Transcript: I Tried 100+ Claude Code Skills. These 6 Are The Best

Source video ID: eRS3CmvrOvA

Transcript

• 0:00 — After spending 400 hours in CloudCode, 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.
• 0:31 — 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 want to make money selling AI in 2026, don’t overthink it. Just start with what the businesses are already paying for. All right, let’s get into the first skill. So, the first skill that I 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.md5. You don’t
• 1:01 — have to figure out the formatting. You just talk to Claude like you’re explaining the job to a co-worker, and you walk away with a skill that runs the same every time. You can even drop in like an SOP, and it would be able to turn that into a reusable skill. So, the problem it solves is very simple. Most people who get into ClaudeCode 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
• 1:31 — 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 factory that builds the product. 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.md5, figuring out the trigger conditions, testing it over and over, and it would 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
• 2:01 — package it. So, it just compresses that learning curve. To install this, you can run /plugin 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 product 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. All right, so quick thing before we get into the rest. You’re going to 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
• 2:31 — 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 ClaudeCode behaves under the hood because of that plugin. Also, because I know you guys are going to say so in the comments, I’m not an AI avatar. This is really me here to speak to you guys. But anyways, for this video, I’m going to be calling some of these [music] 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
• 3:01 — 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 ClaudeCode, which is rushed or sloppy code. You ask for something, Claude sprints to write it, 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
• 3:31 — apart. So, Superpowers 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 ever reaches your
• 4:01 — 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 150,000 stars on GitHub. So, to install it, just run this command right up here. I’ll put this in the description. And by the way, I also made a full YouTube video breaking down this Superpower skill and talking about the
• 4:32 — tokens that it actually spends. So, you can check that out right up here. All right, skill number three is called GSD, which stands for get done. And if Superpowers is about how Claude writes code, then GSD is about the environment that Claude actually writes that code in. Every ClaudeCode user eventually hits that wall, right? You start a session, things are going great, Claude’s nailing everything, and somewhere around halfway through your context window, it starts to fall apart. You know, code gets sloppy, Claude forgets requirements that you set at [music] 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
• 5:02 — this is essentially called context rot. Now, GSD fixes this with context engineering. Instead 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 production 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
• 5:32 — 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 Claude a 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 want to be up front 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 first place. So, Superpowers gives Claude the process, plan, » [music] » test, review, repeat. GSD gives Claude
• 6:03 — 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 /gsd-help inside of ClaudeCode 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 ClaudeCode, and most people just aren’t using it. It’s the /review command and its more powerful cousin /ultra review. So, when you finish writing code and you type /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
• 6:33 — would be flagging. It runs locally, it’s fast, it costs you nothing beyond your usual usage tokens. And then we have /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 going to get a pile of style nitpicks or false positives, you’re going to be getting confirmed
• 7:03 — bugs. Practically, I mean, this fits at the end of any ClaudeCode workflow. You plan with Superpowers, you execute with clean context using GSD, and then before you merge anything important, you run /ultra review to catch the bugs that Claude missed on its own. A few things to know before you run it, you need ClaudeCode version 2.1.86 or later, you need to be signed in with a Claude account, an API key alone won’t work. And when you run the /ultra review, it could take, you know, 10 to 20 minutes, but it runs in the background so that you can keep working. You’ll want to use /review for fast feedback on everything, and you can use /ultra review when you’re about to merge
• 7:34 — 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 ultra review may cost you. And also, quick heads-up, ultra review 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 five to 20 bucks per run depending on size. Once again, you don’t need to go install this if you’re on ClaudeCode 2.1.86 or later, just type /review or /ultra review in your terminal, and it should work. All right, this fifth plugin is context mode. So, here is the
• 8:05 — 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 30 minutes of real work, 40% of your context window is just garbage, 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
• 8:36 — fetches a URL, Context Mode routes that call through a sandbox. So, your code runs in an isolated sub-process, the 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 5 kilobytes total. That’s obviously from their published benchmarks, and you can see your own numbers by running /contextmode:ctx-stats at any point. Now, second, and this is
• 9:06 — 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 [music] intact. Practically, what does this mean for your ClaudeCode projects? Sessions that used to fall apart around the 30-minute mark now run for 3 hours because you don’t hit any slowdown. You stop wasting
• 9:37 — prompts reminding Claude of the context it already had. So, to install this one on ClaudeCode, just type in these two commands right here, and then you restart ClaudeCode, and that’s basically it. The plugin auto-installs the MCP server, the hooks, and the routing instructions. All right, and now this last skill is called ClaudeMem. And if Context Mode keeps your current session clean, then ClaudeMem is what carries knowledge across all of your future sessions. ClaudeCode 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 10 minutes and thousands of tokens just getting Claude caught up. Now, you
• 10:08 — can try to solve this with things like your Claude.md 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 kind of 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, Claude Mem 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
• 10:38 — 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 Claude.md files and updates them as you work. So, your project documentation literally writes itself while you code and it’s automated. Now, the part that actually saves tokens is the retrieval. So, instead of dumping every past observation into your context, Claude Mem 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.
• 11:08 — 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 Claude Code projects, you stop paying the startup tax on every new session. You stop repeating yourself. You pick up a project where you haven’t touched in 2 weeks and Claude 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 Claude remembers about your project and you can look through it yourself. Now, inside of Claude Code, you can install this using the plugin
• 11:38 — 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
• 12:09 — Claude 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 product called Claude Design. But if you do bring that project back into Claude Code, you’re going to want to use this skill. Okay. So, now that you know the 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 10 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.
• 12:39 — That’s what they actually want to 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 business owner a demo. They won’t really care about your experience, they’ll care about the value of what you’re showing them. And once you start to understand how these 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
• 13:10 — experience is going to 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 want to see exactly how I make money in 2026, then check out this video right up here. But anyways, that is going to do it for today. So, if you guys enjoyed or you learned something new, then please give 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.