A strategic advisory council for ADHD entrepreneurs, packaged as a Claude Code skill.
/breakdown-strat <your decision> convenes five isolated advisors, each grounded in the real corpus of a public figure (Naval Ravikant, Alex Hormozi, Paul Graham, Sahil Lavingia) plus an adult-ADHD clinician persona built from clinical literature (Barkley, Hallowell, Dodson). They deliberate blind to each other, debate their genuine disagreements, and converge on a numbered consensus that ends with exactly one irreversible action you can take in the next 24 hours.
Built in June 2026 by an ADHD entrepreneur for his own high-stakes calls: pivot vs no-pivot, which project to kill, pricing dilemmas, any decision that is expensive to be wrong about.
Generic "ask an AI for advice" fails the ADHD brain in predictable ways. This skill counters each failure mode structurally:
- Anti-rumination gate. Every run starts by checking the decision log. If you already asked this question (or a near-duplicate), the council names it out loud: re-asking may be rumination or risk-avoidance wearing the costume of rigor, not a search for new information. It refuses to become another rumination loop.
- One action, irreversible, 24 hours. Every run ends with exactly one concrete next action, never a todo list. Irreversibility (send, publish, pay, announce) is the point: it removes the option to flip back, which is what ends oscillation between options.
- A clinician with a hard lane lock. The fifth advisor analyzes only your brain: why you decide or avoid the way you do, and the conditions under which you can actually initiate, execute, and finish. It gives zero business picks, ever. The decision content belongs to you and the other four; the decision machinery is the clinician's only territory.
- Anti-abandon hook. The consensus names when the abandon pattern will likely strike for this specific decision (usually the 2-to-4-week novelty cliff) and what external accountability counters it.
- Grounded in your real facts. Advisors read a context file you write once (runway, constraints, patterns, non-negotiables) instead of guessing. Missing decisive facts block the run; the council asks instead of assuming.
| Advisor | Lane | Corpus |
|---|---|---|
| Naval Ravikant | Long-term wealth, leverage, specific knowledge, equity over wages | The Almanack of Naval Ravikant |
| Alex Hormozi | Cash now, offer design, pricing, speed-to-revenue | $100M Offers |
| Paul Graham | Founder-market fit, real demand vs hope, default alive | paulgraham.com essays |
| Sahil Lavingia | Minimalist solopreneur, community-first, charge from day one | The Minimalist Entrepreneur, slavingia/skills |
| ADHD clinician | Executive function, dopamine timing, RSD, anti-abandon structure. Never picks the business option. | Barkley, Hallowell, Dodson (clinical literature) |
Each doctrine file distills the corpus into a lane definition, core principles, "when X, do Y" heuristics, attributed verbatim quotes, and a voice profile. Advisors spawn in parallel, blind to each other, and are forbidden to seek consensus.
- Preflight. Scaffold the run directory, check the decision log for repeat questions (anti-oscillation gate).
- Framing. Build
framed.mdfrom your context file plus the conversation: the core decision, verified facts, constraints, stakes, known patterns. Missing decisive facts trigger a question, never a guess. - Council. Five advisors spawn in parallel, each receiving only its doctrine and the framed context.
- Peer review (
--deeponly). The five verdicts are anonymized A-E and re-reviewed by five fresh agents blind to authorship — merit-based critique, no style deference. - Debate. The chair maps where advisors converge (high-confidence signal), names each genuine clash and why it exists, and resolves by reasoning, never by averaging. Contemporary factual claims are web-verified before entering the consensus.
- Consensus. Numbered decisions, resolved disagreements made visible, one 24-hour irreversible action, one anti-abandon hook. Logged to the decision log.
git clone https://github.com/Nazim23i/breakdown-strat.git ~/.claude/skills/breakdown-strat
cp ~/.claude/skills/breakdown-strat/context.example.md ~/.claude/skills/breakdown-strat/context.mdThen edit context.md with your real situation. This file is gitignored; it never leaves your machine.
/breakdown-strat should I keep my agency positioning or pivot to a productized service?
/breakdown-strat --deep kill project A to go all-in on project B?
Run outputs (framing, individual verdicts, debate, consensus, optional HTML report) land in output/<date>-<question-slug>/. The output/ directory is gitignored.
- Your facts live in
context.md(seecontext.example.mdfor the structure). The richer it is, the less the council asks and the sharper the verdicts. - Voice preferences (language, tone, formatting rules) go in the "Voice preferences" section of
context.md; advisors respect them. - Swap advisors by adding or replacing
references/advisor-<name>.mdfiles following the same structure (lane, principles, heuristics, quotes, voice, spawn template) and listing them inSKILL.md.scripts/ingest-doctrine.shdocuments the canonical sources and what needs rebuilding.
The named advisors are simulated personas derived from each person's published, attributed work. Naval Ravikant, Alex Hormozi, Paul Graham, and Sahil Lavingia are not affiliated with this project and have not endorsed it. The ADHD clinician is a composite persona grounded in published clinical literature; it is not medical advice and is no substitute for an actual clinician.
MIT. Quotes in the doctrine files remain the property of their authors and are reproduced briefly, with attribution, for commentary and education.