How I Use Claude Code to Turn Any Expert's Best Thinking Into an AI Skill
Learn how to convert any expert's public content into a reusable AI skill you can run whenever you need it
Reading Ogilvy doesn’t make your copy better.
Having Ogilvy’s principles codified into a grader that scores your copy in real-time does.
Everyone consumes expert content. Newsletters, podcasts, YouTube. With AI, it’s now possible to instantly turn that content into action.
As Naval Ravikant put it:
“Reading is not the same as understanding. Understanding is not the same as being able to use it.”
Here’s the shift: every piece of expert content is now a functional brief for Claude Code. A newsletter, a YouTube transcript, a podcast episode. This system can aggregate an experts lessons, extracts the best lessons, and turns them into a skill.md file you can run on demand.
Charlie Munger called this a latticework of mental models. He spent decades reading voraciously across physics, psychology, biology, economics, but he didn’t just read. He converted every major idea into a reusable mental framework he could apply to investing decisions. Warren Buffett read all the same books. Most investors of their era did too. Munger’s edge was turning that content into a system he built to convert it into something operational.
That’s exactly what this tutorial will do for you. Just faster. Plus I’ve included the system for you to download at the end of this article.
I built it. I pointed it at myself first.
Here’s what happened.
The Claude Code System: 3 Phases
The Skill Creator is a set of Claude Code skills that research any expert’s public content, extract their best lessons, score them, and turn the top lessons into reusable AI skills you can run on demand.
It runs in 3 phases.
Phase 1: Research
Step 1: Run /expert-lessons to research an expert and extract their top lessons
Point it at any expert and it searches across Substack, YouTube, X, podcasts, and web articles. For YouTube, it uses yt-dlp to find YouTube videos the expert appears in, even on channels they co-host where their name isn’t in every title. It downloads transcripts for the top 5 videos by engagement. It runs 4 Perplexity deep research queries to pull content from everywhere else.
Phase 2: Score and rank
From everything it finds, it extracts discrete lessons. Each lesson has to be actionable (something you can implement), discrete (one clear idea), and attributable (tied to a real source).
Then it scores every lesson on 3 criteria:
Actionability: does it contain a concrete framework or is it just an observation?
Clarity: is the methodology well-defined with extractable steps?
Novelty: is this a unique insight or common knowledge?
It surfaces the top 10 in an interactive moodboard you open in your browser, filterable by source type and sortable by any score dimension. You pick the lessons you want to convert and export them as JSON.
Here’s the moodboard it created for my content and content I pulled in for the awesome Elena Verna. The moodboard itself is valuable. It can create you an entire moodboard of tactics aligned to any creator you admire.
Phase 3: Generate the skill
Step 3: Run /create-skills to generate skills from the selected lessons
Now you can pass your JSON to the skill that turns lessons into skills.
Before generating anything, the system analyses whether each lesson should become 1 skill or multiple focused skills. It asks: does this methodology contain distinct sub-frameworks? Are these steps sequential phases of one process, or independent strategies?
Then it scores each resulting skill on Repeatability, Ease of Implementation, Impact Potential, and Universality.
The highest scorer gets recommended. You confirm what you want built. It generates a markdown skill file, ready to run in Claude Code.
Here’s the system recommending what skill I should create from one of Elena’s top scored lessons.
I Ran It on Myself
The system extracted 7 discrete lessons from my public content. It scored them all.
I decided to follow the system and choose the first one:
Before generating anything, the system runs a decomposition step. It looks at each lesson and asks: does this contain distinct sub-frameworks, or is it one coherent process? Should this become one skill or several?
For the Quality Grader, it identified one clean skill and scored it:
Repeatability 9/10
Ease of Implementation 8/10
Impact Potential 9/10
Universality 9/10Universality is what pushed it to the top. Most skills extracted from a marketing expert stay in marketing. This one works for any craft and any domain master. You pick the standard. The skill adapts.
So it generated the skill file and here is what it does:
Context gathering — asks what industry you’re in, what type of work you’re grading, and what outcome it needs to achieve
Pick a domain master — you choose an expert whose principles become the rubric (David Ogilvy for copywriting, Nancy Duarte for presentations, anyone)
Establish principles — researches the master and generates 10-15 principles, each with a name, definition, good and bad examples, and equal weighting toward 100 points
Submit work — paste text, point to a file, any format
Score and feedback — scores each principle (Strong / Adequate / Weak), gives an overall score out of 100, highlights the top 3 improvement areas by impact
Rewrite — generates a full rewrite targeting 100/100, with a change summary explaining what was fixed and why
I chose David Ogilvy. I submitted HubSpot’s homepage.
It scored 67/100.
Here’s the output:
Headline Does the Selling 5/7 ~ Adequate
Specificity Over Generality 5/7 ~ Adequate
Reader-First Language 3/7 ✗ Weak
One Big Idea 4/7 ~ Adequate
Promise a Benefit 4/7 ~ Adequate
Simple Language 6/6 ✓ Strong
Factual Persuasion 6/7 ✓ Strong
Long Copy Sells 3/6 ✗ Weak
Clear Call to Action 6/7 ✓ Strong
Don't Be Boring 4/7 ~ Adequate
Competitive Differentiation 2/6 ✗ Weak
Emotional Connection 3/7 ✗ Weak
Credibility & Authority 6/7 ✓ Strong
Visual-Copy Integration 5/6 ✓ Strong
Consistent Brand Voice 5/6 ✓ StrongThe top 3 weaknesses:
Reader-First Language (3/7). HubSpot’s hero talks about HubSpot, not the reader’s problem. “HubSpot’s Smart CRM” should be “Stop toggling between 12 tabs to find one customer’s history.”
Competitive Differentiation (2/6). Zero reasons to choose HubSpot over Salesforce, Pipedrive, or Marketo. Ogilvy built entire campaigns around competitive claims.
Emotional Connection (3/7). The case studies (pipeline from millions to billions, 300%+ fan growth) carry genuine emotion. They sit buried under 6 product feature descriptions.
The system then offered to generate a full rewrite targeting 100/100. Pretty incredible given the skill was automatically built from a single lesson of mine.
What This Actually Means
Any expert's public methodology can become a reusable AI workflow. The knowledge doesn't have to sit inert in a newsletter you'll never reread. It can become something you run. Every piece of content is now a functional brief.
Three design decisions worth knowing before you build this:
Lessons and skills are scored differently, and that's deliberate. Lessons are scored on educational quality: is this actionable, clear, novel? Skills are scored on operational potential: is this repeatable, easy to implement, high impact, universal? A great lesson doesn't always make a great skill. The two-stage scoring prevents you from converting lessons that sound impressive but don't run well as workflows.
Every generated skill is domain-neutral by design. Skills strip the expert's industry context. They start with a Step 0 that asks about your context: your industry, your audience, your goals. The expert's structural thinking is preserved. Their industry-specific assumptions are stripped. The same skill extracted from a marketing expert works for a fitness coach, a developer advocate, or a sales trainer.
Apply to your internal workflows. If I was going to give one piece of advice to make this more impactful for you, plug it into your internal systems e.g. email, slack, asana etc. Have it recommend skills based on repeated patterns so it can start to automate your work.
Now, the objection I’d have if I were reading this: “This sounds impressive but it’s for people who can build in Claude Code. I’m a marketer, not a developer.”
The skill files are markdown. Plain text with instructions. If you’ve written a Google Doc, you can read a skill file. The Skill Creator generates them for you, you don’t write them from scratch.
Munger spent decades converting other people’s best thinking into reusable mental frameworks. This does the same thing. Just faster.
The kit includes both skills, the moodboard template, and a full setup guide. Free, no email required.
Until Next Time,
Happy AI’fying
Kieran