Transcription
So, I was teaching my buddy James Claude skills this weekend and he did awesome. He's already built five of his own and he's running them already. It's a massive level up for him.
But then he text me this morning. He said, "Elliot, what the hell do I do with Claude projects now? Am I still building them or can I just do everything with Claude skills?"
So, this video is for you, James. It's Claude skills versus Claude projects, the difference, why they work together. So, we're going to go through everything and even how to build your Claude skills and your Claude projects so they work neatly together to level you up and create this AI flywheel running again.
Now, through this video as well, I have my AI recipe vault here, which you can access a link in the description. It's got a full document on Claude skills versus Claude projects. We've even got in here some prompts I'm going to use that you can replicate through this video if you want to build the same sort of stuff as we go as well.
But before we start building, we're going to go into a very simple explanation so you understand clearly from the start the difference between Claude skills and Claude projects. A Claude skill is a repeatable prompt that Claude can run on command. So if you have a prompt library where you keep dipping in and out of every day to go and retrieve certain prompts again and again because you use them regularly, you can turn them into a skill so they run automatically anywhere in Claude.
Whereas a Claude project is a knowledge base. It's a deep library of information. So giving you a deeper breakdown on this Claude skills you can think of your processes whereas Claude projects you can think of a place to house all the information behind your business or your work or whatever it is you're using Claude for. So I like to think of Claude skills as verbs to get real technical. A doing word or a doing thing versus Claude projects which are the nouns. They hold knowledge, overarching instructions about how to behave and chat history.
Now, the big big difference here, not just in what they are and how they work, is that Claude skills are reusable anywhere. So, they run anywhere in the Claude ecosystem, in any chat, in any co-work area, even in Claude code. Whereas a Claude project is very specific. It has guard rails and it only runs in that one area of Claude.
And the big mistake, which is why I'm making this video, is that people try and treat these as competitors. They think, should I do Claude skills or should I do Claude projects? The actual truth is they're complimentary. You should build both to supercharge your Claude ecosystem.
So, I want you to build a mental model of Claude being your AI kitchen. If Claude overall is the chef, the operator, then your projects is a stocked kitchen. It's your pantry of ingredients and information, the general rules about how the kitchen operates, and even a log of all of the dishes you've ever cooked in your chat history. Then your skill is your recipe. It's the repeatable process of how a dish comes out each time. So, the name of the dish, the step-by-step cooking method, and even a side drawer with some supporting sauces, little surprises, or maybe just your little spice rack to jazz things up a little bit.
So, let's dig into the anatomy of each. I've got my dashboard up here, but I'm actually going to dig straight into Claude so you can see this live rather than boring you with me just talking at you. We're going to get into something more practical here and actually build these out.
So, to find Claude skills in your Claude dashboard, you're going to go to the left-hand panel here. This is in the desktop browser. You'll see a very similar setup in the web browser as well. You're going to go to customize. This is for Claude skills. When you open customize, you're then going to see the skills tab on the top left here, or you're going to see create new skills. I usually go into this skills area in the top left because then I can see all of the different Claude skills I have running in my account.
Now, the first one I want you to make sure is turned on is something called skill creator. This is really key, and you don't really need to know too much about it, but you need to click on skill creator in your list of skills. This should be inbuilt in every account. And make sure it's toggled on because this is the skill within Claude that helps you build other skills conversationally. So you don't have to know how to code or write all these custom instructions. Claude is going to help you build these.
Now let's break a skill down a little deeper. So you can see the three different sections that is actually contained within them. Obviously at the top you have the name of the skill so we know what it's called. But then we have the description. And this is literally as you'd expect a short paragraph of information that tells you what the skill's about. So, this one converts YouTube transcripts, voice notes, or rough content into fully optimized SEO articles. You'll also see it contains phrases of how to activate this skill. So, anytime I say a phrase or a similar phrase to turn this into a blog post, SEO content writer, SEO article from this YouTube to blog post, this skill is going to activate. So, Claude always has access to these descriptions. And every time you prompt, essentially, it's checking the descriptions and the titles to see whether we should be activating this skill or not. So it can run automatically without you even realizing just from conversational work in your AI system.
The second area of the skill is the custom instructions. This is essentially your prompt. This is where you are putting your repeatable prompt. So you can see here we've got the description of the prompt. We've then got these or custom instructions. We've then got the step by step of what we're going to do. Stage one, extract and plan. Stage two, draft the article. It's got what to do, some voice rules such as no m dashes, British spelling, vary sentence lengths to help make it more engaging writing. Uh how to wrap it up, stage three, even auditing and checking and giving it an editorial pass at the end. This is quite an in-depth set of instructions or a prompt.
We've also got reference files in here, which is the third stage of a Claude skill. This is where some people get a little bit confused and think like, well, should I be actually just building skills instead of projects? Because they can hold on the left here, you'll see you've got your skill.md file, which is the bit we just saw, which is where the prompt is actually kept, all the custom instructions. Then we've got reference files. So, our reference files here, you can see we've got article template. So, I've got an exact template of how we like this the articles to come out for it to reference. So when I reference the template in the main set of instructions, it goes and checks this file of the template. It's also got a checklist here as well.
Now you can build more knowledge into this area, but it is way more limited than a Claude skill. I tend to keep these reference files on the left here as templates, checklists, any short sharp bits of reference information they might need rather than in-depth company or client or information about me and my business. because actually I want these skills to be transferable across anywhere I use them, not just linked to one specific bit of lightweight knowledge.
So that's the anatomy of a skill. How are you actually going to go and build them? Well, I'm going to show you the quickest way to actually start building a skill from a repeatable prompt. So the first thing I want you to think about is every time you grab a prompt and you use it in Claude, a structured prompt or you do a regular process in Claude, I want you to think about turning that into a skill.
Now, there are multiple in-depth ways you can do this, but the first easiest way is I want you to go and grab one of your most used prompts. For example, this is an SOP, a standard operating procedure prompt that I can basically voice and brain dump information, and it comes out with a step-by-step operating procedure for us to repeat and use. If you don't have many prompts as well, you'll see in my AI recipe vault here, we've got my mega prompt chest, which is actually a Google doc filled with, I don't know, well over a 100 prompts. You can see there's a few people in here checking this out at the moment. But we also have this SOP prompt that we're going to use in our Claude skill here.
So, I'm going to take this prompt and I'm going to go back to Claude. And what I'm going to do is I'm going to say something like I want to create a Claude skill. That is going to invoke the Claude skill creator that we turned on earlier. Now you can also invoke a skill, any skill by hitting forward slash and then typing in the skill you want. You can see we can actually forward slash skill creator. That 100% makes sure the skill is turned on. But most of the time I just use it conversationally because they're pretty intelligent about working out which skill we want to be activating from this prompt. Zoom in a little bit for you guys so that it can run an SOP anytime I need. Then I'm going to add the prompt in. And we're simply going to hit run. And that's it. Claude is going to take care of the rest. It's going to wrap up this prompt and build it into the ecosystem so it can just run on command automatically. Even sometimes when you're not consciously, you might be thinking, "Hey, I'm going to build an SOP." And it invokes the skill before you even think about invoking it. You'll see turning this into a skill means it automatically triggers whenever you need an SOP. No copy and pasting required.
Now, once Claude has gone through the process of building this for you, you'll get a page like this. You can check it through to actually make sure you're happy with it. You can see it gives me that description at the top. It gives me the phrases that I can use to trigger this. And it has literally that prompt built in here, even an output format and template. It's gone and built extra style rules, some quality checks before delivering. So, it actually improves and deepens the prompt we're going to use. Now, we can just hit save skill. And that is then saved to our skill library. You can see at the top the Gerber SOP writer cuz the prompt is about like acting like Michael Gerber, the author of the E-Myth Revisited. And that is ready to go. That prompt or that skill is now ready to run anywhere in Claude on command.
And I'm going to show you how it works. But first, I want to build a Claude project to show you the exact difference and then show you how they link together to become a supercharged version of your Claude workflows.
So, inside of a project, the anatomy of a project, there are knowledge bases, custom instructions, chat history, and that's where it all gets a bit muddled in a lot of people's minds. Now to find your projects and if you haven't built Claude projects before then I suggest this is one of the first places to start in your Claude journey is on the left hand side here you'll see projects. If you open up projects you can see all sorts of got things I've got in here from I mean I've got YouTube consultants I've got some clients in here that I do work with. I've got a powerful public speaking coach to help me improve my public speaking right now. I've also built this. This is a fictional company by the way, Kerno Facilities Operations Manager. They're a commercial cleaning company that I've sort of thought up in Cornwall here. And now you can see this new workspace. You can think of this like a sandbox within Claude. This is where you can train it on specific information. Give it a library of information to work with.
So, if you've ever felt like you keep going back into Claude and you keep having to re-explain what your business is about and it doesn't quite get it right and you've built up some memory but it's a bit loose or you actually work with multiple clients, you have multiple businesses, you have multiple areas of your life or different projects you're working on, you should be distilling them into Claude projects. One because they allow you on the left hand side here, you can see we can add files to this. These are files that will always be referenced in your conversations with Claude in these projects. So, you can see here I've got just general information about the company, just even down to the legal name, the headquarters, some various finance numbers in there, where they operate, etc. Just general information about the company. I've got even down to the client base, all their different clients. We've got their people and team breakdown. We've got a document about what makes them different. You could honestly go way deeper. Some of my projects have dozens of files in there from transcripts of conversations with clients to deep details about ICPs. So this is all the knowledge that you want Claude to know about a particular project you're working on or a particular client or a particular workspace area of your business.
The second area which is where people get confused with Claude skills are the custom instructions. These are similar like this is essentially a long set of instructions that Claude should always follow within this project workspace. Now I like to think of these as overarching guidance. So you are the operating brain of Kerno facilities group. We've got what we always do, what we don't do, but it doesn't have specific breakdown about how to behave for each individual task or workflow we would run in it because that would just be way too much information for it to try and go through.
Now the third part of Claude projects are just a nice filing system. You can see all the chats that you've ever had in a particular project. So again, particular department, particular tasks you're working on or particular clients. You could have every conversation you've ever had with Claude about a client nicely lined up here or about the operations of your business.
Now, to give you an example of one that I ran a minute ago, you could just say something simple and you can see how it shortens the need to actually write long prompts in a project is say something like, "Can you identify in my client list the best targets to actually provide more services to to grow our revenue without having to go and find more business?" So, we can just run a really short prompt there. It's already got all the context and information. Sure, you could give it way more in-depth prompt there to actually go and extract and give you structured information, but what you'll see when I ran that prompt previously in this project, we actually get a breakdown of the different clients and where there is more headroom, you can actually see the best candidate here is the most underserved and we could run some adjacent services for them, etc. So, we're able to synthesize through a lot of this information we've got in our project knowledge. That's where it's pulled the information from to actually run the chat. The fact that I don't have to upload a big spreadsheet of information to get it to work with every time actually helps way way way reduce my token cost. So again, if you're running into usage limits and Claude keeps tapping out and saying, "Hey, you got to wait 4 hours to use me again, building Claude projects is going to reduce that token cost dramatically by I don't even know 90% perhaps in some cases than keeping uploading and analyzing and synthesizing new files.
Once you've built your Claude project and your Claude skill, how do you link them together to actually make them work? Remember, we can use Claude skills anywhere across the Claude ecosystem. So that SOP skill that I just built could be used across any Claude project in co-workver. But let's use it in the Kerno facility operations.
Now all I have to do is say I want to create an SOP. that is going to naturally invoke the standard operating procedure skill. Or of course, I can actually do the slash command and go for /gerber SOP writer, but I tend not to do that because conversationally works pretty well. And then I'm just going to brain dump my thoughts about what I want to go in this SOP. So the SOP is for people using the company credit card. To use the company credit card, if you are not in leadership, it is attached to the whiteboard in the office in a plastic wallet wallet in the bottom left. When taking the credit card, you should sign your name on the board with the time and date that you took it. It should only ever be used for van fuel or cleaning supplies specifically for work. The credit card must be returned at the earliest opportunity. You may continue to do a job before returning it, but it should never be taken to do personal errands or be taken home at the end of a shift. It always needs to be returned the same day at the end of your shift. There is a £100 contactless limit on the card. If you need the PIN number, please ring your shift leader and they will be able to give you the PIN over the phone, which should never be written down or stored with the credit card.
And then that's literally all I have to do. I can just dump my thoughts about the process for using the company credit card etc. Then we can let Claude take over. Then it's going to run that skill that we just created and it's going to pair it with the knowledge base in our Claude project. And out the other end we get it has some bits that need some input in there in terms of who the team leader is and any other rules in there that we need to check. But essentially, we've got our new SOP that we can even download as a PDF there. So, we can take it as a PDF, we can add it to Google Drive, or we can actually add it to a project as well.
And when you think about this, it's taken me a couple of minutes to build an SOP and we can add it to a project. Imagine if we had all of these SOPs, these standard operating procedures, 50 or 100 them across the core things of a business, loaded into a project or some sort of knowledge base for AI to use. Now, even the staff at this company, instead of going to the boss and saying, "Hey, how do I use the company credit card?" They can just go to whatever AI platform you want to build these into and question that to see, can you tell me where to find the company credit card? And it will give them not just where to find it, but actually all the other areas about how to use it. So, onboarding new staff becomes an absolute doodle when you systemize a business like this.
But you can see our SOP here. Beautiful. It's always going to come out the same way. Step-by-step procedure, nice bullet points, quality standards, etc., etc. And you can see here, it's actually saying, "Hey, we could do with some related documents here. Can we get some finance and expenses policies uploaded, an employee handbook, etc." Which, as you can imagine, now you can think about how easy it is to start building all of this stuff using just brain dump voice transcripts plus skills plus projects. The flywheel is just incredible once you start linking all of these together.
So to sum up, inside your skill, inside the customize tab, you have the description at the top of what the skill does. You have instructions and you have some supporting files and you can activate them either slash commands or conversationally in your chat. A Claude project is a library of information, your knowledge base of specific details of a business or whatever project you're working on. You have some custom instructions which you can think of the general house rules rather than individual processes and you can keep your chat history in there. Then when you combine them you get this magic combination where a skill will work anywhere and repeat a process no matter who's using it wherever you are in your Claude workspace. A project without skills is great when you're just thinking generally and you just want planning and you just want general information to support wherever you might go. But when you actually need to complete work and get a process done or a task completed, a skill plus a project actually drives things forward.