team pipeline.
2:26:53Let me first update the transcript file with the Collins marketing discovery call and then run the pipeline. Pipeline just means run the agent team. Now it's saying, hey.
2:26:59Can we replace the transcript m d, which is the one here, with the transcript that you just gave us? I'll just say yes. Allow all edits during the session just because I don't have to then, uh, prove everything.
2:27:08So it it uploaded here, and now it's saying, hey. Can we run the pipeline? I can just press yes and don't ask again.
2:27:14Now it's saying, hey. Let me kick off agent one. Call Insights agent first since the other three depend on its output.
2:27:19And this is great because you get to see exactly how agent one now talks to the other agents. Agent one is a call insights, and based on the call insights, we then feed it to the other agent to make the email and the proposal and the CRM as well. Because now it's telling me that it made the client summary, company overview, core problem, main pain points, desired outcome, timeline, red flags, objections and concerns, key quotes.
2:27:38I can say, yes. I wanna proceed, and now I should get the insights here, clientsummary.md. And by the way, I'm doing this obviously manually to show you how to test it and how everything works, but you can pretty much upload this or publish this to a platform like modal and then all you have to give it is the input and then the whole agent team will work on its own.
2:27:54And one thing I'll also cover later on is the pricing because you can see I'm getting completely rinsed with my tokens. It's going faster than than I've ever seen. So this is one drawback of agent teams, but we'll get into that in a later part.
2:28:06So agent one was complete. It then ran agent two three at the same time. So two agents finished.
2:28:10They built the product proposal from insights and the CRM deal for JSON from insights. And now that agent two and three are done, now we're gonna be launching agent four, which is gonna be creating the follow-up email. So it's saying now that all four agents have been completed.
2:28:22Here's a summary of what was generated. So we have the call insights agent, proposal, CRM, and follow-up emails. We have the files, which you can find emails here, CRM here.
2:28:32We have insights here, and we have proposal here, which I'll go in just a second. And it says all outputs files are ready in the output directory. Want me to open them for review?
2:28:40I can just say yeah. Let's see what it does, sweetie. Yes.
2:28:44Open them. By the way, you can also edit the permissions. So right now if you press shift tab shift tab, you can see that I'm, uh, I'm changing this.
2:28:52Plan mode accepts edits. You can also do bypass permissions, which is letting it just do everything on its own. So client summary, this is a document, project proposal, CRM deal update, and follow-up emails.
2:29:02So I'm gonna make this smaller so you can see. Right? And I'm gonna go through each one.
2:29:06So CRM, this is the deal update that you uploaded in your CRM to be able to include all the fields. So this included the company name, industry, size indicator, growth rate, contact name, title, email.
2:29:16There was no email, so we couldn't find it. The deal stage, discovery, summary main problems, pain points, desire.coms, and a bunch of other things as well.
2:29:23Then we have the emails. So this is the follow-up email that I made, and it said, to Sarah Collins from Ren JM Solutions. Subject line, your road map from 18 to 40 clients.
2:29:32Proposal inside. Thank you for the candid conversation on Wednesday. The way you described the gap between where Collins marketing is today and where you want it to be, 40 clients without the operations breaking, give us a clear picture of what needs to happen.
2:29:43A few takeaways I wanna make sure we capture correctly, onboarding is the biggest time drain. An hour per client with manual folder creation, document sense, and tool setups across Google Drive, ClickUp, and Slack, and things still slip through the cracks. Reporting is costing your VA three to four hours per week.
2:29:56The knowledge bottleneck is real. I've attached a proposal which outlines the two phase eight week engagement and talk soon run.
2:30:02We also ask you to book a call. Now if I go to client summary, I can see the insights. So this is the summary of the actual conversation that we had on the call.
2:30:09And in proposal, we have the proposal that we have that they just made based on the conversation. So understanding your challenges, manual client onboarding, reporting, tools, knowledge, proposed solution, and we have scope of work, deliverables, timeline, investment, why us, next steps, and so on.
2:30:25And these are all things that you could easily turn into like a PDF. So let me go here to Clockcode. I'm gonna say, can you turn the project proposal dot m d file into a PDF that I can use?
2:30:36Make sure the formatting is correct. Obviously, if you would look at this, you'd be like, okay. How do we use this now?
2:30:40You could easily, uh, copy that into Notion or copy that into Google Docs, but I just wanna show you how Clockcode does it for me. Alright. So I just finished making the product proposal PDF, so we can go here.
2:30:50This is the markdown format, and it turned it into a PDF. Of course, if you look at it this way, it also wouldn't make sense. But because it's a file on our computer, I'm just gonna open it up, and we get this right here, which is PDF version with the executive summary, understanding your challenges.
2:31:04Then we have tools, proposed solutions, scope of work, deliverables, timeline, investments that also made the table, very nice formatting, the price, why else, and next steps. And so that's exactly how the whole agent team ran through to make the output, which you can then use for whatever it is.
2:31:17And this right here is a seven out of 10 use case, I would say, because you could still do with sub agents in some way, but the output is better here just because you have the client summary, which is made, and then passes the client summary to the proposal and to make the email and the CRM to then give you the full output.
2:31:32Alright. So there's one aspect of this that I briefly mentioned, but I didn't really dive deep into, which is price. Because if you saw before, for me to run all the four agents, it cost me 10,000 tokens, and it was increasing way too fast for my liking.
2:31:45And so here we have token cost and here we have speed. Obviously, regular Claude will take a lot more time. It'll be cheaper the faster you get and the more expensive it will get until you get to agent teams, which is five minutes and $80.
2:31:56This is because each employee in the agent team has its own context. It has its own kind of place where it runs and so it will cost you a lot more to be able to run this agent team. But if you are building something a bit more complex where let's say you're building a software application And for the software application, you want the main, you know, team leader and then you want the different agents to be able to talk to each other.
2:32:18Because when you're making the front end, the back end, the privacy, the the OAuth, all these things, you'd want the different agents to communicate to build a much better output rather than using sub agents. Now for the most part, if you're working with ClawCode and you want to run different tasks in parallel, you're better off going with sub agents rather than spending six times more but also getting six times the speed.
2:32:41Alright. So we now know how to build sub agents and agent teams inside of ClawCode. But after having built hundreds of Agentsia workflows, I've discovered specific techniques, hacks, really, that make your ClawCode dramatically and exponentially better.
2:32:54And I've divided these hacks into eight different hacks, which is why in this video, I'm gonna break them all down step by step. With that being said, let's dive in. Alright.
2:33:02So we're gonna divide these hacks into three different types of hacks. We have system design hacks, then we have execution and performance hacks, and then we have control and optimization hacks. Alright.
2:33:11So the first hack is actually setting up the folders for your project. Alright. So here we have ClawCode, and the mistake that people make when starting out to build is that they go straight to building.
2:33:20Right? But the problem is that right here on the left hand side where we store all the folders and the files and all these things, we have to structure ClawCode in a way where it knows where to store all the different folders and the files, which is where we go to the three layer agent architecture, which is a way for us to be able to structure the AI agent to be way more reliable.
2:33:37The way that we do this is dropping a file inside of ClawCode, which is called the claud dot m d file. This right here, can think of it as a blueprint that you get to ClawCode and it contains a three layer architecture, a three layer structure that ClawCode needs to use to be able to then build these things and store all the files in the same place.
2:33:55The first layer is the directive, so this is what to do. So we give it SOPs, we define objectives, we list the inputs and the outputs, we specify which tools do we have to use, and we cover all the edge cases of the workflow.
2:34:05It's like a brief that you give to someone working in your company. Then we have layer two which is the orchestration and this is for the decisions. So the actual LLM lives inside this folder.
2:34:14So this is good when giving it instructions as to where to store the right scripts. Right? How to handle errors?
2:34:20How to ask questions when it's stuck it's self healing right that's what we call cloud code because when something breaks it knows exactly what's broken and then it goes back and fixes it so you don't have to and this is the glue between what you want to build and what's actually built And then layer three is the execution.
2:34:34This is doing the work. So this includes the Python scripts, the API calls, all the data that's going within the automation, all the files, all the databases, all the techy stuff that it needs to do to actually do the thing. And this is all code, right, of course, Clockcode is just building code that we use.
2:34:48Now instead of doing this, why don't we let the LLM do everything and structure everything? It's because LLMs are about 90% accurate per step. Right?
2:34:56It sounds good until you realize that when you put step over step, you have 0.9 to the power of five, which is 59% success rate, and it's simply not good enough at all if you have anything below a 99% Brexit, which is simply not good enough when we're building agenda workflows. And so the solution is to give it more frameworks, more guardrails in ClawCode to be able for it to understand how to structure things when it's building.
2:35:16K? Now if that didn't make too much sense for you, I'll show you what it means. So right here, I have the Claude dot m d file that I just mentioned.
2:35:22I'm gonna copy this. And by the way, you want access to this full prompt, second link down below in my community. You can find that in the classroom section.
2:35:28So we go here and we can make a file and name this claud dot m d, and then you can paste this whole prompt, right, which again is just a verbal explanation of what I just said. So layer one, layer two, and layer three, and then you wanna press file and save.
2:35:43And then you can say, initialize this project based on the claud dot m d file that I just put in the project or in the files. There we go. And I put plan mode, and then I press run.
2:35:53K. So now what it's gonna do is it's gonna go inside the claud dot m d file. It's gonna read it, and then it's gonna start structuring all the different folders of what it needs to do.
2:36:01Alright. So I just made a plan here. Plan to initialize project structure from cloud dot n d, the context, the changes.
2:36:06So if you see this right here, don't get too intimidated. What this means is the structure of all the folders here. Right?
2:36:12So this is the project root. We have directives. We have execution.
2:36:15We have dot TMP, and we have dot EMV, and then we have, uh, git ignore. Right? These are all different things that are part of the actual structure that you would need to do.
2:36:22You can just press yes in auto accept. Now we'll have tasks, and then on the left hand side here, you can see that it's adding the different files and folders that it's going to use when we're gonna be able to make the actual workflow to make things more structured. K.
2:36:32There we go. It just finished. So all the folders are here.
2:36:34We have directives. We have readme dot m d, which is a basically a file that exposes exactly what this folder is. So this folder contains marked on SOP, standard operating procedures that define objectives, inputs, tool and scripts, outputs, and edge cases.
2:36:46And the easiest way possible, in this way, a second architect going to build a house. Right? Either he goes to build a and starts building immediately as soon as he goes on-site, or what he does is that he sits down.
2:36:55He basically makes a road map or a blueprint as to how he's gonna build the house, and then we get to the building part, which is exactly what this is. This is the structure and the blueprint that we're gonna use well, the color code is gonna use when we have to make all these things. To give you a practical use case, I actually built this YouTube strategy system for myself, and I made a full video as well.
2:37:14You can check it out here. Uh, but, essentially, we have the cloud dot m d file that we propped, which is the exact same one that we did for the other one. And then it started making all the folders.
2:37:22And if you go to directives, we get to see all the different kind of SOPs, right, explaining the different automations. So for example, this right here, analytics tracking.
2:37:30So it's an MD file. So it's marked down file. Marked down because we have hashtags.
2:37:34Just a way of writing, uh, so AI or computers can understand better. But it's a markdown file that tells the actual cloud code exactly what to do, the SOPs.
2:37:43This is what the objective of this thing is. This is the objective or what it is that we want to do. What is the input and what is the output?
2:37:49Right? The input is the channel ID, and the output is performance dashboard data. So a dashboard with all the data and all the different things.
2:37:55Now to be fair, it might seem a bit intimidating when you're starting out and you're wondering still wondering why is it like this or why is it like that, but just understand that this is the way that we structured all the different folders and files. The second hack is using Cloud Code skills. Now I've made a full video breaking Cloud Code skills down in much more detail, but essentially what they are is instruction manuals that Cloud reads before doing a task.
2:38:13Right? So the same way that you give instructions to a chef to make some sort of food, you give Claude the skills to then make better PDFs, better websites, better presentations, and so on. And so for example, let's say you give Claude a task to make a website.
2:38:23Claude, which is the agent, will look at its own scale that we just added to build effective websites, and it will give us a much better result rather than us not giving it a skill. Right? Because without a skill, God guesses, and it's inconsistent in the results.
2:38:35Whilst with a skill, it follows a pattern and it follows instructions, and it gives us a much better output. Now you can find these skills at aimpl.com. I'm gonna leave the link down below in case you wanna check it out.
2:38:45And here, we have over 683 skills, and there's more added every single day.
2:38:50And these skills, like senior back end, web app testing, senior architect, code reviewer, Git commit helper, are the attachments or the things that we can add to Clockcode for it to be able to do a better job at the thing that it needs to do. So I'm gonna be working with this agenda workflow that I just mentioned, which gives me a full presentation of all the competitors, the analytics, and everything like that.
2:39:07And so I'm gonna be using this tool right here, which is Canvas design, which allows me to be able to give ClawCode the extra skill of making better presentations and better PDFs and visuals for any kind of thing that we're doing. So I actually already added the skill. I'm just gonna go ahead and delete it.
2:39:22I'm gonna delete the folder as well because then you can see the output before and after. I'm gonna say, can you please make the presentation and give me back the result analyzing my competitors and the analytics and everything like that. I'm gonna press go and now it's gonna start to do what it needs to do and I'll catch you when it gives me the help.
2:39:36Alright. So I just finished making the presentation with no scale. And as you can see, it's very plain and simple.
2:39:41No effects, no colors, and it's alright. The actual structure of it is a bit bit off, but nothing crazy.
2:39:47Right? And so ideally, we want something that's branded, that's better in terms of structure.
2:39:52I feel like I want some effects here. I want some colors to make it easier on the eyes, but also to make it appear better when I'm using it, uh, with my team and so on. So ideally, we want something that's a bit more professional, that has colors, structure, uh, and the layout looks a bit better.
2:40:05So what we're gonna do here is go back to the colloquo templates. I'm gonna scroll down to the Canva design. And by the way, if you're looking for a PDF or for a MCP builder, which I use all the time, and for other things, you can just simply search it here and you'll find it.
2:40:16You can for example, we can look for a website, web design reviewer, Google Analytics, SEO optimizer, web experience, a lot more things that are there. So I can look for Canva design Canva's design.
2:40:27And if I go to view details, I will get to this page right here, find out that the skill is actually a prompt. So every skill is a prompt, and the prompt is very detailed, which allows Clawcode to be able to build better presentations based on this full prompt that we have here, which someone made.
2:40:42Now I would recommend looking at the amount of downloads because if it has good downloads, then obviously it's good. Not all the skills are great. Right?
2:40:47Depends on the prompt and prompting depends on the person. So all we have to do here is copy this code. I can go back to Clockcode.
2:40:54I can clear conversation, and I can say install this skill here, and I can press go.
2:41:01And now what this will do is it will start adding the skill inside of ClawCode here. We just have to approve a few things. And as you can see, it's saying that the Canvas design skill has been installed successfully.
2:41:10I can go to Claude. I can go to skills, and here we have Canvas design. And if I go to skill.md, here we have the full prompt that we had here under code.
2:41:18Let's run it again. So let me go to clear conversation. I can say, okay.
2:41:22Can we now make the presentation for the analytics comparing me and my results and my competitors and give me back the full presentation on a slideshow? I'll press plan mode. It will now start working on the presentation.
2:41:32Alright. And after a few minutes, it made this full presentation right here with colors and designs, and it looks a lot more cleaner than the one that we had before.
2:41:40We have colors. We have different types of colors. We have a better structure now, better way that we can look at things with headers, which is a lot cleaner than the one we had before.
2:41:47I can see there's a few mistakes though on the top as well, but it's something that we can fix just by prompting it, uh, to do a better job in those specific areas, and we have until the end. Now ClawCode skills are by far one of the most important things that you need to know when it comes to claw code because claw code is great, but it's even better when you add enhancements.
2:42:04It's sort of like the hour man suit, which is amazing, but it's even better if it has the controller, if it has laser, if it has all these things that make it 10 times more powerful, and it makes you build things 10 times the speed instead of you having to go back and forth and prompting it saying, hey. It's wrong. It's not wrong.
2:42:17Make it better and things like that. The next hack is using the spec to do and code workflow. Okay.
2:42:22So now that we added all the structure of the files, do we go straight to building? Well, no. We have to plan even more if we wanna build something that's actually scalable and that's more structured because, yes, you will spend more time now, but it will save you a lot more time later on when you want it to actually work.
2:42:35And the problem with most people using Clockcode is that they paste the prompt like which is, Claude, build me a family activity finder app. AI is great. Right?
2:42:43But if you're not specific about what you want, then it's not gonna do a good job. And so we replaced this prompt with something that looks like this, which is asking you to make a spec dot empty for a family activity finder app. We have the requirements, which is parents enter city, kids ages, availability, distance willing to travel.
2:42:58The app returns five activities with title description and emojis. The text type that we're gonna use is simple HTML, CSS, and JavaScript. And we have milestone one, which is building the UI with dummy data, and milestone two, which is connect to real activity API.
2:43:09Now the reason why we do this is because we want a detailed blueprint with requirements, tech choices, and clear milestones as to what it needs to do to be able to build this thing before we even start building anything. And so I'm gonna press go. And as you can see here, the spec dot m d is sort of like an SOP.
2:43:24And we told previously to Cloud Code that all the SOPs should be stored under the directives. Now it's gonna ask me to accept this plan. I'm just gonna press out just because I don't want it to do anything.
2:43:33And if I go to the directives, I should be able to now to see a spec dot m v file, which is a full file. Let me zoom out a little bit. It tells us the overview of the exact app that we're gonna use.
2:43:43Right? So we have the overview, tech stack, user inputs, output, UI layout, milestone, and we have them in data examples, milestone two, and so on. This gives a lot more clarity and structure to Cloud Code later on when you are building something because bear in mind that when you talk to Cloud Code, you're talking here.
2:43:58Then when you leave Cloud Code and you come back, this will be empty and so it will need some sort of memory to understand exactly what you were doing in the first place. And so you always wanna make sure that you were structured as possible when building the actual thing.
2:44:09Now we're done with step one, which is creating a detailed spec. Now we're gonna go to step two, is generating to do lists. And so now I can say, create a to do list for milestone one based on the spec, and we'll be able to go through milestone by milestone to make sure that everything is alright and everything is structured, and it doesn't build a full thing.
2:44:24And then we're like, hey, man. Not the thing that we wanted. It's gonna make to dos for that specific milestone.
2:44:29Now it has made all the to dos for the first milestone in specific, and I can see here that it says here's the milestone want to do list, create product directory, build index HTML, it can do x y z, and it gives us all the steps in specific because we right now are acting as a supervisor for the actual developer that's building the thing.
2:44:45And now I can say build milestone one using the to do list. I'm not gonna press enter because I don't wanna build anything, but I can tell you that once it finishes milestone one, you can then go and say, okay. Can you not build a to do list for milestone two and milestone three and milestone four until you're finished?
2:44:57And, yes, you've taken a bit more time and to structure things, but you've made the output 10 times better just because you've been more structured. Remember that clock code is only as good as your input and the prompts that you give it. This is a developer, but a developer is good at coding, but it might not be the best to understand the context behind the code and what it is that you actually want, which is why we wanna be as structured and explicit as possible when building these agentic workflows or apps.
2:45:19Alright. So now we get to the execution and performance hacks. The first one being permission mode settings.
2:45:24So down here when we're talking to ClawCode, we can actually change the mode or the settings of the permissions that we're giving ClawCode. You can either press and change or you can simply go shift and tab, and it will change automatically. So the main ones is bypass permissions, and then we have the plan mode.
2:45:37Bypass permissions is used when you're making a very simple task. This is what we call the yolo mode, which is telling Cloud Code, hey. You do everything.
2:45:43Don't ask me any questions. Don't plan. Just, like, go at it.
2:45:45This is great for something very, very simple, whilst plan mode lets you collaborate with Cloud Code to explore solutions and create detailed structures without wasting tokens of going back and forth. What we usually say is a minute of planning saves you ten minutes of building. The next hack is using sub agents for parallel processing.
2:46:00So what this means in a very non techy way for beginners is that when you talk to Clockcode here, you're talking to the main agent. But what you can also do is talk to the main agent and then have other agents working in parallel as all these ones which means that you can do four times the work at the same amount of time.
2:46:15Alright. So here I have my partnership's email inbox. Let's say I wanna process all of these emails together.
2:46:20Of course, I only have about 84 right now but just let's say I had a thousand or 2,000 or whatever it is because most people do have that, then instead of having one agent trying to process all of them, you can split the work between agents and so what you could do is you can split the task across these four agents. So you can say to this agent, hey, can you label emails from one to 200?
2:46:39This one from 200 to 400 and then 400 to 500 and so on, right? And so when you split the task, they all work simultaneously at the same time and it'll be so much faster. Here's my client ID and client secret.
2:46:49Uh, I wanna create a email classifier system where it goes through my Gmail and it basically classifies the emails on whether there it's it's a full full caps. So editor request or software partnership.
2:46:58So I'm gonna give you my client ID client secret so you can use that and then authorize my Gmail. Like, ask me to authorize my Gmail. I'll let you in and then you can we can start the task.
2:47:05Alright. And these four agents just finished running. I can see here that they've all been labeled either a software partnership, other, or, uh, editor request as well.
2:47:12Now the example I showed you, because it's 80 emails, it wouldn't be realistic. But if you have thousands and thousands of emails, then it would work really well because you have these agents running at the same time doing the exact same thing. And all you have to do is make sure that you're there to be able to conduct them and answer any questions if you need to answer.
2:47:27Alright. The next hack is gonna be the control and optimization hacks. The first one is mastering context management.
2:47:32To think of Claude's brain like a notebook with only 200,000 pages. Right? Every time you chat with Claude, it writes down your message, his response, all the files it reads, every tool uses.
2:47:40As the notebook fills up, Claude gets dumber. It gets more stupid. Right?
2:47:44Because he's trying to remember too much at once from a conversation and files. And so if we keep having a chat with this agent right here, it becomes very sluggish because you've used up 50,000 plus tokens just on conversation history. The solution to this is using the smart reset.
2:47:58Right? The smart reset is when you put slash dot compact.
2:48:02Right? And this compresses your conversation into a summary, but it keeps the most important stuff. Right?
2:48:06And you use this when your chat is getting way too long, but you wanna keep working on the same project. And it's compacting the conversation that we had here. Of course, most of this is just Cloud Code just working on its own, but let's say you were to have a conversation, it would extract all the summary points of this, and then it would give us the summary, which we'll then use to be able to go to the next steps.
2:48:23As you can see here, it says the session has been continued from a previous conversation that ran out of context. The summary below covers the earlier portion of the conversation. And so this right here summarizes the conversation that we had to then be able to then keep going with the conversation.
2:48:35And once you keep talking to it, now you have used about a 180,000 tokens free again, but Claude still remembers what you were working on. And the next command is actually something you've seen me use over time and time again, which is d slash clear conversation. Because when I wanna have a new conversation with ClawCode, instead of having to spin up a new agent, I wanna have the same conversation here, but the conversation will be cleared and you can start from zero.
2:48:54And one more thing you can do within the context management is check your context usage. So if you're worried about how much context you're using within a conversation, you can simply press slash context here.
2:49:05So I actually have to say, hey. Can you give me the context usage? Slash context.
2:49:07And here it says, hey. Here's your current context usage. Model is called Opus 4.6.
2:49:11The tokens is 55,500 out of 200,000. We're using about 28%.
2:49:15Again, 200,000 is the maximum of tokens that you can use within one conversation, and it tells us the system prompt, the system tools, the memory files, the skills, the messages, the free space, basically giving us an overview of how much we can use more than what we're using right now. This context hack is important because it allows you to be able to understand exactly how much more you have left, how much runway you have left, and also how much you've used so far.
2:49:35And the last hack here is using the Pixel agent extension. So there are some of you here, including myself when I was starting out, that look at this and think that is very intimidating and it looks very, very complex. And there is actually a way to view your agentic workflows or your cloud workspace in a much more visual way.
2:49:50We can go here to extensions. We can look for pixel agents, and then you choose this one here.
2:49:56You can press install, and now this will be installed in your workspace that we're able to view our workspace in this manner right here. It's sort of like different emojis or Lego blocks, um, that show us all the different agents working simultaneously, which is actually really, really cool.
2:50:08And the way to find it is to go to command shift p. I can look for pixel agent show panel. There we go.
2:50:13And on the right hand side, we can see here that we have a visual representation of a workspace. And if I press new agent on the bottom, you can now see that this person just appeared, and he is idle because we haven't given him any task to do. And I can press more and more agents, which will add more and more people here, and these are the different agents that we can use to be able to run up and spin up different things.
2:50:30So maybe the first agent will be, hey. Can you research the web on the top AI tools, uh, out there in 2026? And then the second agent will be, hey.
2:50:38Can you research the best work of automation tool past ten years? Hey. Can you do extensive research on human psychology and tell me the top 10 trends that you see over time.
2:50:46Right? And now using agent needs approval, I can see exactly when something needs approval. And I can go back to the agent itself, right, and just give them approval on what they need.
2:50:53But all in all, you're not able to visually see your workspace and the agents working at the same time all simultaneously through a very, very visual representation. I do love this because it also makes it fun for me to work with FallCode and I can see different things.
2:51:06And also in addition to that, I can go to the layout and I can start moving things around. I can put the plan here or I can put the boxes here or I can put the desk here or the computer and so on in case you're bored and you wanna change things around for no reason. Yeah.
2:51:18But you have some you have some options. You can also choose the floor. You can basically create your own game and then work with Clockcode at the same time.
2:51:26Quick one before we get back to the video. If you're looking to work with me one to one to be able to start and scale your own AI automation agency in a ninety day one on one mentorship program, then check out the first thing down below. With that being said, let's get into it.
2:51:37Alright. So you've now got eight proven hacks to build better agent of workflows, but everything we built so far lives in your machine, lives in your computer, which is why in the next lesson, I'm gonna show you the top four ways that you can deploy your Agenta workflows onto the cloud so they run twenty four seven without you having to touch a single thing.
2:51:55Let's dive in. Alright. So first of all, what's deployment?
2:51:57Because I know it's a word that everybody says, but only a few people actually know what it means. But essentially, when you're building an agent in Clockcode so let's say you wanna build an agent that can run every single morning at 3AM.
2:52:08When you're testing, it's hosted on your computer. That means you tell it when to run. You press the button.
2:52:13You extract it to run the actual workflow to give you the desired output. The problem with this method is that if you close your laptop, it stops, and if the Wi Fi goes out, it stops. It's sort of like a plant that only lives in your desk, and it can only survive if you water the plant.
2:52:26And if you want the workflow to run every 3AM in the morning, then you would have to be in your computer turned on with Wi Fi to be able to make it work, or else it just wouldn't run. And so in very easy terms, deployment means that we're taking this exact agent that we built in ClawCode, and we're hosting it.
2:52:39We're putting it in someone else's computer. And this is great because the computer is always on. It's always connected to the Internet, but most importantly, it runs twenty four seven.
2:52:46So you can actually run the workflow at 3AM every single day without you having to be in your home computer. Now getting more in-depth into it, a computer here is called a server. So this right here is a server.
2:52:56And servers live in big warehouses called data centers. So data centers right here, they basically store all these computers.
2:53:03Right? Each one is a computer right here. And a computer hosts your workflow, and it's always on.
2:53:08K? That's essentially what it means. Now the good news about deploying it to these servers is that you don't have to buy a server and you don't have to know exactly how it works because what they do is they rent server time.
2:53:17So you pay for the time that the workflow is in the actual server, and they handle all the complicated stuff for you. So all you have to do is build it on Cloud Code. You have to take the code and then upload it in the app, and then it runs it for you.
2:53:28And these are the companies that we're gonna be talking about today. So here's an agentic workflow that I built in Cloud Code. I'm not gonna show you the full build step by step because if you want the full video, you can check it out up here.
2:53:36But, essentially, it's a strategy system that runs every single week on a Monday, and it gives me a full presentation, basically comparing my analytics with my competitors on YouTube, and then it gives me content ideas.
2:53:47So I'm gonna show you the output first so you can see exactly what it is that we need. I can put slash start. So I just ran the actual workflow.
2:53:55Alright. So the workflow just finished running. It gave me the weekly report right here that I can copy, and then I can paste in my browser right here.
2:54:01And this is a presentation that we get every single week. So I get analytics report, which tells me my stats for my channel, my top performing video right here, my benchmark comparisons against other competitors in my space with views, engagement, and so on.
2:54:15Key insights from my channel that I can see, then I get five ideas. So one, two, three, four, five that I can use, and then finally, I get next steps as well.
2:54:22So now we get to the part where we deploy the actual workflow. So there are four main ways to deploy the agent of workflow. The first way is using modal.
2:54:29The second way is using railway. Then we have render, and then we have replant. So we're gonna start with modal right here because it is one that a lot of people in the cloud community are already using and for good reason because it was literally built for AI workflows.
2:54:41So first thing we wanna do is press get started. You can then sign up to Google right here. You can choose your account.
2:54:47Push continue. You can do personal, and you can get started. And one important thing here is that we get $5 of free credits, but you can also have $30 of free credits if you add your card details.
2:54:57So once you add your payment method, you should be able to see that you have now $30 of free credits right here. Now you do get charged 50¢ for the actual verification, but it's worth it because then you get six times the amount of money, uh, that you can use on your own account.
2:55:10So the first up here is going to ClawCode and saying, hey. I wanna host the, um, agentic workflow to run every morning on a Monday at 9AM London time, and I wanna host it in modal.
2:55:22And all I have to do is go back here and I can run this. So I can copy this and then simply paste it here, and I can press go. We could go down different ways, but this is the easiest way to just stall Cloud Code to figure it out before it starts adding it to the model deployment.
2:55:37So after a few minutes, you will get Cloud, which will make a plan and will go step by step as to how it needs to go through deploying the workflow. All you have to do is press yes and auto accept. And now it's gonna start executing on the actual plan.
2:55:47Alright. So the first thing it does is that it creates tasks that show us step by step what it has to do before starting the plan of the deployment in model. Alright.
2:55:55And here's the plan that Cloud has to deploy the system on model. We have step one until step four.
2:56:02All we have to do is press yes in auto accept. You can also read through this if you want, but, of course, everything is in code, so you won't really understand exactly what it is. But one important part is, as you can see, we just made a file called modal app dot p y, which is gonna be the Python file.
2:56:16Now they should know exactly what it says here, but it'll be the Python file that it uses to then deploy the actual workflow on model. And we also see that it made tasks to then deploy to model itself. Alright.
2:56:24So once everything is done, you should be able to see that it finished all the to dos and it says test runs results all passed. And now you can press this link right here, which will take you to the actual app in the place as well. And if you wanna see all the apps, all the different works that you have, you can simply just go to apps and then press on each one in specific.
2:56:39Now here we have the overview, which is the app function. So the thing that we actually need to do. Then we have the deployment history.
2:56:45So let's say you wanna change something in the actual agenda workflow, you don't actually go here and change it. You go back to cloud code. You go right here.
2:56:52You then tell it to change things, and then it automatically updates and it makes a new version of the actual agent workflow. And that's what we can see v one, v two, v 1.3, v 1.4, and so on. And you get to see all the different deployment histories here, so you can always come back to this, uh, and maybe change to the previous version.
2:57:06Then we have app logs. And by the way, replace the word app with workflow, so workflow logs. So you can see all the different logs, all the different runs that we had before.
2:57:14And then we have run weekly YouTube report, which is the action that the actual workflow needs to do. We get to see all the function calls. And then up here, we get to see that it is scheduled for 9AM and only on Mondays, London time.
2:57:26And the next one will be February 16 as today is Wednesday at 1PM. Now to show you how it does look like when it runs, we can press run now, and what this will do is it will schedule an immediate run, and we should be able to see the run show up somewhere here. As you can see, it just showed up.
2:57:41So queued started, start up, and running. So now means that it's executing the actual workflow. Workflow.
2:57:45And just so you know, the workflow finishes by sending myself an email with the actual report every single week. So we should be able to get an email, uh, if it succeeded. Alright.
2:57:53As you can see here that it took fifty six seconds for the presentation to actually run, and I can see the status is success. We can see the time here. And if I go to log, I can see all the steps that it took to be able to make the actual presentation.
2:58:04Now to show you that it did actually make the presentation, I can go to my email, and I can see that the latest email we got is weekly YouTube analytics report, and I get an email saying, hey. This is your report. I can press view presentation, and this will take me to the actual report with all the different ideas and so on.
2:58:18As you can see, it took us less than two minutes to set everything up, and a good thing is that you were now able to run it every five days. So it will run every single Monday, but you can also test it the same way that you're in a workflow platform like any 10or make.com.
2:58:31You can simply just manually press this and see if it works in real time. Now there is a chance that when you do run this, it doesn't work, which is why you can go inside here and it will show you the, uh, message as error, there's so something wrong with it. And you can simply copy this.
2:58:44So copy the whole thing, bring it back to ClockCode and say, hey. I just ran this on modal. It didn't work.
2:58:49It gave me this error. And you can paste it here. K?
2:58:51Now, of course, this is success, so I'm not gonna run this. But if you have an error, that's what it will show you. You copy this, you paste into a ClockCode, and you let ClockCode debug and self heal its own workflow.
2:59:01So in terms of pricing, the way that model works is that we are actually able to pay per second of the workflow running. Now there are three plans.
2:59:09The first one is starter, team, and enterprise. The good thing about model is that it is very easy to set up. Plus because we get $30 of free credits, we're pretty much able to run this every single month for completely free unless you have some astronomical automation and you have, like, 50 agents that are all running at the same but you're most likely not.
2:59:25You got three workspaces, you get a 100 containers, and 10 GPU concurrency. Now what this means basically is the size of the automation, so the space that you have.
2:59:32So I definitely recommend starting with the plan right here and just know that it is free because you get $30 of free credits. So who is modal for? Now modal is great for beginners who want the easiest path to deploy and maintain a workflow built in Cloud Code.
2:59:45It's also best for people who have a workflow that is Python based, which it probably is since you're using Cloud Code. It is also best if you have a workflow that is not running twenty four seven, but it runs in bursts. So maybe it starts the automation when it gets triggered by something.
2:59:58So it can be every single day at 9AM, It gets triggered. And, also, it's great if you wanna pay only for what you use.
3:00:03So the second way that you can host your workflow is using a software called Railway. Now Railway is a slightly more traditional hosting platform, but it is still incredibly simple to use. Think of it as a very friendly modern version of what developers used to use when deploying workflows.
3:00:16All we have to do is press deploy, and we can see here that we have the option of GitHub, database template, docker image, function bucket, and empty projects as well. So we're gonna be using GitHub.
3:00:25Now GitHub is a software that we now have to make an account as well, which is a software where we basically upload the code there and it's gonna be the middleman between Cloud Code and Railway itself. So the first step is going here and going to github.com. Alright.
3:00:39So here's GitHub and think of this as your Google Drive folder. So we're gonna host all the different files in the Google Drive folders, and then we're gonna connect the Google Drive folders to Railway so you can look inside the code and then be able to host it there. So first things first is you wanna sign up.
3:00:52You wanna continue with Google. Choose your email. Press continue.
3:00:57Create an account. And once you've made an account, you will get on a page that looks like this. So the first step that we wanna do is you wanna create a repository because we're making the Cloud Code agent.
3:01:06We're hosting it on GitHub, and we're deploying it on Railway. We can create a repository. We can name this, uh, agentic workflow.
3:01:13We can do public. Everything else is fine. We can create the repo, and then you will get on a page that looks like this.
3:01:19All we wanna do here is you wanna copy this URL, bring it back to Cloud Code, and say, hey. Can you, um, upload all the code to this GitHub repo?
3:01:28And then you wanna paste the link. K. So we're saying, hey.
3:01:31Take all the code of this and connect it to GitHub repo so that we can then use Railway here so that we're able to see the code in GitHub through Railway because there's no way of connecting it directly to Cloud Code. That's why we're using this, and I can press go. Alright.
3:01:44So here we have the plan to push code to the new GitHub repository. We can press yes auto accept. Alright.
3:01:49So now it's telling me that I need to get something called an SSH key. Now if you've heard of API keys or passwords, then this is exactly what it's asking us. So you can press add key to account, submit answer.
3:02:00So what you wanna do here is you wanna go to github.com/settings/keys, and you will see the SSH and GBG keys.
3:02:07All you have to do is press new SSH key. You can name this MacBook. Authentication key is fine.
3:02:13And then make sure to ask ClawCode to give you the actual key that you can paste. Once you have the key, you can copy it. You can paste it here, and then you can add an SSH key.
3:02:21And that's pretty much all you have to do here. I can go back and say, okay. I just finished and added it there.
3:02:25Keep going. And it says that all my code has been pushed to my agenda workflow, so I can press this link right here, which will take me to my repo, which is this one right here.
3:02:33Or if you're stuck on here, you can simply go to three dots. You can go to home, and you'll be able to see that you have the agenda workflow here that you can then see. So all the files that we have here are all the files that we have here.
3:02:42And now I can go back to Railway. I can press GitHub repo. I can log in with GitHub.
3:02:47I can authorize the Railway app. And here you can press I agree. I will not deploy any you can press this button right here.
3:02:53And you wanna press configure GitHub app. All repositories, press install and authorize, and then you wanna press new GitHub repo.
3:03:01And now you'll be able to see the Mikaela partnerships one. Press this right here. It's deploying the repo in Railway.
3:03:06As you can see now, there was an error. So I can press this right here. It's error creating build plan with Railway.
3:03:11So what I would do here by instinct is go back to Cloud Code and say, hey. I've tried to upload it to GitHub, and I got everything working.
3:03:19I got all the all the files. Then And I tried putting it everything into Railway, so connecting Railway to GitHub. And it said, error creating build plan with Railway.
3:03:26So I don't know what to do here. It's gonna ask me what do you want Railway to do with this project. This will determine how we set it up.
3:03:31We can run it on a schedule. So we can just yes and auto accept the plan that he uses to then fix the railway. Alright.
3:03:39So I just fixed this so I can go back here and I can see that this is now building. So it just fixed the thing for me by actually modifying the GitHub repo, which automatically updates the railway app. If And I go here, I can see that I have building status.
3:03:50As you can see here, the deployment was successful. And now we have the agentic workflow, which is online, which should now be triggered automatically and is uploaded and deployed. And if I go back here, I can see that now I have the actual app in production.
3:04:01Now bear in mind that this is also running on a Cron timeline, which means that we are running it every Monday at 9AM UTC. Up here, have the architecture, which is where you can actually edit the agent workflow. Then we have this tab right here, observe the environment so you can monitor different projects and look at the metrics.
3:04:17Then we have log, which is very similar to the one that we're talking about before, which is essentially all the logs that we have here. And whenever something goes wrong, whenever there's an error, you can always go here and copy this and then paste it into ClockCode, and it will fix it automatically. And then we also have settings here as well.
3:04:29Alright. So in terms of pricing, we have the $0 per month plan, which actually isn't a $0 a month plan because we start with a thirty day free trial with $5 credits, and then it's $1 per month. Now if you're confused about vCPU, what that means is basically how fast the computer can think.
3:04:43So one vCPU means that you only get one brain for working on your tasks, whilst 0.5 gigabytes RAM is how much the computer can remember all at once. Basically, like a short term memory or the space that it has to contain everything. And I would recommend this plan to start out, but as you start to build more and more agents, you ideally wanna go to the hobby plan just because there isn't enough RAM and CPU, but also there's rate limits on the things that you can actually build.
3:05:06So for one agent, it's okay. For multiple, I'll recommend going to the hobby, but you also get a $20 a month plan as well. Now when it comes to Railway itself, it's best if your workflow needs to run continuously.
3:05:15So maybe it needs a bot that runs and checks the workflow twenty four seven. Railway is good for that. It's also good if you want the database support built in in the software, and it's also good if you like the idea of putting everything in GitHub and then automatically deploying it on Railway.
3:05:28Now in terms of debugging, so when something goes wrong, the way that it tactically works so you understand is that so let's say there's an error. We go to Cloud Code. We say, hey.
3:05:35There's an error. Cloud Code fixes the code, which then goes to GitHub and updates the code with the new updated version, and then it goes to Railway and redeploys the actual Jentic workflow. Alright.
3:05:45So the next platform is called Render. Render is often described as the modern Heroku. And if that means nothing to you, then just think of it as a really reliable, no surprise hosting platform.
3:05:54So all we have to do here is you wanna get started. You want to go to Google, choose your account, continue, and you will get on the page that looks like this.
3:06:01You wanna press some background workers as this is the agent workflow, uh, builder, and then you wanna connect this to GitHub. So this is very, very similar to Railway in the sense that we use GitHub as a middleman between ClawCode and the deployment. We press GitHub here.
3:06:14We can authorize this. So authorize render. And when we go back, we can install.
3:06:18And right here, I can see that now I can choose my repo. So choose this here. I can name this here, Python three, leave this all the same.
3:06:25And I now have to choose between one of the plans. Because Render does not have a free plan, you either start with a $7 a month plan or you don't start at all. So Render, in contrast to the other ones, does not have a free plan, so you have to pay for the $7 a month plan if you wanna use the software right here.
3:06:38We can go here. We can deploy back worker. And now this will take you to this right here, which is now going to GitHub.
3:06:44It's extracting all the code. So it's taking all the code, and it's building the actual workflow and hosting it and deploying it inside render right here. Alright.
3:06:51And after a few minutes, we can see that it is now successful. It says the service is live, and we're now able to use this workflow automatically. And now this is hosted easily on render right here.
3:06:59If I go to logs, I'll be able to see all the logs of the actual workflow. And metrics here is where you can track the memory and CPU that the workflow is using over time. You also have other things as well here.
3:07:08And now the workflow is gonna be triggered at 9AM every single Monday every week as that is what we added to GitHub. The process of debugging is the exact same as the one that we had on Railway. You basically go to call code, you fix the code, which then updates the code on GitHub, and it finally redeploys on render right here.
3:07:24So who is render for? Renderer is great if you want predictable flat rates pricing for your workflows. It's also good if you need background workers for your agents.
3:07:31So, basically, sub agents for your main actual agent to work. It's a platform that has been around for a while, and it has been battle tested. So if it's important to you to have a platform that has a good reputation and it's been for a while, the render is a choice.
3:07:42And, also, it's great if you value automatic rollbacks when something breaks. Alright. And the last platform here that we're gonna be using to deploy is called the Replit.
3:07:49Replit is one of the most popular platforms as well because it does actually have the agentic feature. Very, very good. So first thing you wanna do is first create an account, choose your email, continue, put your name, press next, personal project.
3:08:03That's fine. And now we can go straight to pricing. So we have the free version here you can publish one app.
3:08:09Then we have the replicore, which is the most recommended one. It's the one that most people use where you can publish more apps and you have more money as credits that you can use. And then finally, we have the team's plan as well.
3:08:17And it's always something that you use if you're trying to build something more complex and you have more and more users in your organization. So I'm gonna go ahead and start with the starter, and here we have the app right here.
3:08:26I can press import code or design. I can go to GitHub. I can connect my GitHub account.
3:08:31I can press continue to GitHub source control. I can authorize rapid integration. You just connected my GitHub here.
3:08:36I can choose a Junta workflow, and I can simply press import from GitHub. As you can see now, the workspace is loading, and it's now importing all the code from GitHub to Replit to then be able to deploy the agent here.
3:08:47As you can see now here, it says status okay. Service is YouTube strategist, and it says the app is running successfully. Let me verify it responds correctly and then set up the deployment.
3:08:55Alright. So now it's finished, and it says everything is set up and running. Let me finalize the import.
3:08:59And it basically imported all the different files here. And all we have to do is press published. And if I go back to the home page, I can see that I have the agenda workflow, which is the one that we have right here.
3:09:07If I go to apps, I can see that I have the agenda workflow, which is private, and this will now be able to run every 9AM in the morning on a Monday. Replit is great if you want to code and deploy in the exact same place. It's also best to use if you're building a quick prototype or a proof of concept for something because it has the agentic feature which allows us to bype code different softwares as well.
3:09:25And it's also great if you don't wanna deal with GitHub itself or terminals or any setup at all. So those right there were the four main ways that you can deploy and maintain your agentic workflows that you build on ClotCode. Now one important thing is that no matter what platform you choose, the debugging workflow is the exact same.
3:09:39So here's the process. Your workflow breaks. You go to the platform's dashboard.
3:09:43You open the logs because every platform has logs. You find the error message, which is usually in red. You copy the error message.
3:09:48You go to Cloud Code. You paste the error message and say, this is the error from my deployed agent. Fix it.
3:09:54Cloud Code fixes the code. You redeploy, which means that you push to GitHub or you run the deploy command, and you check if it works.
3:10:00If not, go back to step one, which makes it so much easier for you as you don't have to go in the code yourself. You can simply let Cloud Code take care of the whole thing. So if you're still not sure as to which software you should go with when it comes to deploying, then here's the thought process.
3:10:11If you're someone who says I want the absolute easiest option to deploy, well, modal is your best option. If you're someone who says my agent needs to run twenty four seven like a background process, then Railway and Render will be good for you. If you're someone who says I want predictable monthly builds with no surprises, then Render is good.
3:10:27If you say I want to code and deploy without leaving my browser, then Replit is the way. And if you say I just want to test something quickly for free, then you can use modal. Now we do also have some other options like AWS, GCP, or Azure.
3:10:38The only reason why I did not include them is because they're more for enterprise grade level, uh, deployment, and they're more advanced as well. And that right there is a complete picture because building the agenda workflow was just the first half. Deploying is everything that comes after you build it, and it's arguably the most important step as building it is fun.
3:10:54Right? But how do you actually give it to the client? How do you deploy it so it actually works without you being there all the time?
3:10:59And a good thing is that to deploy an agented workflow with all this code, you don't have to be a developer. All you have do really is just follow step by step what I did in this video and let it run on its own.
3:11:12Right. Congrats on finishing module four. At this point, you have everything.
3:11:16I mean the foundations, the customizations, the prompting, the agent workflows, and the deployment itself. So at this point, you're not just someone who uses Clockcode, but you're someone who understands how it works but also how to build with it.
3:11:29Now it's my turn to prove it to you because in the final module, you're gonna watch me build real products live from scratch using everything we covered so far. There's no theory, just building, making it really really practical for you. Let's dive in.
3:11:41Alright. So when it comes to my YouTube content pipeline, there are four different stages. The first one is the ideation, which contains three different types of we call them skills or automations that are built inside of Glockcode.
3:11:50The first one here is called the comment scraper, which is where the automation itself will go to x amount of videos posted by x amount of people in x amount of time and then extract all the comments and then give me ideas for comment. Twitter monitoring is where it's going to the actual profiles. It's extracting all the tweets and gives me ideas, and not a lot of videos is the videos that have gone well for other channels.
3:12:09Why have they gone well, and how can we apply the same idea to my channel? Then we have packaging, which is title and thumbnail. So for example, this video right here, we have a title.
3:12:17There's a picture before that made you wanna press on the actual video, and that's the packaging. Then we have the video itself, which is the hook. So the first thirty seconds of this video, that was a hook.
3:12:26That was scripted. There's a script, which is what I say in throughout the video. Of course, a video like this, there's no script.
3:12:31I go off the rip. Right? But something that's more like a head talking video where I go through a concept.
3:12:37Right? Where it's me standing up. Well, that's scripted, so we have a script.
3:12:40And then we have visual diagrams. So this right here, for example, is built by an automation. I didn't actually build it myself.
3:12:46That's diagrams. And then lastly, have post video, which is looking at the performance of my videos but also views to clicks. Views to clicks is looking at the amount of people who clicked to watch my video going through my program versus the amount of views.
3:12:58Right? Because obviously, on the back of my videos, I want people to work with me one to one. I wanna help them.
3:13:03I wanna help them scale their agencies. So I wanna check which videos do people like the most to make them want to kinda wanna work with me one to one. That's what it is.
3:13:10So with that said, here is my ClawCode workspace. It's called the YouTube workspace. It's consisted of different folders that we have.
3:13:16The first one is dot cloud, then we have context, executions, outputs, references, scripts, dot EMV, and then the main one here which is cloud dot m d. Now I actually separated the folders into four different The first one is commands, then we have skills, then we have context, and then we have the cloud dot m d. I'll explain what these are as well, but not a main priority, like, are in main folders that we use.
3:13:36K? So the first one here is called commands. So commands are reusable slash commands, like slash competitor analysis or slash create plan or slash something else that allow you to trigger to start any workflow just by calling it using a slash.
3:13:50Right? And when I say workflow, I mean it triggers a skill. So if I want to run the Excalidraw diagram, for example.
3:13:55Right? Then I would do slash ExcaliDraw diagram, and that would call that specific skill instantly.
3:14:01Now if you're not sure what skills are, first, I recommend you watch this video up here. Um, but in case you are aware, skills would be modular instructions that we give Claude to explain or to give it an overview of how to do a specific task step by step.
3:14:15So for example, Excalidraw is a platform like Miro where you build kind of visual diagrams like I showed you before. Well inside the skill, there are step by step instructions as to how to do that specific thing and so all in all, our Claude is gonna be our kind of company. Right?
3:14:29And inside the company, there's different employees which are responsible for different things and those employees are skills. Then we have context. So everything that Claude needs to know about my business.
3:14:38So first of all, who I am, my strategy when it comes to, uh, the content, my data, and my content guidelines. Because for example, let's say we're using the scale to script my videos. Right?
3:14:48I'm making a script. When I wanna make my script, I also want Claude to use my YouTube content guidelines and business info to check, hey. This is the script.
3:14:56This is the guidelines. Are we sure we're following the right guidelines? That's all it is.
3:15:00And then finally, have the claude.md, which is a master file here that ties the whole workspace together telling Claude how all the pieces work in one system. K. So it tells Claude, hey.
3:15:09We have the folder here, which is called context, the folder here, which is called skills, and it gives it an overview of the whole organization. The reason that is is because it will look at the Claude. M d every single time before it does anything.
3:15:19That way it has context as to your workspace and how it all ties together. Now first things first is the ideation. So ideation consists of the comment analysis report.
3:15:29It consists of the competitor analysis, which is the outlier videos, the videos that have done really well for other channels, and then we have Twitter cloud monitoring. I'm I'm gonna run it first, and then I'll walk you through how it's set up as well. So dot comment analysis report, and then we can do bypass permissions because I don't wanna tell it, yes.
3:15:45Approve. Approve. Approve.
3:15:46I just want Cloud Code to do its own thing because it knows everything already. As you can see here, it says that it fetched 1,369 comments across five different channels.
3:15:54Now it's reading through all the comments. It's looking at the YouTube content guideline. Very important.
3:15:57Alright. So now the skill finished running. It said it found 1,369 comments.
3:16:01It made the PDF that I can see, and it gave me the top five content opportunities that I can do. So the PDF looks like this where we have my channel, top four content ideas requested by your audience. And by the way, this is the comments that you guys add to my videos.
3:16:13And then what I also did is I looked at Nate Herc, looked at the top four content ideas from his audience, looked at Lee Motley, looked at, I believe, Nick Sariaev, and then Jack Roberts as well. Right? And we have, uh, the ideas as well.
3:16:24And then it gave me the cross channel themes. So what are the common request or ideas that people want across all channels? For example, how do I actually deliver AI work to clients?
3:16:33This came from my channel, Liam and Nate. Or token management cost optimization for ClawCode, Nate, Jack, and Liam, and so on.
3:16:40Right? Plus, it also gave me content opportunities specific to me. So how to deliver AI products to clients?
3:16:45Complete guide. This is the biggest gap in the market. Nobody's making comprehensive content about the delivery side, account setup, hosting, handover documentation, maintenance agreements, access management.
3:16:55You've done this for real clients. Make the definitive video. Right?
3:16:58And we go step by step. Then we have the stats at the end with subscribers, videos, comments, and average comment per video. Guys, I believe you should be commenting more of my videos so we can get the numbers up.
3:17:08But, hey, comparison is the thief of joy. Am I right? Anyways, we have the full PDF, which I can use now as my way of generating ideas for content, which are really, really good.
3:17:18And inside the skill, we have the skill.nd, which is the full instruction manual for the skill, which goes through step one, fetch comments. Step two, analyze the report.
3:17:26Step three, generate PDF, and that's it. And it gives you the PDF itself.
3:17:30Then we also have scripts, which are the different Python files, the code that basically allows the automation to run. Think of this, like, if you know 10, if you've heard of n 10 or make.com, think of the script, each script as like a workflow.
3:17:41Right? And so we are running workflows. The first one is fetching comments.
3:17:44So we're running the workflow to fetch comments, to extract comments from videos. Then we run the generated PDF as another workflow to basically make a PDF that we can use. And then we have references, which is the report format.
3:17:55It's giving the ClawCode a a way to understand what the format is of the report itself. Right?
3:18:01So we have raw stats. We have cross channel themes, which is how we make the report look like this. How does it know the structure, and how is it consistent with the structure?
3:18:08Well, it has instructions for that. Alright. So now that we're done with that, we can go to the other videos.
3:18:12So here I can press competitor analysis. Uh, there we go.
3:18:17And then I can do bypass permissions so I can let it run. So the first thing it does is that it looks at the CSV file, is called competitors dot CSV.
3:18:25And this is a CSV file that I have down here, which is a list of 43 channels, including mine, which are gonna be the channels that we use to extract the past 10 videos that they've done, and then we can start doing the math as to what videos have done really well for them compared to the other videos that they've done.
3:18:41Alright. Now it's saying that it just finished and exported everything to the Google Sheet. I'll show you exactly what that means.
3:18:46We found four Alire videos across 42 channels, a lean week, all scored 10 out of 15. Cool. And then we have key patterns most actionable for you.
3:18:54But if I go here to my views to clicks Google Sheet where I host everything, we can see that we have the channel name, subscribers. I think it's a bit blurry, but it's fine.
3:19:02Video URL, the amount of views, the upload date, the duration of the video, the medium views, so the average views that they got over the past 10 videos, and then we have the outlier multiple. So the outlier multiple is a mathematical equation that we use, which is how many views did this video get over the average views of their channel to see if this exceeded that.
3:19:21Right? And then it also gives me a relevant score. So for example, this video over here, which was about MCP is dead and nobody saw it coming.
3:19:28Yep. Gave a zero relevant score because it's not aligned with my brand or audience.
3:19:33That's what I mean by cloud code using my brand guidelines and my the person that I'm targeting, what they care about, and what that looks like. And what we have, relevant scores of three, three, and two. So not only does this save me hours and hours and hours of time of me having to go through their videos, having to look at the views, having to average it out, having to make the equation, but it also gave me the relevance that I could use for my channel.
3:19:53And so here I could do z to a. I can look at these two on the top ones.
3:19:58Let me look at the ideas, how to use Clockwork, how to get unlimited website clients. And so I could do a video based on that in my kind of way. And this skill right here is consisted of references.
3:20:06So this is the scoring and output, which is a kind of document that goes through how to score an outlier video. What does that look like? And then we have, again, the scripts, which are the different workflows.
3:20:14And then we have the skill dot m d, which is the step by step instructions that we go through to be able to actually make this work. And then we also have Google. Token JSON, which is the way that we are able to access the YouTube API, but also my Google Sheet account.
3:20:27Then the last one we have is called Twitter Cloud Monitor. So Twitter Cloud Monitor is a workflow that goes to configuration profiles.
3:20:34Yeah. These profiles in Twitter, and it might look a bit kind of weird, but, uh, by the way, Claude wrote this, not me. And it looks at these channels, not channels, these Twitter profiles.
3:20:45Right? And what it does is that it goes through each profile, it extracts the tweets from the past twenty four hours, Right? And we'd have this workflow running every single day and it gives me the new updates on ClawCode because honestly, I can't keep them up myself and I'm sure you cannot too.
3:20:57So that's exactly how a lot of YouTubers do get their ideas is through Twitter. Right? So why not make an automation that can automate that process of you having to find the tweets yourself?
3:21:05So I can go here and put slash Twitter monitor cloud monitor.
3:21:09I can do bypass permissions. Right. It just finished running.
3:21:12It said it scanned through 63 tweets. The relevant score above three is just two, and then we have the key findings. And it also made the PDF which was sent to my email.
3:21:20If I go to my email, I can see that I have a new PDF. I can open it up, and it can give me now the new features that we have here, and then we have the content opportunities that we can use.
3:21:31Right? Now are all of them gonna be great ideas?
3:21:34No. Right? But it filters them out.
3:21:36Because not every day Clockcode releases something, maybe every two days or every three days. So it looks at the tweets, it checks the relevance, and then it gives me the report as well. So I basically keep myself updated without having to go to these 12 different profiles every single day and check what they've posted.
3:21:49That's how you get the YouTubers making videos about the new slash command or the new slash body or the new slash executions or the new slash whatever. Right? Uh, it's because they look at these exact profiles that I walked you through.
3:21:59Now that we're done with ideation, let's go to the next one, which is packaging. So packaging is consisted of thumbnails and titles. So let's go through thumbnails first.
3:22:06So the first one is thumbnail creation. So the way that this skill works is that we take a video. So let's say we have a video, uh, like this.
3:22:12Let's copy the video, copy the link, and then we go back to ClockCode, and I can put slash thumbnail creation.
3:22:22There you go. And then what we do is I took the link of the video with Alex or Mozi. It will look at my face, right, that we have a picture of, which is gonna be here.
3:22:29Reference is right here. That's my face. And then it will take that face.
3:22:33It will replace Alex Sramozi's face in that thumbnail with my face and then give me back the thumbnail, basically. What source would you like to use a thumbnail? I'm gonna put the link here of that video that I copied.
3:22:42What it's gonna do now is it's gonna go inside this video. It's gonna download the thumbnail, and then it's gonna go to my picture, swap the face, and then give me back the image. And this is using a server called fowl.ai because that's where we're gonna use the nano banana model to be able to make this happen.
3:22:58Alright. So it just finished the thumbnail. I'm gonna say, can you open up the image?
3:23:03And to be honest with you, it didn't do the best job. Right?
3:23:09And replace my face with his face. Obviously, that's what he looks like. I don't have a beard as you can see, but it copied the words.
3:23:16It made the words good. It made these massive traps that he has, and it put my face there. So ideally, now we can go and say, hey.
3:23:23Can you can you remove the beard from my face? But it swapped the face initially. And by the way, this is all prompting.
3:23:29So one thing I could say is like, ClawCode. It actually added a beard. Can you remove the beard?
3:23:34Hey, ClawCode. Can you remove the beard from my face in that thumbnail and give me back the thumbnail? And, also, replace the body with, a black shirt as well.
3:23:41Keep everything else the same. Alright. I just changed it, and here we are.
3:23:43Much better thumbnail here. It had no beard, and it put me in a better place as well. So that's the thumbnail creation, which I think is nuts because you can keep iterating it.
3:23:50You can keep saying, hey. Can you change the text? Can you change this?
3:23:54Can you change that? But, essentially, we're taking inspiration from other channels and then using it for our own. Now I don't recommend copying the exact same as someone else's thumbnail, but you can always copy the the way of thinking or taking inspiration on the idea of the packaging.
3:24:08Right? Now with that said, we go to the next one, which is the title generation. Let's say I was gonna do the title generation.
3:24:15I'm gonna wait for it to ask any questions. So now it's saying, what's the video content you like the titles for? Any specific angle or hook you want, how many titles?
3:24:22You can do 20 titles. That's fine. So the video is about how I'm using Clawcode to automate my YouTube channel that I've grown to 51,000 subscribers, about 2,000,000 views in the past seven months.
3:24:33That's what the video is about. I'm gonna go to bypass permissions just to make sure that it does everything. As you can see here, it's looking at the title frameworks because we have a list of title frameworks that work really well.
3:24:42So 300 that it's using. It's then looking at the YouTube content guidelines, which is my MD document, goes through my guidelines and my kind of avatar that I'm looking for, and then it's gonna give me 20 different titles.
3:24:54So how I automated my entire YouTube channel with Clockcode? I use Clockcode to grow to 51 subscribers. Here's how.
3:25:00How ClawCode runs my YouTube channel, 51 cut 51 k subs in seven months. I let AI run my YouTube channel 2,000,000 views later, and it does a lot more, which I think are really, really, really, really good. And then it gives me the top titles based on curiosity, specificity, but also emotion.
3:25:14Alright. With that said, we're done with packaging. Now we can go to the video itself.
3:25:17Alright. So for the video itself, we have script writing, which is the hooks plus the script of the video, and then we have two different Excali draw diagrams and visuals that I'm gonna walk you So the first one is script. So script writing, let's go here.
3:25:29Let's wait for it to ask me any questions about the video. Target length, any specific key points to cover, hook style and preferences, and CTA goal. I'm gonna start writing.
3:25:38There we go. So with all this said, I can press go, and then I can do bypass permissions. Let it do its own thing.
3:25:44And while this is running, I can show you the actual references which we have here. So here we have the two different videos that I've done.
3:25:52So head talking videos. So previous examples of my scripts that have worked pretty well. Then the intro framework.
3:25:57So so what's a good hook? What does that look like? So we give it references.
3:26:01Script writing framework. So this is a YouTube video which talked about the best storytelling or the best ways to write scripts. And then we have the skill.m d, which again is a step by step instructions as to what it needs to do.
3:26:13So step one, gather information. Step two, load context. Step three, build the audience avatar.
3:26:18Step four, write the hook. Step five, write the body. And then step six, write the mid roll CTA.
3:26:24And step seven, write the closing CTA. Step eight, add production notes. Step nine, quality check, and then we have the output, which is insane for AI to do all of this because previously, we had to spend so much time on everything, but now it does it in seconds.
3:26:37There we go. So now we had the script. I can go here, and I can see that this is the full script.
3:26:41What I typically do, because this looks very confusing, is I copy this and I paste it in Notion like this. And now we have the hook.
3:26:48It took me five months to get my first AI agency client, five months of building systems that nobody asked for, sending messages that nobody replied to, and, honestly, questioning whether the whole thing was even going to work. And then in fourteen months, I scaled the same agency to 6 figures, working with over 50 businesses and teaching over 22,000 people in the process, all starting with zero business or technical knowledge.
3:27:06I can read the whole thing. Obviously, I'm not going to in case you were wondering. But we have section one, which gives me the script itself, and I'm reading this.
3:27:12It's very, very good. And you can see it also gives the editor a kind of a a reference of what edits would need to be here. Text on the screen, b roll, pause, section two, section three, closing CTA, and that's it.
3:27:25Now bear in mind that you can use a general AI to make your script. There's a difference between a general slop AI script and actually making a good script, and that's typically using references, using what's a good script look like, using human language, using good transitions.
3:27:38And so that's why we feed a lot of information to ClockCode to tell it, hey. Here's a video that worked really well in the past. Here is a framework of how to make good scripts.
3:27:47Here's the idea. Now give it back to us. And this also looks like the context.
3:27:51So the personal info, the strategy, the content guidelines, like I mentioned before. Alright. So now that we've made the script, there are times in a video where we need supporting visuals.
3:27:58That would be a diagram. That would be some sort of, uh, yeah, a diagram or an illustration of a concept. So for example, here in the ClawCode skills video, we have diagrams that look like these.
3:28:08We have this, all of these where skills are stored, this, and it keeps on going. Right?
3:28:13More and more diagrams. And so here I can say, hey. Can you make an Exceledraw diagram based on a video that I'm doing about sub agents?
3:28:20So make a diagram as to, like, what sub agents are, why they're important, how they're different from agent teams, and that's it.
3:28:27Make them really good, sophisticated, and give it back to me. I'm gonna go and bypass permissions. There we go.
3:28:32Cool. So now what it's going to do is it's going to read the skill first. So let me read the Excalidro diagram first.
3:28:38As you can see, Claude skills Excalidro diagram, Excalidar diagram dot n d, is looking at the instructions, and now it's gonna give us back the code, which is essentially the diagram itself. Alright. So I just finished making the diagram.
3:28:48I can go here. I can copy this. So command a and then bring it back and paste it here.
3:28:53Goddamn. Look at this. Beautiful.
3:28:55So sub agents, architecture, and purpose. What are sub agents? Why they matter?
3:28:59And sub agents versus agent teams. I mean, guys, come on. This is insane for it to be made in a few minutes.
3:29:04It made some errors here, but nonetheless, this is crazy. And you can instantly make a video about sub agents as soon as you've made the diagram.
3:29:12Now this saves a lot a lot of time because sometimes when I'm explaining a concept, I want to show a diagram. I wanna show some visual representation so you can hear but also see at the same time because you retain more by doing both.
3:29:23So that's where I use this. Now let's say, can we also add some visuals using the Excalidraw visual skill?
3:29:29Yeah. Add some visuals there as well. There you go.
3:29:31So now it's gonna take the diagram. It's going to then add visuals, and then you will see the difference as well. Alright.
3:29:35So you just finished making the three different images. So I can go to the AccelaDraw, and I can drag them across from here to here. And here we have the three different visuals.
3:29:44So you see how this is a diagram, but these are images? That's exactly what the Excalidrop visual scale does.
3:29:53Which one's better? It kinda depends on you. Sometimes I like to have a mixture of both, but nonetheless, they save me a ton of time whenever I have to make a video and make diagrams for that video.
3:30:02Alright. So now that we're done with the video assets, let's go to the last one which is post video data. So here on Google Sheets, I have a list called the views to clicks, which is out of this many views, I get this many clicks.
3:30:13And this is a CTR. And then we have the type of video. Now the reason we do this is because we wanna see out of this many views, how many people click to see the video of my one to one program to then be able to work with me.
3:30:24Right? And we do this because typically the more value that you provide, the higher the CTR is. K?
3:30:29So in this case, if I have more clicks and I have more data as to what videos people like or people want, then I can provide a shit ton of value to you guys and then be able to help you guys even further. Right? So now you might be asking yourself, how do you do this?
3:30:42So manually, the way that this works is that if I go to one of my videos, you can press this link. Now this link, you can see Bitly. Bitly is a software that pretty much all the YouTubers, um, that I know kinda use to be able to track how many people click to the next thing.
3:30:54And then, obviously, you click to kinda watch the video of, like, the full one to one program, how it'll help you to be able to start and scale your own AI agency. Right?
3:31:02And so here, I see how many people clicked out of this many views. So out of this many views, how many people press this button? And it's very tedious because I have to go to Bitly.
3:31:10I have to see the views, and it takes me a lot of time, especially if I have to do this weekly for this many videos. And the videos just keep on adding and adding and adding. Right?
3:31:18And so what we do here is I have a automation that I can do. I can put slash views to clicks analysis. And what this will do is that it will look at my videos.
3:31:28So all my videos in my channel since a specific date where I started tracking. So now it's running the automation, and now it said that it updated the sheet successfully. So if I go back, I can see that this is now all in different places, and I can delete these two columns.
3:31:41I don't know why they're here. Yeah. And I can do sort z to a, and I can see the CTR here.
3:31:45So this is the videos where it had a higher CTR rate. Right? There's actually b t c hot talking and b t c hot talking.
3:31:51There you go. And I can start to see, hey. These many videos on the bottom, they don't do too well in terms of the CTR, but these videos do.
3:31:56And so let me make more of those videos and less of these videos. And that's one of the ways where we strategize what content idea works better and what content ideas don't work better based on my audience, my offer, and who I'm targeting.
3:32:11Now let's take it one step further because in the final lesson, I'm gonna show you how you can build and deploy $10,000 websites using Clockcode. Less Devon.
3:32:20So to give you, like, a full recap, high level of what we did, we made an image that looks like this of this Ferrari f one. Then we made a video that looks like this where we have the components of the car breaking up to then make a website that looks like this where we have born to perform, seven decades of fury, forged in Maranello.
3:32:35Then we go down. You can see the car breaking up and going back in, which is super, super, super sick. And by the way, this right here is a video.
3:32:42So we have the video incorporated in the websites. You can also see here prancing, which is the the horse of the Ferrari.
3:32:49So because now we can see maximum speed, we can see it's zero to a 100 kilometers per hour, Top speed here. We also have the final page, which is experience the legend. Reserve now.
3:32:57So reserve the spot for the for the actual car. K. So that's the full website that we're gonna build.
3:33:00So we're gonna go through everything step by step. Now there's lot of moving components for this exact websites because there's different places where we get data, make the image, make the video, which is why we're gonna break it all down into a four step framework. The four step framework is called plan, execute, review, and deploy.
3:33:15Alright. So the first one here is plan. Okay.
3:33:16So here's the product that I used to be able to create the website that I showed you. As you can see here, I have the video of the Ferrari that we used inside the website. And, essentially, to be able to start everything, we need two different things.
3:33:26We need something called the claud dot m d file, and we need the second thing, which is two different skills. So first thing is the Claude dot m d file. Now what is it?
3:33:33This right here represents the definition quite in a in a very quick way, which is, are you tired to repeat yourself? Tell Claude to remember what you've told it using the Claude dot m d file. So, essentially, it's a way for us to give Claude memory of the project.
3:33:45Because every single time that we're talking to Claude, it needs to have context as what it is that we're doing. So this right here describes the project, which is a landing page for Ferrari. Then we have the file structure.
3:33:54So when it makes the landing page, where does it store everything? What folders? Where do the files go inside the folders?
3:33:59And it gives structure to the whole project. Then we have more things specific to the actual landing page, the design. So these are the different designs that we use.
3:34:06Animations and external dependencies, coding conventions, constraints. Now a lot of the things that are here, don't fully understand, right, because I'm not super technical, but we use AI to craft the Cloud.
3:34:16Md file, which makes it easier for us to be to do it. So once you have the Cloud. Md file, which gives Cloud Code the memory, then now we have the two different skills.
3:34:23So skills are basically ways for us to give Cloud Code a series of instructions to make something better. So for example, we need to make a landing page, and we need to make a video that fits well inside the landing page. So the first skill that we need is called front end design.
3:34:37Now a skill, again, is just a list of instructions. So this skill right here tells ClawCode, hey. Here's how to create a nice looking UI, so user interface, basically a landing page, which increases the output for the landing page that we're gonna make.
3:34:50And so every single time we mix a landing page, it looks at the skill, it looks at all the components, how it needs to think about design, the front end guidelines, and so on. And so it will use the skill when making the landing page. Now your question might be, how do I even get the skill?
3:35:02If you go to slash manage plugins, you'll be able to see the skill right here called front end design. You can press install and you have three options.
3:35:11Either you install it available for all your projects, which means that you don't have to install it again for any other project or install it for this project in specific. But bear in mind that if you do that every single time that you start a new project, you will not be able to see the skill, and you'll have to install it again.
3:35:24Once you've installed it, you should be able to see either here or here. Make sure you refresh your claw code, and you'll have the skill that I'm doing, which is the instructions itself. The second skill is called the video to website skill.
3:35:34And this right here is not something you'll be able to find in the plugins. So you can find it in the resources that are left below. But, essentially, this gives ClawCode instructions as to how it should take the video and turn that into something that you can use within the website.
3:35:48And it's very, very long as you can see here. List of instructions. So once you have those two things, we are good to go to the next steps.
3:35:54Before we do that, I made a new project with the two different skills in a cloud. M d to keep it very, very simple as you can follow along with me when I do this. Now that we're done with planning, we can go to step two, which is execute.
3:36:03Okay. So when it comes to execute, we have to ask ourselves what do we need to be able to make this landing page happen? So I firstly need an image, which then turns into a video.
3:36:11And then also one more thing that I wanna use is a reference landing page so that we can feed it all into Clockcode, and it uses the reference landing page. It uses the video, and it uses the skills to be able to craft us the beautiful landing page that we're gonna use. Now to be able to make the image in the video, we can use a software called Keyed API, which is one of the best softwares that I've used because it gives us access to all the video models, image models in a fastest way, and it pretty much has all the options that you need.
3:36:33So first thing is to make an account, and then you can go to market to be able to add your API credits because this is not free. It costs credits to run these images and videos. So first thing is to make an image, and I'm gonna be using the Nano Banana two model, which is a new model that, uh, Google Gemini made because it's fast and it's cost efficient.
3:36:49And down here, we can write the prompt that we need. So I'm gonna be copying a prompt that I already made. There we go.
3:36:54Which says that I need a high resolution studio grade image of a Formula one racing car positioned against a completely plain black background. The scene should be a minimal and isolated with no shadows, reflections, hands, or additional objects visible. The car should appear sharply detailed and professionally lit as if photographed for a premium automotive studio shoot.
3:37:11And then we can remove this. We can go down to four k. Aspect ratio, we can do auto.
3:37:17Automatic is fine. And I can press run. Bear in mind that if you're wondering how much does this cost, well, you paid $5 for 1,000 credits, so it costs you 18 to be able to do it, which means that you're paying about $0.09 to make the image.
3:37:29Alright. So it just finished making the image. As you can see here, very, very clean.
3:37:33Now we can download this, and we can go to the next step, which is making the video. So I can go to AI video API. I can go to Cling three point o.
3:37:39There's also other image video models that you can use, but this one is by far the best that I've seen. I can take this out, and I can drag the image across. Boom.
3:37:49So this right here is the input image. I can take out the sound. We're not gonna add any end frame.
3:37:54So let's say you're making a a website about a cup. Right? Just some sort of cup.
3:37:58You would have a cup with no water and then a cup with water. That's a start frame and then end frame. So you can basically see the animation of the water going inside the cup.
3:38:05But in this case, we don't need it. And now when we get to the prompt, we can go to chat gbt. I can paste this prompt right here, which says that it's an AI video prompt engineer.
3:38:13You can stop this video and read it, but I'm gonna then give it the image of the Formula one racing car as an input, and then I'm gonna press run. And now it's gonna give me the full prompt that I'll need to be able to make the video. Now one thing, I do like this prompt.
3:38:27I just think it's a bit long, so I can say, can you shorten this a little bit and make it more concise?
3:38:34But sometimes when you provide too much context or when the prompt is too long for an AI, it does a worse job. Okay. Cool.
3:38:40That's fine. So I can go back here. We have the clean three point o, and I can paste this prompt.
3:38:45And now the duration will be five seconds. That's fine. And I wanna make sure that the mode, so the actual resolution is higher.
3:38:52So I can press pro because the resolution is basically how clean or how nice does the video look. We can press run now. As you can see, this is taking a 135, which is about $0.625, uh, that we're spending on the video.
3:39:03Alright. So just finished making the video. Let's see the video here.
3:39:08It's even better than the one I had before. Cool.
3:39:10So now we can download this, save this to your desktop. I can go back to Cloud Code.
3:39:15I can then give it the video as the input. K? And I can rename this Ferrari video.
3:39:24There we go. Now that we have the video, there's one more thing that we have to do, which is getting a reference landing page. Because, yes, we can make a landing page, but if we give it stricter guidelines, it does a better job.
3:39:34So if you go to dribble.com, you can find the charger product landing page right here, which is a very, very nice landing page that we can use as reference. So you can go here. You can save video as reference.
3:39:45I can press save, and then I can drag it across the Claw Code as well. So now we also have the video of the landing page as a reference.
3:39:52Finally, I can go to Claw Code. Make sure this right here is in plan mode, and then you can paste this here, and you will say, I just dropped in a video, which is the Ferrari. Where is Ferrari video?
3:40:03Here we go. Create a one product landing page for this product. The site must feel modern, premium, and highly professional with smooth animations and polished UI design throughout.
3:40:11All text should be clean, minimal, and easy to read using high contrast typography on a pure black background dark mode. The layout and visual styling should be seamlessly blend with the Ferrari image, making the entire page appear continuous and fluid with the background. And then I will be able to say, I also added a video, which is a reference video that I want you to look at in terms of the actual design of the landing page itself.
3:40:32And now I can press go. Alright. So as ClawCode is actually making the plan, you can see that it's reading the skills.
3:40:37So skill two, skill three when making the plan itself. That's what we use in skills. Alright.
3:40:42So I just finished making a plan. Context, framing extraction, uh, file structure, design tokens, page architecture, and you're looking at this, you're like, what the hell even is this?
3:40:49The same here, so don't worry. JavaScript initialization order. I'm wondering why they have to make such complex words for all of these, man.
3:40:55So let's go and press yes and auto accept. And then we can turn this to bypass permissions because I don't wanna be approving every single thing.
3:41:03Bear in mind that bypass permissions is a bit more dangerous because it takes over and it does everything, but I usually use it when I'm actually looking at Clockcode myself. I never go to sleep and then let Clockcode do its thing and then come back. Right?
3:41:13Alright. So I just finished making version one of the landing page, and bear in mind that on the left hand side, you you will be able to see new folders that were created. And if I go to the Cloud.
3:41:21Md file that we had before, we had CSS folder, we had JS folder, and we had frames folder. And here you can see CSS folder, frames folder, and JS folder. So that's what we mean by file structure.
3:41:31Where are the files gonna be stored? CSS is gonna be all the different styles that we're gonna use, so all the colors and so on. I'm only saying this because a lot of you guys are beginners and are getting into this and don't really know what all this code stuff is.
3:41:41On a fundamental level, that's essentially what it is. Putting files in the right folders and each one doing a specific thing. You don't have to know what the code actually is.
3:41:48So I can copy the link. I can paste it in my browser.
3:41:55That's nice. Born to perform. So we can see we have the the little page before we get in the website.
3:41:59We have this here, the animation, the car being broken into different pieces, and then coming back in.
3:42:06400 follows function. Okay. The speed, maximum power, just sick.
3:42:13So a bit of writing giving us some context, and that's it. To be fair, it did actually a very good job for it to be like a first time kind of thing.
3:42:20Alright. So now we can go to the next step which is called review. Now reviewing it means that we have to look at the website and say, Do we want any changes with the way that it's laid out?
3:42:29I can already see a few changes that I want. So for example, I want something happening in the background here. So some sort of lines or or dots.
3:42:35I have a vision of it, but I have to see. Then right down here, I would want the car to be positioned on the right hand side and then the writing to come only from the left hand side, not the right hand side because when it goes over the car, it doesn't look nice. So we can take a change by change.
3:42:48The first change is having something within the background here. So I can go to 20first. Dev which is a software that we use to pull bits and pieces of different website animations, motion graphics, and so on that we can use for our website.
3:42:58So I said I wanted something in the background. So what we could do can look here, shaders. We can try this one.
3:43:04Background paper shaders. We can copy the prompt and I can go back to ClawCode.
3:43:09So we can see we're using 54% of the context. And by the way, the more context you use, the worse the quality gets. So I can put clear conversation.
3:43:16Go to bypass permissions and I can say, okay. So I'm looking at the website and I want you to implement the following thing at the start of the website and I can paste the prompt. So essentially twenty first dev, to be able to get this, we have the prompt that allows Cloud Code to understand exactly what it needs to do and how it needs to update the code.
3:43:33And as you can see here, it's a full full thing. I can press bypass permissions, and we can see here.
3:43:38Because, honestly, I just wanted to update, and that's it. It just finished implementing the change. If I go here to the website, I can refresh.
3:43:45We can see now in the background, it's very very subtle which I really like but if you zoom in, you can see that we have different things changing in the background which makes it a lot nicer to to look at visually. Now that's the first thing. The second thing is putting the car to the right hand side and having the text go on the left hand side.
3:44:01Okay. So second thing that I want you to change is on the website itself, the car takes up the whole space. What I want to happen instead is I want the car to be on the right hand side.
3:44:11So I want it to take, like, basically 75% of the whole screen from the right to left, 75%.
3:44:17And then the other 25% that's left on the left hand side, then I want that to be the space where text comes through because I don't want text to overlay over the car. That makes sense.
3:44:26Keep the animation, keep the video, but make the car smaller. Yeah. Position it on the right hand side of the screen.
3:44:30I'm gonna go on plan mode for this one just because I wanna make sure that it understands what I'm talking about. And a lot of this is just talking to Claude, going back and forth as to what it is that you want. Alright.
3:44:38So here it made the plan for the change. The context implementation plan, 25% to the left. Okay.
3:44:45Good. Moving everything to the left, the card to the right. Everything looks good.
3:44:48Then I can press yes and auto accept, and we can go back to bypass permissions to have Clockwork make the changes. Alright. So now I just finished making the change.
3:44:56I can refresh, and we can see that all the text has been on the left hand side. And then this also is fine. Um, I can go down.
3:45:02And as I scroll through, we can see that all the text is on the left hand side, which looks a lot better than us having to have the text over the car, and that covers the whole website step by step. Now that we're done with step three, we can go to the last step, which is deployment. So the way that we deploy ClawCode websites is through a two step process.
3:45:18The first thing that we do is we push all the ClawCode into GitHub. And then from GitHub, we deploy it to Vercel. So the first thing we wanna do is go to github.com and want to make an account.
3:45:29Once you've made an account, you can press new. We're making a new repository, and we can name this website Claude Code.
3:45:36We can press create repo. And so, basically, what's happening is that we're gonna be using Claude Code, and we're gonna be giving all the code here into GitHub.
3:45:43And from GitHub, it's gonna go to Vercel to then deploy it there. All I have to do here is press copy, and I can go I can clear the conversation. I can say, okay.
3:45:53Now I want to push this to GitHub. Here's the GitHub link, uh, repo that I've just made. I can paste this, and I can go bypass permissions because I know it will do a good job because it's a pretty fast task.
3:46:01It should tell us to log in into GitHub because I didn't actually give it any credentials here. So you can see here GitHub needs your identity configured, run these two commands in your terminal, replace with your actual name and email.
3:46:10I don't know how to do that. Can you tell me how I should do it, what I should give you to make that happen? I do this just because a lot of you guys don't know how to use a terminal, so that's fine.
3:46:17But in this case, what I'm gonna do is give it my name and then the email that I'm using.
3:46:25Alright. Now it's saying that it's all live. So if I go here, I can refresh, and I'll have all the different files and folders of this product.
3:46:32They are automatically added here. And the last step here is to deploy it on Vercel.
3:46:37I can press start deploying. I can continue with GitHub. And once you connect to GitHub, make sure to choose the correct product that you're working with.
3:46:43In this case, for me, was website clock code. I can import, and I can press deploy. And as you can see here, well, now we have our project, so our website deployed.
3:46:51And now it says congratulations. You just deployed a new project to Mikaela Pass Projects. I can press continue to dashboard.
3:46:57And if I use this link right here, you can see that I have the website itself, which is sick. Boom.
3:47:06There we go. And this literally took two minutes to deploy, which is absolutely insane. And now for the final thing is answering the question, what if you want changes?
3:47:14What if you're here and you wanna change to a website? You have to go to Vercel. You have to go to GitHub.
3:47:17Well, the beautiful thing about GitHub and Vercel and the whole process is that you just have to go to Clockcode. You have to tell Clockcode what you want to change, and that will automatically update the GitHub, which will automatically update Vercel. And that's the whole flow of things to make sure that you don't have to go to two different platforms at the same time and do any extra stuff.
3:47:36You can just talk to ClawCode simply and it will automatically do everything for you. Alright, guys. So congrats on finishing the full ClawCode masterclass where you went from a complete beginner in ClawCode all the way to building a gentle workflows, deploying them, and using ClawCode to build actual products from scratch.
3:47:53Now everything, all the resources and templates of this video are gonna be stored in the second link down below. In my private community, you can go to the classroom section, go to the templates vault, and you'll find everything there. Alright.
3:48:02So now you've watched how to build with ClockCode. Your next question might be, how do I build a business with it? Which is why I put together a full AI agency master class that you can check out on the screen, which goes through everything that you have to know as a complete beginner to make money with AI.
3:48:15With that being said, I really hope you found value from this video, and I'll see you in the
The Hook
The bait, then the rug-pull.
Four hours, five modules, and a single promise: go from never having coded to shipping agent workflows that run without you and building products you can sell. Michele Torti runs a six-figure AI automation agency on Claude Code, and this is the course in the order he wishes he'd had.
Named ideas worth stealing.
00:30list
The 5-Module Roadmap
- Foundations
- Customization
- Prompt Engineering
- Agentic Workflows
- The Build
The full course structure, taught in the order he wishes he'd learned it.
Steal forany beginner-to-pro course outline
21:40list
CLAUDE.md Four Pillars
- Product description
- Tech stack
- Rules & guardrails
- Folder structure
The four sections every CLAUDE.md should cover so the agent has full project context on load.
Steal forsetting up any new Claude Code project
2:35:00model
Three-Layer Agent Architecture
- Directives (what to do)
- Orchestration (decisions & error handling)
- Execution (code, APIs, data)
A CLAUDE.md folder pattern that separates responsibilities to fight the compounding-error problem of chained LLM steps.
Steal forstructuring reliable agentic workflows
1:14:10list
Four Hook Lifecycle Events
- PreToolUse (before a tool runs)
- PostToolUse (after code is written)
- Stop (when Claude finishes)
- Notification (when Claude needs you)
The moments a hook can fire, all configured in settings.json at the org level.
Steal foradding guardrails and automation to any Claude Code setup
1:25:50concept
Skills vs Hooks vs MCP
- Skills = instructions
- Hooks = guardrails
- MCP = new abilities
The clean mental model for the three customization layers and how they combine.
Steal forexplaining Claude Code customization to beginners
2:05:00list
Sub-Agent Tool-Access Tiers
- Read-only (look but do not touch)
- Balanced (read, edit, run, no delete)
- Full access (read, edit, write)
Three permission levels for sub-agents, trading power for safety.
Steal forsafely scoping any delegated agent
2:08:20list
Three Claude Models
- Haiku (fast, cheap, simple lookups)
- Sonnet (balanced, most daily work)
- Opus (most powerful, hard debugging)
When to pick each model for a sub-agent.
Steal forcost-optimizing agent model choice
2:23:20list
Agent-Team Repeat-and-Improve Cycle
- Collect
- Organize
- Execute
- Review
The four-step loop an agent team runs, just like a real company department.
Steal fororchestrating multi-agent projects
2:33:20list
8 Hacks (three buckets)
- System design: folder architecture, skills, spec-to-code
- Execution/performance: permission modes, parallel sub-agents
- Control/optimization: context management, Pixel Agents
Eight workflow hacks grouped into three categories.
Steal forleveling up an existing Claude Code workflow
2:41:40model
Spec to Todo to Code Workflow
- Write a detailed spec.md
- Generate a to-do list per milestone
- Build one milestone at a time
A structured build process that trades upfront planning for far less rework.
Steal forbuilding any scalable app with Claude Code
2:53:20list
Four Deployment Platforms
- Modal (easiest, bursty, Python)
- Railway (continuous, GitHub, DB)
- Render (battle-tested, flat rate)
- Replit (code+deploy, agentic)
The four ways to host a Claude Code workflow so it runs 24/7.
Steal forchoosing where to deploy an agent workflow
2:54:10list
Universal Debug Loop
- Open the platform logs
- Copy the red error message
- Paste it into Claude Code
- Redeploy and re-check
The identical debugging process across every deploy platform.
Steal forfixing any deployed workflow without touching code
3:12:30list
YouTube Pipeline Four Stages
- Ideation (comments, competitors, Twitter)
- Packaging (thumbnails, titles)
- Video (script, diagrams)
- Post-video (views-to-clicks CTR)
His real content production system, each stage a set of skills.
Steal forsystematizing a content operation
3:33:20list
Website Build Framework
- Plan
- Execute
- Review
- Deploy
The four-step framework used to build and ship the Ferrari landing page.
Steal fordelivering client web projects with Claude Code
How they asked for the click.
VERBAL ASK
00:10link
“Get my 1-1 support to start and scale your AI agency; get my free resources and join my community in the second link down below.”
Two mid-roll pitches for a 90-day 1-on-1 mentorship plus a persistent push to the free Skool community and resource vault; soft and consistent rather than aggressive.
MENTIONED ON CAMERA
Visual structure at a glance.
