Did This Claude Feature FINALLY Kill Prompting For Good?

Everybody's making the same claw skills video right now, but the problem is they're all starting at the wrong layer. So, one way this video is going to be different from the rest is I'm going to show you why this update is actually relevant and what really makes it so special and how you can build on top of it and get ahead of the next update so you don't have to wait for another YouTube video to come out to let you know what you can actually accomplish using Claude. Yes, skills are a big deal. Yes, Anthropic moved the platform forward in a big way with this new update inside of Claw Code. Anthropic integrating skills into their platform as a native feature is more like them replacing everyone's torches and candles with flashlights and in some cases spotlights. It's a more efficient way to manage the workflows that you've already built. The problem with prompts is that you constantly have to reexplain context. repeat yourself with AI. And that's where context engineering comes in. and skills builds largely on that framework of thinking because you're able to capture context. expertise and you're able to use these repeatable workflows, this reusable behavior on a regular basis very quickly, very easily, and get the same quality output every single time. That's what makes Claw Skills special. So, why

is everyone talking about Skills 2. 0 and how does it actually apply to you? because it's an upgrade to claw skills inside of claw code. And I understand that you're probably not someone who uses claw code. So don't check out on the video yet. I'm going to show you how to take advantage of it without claw code. This update allows claw to test, measure, and refine skills by monitoring how the skill actually carries itself out. Where this really shines is that Anthropic once again have integrated user behavior natively into the system into the platform so that it frees you up to be more creative and not have to stop and switch modes and evaluate as much. And again, this is typical behavior of people who use these systems on a regular basis. We identify problems, we look for solutions with the help of AI, and then we integrate them into our workflows, into our skills already. But now they've made it native to the platform, at least for claw code users. So it's great that there's so much content coming out right now about claw skills because it's not just a platform feature that you need to learn about, but it's a user behavior that you need to adopt no matter what your preferred AI system is because claw skills are not just something that you can use on their platform. You can use skills on chat GPT if you simply create reusable behavior workflows, documentation, etc. I have several videos covering that. You can do the same thing on Gemini as well. But the terminology skills is something that the industry is starting to adopt. Google anti-gravity uses it. Codeex by OpenAI uses it. So let's just use it. And so

everyone including Anthropic talks about how to package a skill. They even have a Claude skill creator that will help you create a skill most likely based on the general knowledge that Claude has. But we need to have a conversation about what goes inside of a skill which Anthropit doesn't talk about because a polished markdown file is not going to improve Claw's capabilities in any significant regard. If you have a bad prompt, then you just get a bad response. But if you have a bad skill now you have bad results all over your account. It's a much more dangerous situation and the stakes are higher. So we should slow down right here before we go forward building skills and think about how should I create one? What should go inside of one? What approach should I take? So remember when I told

you about the Makoshi protocol? Let me take you back to it and let me give you a brief glimpse of what was inside of it. I'm not going to read it word for word and bore you, but I want you to see with your own eyes so that you can understand how to build out skills that are actually valuable, effective, and powerful, and that give you the results that you want. We're inside of my chat GPT account, and this is my Makoshi project. One of the first things I want you to notice is that I've always used projects in the same way that Claw defines skills. I've always had a primary prompt with tools and other items that accompanied it. My logic and library system is something that I've been doing for almost a year now. The Makoshi protocol is the primary logic. Every other prompt, every other template is just something inside of the library to help this one achieve its goal. Let's open it up and let's see what's inside of it. Rather than trying to cover this entire extensive workflow, let's just focus on this one section because it captures the nature of what I was doing here. Number one, the goal was to solve 10 domain defining constraints with irreversible logic and store them as lifelong memory blueprint. This was the expertise that we were looking to capture around certain topics and ideas. Because unless you can verify that the skill you're creating actually has the expert knowledge to do what people are already doing in the real world, then you have no idea whether your skill is actually going to work for you. And so yes, you can create a skill with claw, but it's much better if you sit down and do the prerequisite research yourself to gather that information. And that's where platforms like Notebook LM come in handy. That's where my notebook LM wizard comes in place. So you can do desk research and get the information you need to actually build out a useful and valuable skill. Then there was the core law. A construct must earn the right to exist by resolving every constraint to CIQ more than or equal to 99 and operator approval. So with every single construct that I created, which is what I called them back then, there was a rubric that it had to pass. And CIQ simply stands for construct intelligence quotient. It had to have that IQ score. And I'll come back to that in just a moment. The third item was the quadrant system. Every item was labeled as either alignment, path, detour, or misalignment. That way, I could track false and accurate reasoning. Whether it was a solution or step in the wrong direction, I wanted to make sure that everything that was said was documented. And this is probably one of the most important things I did because I wanted to be able to trace the source to the answer. It's almost like when you were in school and the teacher would say, "Show your work. " Some people could look at the problem and just give you the answer. And sometimes AI does that. But the way the teacher got firsthand knowledge and confidence that you knew the work was by having you show your work. And so this way I could actually trace the logical steps and see how we arrived at said conclusions. And then the fourth part was the construct diff logs. Now what this was is a compilation of the entire diff logs from the entire conversation. Massive document but it was essentially the brain. It was every wrong thought, every misaligned thought, everything they thought would work but didn't work, everything that was simply flattery on the surface but wasn't actually the solution, every algorithm that was logically weak or not sound, as well as everything that was good, as well as the breakthroughs, the good ideas, the things that were in the right direction and the solutions. So, how do

we actually get to this result? How do we reach this place? All I have to do is type in start and then it's going to ask me a few questions about my domain and different things like that. But then it's going to print an primary prompt that has been configured with the inputs that I gave it. And it's also going to print an evaluators prompt that has been configured to activate another AI system to evaluate this primary system. So the domain we're choosing is how to create claw skills better than 99% of people. And if you look at the message, it says that it's loading the two prompt files now, adapting them to the domain you chose to so the constraint cycle starts on the right footing. So now this is the primary prompt. This prompt is what I would take and I would run it with chat GPT Claude or Google Gemini Grock, whoever it does not matter. But what really makes Makoshi special is the evaluator prompt. This is the purpose of the evaluator. You are an independent construct evaluator AI. Your task is to assess the quality of a proposed constraint resolution using the CIQ rubric and enforce the standards required for breakthrough grade output. You do not assist the construct. You evaluate, score, and if needed, reject. I would then copy this prompt, go to a different AI system, and then run the prompt over there. And whenever the primary system responds, I would then copy that response and paste it with the evaluating AI and allow it to grade the response. And this was the beginning of me realizing that almost every AI gives you a general quality with each response. And you have to develop unique approaches if you want to increase the quality of your outputs. And so what I did was I would run my primary prompt with chat GPT and then I would run my evaluator prompt with Claude. So the next question became, well, I have Claude keeping Chat GPT honest, but who's going to keep Claude honest? So what I started doing then was bringing Gemini into the sequence. And when I would share with Gemini the arguments that Claude made, Gemini would always say something along the lines of those are some great points that I completely overlook. I acquies and I go along with it. And I realized then that Gemini, although it's a great reasoning model to think with you if you know where you're going, it's not the best problem solving model in my opinion because it has no push back. None at all. I'm not saying it's a bad model. It's just not good for that. So it's great that you can create Claude skills with Claude to automate certain tasks on a surface level, but if you're doing anything serious, you need to take the time to develop your own system. Even if you don't take the approach that I took, if you take another approach, which I still use that approach today, it's just a little bit different, but come up with a system to research first and find out what works in the real world and then evaluate that, to doublech checkck that knowledge, but use an independent system. And then thirdly, find a way to document that before you ask Claw to build the skill because once you do that, it's very simple to build it from

there. On the first layer we have the features claw can do something. On the second layer we have the capability pack. This is where we start packaging behavior into a skill. Number three we have the skill set architecture. So this new feature that Enthropic has added into the platform is great. Being able to create a skill is great. But what we want to start looking at is how can I create skill sets? And then after you learn how to create skill sets, how can I create skill trees? because it's at this level where we really begin to leverage the capabilities and features of these systems. So, even though this was a huge update for Anthropic, and I know that YouTubers are typically over excited about updates and news releases, I wasn't that excited about it because I understood that this is simply user behavior that people need to start adopting now. And they don't need a platform to release a feature for them to start thinking this way and benefiting from it. They don't have to learn claw code to actually start using this type of behavior. You don't even have to switch over to claw from chat GPT to start using skills or the approach that skills is based on. Like I've shown you already, what I've been doing since June and since chat GPT released projects can easily be classified as a claw skill. Even though people told me I was wrong for using the custom instructions as a router prompt or as a system prompt as some people like to say. Now, let's answer the title the right way because prompting is dead is not the most interesting takeaway that you're going to get from this video. The truth is prompting didn't necessarily die. You still use prompts on a daily basis, but it has definitely been demoted. It's not the number one way to use AI anymore. Instead of thinking about asking for information, you should be thinking about giving or installing capabilities and installing superpowers into your model. Most of the time when you're prompting, you're asking questions. You're looking for information. You're a learner. But when you start creating skills, you're someone who's taking action. You're a doer. You want the system to do a certain thing every time the right way. And so, even though Anthropic improved their skills creator, I still think that you should be the primary person creating your skills. Because when you've done the work yourself to gather the necessary information, to do the research, to stress test it, then you have confidence that you've done your due diligence to create something that you actually want. Now, let me show you

just how easy it is to build out skill trees, skill sets, and skills when you've done the prerequisite research. We're going to open up Claude on desktop, and you can see that there's chat, co-work, and code. We're going to go to co-work, and we're going to add a new folder. And if you haven't done so already, I highly suggest that you create a folder for co-work. And you put all of the folders that you're going to be working in within that folder so it's easy for you to find your work when you're done. We're going to select Makoshi and press open. I have two other skills that are closely related to this. So I'm going to tell Claw to use those two skills as inspirations for creating a third new skill. This is the origin of the librarian and the kit builder. Use builder as inspiration to recreate or create a new the Makoshi skill. I've included a folder with the Makoshi protocol which is the logic and several other documents which are the library. So if you look on the screen you can see that co-work has completed my skill based on my months of work. And so right here is the skill. This is the actual logic in my system. And then there's a library. There are references and there are templates. It only took co-work about five to 10 minutes to complete that. And now I can iterate upon it very easily and perfect this by tonight. But you can see that there's a skill. Then there are references and templates. And so the skill would be my logic. And the references and templates that's my library. All I have to do now is open and claude add to library. And just like that, the Makoshi protocol is not just a chat GPT project. It's a claw skill that I can activate simply by telling Claw to run the Makoshi protocol upon an idea, etc. And then I can stress test that idea with other models, which is something I still want to do. I can even take my Liberian skill, which uses desk research methodologies to do real research for me, and use it to actually get the right data to create any expert

that I want. So, while everyone's excited about being able to create skills, what I've always been excited about ever since Chat GPT projects came out until this day is the ability to create infrastructure and architecture around these systems because that's where the real power comes in. Now, the fact that Anthropic is building this natively into the platform, I think puts them head and shoulders above every other platform and it's partly why it's my primary platform now and also because of co-work. But if you want to learn more about skill sets and skill trees, then definitely make sure you check out the next video where we do a deeper dive into those since we really didn't cover them in this video. Or you can check out this other video where I show you how I automate my research because you can't start building real systems until you have real data. The last thing you want to do is build a skill with bad logic and then a skill set based on that tree still based on the same bad logic. So I check you on that other