§ 01 · The Hook
The bait, then the rug-pull.
Nate opens by burning the rest of the genre — '100+ skills tested, most don't matter' — then immediately narrows the field to six. The title promises a filter and the first 22 seconds deliver the filter's criteria: save time, save money, or remove mistakes. From there it's a pure curation play built on his 400-hour credential and his portfolio of real-estate, HVAC, coaching and marketing clients.
What the video promised.
stated at 00:22“In this video I'll show you what they are, why they sell, and how you can use them to sell AI automations.”delivered at 22:50
Where the time goes.
00:00 – 00:47
01 · Cold open: 400 hours, 6 skills
Pattern interrupt that says most YouTube-famous skills are useless to real clients. Names the criteria: save time, save money, remove mistakes — and promises the six that pay.
00:47 – 01:30
02 · Credential drop + promise
Lists the verticals he's shipped to (real estate, HVAC, coaches, marketing) to earn permission, then restates the value: 'if you want to make money selling AI in 2026, start with what businesses already pay for.'
01:30 – 04:00
03 · Skill 1 — Skill Creator (Anthropic)
Official Anthropic skill that writes other skills from plain-English descriptions. Frames it as 'the factory that builds the product' — not what a client pays for, but what makes every other paid skill possible. Install command: /plugin install skill-creator@claude-plugins-official.
04:00 – 04:37
04 · Skill vs plugin disclaimer
Quick aside: a skill is a markdown file teaching Claude one job, a plugin is a bigger package that can bundle skills + hooks + MCP servers. Also addresses comments-section conspiracies that he's an AI avatar.
04:37 – 07:05
05 · Skill 2 — Superpowers
Community skill (150K GitHub stars) that forces Claude to plan, test, review in two stages, and work in an isolated environment before writing code. Pitch: gets you from 60% to 80% one-shot quality, cuts debugging cycles and token costs. Tied directly to client-grade production work.
07:05 – 07:59
06 · Skill 3 — GSD (Get Shit Done) intro
Names 'context rot' — the mid-session collapse where Claude forgets requirements and cuts corners — as the wall every Claude Code user hits.
07:59 – 10:15
07 · Skill 3 — GSD context engineering
Solution: spawn fresh sub-agents per task, each with a clean context window. Built-in quality gates catch scope reduction and enforce security. Autonomous mode lets it plan/execute/commit unattended. Pairs with Superpowers — Superpowers gives the process, GSD gives the clean environment to run it.
10:15 – 12:50
08 · Skill 4 — /review and /ultra-review
Built into Claude Code, not installed. /review runs a fast local structured code review; /ultra-review (launched with Opus 4.7) uploads your branch to a cloud sandbox, spins up parallel reviewer agents, and only surfaces independently reproduced bugs. Requires Claude Code 2.1.86+ and a Claude account. Pro/Max get 3 free runs, then $5–$20 per run.
12:50 – 15:20
09 · Skill 5 — Context Mode (the garbage filter)
Every tool call dumps raw output into the context window — Playwright snapshots, GitHub issues, logs. After 30 minutes, 40% of context is junk. Context Mode routes tool calls through a sandbox subprocess and returns only the slice Claude needs: a 56 KB Playwright snapshot becomes 299 bytes.
15:20 – 16:50
10 · Skill 5 — Context Mode session memory
Second half of Context Mode: tracks every meaningful event (file edits, tasks, decisions, errors) in a local SQL DB so when Claude compacts, it rebuilds the session snapshot and injects it back. Sessions that died at 30 min now run 3 hours. Install with two slash commands + restart.
16:50 – 19:40
11 · Skill 6 — Claude Mem (cross-session memory)
Where Context Mode keeps the current session clean, Claude Mem carries knowledge across sessions. Hooks into the lifecycle, auto-captures edits/decisions/bug-fixes via Claude's own agent SDK, compresses into semantic summaries, stores in local SQLite with vector search. Auto-generates folder-level CLAUDE.md files. Three-layer retrieval claims ~10× token savings vs dumping everything at session start.
19:40 – 20:35
12 · Bonus 7 — Anthropic Frontend Design
Official Anthropic frontend-design skill — installs globally, makes Claude Code's UI output look 'a lot less AI generated.' Notes this is now baked into Claude Design (Anthropic Labs) but you still want the skill installed if you bring Claude Design output back into Claude Code.
20:35 – 22:50
13 · Sell the outcome, not the workflow
The actual business lesson: stop pitching 'I'll build you an AI workflow.' Pitch saving 10 hours/week, cutting admin mistakes, speeding up or growing leads. Beginner playbook: pick ONE of the six, build a few workflows with it, show a demo — the business owner doesn't care about your experience, they care about value. CTA to free Skool community resource guide + next video.
Visual structure at a glance.
Named ideas worth stealing.
00:18list
Three filters for client-payable AI skills
- Save time
- Save money
- Remove mistakes
Nate's whole curation criteria. Anything that doesn't pass one of these three doesn't make the list — and these are also what the business owner is actually buying.
Steal forany pitch deck or sales page where you're selling productivity tooling — replace the feature list with which of the three filters it satisfies
01:00list
The 6 (+1) Claude Code skill stack for AI agencies
- Skill Creator
- Superpowers
- GSD (Get Shit Done)
- /review + /ultra-review
- Context Mode
- Claude Mem
- Frontend Design (bonus)
The full curated stack — Skill Creator is the factory, Superpowers gives Claude the engineering process, GSD gives it clean sub-agent context, /review and /ultra-review catch bugs pre-merge, Context Mode trims tool-call garbage in-session, Claude Mem carries memory across sessions, and Frontend Design fixes the AI-generic UI look.
Steal fora one-page 'starter stack' lead magnet for any creator audience — Joe could do the exact same listicle for video-editing or content-creator tooling
09:40concept
Superpowers + GSD pairing
Superpowers controls HOW Claude writes code (plan → test → review). GSD controls the ENVIRONMENT it writes code in (clean sub-agent contexts). One gives the process, the other gives the substrate the process runs on. They compose.
Steal forany 'two tools that look competitive but are actually complementary' explainer
11:30concept
Pre-merge review ladder
Daily flow: /review for everything (fast, local, cheap). /ultra-review only for high-stakes merges — payments, auth, DB migrations — where production bugs cost more than the $5–$20 run.
Steal forany 'when to use cheap vs expensive AI tier' explainer
20:35list
Sell the outcome ladder
- Save 10 hours/week
- Cut admin mistakes / human error
- Speed up leads
- Get more leads
- More profit
The pitch frame for AI-automation freelancers. Don't sell the workflow, don't sell the skill, don't sell the AI — sell what the business owner already wants to buy.
Steal forevery single offer Joe builds — Mod Producer, Clip Lab, MCN+ — the pitch should always lead with the outcome ladder, not the toolkit
21:30list
Pick one + demo playbook
- Pick ONE skill
- Build a few workflows with it
- Show a business owner a demo
- They don't care about your experience, only the value
The beginner cold-start sequence. Resists the temptation to learn all six before pitching anything.
Steal forany 'how do I get my first client' content — the demo-as-proof loop beats the credentials-as-proof loop
Lines you could clip.
00:12
“They're simple, they're boring, but they are effective.”
01:00
“If you're watching this and you wanna make money selling AI in 2026, don't overthink it. Just start with what the businesses are already paying for.”
02:35
“The skill itself isn't what a client pays for. But every single skill that you're going to sell... they all come out of this one. It's the factory that builds the product.”
07:02
“Superpowers slows Claude down just enough to actually think the problem through, and it doesn't let Claude skip steps.”
07:40
“If you can get somewhere like 80% of the way there rather than just like 60% of the way there, that's a win.”
09:05
“Somewhere around halfway through your context window, it starts to fall apart. And this is essentially called context rot.”
14:10
“After about thirty minutes of real work, 40% of your context window is just garbage.”
15:00
“A 56 kilobyte playwright snapshot becomes 299 bytes.”
16:10
“Sessions that used to fall apart around thirty minute mark now run for three hours.”
20:50
“You need to be selling the outcome. You need to be selling saving the business owner ten hours a week or cutting their admin mistakes.”
How they spent the runtime.
Hook length47s
Info densityhigh
Filler8%
Things they pointed at.
09:40toolGSD — Get Shit Done plugin
10:15tool/review (built into Claude Code)
10:40tool/ultra-review (built into Claude Code 2.1.86+)
12:50toolContext Mode plugin
16:50toolClaude Mem plugin
19:40toolAnthropic Frontend Design skill
20:10productClaude Design (Anthropic Labs)
22:20linkNate Herk free Skool community + resource guide
How they asked for the click.
22:20newsletter
“I threw all of this into a resource guide that you can access for completely free inside of my free school community. You can check that out using the link in the description.”
Soft, low-pressure pitch tucked after the big 'sell the outcome' lesson. Followed immediately by an end-screen video card and a like-the-video ask. Note: he says 'school' but means Skool — the spelling drift is verbal habit.
Word for word.
HOOKopening / re-engagementCTAthe pitchmetaphoranalogy
00:00HOOKAfter spending four hundred hours in Cloud Code, I noticed something. Most people online are developing fancy skills for the sake of a cool video, but businesses don't actually want that. They want six types of skills. They're simple, they're boring, but they are effective. Skills that save time, save money, or remove mistakes. And the crazy part is these six skills are the ones that clients will pay the most for because they allow you to build the best agents at the lowest cost. So in this video, I'll show you what they are, why they sell, and how you can use them to sell AI automations. But before we jump into the first one, let me show you how I figured all of this out. I've worked with real estate agencies, HVAC companies, coaches, marketing agencies, and more, and I've seen the same patterns across every business. They're different industries, sure, but they face the same problems. And these skills keep showing up again and again. So if you're watching this and you wanna make money selling AI in 2026, don't overthink it. Just start with what the businesses are already paying for. Alright. Let's get into the first skill. So the first skill that I'd install is the one that builds every other skill, and it's called the skill creator. It's one of the official skills from Anthropic. You basically just describe what you want in plain English. Claude drafts the skill, tests it, iterates on it, and it packages it into something that you can use forever. You don't have to manually touch or edit a skill dot m d file. You don't have to figure out the formatting. You just talk to Claude like you're explaining the job to a coworker, and you walk away with a skill that runs the same every time. You could even drop in like an SOP, and it would be able to turn that into a reusable skill. So the problem that solves is very simple. Most people who get into clogged code try to write skills manually, and they immediately hit a wall. They don't know the structure. They don't know what makes a skill reliable versus flaky. So they either give up or they cobble something together that breaks the second a client touches it. So the skill creator removes all of that headache. Now, honestly, the skill itself isn't what a client pays for. But every single skill that you're going to sell to an HVAC company, a real estate agency, a coach, whatever, they all come out of this one. It's the . So indirectly, this is the skill that can help make all of the other dollars possible. So here's a simple example. Let's say a real estate agency tells you they waste hours every week writing property descriptions. Without Skill Creator, you'd be writing out the whole skill dot m d file, figuring out the trigger conditions, testing it over and over, and it will probably break a few times, and you could obviously get to a spot where you have a good working skill. But with the Skill Creator, you can just explain it more naturally, and it will help you build, test, and package it. So it just compresses that learning curve. To install this, you can run slash plug and install skill creator. I'll put the full command right up here and in the description, and I'll include all of the actual commands that you'll need in the description so you can just copy and paste them. I personally installed this skill globally on a user scope, so no matter what project that I'm actually working in, if I ever need to build a skill, it just automatically invokes this, and I don't even have to think about it at all. Alright. So quick thing before we get into the rest. You're gonna hear me use the word skill throughout this video, but technically some of these are plugins. A skill is basically a package that you install that teaches Claude how to do a specific job better. It is the markdown file, but a plugin is a bigger package around that skill because it can contain multiple other skills plus extra things like hooks or MCP servers. And all of that changes how ClawdCode behaves under the hood because of that plugin. Also, because I know you guys are gonna say so in the comments, I'm not an AI avatar. This is really me here speaking to you guys. But anyways, for this video, I'm gonna be calling some of these skills because the point isn't the technical category, it's that each one of these makes Claude noticeably better at its job. Anyways, the next skill is called superpowers, and this one literally gives Claude superpowers. It's a great name. It forces Claude to work the way that a senior developer works. Meaning, instead of jumping in and writing code right away, Claude will now step back and plan the whole thing first. It works in an isolated environment so nothing breaks your main project. It writes tests before it writes code. It brainstorms and it reviews its own work in two stages. Once for whether it actually matches the spec and once for code quality. The problem it solves is the number one failure mode when people use Claude code, which is rushed or sloppy code. You ask for something, , and on the surface it looks fine, but then when you start to run it and use it, it falls apart. Or worse, your client runs it and it falls apart. So superpower slows Claude down just enough to actually think the problem through, and it doesn't let Claude skip steps. So really any project where you're building software or automations that will actually be put into production needs this skill. An HVAC company paying for a dispatch system, a marketing agency paying for a custom reporting tool. These clients don't care how many lines of code you wrote. They care that the system works when their business depends on it. Without a framework like superpowers, Claude might write the whole automation in one shot and miss an edge case. But with superpowers, Claude can plan it out. It can write the tests. It can find edge case scenarios, and it can catch those gaps before it ever reaches your client. Now it's not realistic to say that it's still going to be able to one shot everything. Ultimately, when you're building, a large part of your job is QA, quality assurance. And on your first pass, if you can get somewhere like 80% of the way there rather than just like 60% of the way there, that's a win. Fewer debugging cycles means less time spent fixing things and lower token costs. This is also why superpowers is one of the most popular community skills out there with over a 150,000 stars on GitHub. So to install it, just run this command right up here. I threw this in the description. And by the way, I also made a full YouTube video breaking down this superpower skill and talking about the tokens that it actually spends. So you can check that out right up here. Alright. Skill number three is called GSD,
04:37HOOKwhich stands for shit done. And if superpowers is about how Claude writes code, then GSD is about the environment that Claude actually writes that code in. Every Claude code user eventually hits that wall. Right? You start a session. Things are going great. Claude's nailing everything. Then somewhere around halfway through your context window, it starts to fall apart. You know, code gets sloppy. Claude forgets requirements that you set at the beginning of the session. It starts skipping steps, and it actually cuts corners, and it might tell you things are done when they're not. And this is essentially called context rod. Now GSD fixes this with context engineering.
05:06HOOKInstead of letting one big session slowly degrade, it spawns fresh sub agents for each task, and each one has a clean context window focused on only what they need to know. So your main session stays clean, each task gets the full power of Claude's context window instead of just the leftovers. It also has built in quality gates that catch real problems during execution. Scope reduction detection catches the planner when it silently drops a requirement that you asked for. Security enforcement anchors verification to your threat model. These aren't things that you'd manually think to go check through, especially if you're not coming from a software engineering background, which is exactly why having them automated really matters. There's also an autonomous mode if you want to hand claw to spec and just walk away. It'll plan, execute, commit, and work through the whole project without you having to babysit it as much. Now one thing I wanna be upfront about is that GSD isn't really a token saving plugin. All of these sub agents cost tokens, but what it saves you is the hours you'd otherwise spend redoing work that Claude broke because it forgot what you asked in the place. So superpowers gives Claude the process, plan, test, review, repeat. GSD gives Claude the clean context so that processing actually works every time. So to install this one, just run this command in your terminal, and then you can also type slash g s d dash help inside of Claude code to see what other commands are available within that plugin. Okay. So the fourth one isn't actually a skill that you install. It's already built into Claude code, and most people just aren't using it. It's the slash review command and its more powerful cousin slash ultra review. So when you finish writing code and you type slash review, Claude does a structured code review on what you just built. It looks for bugs, edge cases, and design issues, all of the stuff that a careful reviewer would be flagging. It runs locally. It's fast. It costs you nothing beyond your usual usage tokens. And then we have slash ultra review, which launched alongside Opus 4.7. Now this one, instead of running locally, it uploads your branch to a cloud sandbox and spins up a fleet of reviewer agents in parallel. Each agent will attack your code base from a different angle, logic, security, performance, edge cases. And here's the part that's really cool. Before any bug shows up on your list, it has to be independently reproduced and verified. So you're not gonna get a pile of style nitpicks or false positives. You're gonna be getting confirmed bugs. Practically, I mean, this fits at the end of any Claude code workflow. You plan with superpowers, you execute with clean context using GSD, and then before you merge anything important, you run slash ultra review to catch the bugs that Claude missed on its own. A few things to know before you run it, you need Claude code version two point one point eight six or later. You need to be signed in with a Claude account. An API key alone won't work. And when you run the slash ultra review, it could take, you know, ten to twenty minutes, but it runs in the background so that you can keep working. You'll wanna use slash review for fast feedback on everything, and you can use slash ultra review when you're about to merge something that actually matters, like a big refactor or anything touching payments or auth or a database migration. The kind of commit where a production bug costs way more than the time and the tokens that UltraView may cost you. And also a quick heads up, UltraView isn't free. Pro and max plans get three free runs to try. It might have changed by the time you're watching this. But after that, it might cost 5 to $20 per run depending on size. Once again, you don't need to go install this if you're on Claude code 2.186
07:59or later. Just type slash review or slash ultra review in your terminal, and it should work. Alright. This fifth plugin is context mode. So here is the problem. Every tool call that you make, Claude dumps raw data into your context window. So a playwright snapshot might be 56 kilobytes. 20 GitHub issues might be 59 kilobytes. And after about thirty minutes of real work, . Log output, raw data, you know, stuff that you never actually need Claude to look at. And when Claude runs out of space and compacts the conversation, it forgets what files it was editing, what tasks were in progress, and what you last asked it to do. So context mode's job is to fix both halves of these problems. First, it keeps the garbage out. When Claude runs a command or fetches a URL, context mode routes that call through a sandbox. So your code runs in an isolated subprocess,
08:41HOOKCTAthe raw output gets captured, and only the part that Claude actually needs comes back into the context window. Now according to their own benchmarks, a 56 kilobyte playwright snapshot becomes 299 bytes. A 46 kilobyte access log becomes 155 bytes. So over a full session, 315 kilobytes of raw output becomes five kilobytes total. That's obviously from their published benchmarks, and you can see your own numbers by running slash context mode colon c t x dash stats at any point. Now second, and this is the part that I didn't even know I needed until I had it, context mode tracks every meaningful event in your session in a local SQL database. Every file you edit, every task you create, every decision you make, every error that comes up. When Claude compacts the conversation, it doesn't lose any of that. Context mode rebuilds a session snapshot and injects it back in. So the model picks up exactly where you left off with your files, your tasks, and your last prompt intact. Practically, what does this mean for your Claude code projects? Sessions that used to fall apart around thirty minute mark now run for three hours because you don't hit any slowdown. You stop wasting prompts reminding Claude of the context that it already had. So to install this one on Cloud Code, just type in these two commands right here, and then you restart Cloud Code, and that's basically it. The plug in auto installs the MCP server, the hooks, and the routing instructions. Alright. And now this last skill is called Claude Memm. And if context mode keeps your current session clean, then Claude Memm is what carries knowledge across all of your future sessions. Claude code starts every new session from scratch. So you open up a session, and you're explaining the project for the fifth time this month. Because every new conversation costs you maybe ten minutes and thousands of tokens just getting Claude caught up. Now you can try to solve this with things like your Claude dot m d files and your memory files, which obviously are important, and these are plain text files that you write to give Claude persistent project instructions. And those work, but you do have to kinda think about maintaining them by hand. So if you miss an update on one of those files, then Claude will also miss that context. So ClaudeMem basically takes this completely over. It hooks into Claude's session life cycle, and it automatically captures what happened during your session. It'll look at the file edits, the decisions, the bug fixes, the commands, things like that. And then it will use Claude's own agent SDK to compress all of that semantic summaries and stores everything into a local SQLite database with vector search. So when you open up a new session, the relevant parts get injected back in automatically, and it also auto generates folder level claw dot m d files and updates them as you work. So your project documentation literally writes itself while you code, and it's automated. And the part that actually saves tokens is the retrieval. So instead of dumping every past observation into your context, CloudMem uses a three layer search system. It first returns a compact index of observations, and then you can pull a timeline around the ones that matter. And only then does it fetch the full details for the handoff that you need. So this repo reports about 10 times token savings on retrieval compared to dumping everything in at the session start. So practically, all of your Cloud Code projects, you stop paying the . You stop repeating yourself. You pick up a project where you haven't touched in two weeks, and Cloud already knows exactly what you're working on and what you've decided and where you left off. And because everything runs locally on your machine with a web viewer, you can actually see what Cloud remembers about your project, and you can look through it yourself. Now inside of Cloud Code, you can install this using the plug in marketplace with the commands right here on screen. Now one quick warning from the repo itself, you don't have to run this NPM install command. That installs the SDK library only, and nothing actually works because the hooks never register. So stick with the two commands above, and that ladies and gentlemen was number six on our list today. But real quick because I know a lot of you guys are doing things with website design or you know building out slide decks and whatnot. Bonus skill number seven is the official front end design skill from Anthropic. You just install it using the command on screen and I would go ahead and just install that globally. So whenever you're doing things like actually designing things in Cloud Code, it will make them look a lot less AI generated because of that front end design skill. Now this type of stuff also is natively baked into Anthropic Labs new products called Cloud Design, but if you do bring that project back into Cloud Code, you're gonna wanna use this Okay. So now that you know these skills and what they do, here's how to actually sell them. Because I see so many people starting out trying to sell the actual workflow, but you need to be selling something else. You need to be selling the outcome. You need to be selling saving the business owner ten hours a week or cutting their admin mistakes and their human error. You need to be selling the results like speeding up their leads or getting them more leads. That's what they actually wanna pay for. So instead of offering to build an AI workflow, you're actually just offering to save time, cut costs, and help them focus on making more profit for their business. And there's actually a simple way to start doing this. If you're brand new, don't try to use all six of these right away. Just pick one, learn it, build out a few workflows using it, and show a business owner a demo. They won't really care about your experience. They'll care about the value of what you're showing them. Once you start to understand how skills work and how they piece together, you're able to build better automations quicker and hopefully for cheaper, which means you can make more demos. You can make more content. You can automate more parts of your own business. All of that experience is gonna help you communicate the value of what you know to business owners much better. So I know that we covered a ton of information in this video, so what I did is I threw all of this into a resource guide that you can access for completely free inside of my free school community. You can check that out using the link in the description. And if you wanna see exactly how I make money in 2026, then check out this video right up here. But, anyways, that is gonna do it for today. If you guys enjoyed or you learned something new, then please give it a like. It helps me out a ton. And as always, I appreciate you guys made it to the end of the video, and I'll see you all in the next one. Thanks, everyone.
§ · For Joe
Steal the format.
Toolkit-curation listicle playbook
The whole video is one move — burn the field of 100+ options down to six, name the criteria upfront, and tie each item to a paying business pain.
- Lead with a credential-plus-curation hook: '400 hours in, here are the only 6 that matter.' This works for ANY toolkit Joe owns — '14 versions of ModBoard later, here are the 6 features that survived.'
- State the three filters at second 18. Nate's are save time / save money / remove mistakes. Joe's $6 Stack version could be: cuts your bill / runs offline / you own it.
- Give each item the same 4-beat structure: name + 'the problem it solves' + concrete example + install/use command. Repeatable, scannable, edit-friendly.
- Mid-video re-hook every 2-3 minutes. Nate restates the thesis at the top of each new skill ('the next skill is...') — keeps mid-roll retention up without feeling forced.
- Plant pairing logic between items. 'Superpowers gives Claude the process, GSD gives it the substrate' — turns a flat list into a composable system, which is a memetic upgrade.
- Close on the outcome ladder, not another tool. The last 2 minutes pivot from WHAT to HOW-TO-SELL, which gives the value-promise viewer their reason to stay through the CTA.
- Bonus the 7th. Adding a 'and one more' bonus item after the listed count feels generous and gets a second engagement spike at the end.
§ · For You
What this could mean for you.
If you're new to Claude Code and want to skip the trial-and-error
You don't need to learn 100 skills — these six are the ones that actually pay off the time you spend setting them up.
- Install Skill Creator first (/plugin install skill-creator@claude-plugins-official) so every future skill you build comes out reliable instead of flaky.
- If you write any production code with Claude, add Superpowers — it forces the plan → test → review loop that catches bugs before they reach a user or client.
- If your sessions die around the 30-minute mark with Claude forgetting what you asked, that's 'context rot' — install Context Mode to filter tool-call output and Claude Mem to carry memory across sessions.
- Before merging anything that touches payments, auth, or your database: run /ultra-review. It costs $5–20 but catches verified bugs in parallel, not nitpicks.
- Add the Frontend Design skill globally so anything UI-related stops looking like default-AI gradient soup.
- If you're trying to make money with this: don't pitch 'I'll build you an AI workflow.' Pitch 'I'll save your team ten hours a week' or 'I'll cut your admin mistakes.' Then pick one skill, build a demo for one real business, and show the demo — that beats every credential you could list.
