#begin

So this one is a bit different from my usual content. No book reviews, no Unity3D deep dives, no event bus patterns. Today I want to talk about something I built together with Claude that I found genuinely interesting from both a conceptual and a technical perspective. Bear with me.

The Background

Some time ago I was reading Think and Grow Rich by Napoleon Hill. You know the book. Published in 1937, it’s been on the “must read” list of every self-help enthusiast and entrepreneur since roughly forever. Now I’ll be honest: parts of it are dated and a little weird. But some of the core ideas are actually quite sharp.

One of those ideas is what Hill called the Master Mind. The concept is simple: when two or more people align around a shared goal in a spirit of genuine harmony, they produce a collective intelligence greater than the sum of its parts. Hill observed this in people like Carnegie, Ford and Edison. Each one surrounded himself with a tight inner circle. Not yes-men, but actual thinkers who would push back and contribute.

That part is well known. Less known is something else Hill described: his practice of convening an imaginary council in his own mind each night. He would mentally invite nine historical figures (Lincoln, Napoleon, Edison, Darwin and others) to a kind of roundtable. He’d pose a question and listen for their answers, each speaking in the spirit of their documented philosophy. Hill claimed that over time these figures took on a life of their own and became a genuine source of insight.

Now is this meditation? Roleplay? Mild hallucination? The Flying Spaghetti Monster only knows. But the underlying principle, forced perspective-taking using deeply internalized mental models, is actually a well-established cognitive technique. You’re essentially simulating how a specific mind would reason through a problem, which forces you out of your own defaults.

So I thought: what if we modernized this with AI?

The Idea

Instead of meditating on imaginary council members, you could actually talk to them. Claude, given a well-researched profile of a real person’s philosophy, decision-making style and characteristic voice, can do a pretty convincing job of simulating how that person would reason about your question.

The key word is researched. The quality of the simulation is entirely dependent on the quality of the profile. A vague profile produces generic advice. A detailed profile (covering not just what someone believed but how they thought, what their blind spots were, how they spoke) produces something actually useful.

And since this is a Claude skill, I wanted it to do the research itself.

The Skills

I ended up building two skills that work together as a pipeline.

mastermind-council-advisor

The first skill helps you design the right council for your specific challenge. You don’t just get Hill’s original nine by default: the skill searches the web and recommends real, contemporary people who are credible in the domains most relevant to your situation.

It asks you two questions: what is your challenge, and what kind of advice matters most to you. From there it identifies 4–6 relevant domains, searches for credible people in each, and recommends a council of 5–7 members with explicit reasoning. It also tells you who it deliberately left out and why, which I found to be one of the more interesting outputs. It forces an actual decision rather than just a list.

For example, when I asked it to design a council for marketing and monetising an open-source developer CLI tool, it came back with Pieter Levels, Adam Wathan, Rob Walling, DHH, Courtland Allen and Justin Jackson. That’s actually a thoughtful list. Each person brings a genuinely distinct perspective. There’s no redundancy.

mastermind-council

The second skill is where the consultation actually happens. It reads the profile files from your working directory and runs a structured multi-round deliberation on your question.

The flow is:

1. Each council member answers your question anonymously (Member A, Member B, etc.)
2. Each member gives brief anonymous feedback on another member’s answer
3. Claude synthesizes everything, noting where the council aligns, where it disagrees, and what the sharpest single insight was

The anonymity is intentional. Member labels are assigned randomly so you can’t trivially figure out who said what before you’ve actually engaged with the content. It prevents you from anchoring on a name and dismissing an answer before you’ve read it.

Profile Persistence

One thing I wanted to get right from the start was profile persistence. The profiles are stored as Markdown files in your working directory with a council_member: true frontmatter field. On each session start the skill scans for existing profiles and skips re-research if they’re already there.

This means if you consult your council regularly (which is the point) the startup time is near zero. You just ask your question and the council is ready. The first session is slow because research takes time. Every session after that is instant.

You can also maintain multiple councils in different directories. One for business decisions, one for health, one for creative projects. Each directory is independent. Simple and clean.

Does it actually work?

I tested it with a real question: whether I should leave my permanent job to pursue freelancing or start my own business. That’s a genuine question I’ve been thinking about, by the way. Not just a test prompt.

The council I used for that was: Elon Musk, Commander Mark Divine, Uncle Bob Martin, Stephen Hawking and Arnold Schwarzenegger. An unusual group for a career decision, admittedly. But that was also part of the experiment: what happens when you force genuinely diverse perspectives on a personal problem?

The results were interesting. Each member gave a distinctly recognizable answer. Uncle Bob went straight to “test before you ship”: don’t quit the job before validating that there is real demand for your consulting. Arnold told me to stop waiting for permission and start the reps. Hawking reframed the entire question around the asymmetry of regret. Divine asked whether I had done the inner work before making an outer move.

The peer feedback round is where it got genuinely useful. One member pushed back on another’s emphasis on urgency, pointing out that certainty and correctness are not the same thing. Another challenged the “build an audience first” framing, arguing that stars and followers don’t pay for hosting time or motivation. These were not generic observations; they were the kinds of tensions that actually matter for this kind of decision.

The synthesis then called out the sharpest disagreement explicitly: should you charge before you have an audience, or build the community first? And rather than smoothing it over, it named both positions and explained why both had merit in different situations.

That’s actually how you want an advisory council to behave. Not consensus. Productive tension.

The Code

The skills are just Markdown files with structured instructions. No external dependencies, no API keys to manage, no hosted service. Two .skill files and a pile of .md profiles per council. The whole thing lives in your local directory and everything is automatically tracked in git if you want it to be.

The repo is on GitHub at Napoleon Hill’s Imaginary Master Mind Group. The README describes the full setup and the CLAUDE.md provides additional context for Claude itself when working in the repo, including what behaviours to preserve when the skills are modified.

Conclusion

I’ll be honest, when I started this I wasn’t entirely sure it would produce anything more than generic advice dressed up in famous names. I was wrong. The combination of well-researched profiles, forced anonymity during the deliberation phase, and an explicit synthesis that doesn’t paper over disagreements produces something qualitatively different from just asking Claude “what should I do.”

The perspective-taking is real. The tensions between members are real. And the synthesis is genuinely useful as a way of mapping the decision space before you have to make a call.

Is it Napoleon Hill’s meditation practice? No. Is it a useful tool for structured multi-perspective thinking? Yeah, I think it actually is.

If you try it, let me know how it goes. And yes, the single other reader of this blog besides myself: I’m talking to you.

Cya 😉

#end

01010010 01110101 01100010 01100101 01101110