The Skill That 10x’d My Claude Code Projects

@nateherk
ENGLISH2 months ago · Jun 04, 2026
113K
176
10
7
538

TL;DR

This article introduces the grill-me skill for Claude Code, a method where the AI interviews the user to capture deep project context, leading to higher quality first-draft outputs and better documentation.

Getting What's In Your Head Is the Hard Part

Everyone runs the same model.

If you're on Claude Opus 4.8, so is everyone else. Same model, same prompts, same outputs.

What actually separates your output from everyone else's is the context you feed it. Your taste. Your voice. Your decisions.

The hard part was never the prompt. It's the extraction. Getting everything out of your head and into the system so your skills can actually use it.

Nate Herk - inline image

So I started using a skill called grill-me from @mattpocockuk, then I upgraded it.

Why Extraction Is the Bottleneck

If you've watched me talk about discovery calls or scoping a project, you know the drill.

You ask the client so many questions they almost get annoyed.

That's the job. It's the difference between a system that works 95% of the time and one that works 80% of the time.

A 5-minute brain dump into Claude Code feels like enough. It never is.

What grill-me Actually Does

It takes what's in your head and turns it into reusable context for your AI.

Nate Herk - inline image

Here's the loop:

1️⃣ It asks you a question, one at a time.

2️⃣ You answer.

3️⃣ It checkpoints and writes everything back to a knowledge doc.

4️⃣ It keeps going until there are no gaps or holes left.

Sometimes it asks 5 questions. Sometimes it grills you for an hour plus. That's a good thing.

The result is better skills, better context, and better projects.

The Original Was Four Sentences

This skill was originally built by Matt PCO. The original is almost nothing:

Nate Herk - inline image

"Interview me relentlessly about every aspect of this plan until we reach a shared understanding. Walk down each branch of the design tree, resolving dependencies between decisions one by one. For each question, provide your recommended answer. Ask questions one at a time. If a question can be answered by exploring the codebase, explore the codebase instead.

That's it.

It's a good reminder that a skill doesn't have to be complicated automation. It can just be a prompt you don't want to say every single time.

The One Change I Made

I added something that makes it better.

I worked in checkpointing after every single question.

Here's why. When it grills you for an hour, the context window starts to fill up. I'd get worried it was going to misremember an answer I gave 40 minutes earlier.

So I kept finding myself saying "write this to a doc, checkpoint every time." Eventually I just built that into the skill.

Now it creates a brainstorms folder at the root of your project. If you already have one, it drops a fresh markdown file in.

Open one up and you get the key decisions, a full step-by-step Q&A log of what it asked and what I answered, and the highlights.

Where It Got Interesting

When I finished a packaging grill session, it noticed something.

It said I already had a packaging guide and a packaging skill, and that a lot of the nuance we just talked through wasn't in either one. It asked if I wanted to update both.

I said yes. Those skills and docs are now far better.

I did another one where I told it to understand everything about the business. We walked through every decision and every process from start to finish. Now my OS just feels like it knows how the business actually works.

Why This Saves You 20 Iterations

Nothing is perfect on the first try.

The old way of building a skill looks like this. You brain dump, you build, and iteration one lands around 70%. You run it, make a small improvement, get to 75%. You keep going until you cap around 95%.

That might take 10 iterations. It might take 30.

You never really hit 100%, because as your business evolves and you evolve, the skill keeps evolving too.

The grill-me way is different. Because you spent the extra time up front, iteration one starts around 90%.

It's still not perfect. You'll keep refining. But you're there a lot faster, which gives you more room to find better ways to improve it.

If I had six hours to chop down a tree, I'd spend the first four sharpening the axe.

The upfront part feels boring and repetitive. It pays off downstream every single time.

The Part I Didn't Expect to Love

It flags what you don't know.

When I ran this on a funnel map, there were processes in the business I can't explain as well as the person who actually runs them.

So it flagged them. It told me to reach out to a specific person, get the info, come back, and drop it in so we could update the brainstorm.

And because every session saves as a doc, you can reference it later. Find a breakthrough in how you package content, and you just come back and say "grill me again, here's what's new."

Wrap

Then it's as simple as saying "grill me about this," or hitting the slash command.

Go grill yourself about the skills and processes that live in your business. Your OS gets better every time.

I walk through the whole thing, including a live grill session, in the full video. Link in the first reply.

One-click save

Use YouMind for AI deep reading of viral articles

Save the source, ask focused questions, summarize the argument, and turn a viral article into reusable notes in one AI workspace.

Explore YouMind
For creators

Turn your Markdown into a clean 𝕏 article

When you publish your own long-form writing, images, tables, and code blocks make 𝕏 formatting painful. YouMind turns a full Markdown draft into a clean, ready-to-post 𝕏 article.

Try Markdown to 𝕏

More patterns to decode

Recent viral articles

Explore more viral articles