Brain of Naval — a Claude Code skill

An unofficial, AI-synthesized persona of Naval Ravikant (optionally Kapil Gupta) that you can consult before making a decision.

Built as a Claude Code skill. Works anywhere Claude Code runs: any project, any repo, any directory.

⚠️ Unofficial. Not affiliated with, authorized by, or endorsed by Naval Ravikant or Kapil Gupta. This is an AI synthesis of their public interviews. Use for personal reference only.

---

What this is

You: naval should I quit my job to start a company?

Naval (via Opus sub-agent):
## Reframe
Question stands.

## Verdict
If you have to ask, don't. The people who should start companies are already
doing it in their evenings. Join a small team with specific knowledge first,
compound for 3-5 years, then decide.

## Why — axioms and models
- Specific Knowledge: you build it at the smaller startup, not from scratch.
- If You Can't Not Do It test: your hesitation is the answer.
- Long-term games with long-term people: a first company built from fear fails by year 2.

## Heuristics triggered
- IF you're asking permission THEN you're not ready. (Wealth)
- IF you lack specific knowledge THEN build it before founding. (Wealth)
- IF the decision is reversible THEN decide quickly; this one is. (Judgement)

## Next move
This Monday: list 5 small companies whose product you'd use daily. Apply to three
by Friday.

Play long-term games with long-term people.

Saved to ~/.naval/consultations/2026-04-19-should-i-quit-my-job.md

The skill reads a pre-built "brain" — axioms, mental models, decision heuristics, signature phrasings — distilled from 34 long-form Naval interviews, then dispatches an Opus sub-agent to answer in his voice.

Install

git clone https://github.com/YOUR_USERNAME/naval-skill.git
cd naval-skill
./install.sh

Or from a tarball:

tar -xzf naval-skill-1.0.0.tar.gz
cd naval-skill-1.0.0
./install.sh

The installer:

• Copies skill/ → ~/.claude/skills/naval/ (brain bundled inside)
• Creates ~/.naval/consultations/ for your decision history
• Safe to re-run — backs up existing installs

Requirements: Claude Code installed. That's it.

Usage

Modes

Command	Purpose
naval <question>	Full consultation — 4-question clarity gate → structured verdict → saved
naval --quick <q>	Compact verdict, no save, fast
naval --kapil <q>	Ask Kapil Gupta instead (spiritual, surgical)
naval --debate <q>	Naval + Kapil in parallel, then synthesis
naval --pushback <slug>	Contest a prior verdict (Naval defends, never flips)
naval --retro	Fill in follow-ups on past consultations
naval --eval	Regression-test voice against 8 canonical Q&A pairs
naval --context <path> <q>	Attach a file (MD/TXT/PDF/JSON/YAML)

Triggers

Any of these fire the skill in Claude Code:

• ask naval ..., consult naval ..., what would naval say ..., naval check, naval take
• naval --debate, ask kapil, naval vs kapil
• Korean: 네이벌, 네이벌한테 물어, 카필, 카필한테 물어

Or just /naval + your question.

When to use

Use when:

• Making a career / life / business / investment decision
• Feeling stuck or rationalizing
• About to commit to something irreversible
• Suspecting a status game
• You want a sanity check before a hard call

Do not use for:

• Pure technical questions (code, API, infra, medical, legal)
• Venting — listen, don't consult
• Post-hoc validation — if you've decided, stop asking

Configuration

Optional environment variables (set in ~/.zshrc or ~/.bashrc):

# Point to a custom brain location (e.g., your own curated version)
export NAVAL_BRAIN_PATH="/path/to/brain"

# Custom save location for consultations
export NAVAL_CONSULTATIONS="$HOME/Documents/naval-logs"

# Your name/wikilink in saved consultation frontmatter (for Obsidian users)
export NAVAL_USER_LINK='[[Your Name]]'

If none are set, defaults are:

• Brain: bundled at ~/.claude/skills/naval/brain/
• Consultations: ~/.naval/consultations/
• Author: system username

How the brain was built

34 public long-form Naval interviews (and his conversations with Kapil Gupta, David Deutsch, Tim Ferriss, Eric Jorgenson, Joe Rogan, and others) were processed by 6 parallel Opus agents, each extracting one thematic slice:

1. Wealth — leverage, specific knowledge, angel investing, hiring
2. Happiness — desire, peace, 5 chimps, Kapil's prescriptions
3. Philosophy — Popper/Deutsch epistemology, hard-to-vary explanations
4. Judgement — reading, decisions, compound knowledge, clarity
5. Life — relationships, harsh truths, meaning, death
6. Technology — AI, crypto, permissionless innovation

A synthesis pass produced 00-brain-of-naval.md — the master persona document (~7,000 words) containing: 7 prime axioms, 24 mental models, ~90 decision heuristics, 9 rhetorical moves, 25 verbatim signature phrases, 40 anti-patterns, a 9-step meta-algorithm, and a system-prompt block.

The sub-agent reads the master + 1-2 relevant domain files per consultation.

Verify voice purity

The skill includes a regression-test harness. Run after any brain edit:

naval --eval

This runs 8 canonical questions (hustle culture, toxic coworkers, MBA, startups, anxiety, marriage, politics, etc.) through the skill. Each answer is checked against:

• Must contain — Naval-vocabulary markers (e.g., leverage, specific knowledge, long-term games)
• Must NOT contain — generic self-help phrases (e.g., grind it out, trust your gut, stay informed)
• Judge score — a second Opus agent rates position-match 1-5

Drift score of 0-1 fails = healthy. 2 fails = degraded. 3+ = regressed.

Customize the brain

You can edit the brain files in ~/.claude/skills/naval/brain/ directly. After edits, run naval --eval to check for drift.

To fork the brain and build your own persona (e.g., "Brain of Charlie Munger"):

1. Copy this repo
2. Replace transcripts with your subject's public interviews
3. Re-run the 6-agent extraction + synthesis (see /docs/HOW-THE-BRAIN-WAS-BUILT.md — TODO)
4. Rename the skill in SKILL.md frontmatter
5. ./install.sh

Architecture

~/.claude/skills/naval/          # Installed skill (global, invokable anywhere)
├── SKILL.md                     # Dispatcher: 8 modes, path resolution, prompts
└── brain/
    ├── 00-brain-of-naval.md     # Master persona (~7K words)
    ├── 01-wealth.md             # Domain extractions
    ├── 02-happiness.md
    ├── 03-philosophy.md
    ├── 04-judgement.md
    ├── 05-life.md
    ├── 06-tech-future.md
    └── _evals/
        ├── canonical.md         # 8 regression-test Q&A pairs
        └── reports/             # Eval run outputs (YYYY-MM-DD-eval.md)

~/.naval/consultations/          # Your saved decisions (git-ignorable)
├── INDEX.md                     # Auto-maintained index
├── 2026-04-19-<slug>.md         # Each consultation, with Follow-up section
└── ...

FAQ

Q: Does this replicate Naval exactly? No. It's an AI persona that reasons like Naval — same axioms, vocabulary, heuristics, voice — but it can still produce outputs Naval would disagree with. The --eval harness is there to detect drift; it's not a guarantee of fidelity.

Q: Why Kapil Gupta? Kapil is Naval's closest interlocutor on the inner-life side. Where Naval gives pragmatic leverage-math, Kapil makes surgical spiritual observations. For personal decisions, the dual lens is often more useful than either alone.

Q: Can I use this in production? You can run it in any Claude Code context. But the outputs are AI-generated opinions about personal decisions. Don't outsource real decisions to it. Use as a lens, not a verdict.

Q: Is the brain static? Yes — v1.0.0 is a frozen snapshot as of April 2026. Future versions will retrain on new interviews. You can also override with $NAVAL_BRAIN_PATH to point at your own curated version.

Q: Does it work offline? The skill runs locally, but Claude Code requires a Claude API connection for the sub-agent. No internet = no consultation.

Q: Privacy? Consultations save to ~/.naval/consultations/ on your machine. Nothing is uploaded unless you do it yourself.

License

MIT — see LICENSE.

The brain content (synthesis of public interviews) is provided under the same license with the explicit caveat that it is unofficial and not endorsed by Naval Ravikant or Kapil Gupta.

Credits

• Naval Ravikant — the source of the thought process being synthesized. All original ideas are his. Find his actual work at nav.al and via The Almanack of Naval Ravikant by Eric Jorgenson.
• Kapil Gupta — kapilgupta.com. Author of Atmamun.
• David Deutsch, Tim Ferriss, Joe Rogan, Eric Jorgenson, Arjun Khemani, Cory Levy, Brett Hall — interviewers whose long-form conversations with Naval were sources.
• Claude (Anthropic) — ran the 6-agent parallel extraction and synthesis.

Contributing

This is a personal tool, not a maintained library. Fork freely. PRs that fix bugs or improve the install UX are welcome. PRs that change Naval's "positions" in the brain will be rejected — the brain is a snapshot of public interviews, not a debate club.

---

Play long-term games with long-term people.