The argument in one line.
The most effective Claude workflow is not about better prompts but a three-layer system that trains the AI on your data, pressure-tests outputs against cloned experts and audience members, then ships through a structured engineering process.
Read if. Skip if.
READ IF YOU ARE…
- You use Claude Projects or Claude Code regularly and keep starting each session from scratch with no persistent context.
- You have tried prompt engineering but find yourself tweaking the same prompts manually every time something breaks.
- You want to test ideas against your actual target audience before shipping, without waiting for real feedback.
- You are building a product solo or in a small team and want a repeatable process for going from idea to shipped code.
SKIP IF…
- You are looking for a deep technical tutorial on any single plugin — this is a system overview, not a plugin deep-dive.
- You do not use Claude specifically — the skill architecture here is Claude-native and does not transfer to other AI tools without rework.
The full version, fast.
Claude outputs are only as good as the data and process behind them. The video teaches a six-skill stack in three layers: first, build a persistent knowledge base using web scraping and ingestion so Claude stops starting from scratch; second, run a self-improvement loop and simulate expert and audience feedback before anything ships; third, apply a five-step engineering process so the AI builds toward your actual goal instead of just solving the nearest problem. Each skill compounds on the one before it.
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 – 00:59
01 · Skill 1: Web Scraping
Introduces the two core problems with Claude web search (no JS support, keyword-only) and presents /web-scraping as a utility skill using Fire + Exa plugins.
00:59 – 02:13
02 · Skill 2: Ingest Source
Explains why raw data ingestion is not enough — pre-analyzing data creates a table-of-contents structure so Claude knows exactly where to look in future sessions.
02:13 – 04:11
03 · Sponsor: Cantina
Mid-roll sponsor for Cantina, an AI character creation platform built by Sean Parker. Framed around inside-joke personalized content rather than viral reach.
04:11 – 06:04
04 · Skill 3: Improve System
Five-mode self-improvement skill: Audit (stale info), Skill Review (optimize from history), Experience (capture lived feedback), Historical Review (mine past sessions), Foundation (interview for gaps).
06:04 – 07:49
05 · Skill 4: Ask the Board
Clone professional experts by ingesting their public profiles; create an advisory council. BuildPartner.ai mentioned as pre-built alternative.
07:49 – 10:39
06 · Skill 5: Internal Focus Group
Clone real named people from your target audience as individual agents. The side effect: building this skill forces you to define exactly who you are building for.
10:39 – 14:05
07 · Skill 6: Modern Engineering
Five-step engineering process (brainstorm, plan, work, code review, debug) via Compound Engineering plugin. Three reasons it beats YOLOing: AI ignores big picture, bad output review costs more than planning, and the habit compounds.
Lines worth screenshotting.
- Claude starts every session from scratch unless you deliberately ingest and pre-analyze your data into a persistent knowledge base.
- Pre-analyzing ingested data is like building a table of contents — Claude knows exactly where to look instead of scanning everything.
- Never fix a problem more than twice. If it happens once it might be a one-off; if it happens twice, build a skill to prevent it.
- Manually tweaking prompts to improve output is too much work, which is why almost nobody does it consistently.
- Cloning experts works best when the person has an extensive online profile, because that public writing is the training data.
- The real value of building an internal focus group skill is the side effect: it forces you to name exactly who you are building for.
- AI defaults to solving the immediate thing in front of it. It does not care about your big picture — you have to supply that.
- Reviewing bad AI output takes more time than a proper plan would have required in the first place.
- The same five-step engineering process works for code, proposals, and product launches.
- A utility skill like web-scraping enhances every other skill you create — build those first.
- Keyword search and semantic search return fundamentally different results; semantic understands intent, not just matching words.
- Each person in your internal focus group should be a real named individual, not a persona archetype — specificity is what makes the simulation useful.
Six skills that turn Claude into a system, not a chat.
WHAT TO LEARN
Every Claude session that starts from scratch is wasted — the six-skill stack here turns a stateless chat tool into a compounding system that learns from your data and pressure-tests your work before it ships.
- Raw data ingestion is not enough — pre-analyzing sources into structured metadata is what lets Claude retrieve the right information instead of scanning everything.
- If you are fixing the same problem more than twice in Claude, it is a signal to build a skill, not adjust a prompt.
- The five-mode improve-system skill addresses five distinct types of knowledge debt: stale facts, underperforming skills, uncaptured lived experience, missed learnings, and missing foundational context about your goals.
- Cloning experts works best when the expert has an extensive public profile — their writing and interviews are the training data that makes the simulation useful.
- Building an internal focus group skill forces you to name the real people in your target audience, which is the thing most builders skip and the thing that determines whether the product lands.
- AI optimizes for the immediate task in front of it, not your larger goal — the engineering layer (brainstorm, plan, work, review, debug) is how you supply that context before the AI runs.
- Reviewing bad AI output is more expensive in time than a structured plan would have been — the cost of skipping planning is paid during debugging, not before it.
Terms worth knowing.
- Utility skill
- A Claude skill that runs automatically and enhances every other skill in a project, rather than being called for a specific task.
- Ingest source
- A skill that pulls content from a PDF or URL, pre-analyzes its concepts, and stores structured metadata into a Claude project for persistent reference.
- Semantic search
- A search method that understands the meaning and intent behind a query rather than matching exact keywords.
- Improve-system skill
- A multi-mode Claude skill that audits your knowledge base, reviews skills, captures lived experience, mines session history, and interviews you to fill foundational gaps.
- Ask the board
- A Claude skill that routes questions to cloned expert agents, each built from the public writing and profile of a real professional.
- Internal focus group
- A Claude skill where real named people in your target audience are cloned as individual agents so you can test work against them before it goes live.
- Compound Engineering
- A five-step structured build process (brainstorm, plan, work, review, debug) applied to AI-assisted projects to prevent the AI from optimizing for the immediate task at the expense of the larger goal.
Things they pointed at.
11:00toolCompound Engineering plugin
11:00toolSuperpower plugin
Lines you could clip.
04:28
“You should never fix a problem more than twice. If it happens once, okay, it might be a one-off issue. But if it happens twice, it is time to fix it.”
01:47
“We are not just ingesting raw data because that is not very useful to Claude. We are preanalyzing data so that it makes it really easy to find specific information. Think of it like creating a table of contents.”
09:30
“In order to create the skill, it forces you to identify who you are actually building for. You cannot create your focus group unless you know who is in the focus group.”
12:03
“Reviewing bad AI output takes more time than creating a proper plan would have ever taken.”
11:56
“AI defaults to solving the immediate thing in front of it as efficiently as possible. And this means that it does not care about the big picture, which is something you do.”
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.
analogy
00:00Skill number one is web scraper. The reality is that AI outputs are only as good as the data you put in. So to help us get better data, we're going to fix Claude's ability to search the Internet because it currently has two major problems.
00:11The first is that it can't get data from a lot of modern websites because they use JavaScript. And the second is that Claude uses keyword searching, but to get better data, you wanna use semantic search, which understands the meaning and intent behind a user's search rather than just matching keywords. To create the skill, we're going to use two plugins.
00:29Fire to access modern web pages and exit to run semantic search. Once those two plugins are set up, which are linked below, I run this prompt to create a skill called web scraping. You can see the prompt on screen for how to create this skill and this is what I call a utility skill.
00:43Anytime I use Claude, this skill can get triggered and it enhances every other skill I create. Now I have access to data all across the internet that I previously wouldn't have access to with Claude. Now that's just the first of six skills I'll walk you through today that you can immediately apply to whatever you're working on.
00:58Skill number two is ingest source. In skill number one, we enhance the data that we have access to. But the problem with that is that we're not storing any of the data so we can't access it later.
01:07And that means that each new session starts from scratch. So let's not do that and instead build our own knowledge base with a ingest source skill. On screen is a prompt to help you create this, but what exactly does this do?
01:18This skill will take any source you attach to it, that's a PDF, a link, whatever, pull the concepts from it, and ingest it into your project for future reference. Reference. And the key with this is we aren't just ingesting raw data because that isn't very useful to Claude.
01:30We're preanalyzing data so that it makes it really easy to find specific information. To visualize how this works, imagine you had a textbook. If you don't preanalyze the data, Claude has to look through the whole thing to find something that's valuable.
01:42But if you preanalyze it, think of it like creating a table of contents. Claude knows exactly where to look. And throughout this video, I'm gonna use a personal trainer example just because it's something that everyone can immediately apply.
01:52So, if you wanna use these first two skills, could say, find the top fitness research in the past three years and pull all of the historical data using the web scraper skill. Then use ingest source to save all the information to my project. It will then ingest the data that you just researched into your project for future reference, so it doesn't have to look for it again.
02:09Now when you ask what should I do as a 36 year old getting back into cardio, Claude has a specific answer based on the information you provided. So we've set up a foundation, but now we have to set up a process to make the system improve over time.
02:22But before we get into any of that, we've covered the massive change in how to think about your data. And there's another massive shift happening right now with AI video, which is exactly where today's video sponsor, Cantina, comes in. Everyone watched the first wave of AI content with Sora.
02:35Short form AI generated content built to go viral with strangers. That came and went. And personally, I don't think that's the future of AI content.
02:43I believe it's in character creation where you are building out an AI persona that resonates with people. And, you can create personalized content that's just for those people. Think about using these characters for inside jokes with friends, not viral clips for algorithms.
02:56And, Cantina figured this out. It's a mobile first platform where you build AI characters that talk, perform, and hold real conversations one on one. Just write out the personality and the voice you want the character to have, which could be human or nonhuman, and then your character shows up in a video message in under a minute.
03:12This is original characters and intellectual property that you own. So for me, I built one that imitates an old Italian character that me and my girlfriend always imitate, and I showed it to her and she was dying. This is just an inside joke that I'm essentially able to create a personality that's a reusable meme with resonates with people in my life.
03:29In my case, it's not about going viral. It's about creating something that's fun with the people I love. It could work for a number of different things.
03:35Right? A fitness coach checking in with a client by name, a creator building a recurring cast for inside joke content. The list goes on and on.
03:43It's all built on the same engine and ultimately, you get to elect the personality that you create. They're actually built by Sean Parker, the creator of Napster, which was one of the most impactful businesses in the music industry.
03:54And now, he's betting on what comes next for AI video content and interactive social media. Cantina is early and creators who experiment now get the opportunity to own a new category. To get started, click the first link in the description and create your first character today.
04:08Now, we've gone through the first two skills, which is about getting better information. Skill number three is about improving your system. As you use Claude, you've likely found yourself running into the same issue multiple times, and it's pretty frustrating, so we want to avoid this.
04:19And, one thing I always say to people I work with is you should never fix a problem more than twice. If it happens once, okay, it might be an one off issue. But, if it happens twice, it's time to fix it.
04:29And, the beauty with AI is that it can learn based on past experiences and build on its past training data so it doesn't make the same mistake again. And, this tweaking of trying to improve the output has existed since ChatGPT came out, but manually trying to tweak prompts so the output is better is just way too much work.
04:46And, what happens is you just don't end up doing it. So the key is making it really easy for you to improve your system.
04:52And that's why you need a improved system skill. On screen is a prompt to help you create this and there's a lot going on here. So how does this actually work?
04:58Well, there's five modes. Mode one is audit. This scans your knowledge base for any outdated piece of information.
05:04Mode two skill review will review your skills and recent chat history to identify ways to improve them the next time they run. Mode three is experience. This is a way to bring into the system your own lived experience, which in my eyes is the biggest differentiator.
05:18Mode four is historical review. This looks at all of your recent Claude sessions for learnings that you might have missed. And mode five, foundation, this reviews your entire setup and interviews you to see if there's anything you should add so the AI knows more about your goals.
05:32For a personal training example, this is what it could look like after week one of using it. You run improved system, here's my feedback on this past week, and then you add feedback about your experience. You run improved system, review my fitness knowledge base, flag any research that agrees or disagrees with what has been working for me, and share the original source.
05:50After doing this, when you go and ask what should I do this week, it'll have a sharper answer based on its previous training data. Now, that's the compounding stage. But how do we know if outputs are actually any good?
06:00That's where the next two skills come in, which is the evaluation layer. Skill four, ask the board. This skill will help you create your own board of advisors to help you make better, more confident decisions.
06:11This is specifically about cloning experts and the idea here is simple. You wanna clone professional experts so you can make decisions with confidence. For example, if you're working on a business, maybe you wanna clone Alex Ramosy, Elon Musk, Mark Cuban.
06:22If you're working on content, maybe mister beast. You're creating your own board of advisors to help you with your decision making process. So step one of this is identify the experts that you want in your project.
06:32If you don't know who would be a good expert to have on your board, here is a prompt that can help you get there. Generally, bias towards experts who have an extensive online profile because that serves as the training data.
06:44For example, if you wanted to clone an AI expert, this channel would be a great place to start. Step two, ingest their training data using the ingest source skill from skill number two. Step three is you create a skill called ask the board so Claude knows to loop through each board member when you ask specific questions.
07:01Here's a prompt to actually use this. Step four is go ahead and ask your board a question. So, you would type slash ask the board and whatever question you have.
07:08So, let's say on your fitness board you had Andrew Huberman and Joe Rogan. You could type slash ask the board. I'm 36, getting back into cardio.
07:15Want to add muscle without losing endurance. What should my first four weeks look like? So then, you would have like Huberman giving you this in-depth scientific answer and then Rogan giving you conspiracy theories and jujitsu advice.
07:26Jokes aside, you get the idea and it's a massive game changer. And if this sounds a little bit complicated, I built a plugin called buildpartner. A I which has a whole board of advisors preloaded, so you don't have to do this.
07:37That's linked in the description and you just call slash expert dash advice. And based on whatever question you're asking, it'll route it to a specific expert. Within that, there also is a slash improved system skill if you want that as well.
07:49Skill number five is internal focus group. This idea is inspired from Amol, who's from Anthropix growth team. Here's Amol talking about how he cloned his manager to help him become a better employee.
08:01I do that for myself as well. So
08:03I basically you know, one of my manager, Ami Vora, was, I think, a podcast guest of yours.
08:10Right? So I say, hey. Based on what you know of Ami, both publicly, she's written extensively about product and then internally, and then our discussions,
08:19what what is every based on everything that I've done or not done this week, what feedback do you have for me as army? And, like, I get that every week. So he described cloning his boss inside Claude so he could get her clones feedback before showing anything to a real person.
08:33And, that's exactly what internal focus group does. This is a skill that lets you run any idea or final product past a group of people who mirror your target audience. Unlike the ask the board skill, which is asking experts for advice, this is asking the actual people who will get the final product.
08:49So, how can you actually set this up? Well, step one is you identify the real people who are on your focus group. The goal of the people in this group is that they emulate the people that will be receiving the final product that you're creating.
09:00And the key here is being specific. It's not person a, it's Brian from college. Step two is you build the internal focus group skill using the prompt on the screen.
09:09Each person will get its own agent with its own context, voice, and biases. And over time, as you learn more about these people and get real feedback from them, you'll ingest more information to enhance their clone over time. And then step three, once you've actually created them, ask your focus group a question to get feedback before it ever goes live.
09:27And what that looks like is slash internal focus group, run this project proposal through our focus group to get feedback prior to my presentation. I do this a ton myself. I have an audience focus group that is the exact group of people that I want my content to resonate with, and so I'll run it through them to make sure that it hits.
09:43And this skill alone has transformed my entire business, but it's not just the skill. It's actually the side effect that comes from it. In order to create the skill, it forces you to identify who you're actually building for.
09:55You can't create your focus group unless you know who's in the focus group. And if you can't actually do that, then you're missing the most important thing when knowing how to build what you're building, and that's the target audience. So if you're able to successfully create this skill, you know who you're building for, which is required to have a successful product.
10:12Now, before we get to the last skill, which will transform how we actually build the final product, if this is your first video, welcome to the channel. But if this is your second or more, here is our anti slob agreement. The visuals, the testing, the time I put into this, this is all for humans.
10:26It's not for AI clankers and robots scraping it. So all I ask is that you subscribe as part of this agreement to help this content reach more people so I can keep doing it. Also, every couple of weeks, I give a Claude Max subscription.
10:36So comment below with what you're building to enter. Skill six is modern engineering. Skills one through five loaded your system and pressure tested the work.
10:45Skill six is the one that actually turns it into something that you can build successfully. Successfully. And there's a broader concept with this skill that most people don't realize.
10:52Everyone watching this is becoming an engineer and I call you a modern engineer. So, I actually went to school for mechanical engineering. I graduated with a four point o.
11:00And after that, I self taught myself software engineering. So I've spent years building things on both the physical and digital side. And the thing that connects everything I've ever built, technical or nontechnical, is a similar process.
11:11One, you plan what you're doing. Two, you define what done actually looks like. Three, you work through clear delivery phases.
11:18Four, you review the work and you troubleshoot if needed. And then five, is you mark it as done and you move to the next thing. And if you apply the same engineering approach to anything you do in the AI world, you'll be amazed at how effective it is to build successfully.
11:30To help you walk through each individual step, a plugin that's helpful is called Compound Engineering. This comes with a bunch of skills out of the box, and there's other plugins like Superpower that you may have heard of. They do similar things.
11:40But what matters is the actual process. So a skill is slash brainstorm is the process of brainstorming ideas to solve the problem at hand. You shouldn't assume that your initial idea is the best solution.
11:51Then there's plan, which will help you turn an idea into a clear structured plan. Then there's work that actually executes the plan for you. Then there's code review, which reviews the output and solution as if it's a senior employee.
12:02And then there's debug, which will help you find the root cause of any issue you run into. And this process of going through each step will work whether you're writing code, writing a proposal, or planning a product launch. And this mindset is so much better than just yolo ing for three clear reasons.
12:16The first is that AI defaults to solving the immediate thing in front of it as efficiently as possible. And this means that it doesn't care about the big picture, which is something you do. The second is that reviewing bad AI output takes more time than creating a proper plan would have ever taken.
12:30And the third is this engineering approach flexes a muscle that will enable you to build progressively more complex solutions, which is a skill that will differentiate you in the AI era. One through each step for the personal trainer example, here's how you could use it. Step one, you would use Brainstorm and say, wanna build a proof of concept web app that allows me to log my workout that I can access on my phone.
12:53Interview me to come up with ideas for features. Step two, plan. Based on what we just discussed, can you create a delivery plan that biases towards the simplest solution?
13:03Assume I will be the only person to use it to start. Step three, you then take that plan and then you execute it. So slash work, complete the plan outlined here.
13:12Make sure to verify that the app works before saying it's done. That last part about verifying forces the AI to think about the verification approach, which will make the product that you actually then review that much better.
13:23Step four, code review. Review the code and flag any issues that may exist. Simplify the explanation so I can understand it.
13:30Then if you run into an issue while you're using it, step five would be slash debug, fix the issue. Take a screenshot, copy and paste it, let AI do what AI does best, is solve problems. So that is the six skills in the exact order that I actually use them.
13:44Three skills to train your system, two to evaluate the work, and one to actually ship it. If you start following this process, you will start shipping much higher quality work at a much faster rate. Now, you like this video, you will love this video about the nine Claude code plugins that will enhance your Claude code setup to build 10 times faster.
14:02It builds on a lot of what I covered here. I'll see you over there. Peace.
The Hook
The bait, then the rug-pull.
The title promises six skills. The video delivers something more useful than a list: a three-layer system where each skill feeds the next, built live around a personal trainer example you can swap for any domain.
Named ideas worth stealing.
10:40model
Three-Layer Claude System
- Layer 1: Train your system (Skills 1-3)
- Layer 2: Evaluate the work (Skills 4-5)
- Layer 3: Ship it (Skill 6)
The six skills stack into three functional layers: data foundation, quality evaluation, and structured execution.
Steal forAny AI workflow or course framework that needs a memorable three-act structure
04:50list
Five Modes of improve-system
- Audit
- Skill Review
- Experience
- Historical Review
- Foundation
A structured self-improvement skill with five distinct modes, each targeting a different type of system debt.
Steal forBuilding a personal AI ops routine or weekly review process
11:00list
Modern Building Process
- Plan what you are building
- Define what done actually looks like
- Work through clear delivery phases
- Review the work and troubleshoot if needed
- Mark as done and move to the next thing
A universal five-step engineering process applied to AI-assisted work to prevent the AI from optimizing for the immediate task at the expense of the goal.
Steal forAny build workflow, project planning session, or AI coding tutorial
How they asked for the click.
VERBAL ASK
13:13next-video
“If you like this video you will love this video about the nine Claude code plugins that will enhance your Claude code setup to build 10 times faster.”
Clean end-card pivot to a related video. Subscribe ask embedded in the mid-video anti-slob segment rather than tacked on at the close.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
Visual structure at a glance.
