The Skill That 10x’d My Claude Code Projects

0:00The toughest part about building good

0:02skills and building a good operating

0:03system is trying to get everything from

0:05your brain into your system. So, for

0:07example, what you're looking at here is

0:09after months and months of me building

0:11up all of the knowledge that lives

0:12inside of my AIOS. It's basically just

0:14the idea that if everyone's using the

0:16same model, so if everyone's using

0:17Claude Opus 4.8, then everyone's going

0:19to be using the same prompts and getting

0:21the same output because the model is

0:23fundamentally the same for everybody.

0:24So, what really makes the difference is

0:25when you add context into that model and

0:27you give it your taste, your voice, your

0:29decisions, and that's how you get

0:30outputs that actually sound like you.

0:32But once again, the real challenge is

0:33still the extraction, getting everything

0:35from your head into the AI system so

0:38that your skills can use it and that

0:40your context is better. And if you guys

0:41have been following me for a while and

0:42you've seen videos I've made about like

0:45discovery calls and and scoping out

0:46projects, that's the toughest part. It's

0:48especially if you're working with a

0:49client, asking them so many questions

0:51about this process to the point where

0:53they might even get annoyed because

0:54you're asking so many questions, but

0:56that's just what you have to do. It's

0:57the difference between a system that is

0:59successful 95% of the time and one

1:00that's only successful 80% of the time.

1:02So, this one skill we're going to look

1:04at today is called grill me. It

1:06basically takes what's in your head into

1:08reusable context for your AI. So, what

1:10happens is all of that knowledge that's

1:11in your head that you might think,

1:12"Okay, I'm just going to brain dump into

1:14Claude code for 5 minutes and it will be

1:16good enough." It's not ever good enough.

1:18So, what this does is it basically

1:19relentlessly asks you questions. It

1:21grills you until it knows pretty much

1:23everything about the process. It'll ask

1:25you a question, you answer it, and then

1:27it basically will checkpoint and it will

1:29write everything back to a knowledge

1:30doc, and it will just keep going over

1:32this loop endlessly until the knowledge

1:34doc is good enough and there's no gaps

1:36or holes in that knowledge. And so, like

1:38I said, this results to better skills,

1:40better context, and better projects. And

1:42originally this skill was built by Matt

1:44Pocock, and what's cool is if you look

1:45at it, it is a super simple prompt. It's

1:47like four to five sentences. "Interview

1:50me relentlessly about every aspect of

1:51this plan until we reach a shared

1:53understanding. Walk down each branch of

1:55the design tree, resolving dependencies

1:57between decisions one by one. For each

1:59question, provide your recommended

2:01answer. Ask questions one at a time. If

2:02a question can be answered by exploring

2:04the codebase, explore the codebase

2:05instead. And I like to look at that

2:07because it makes you realize that a

2:08skill doesn't have to be super

2:09complicated automation. A skill can just

2:11be a prompt that you don't want to have

2:13to say every single time. And of course,

2:15naturally, what did I do? I destroyed

2:17that. I ruined the skill. I made it a

2:18little bit more complex, but I added

2:20something that I think makes it much

2:21better. So, if I go into my dot Claude,

2:23I go down to my skills, and we look for

2:25the grill me right here, and I open up

2:26the skill that I'm D, you can see it's a

2:28little bit longer now, but basically

2:29what I did is I worked in that whole

2:31element of checkpointing after every

2:33single question. Because originally, the

2:35skill doesn't do that, and what happens

2:37is if you are talking, you know, if it's

2:39grilling you for an hour plus, which

2:40sometimes it will, and that's a good

2:42thing, then as the context window starts

2:43to fill up, I started to get worried

2:45that it was going to misremember some of

2:47my answers from earlier. So, I just

2:49found myself telling it manually, "Hey,

2:50write this to a doc. Write this to a

2:52doc. Checkpoint." every time. And so, I

2:53figured, "Okay, why not just work that

2:54into the skill?" So, now what the skill

2:56does is it creates a folder called

2:59brainstorms, and it does this at the

3:00root of your project. So, if I go down

3:02here, you can see I've got a brainstorm

3:04file, or sorry, a brainstorm folder

3:05right here with these four brainstorms.

3:07And so, it will create that for you if

3:08you don't have it, but if you do have

3:09it, it will just chuck a doc in there, a

3:11markdown file, right away. And so, then

3:14if I open up, like for example, this

3:15packaging one, which I was doing, it

3:17will find like the algorithm, the key

3:19decisions, but then it will also show

3:21you the step-by-step Q&A log of the

3:23questions that it asked and what I

3:25answered with and the key highlights.

3:26And then as soon as we finally got to

3:28the end of that packaging grill me

3:29session, it said, "Hey, I notice you

3:31have this packaging guide, and you have

3:33a packaging skill, and there's a lot of

3:35nuance here that we talked about that's

3:36not in there. So, do you want me to

3:37update both of those?" And then I said,

3:39"Yes." And now those skills and docs are

3:41so much better. I also did one where I

3:43said, "Hey, I want you to understand

3:44everything about the business." And we

3:46walked through from beginning to end all

3:48the decisions, all the processes, and

3:49now my OS just feels like it knows even

3:51more about the way the business works.

3:53And so, if you think about it like this,

3:55right? Like nothing is going to be

3:56perfect on the first try. And so, let me

3:58just do a quick visualization. This is

4:00kind of the old way when you're building

4:02a skill, right? So, we've got iterations

4:03down here. Let's say by iteration one,

4:06after you've knowledge dumped in your

4:08brain and you want to build a skill, you

4:10maybe get somewhere, let's just say

4:12around here, where you're about like 70%

4:14successful on iteration one. And then

4:16what happens is you run the skill and

4:17you make a small improvement and now

4:19you're up about I don't know, right

4:21here. Like you go up from 70% to 75. And

4:23then every time you iterate, you get a

4:25little bit better with each iteration

4:27until maybe you cap at the point where

4:28you're about like 95% good. And this

4:32could be 10 iterations, it could be 30

4:33iterations, it however many it takes for

4:35your skill to feel a bit more

4:36battle-tested. And honestly, I don't

4:38think you ever get to 100% because as

4:40your business evolves and as you evolve,

4:42the skill keeps evolving. So, like all

4:44my skills that I've been using for

4:45months and months,

4:47I'm pretty much still changing a lot.

4:49But, the whole idea is, what if on

4:51iteration one, because you do this grill

4:53me and you spend extra time up front,

4:55you're able to jump right up here to

4:57like 90 at the beginning? And yes, it's

4:59not perfect, you're still going to

5:00iterate a little bit, but you're just

5:01there a lot quicker, which gives you

5:03more opportunity to find better ways to

5:06iterate on it. So, that's my horrible

5:07visual of why I think this is valuable.

5:10It just goes back to that whole idea of

5:12if I had 6 hours to chop down a tree, I

5:13would spend the first four sharpening

5:15the axe. Where up front, yes, maybe it

5:18feels boring or repetitive, but that's

5:20what you need to do is get all that

5:21context in there because it helps

5:22downstream so much more. So, anyways, if

5:25you guys want to grab the grill me

5:26skill, you can look it up here from Matt

5:27Pocock. Or if you want my version, you

5:29can come to my free school community,

5:30the link for that is down in the

5:31description. Just join the community, go

5:33to the classroom, click on all YouTube

5:35resources and it will be right in there

5:36for you to find along with all my other

5:38free resources. And then it's as simple

5:39as saying, "Hey, grill me about this."

5:41Or of course, you can invoke it with a

5:43slash command right there, you can see

5:44grill me. But, I could just say

5:46something as simple as, "Hey, I need you

5:48to grill me about the way that I think

5:49about applying AI to my own business

5:52internally in a safe way that won't

5:55damage the business." You could see

5:56it'll obviously load up that grill me

5:58skill. We're going to see in a second

5:59that it's going to create the capture

6:01file, so nothing gets lost, right there.

6:03We have applying AI internally, and this

6:05is what it looks like. It's going to set

6:07up the discovery nodes, the summary key

6:09decisions, um Q&A log, and any open

6:11flags. And what's cool about this is

6:13it'll flag things that you need to go

6:15find. So, when I was running through

6:16this funnel map, you know, there were

6:18some things going on in the business

6:19that I don't actually know super well.

6:21Like, I can't explain the same way as

6:23the actual stakeholder or operator that

6:24does that process can explain it. So, it

6:27said, "Hey, here's some things to flag.

6:28Go reach out to this person and have

6:30them send you in information, and then

6:31come back and drop that into me, and

6:33then we'll update this brainstorm." So,

6:35that's basically how it works. It might

6:36ask you five questions, it might ask you

6:3830. It's just going to go until you guys

6:40feel like you have the same shared

6:41knowledge, and that it is a good

6:43stopping point. And the cool thing is

6:44because these are saved as docs, you can

6:46reference them later, but you can also

6:47come back to, like for example,

6:49packaging. Let's say I find a major

6:51breakthrough in the way that I package

6:53my content, I would just come back to

6:54this doc and say, "Hey,

6:56grill me again. Here's some new things I

6:58found. Let's update all this

6:59information." So, anyways, I do know

7:01that this one was super quick, but I

7:02thought that, you know, there's no need

7:04to really make this video that much

7:06longer. So, go grab the skill, go start

7:09grilling yourself about different skills

7:11and processes that live in your

7:12business, and just start making your OS

7:14and your skills better. So, that's going

7:15to do it for this one. If you guys

7:16enjoyed or learned something new, please

7:17give it a like, helps me out a ton. And

7:18as always, I appreciate you guys making

7:20it to the end of the video, and I will

7:21see you all in the next one. Thanks,

7:23guys.