convez.ai
SkillFinanceiroAvançado
zpphxd

Munger Check

zpphxd/munger-check

Auditoria de viés para decisões críticas usando framework de Munger

Ver no GitHub
InicialPluginScore · de 0 a 100

Skill que analisa decisões (negócios, contratações, investimentos) através dos 25 Padrões de Má Julgamento de Charlie Munger. Executa auditoria equilibrada em duas pistas, identifica vieses reais, calcula base rates, detecta o efeito Lollapalooza e fornece veredito ajustado por viés com ações debiasing concretas.

Instalação

$npx convez-ai zpphxd/munger-check

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
2
Forks
0
Acompanhando
0
Saúde
42%
Último commit
há 5 meses
Versão
—
Issues abertas
0
Contribuidores
1

Casos de uso

  • Avaliar propostas de investimento e M&A antes da decisão final
  • Validar contratações executivas e aquisições de talento
  • Revisar grandes compras ou negociações com fornecedores
  • Detectar stack de vieses (Lollapalooza) em decisões de negócio

Veredito

Ferramenta sofisticada que transpõe rigorosamente a palestra clássica de Munger (1995) em prática operacional. Diferencia-se por separar legitimidade de viés mascarado, fornecer base rates reais, e nomear seus próprios vieses — exigindo usuário com apetite por franqueza desconfortável, mas entregando clareza que tools genéricos de decision-making não dão.

README

A tradução para o português deste README ainda está na fila.

Munger Check

A decision-hygiene skill for Claude that runs any decision, deal, hire, investment, purchase, or negotiation through Charlie Munger's 25 Standard Causes of Human Misjudgment — and tells you, bluntly, where bias is distorting the call.

In 1995, Charlie Munger — Warren Buffett's partner at Berkshire Hathaway — gave a Harvard talk called The Psychology of Human Misjudgment. No slides, no notes: just a plain-spoken list of the standard ways human beings predictably fool themselves, and the "Lollapalooza effect" — where several of those biases stack in the same direction at once and otherwise-smart people do genuinely crazy things. He built a fortune, in his own words, by being "consistently not stupid."

This skill turns that speech into a tool you can point at a live decision. It doesn't recite biases at you — it runs a balanced audit: it gives the rational case its due first, then finds the few biases actually biting this decision, checks where you have a real edge, scores each tendency, supplies the outside-view base rates, separates your bias from bias being worked on you, finds the Lollapalooza stack (where blow-ups and manipulations live), and ends with a bias-adjusted verdict plus concrete debiasing actions — while staying honest about its own slant.

What's new in v2

  • Two-track analysis — separates legitimate reasons from bias dressed up as a reason, so a genuinely good decision doesn't get talked out of, and a weak one doesn't get rationalized.
  • Edge first — an explicit circle-of-competence step: where do you actually have an edge, and is your leaning pointing away from it?
  • Score & rule out — each live tendency gets a severity (1–5) and a confidence, and the tool lists what it considered and rejected (so it isn't horoscope pattern-matching all 25 onto everything).
  • Base rates, supplied — the actual outside-view numbers for decisions of this type, not just "consider the base rate."
  • Self vs. engineered — splits your own wiring from tactics a counterparty is running on you (scarcity framing, staged concessions).
  • What would change my mind — captures the falsifiers as an artifact and offers to schedule a revisit, since commitment bias is best caught later.
  • Analyst self-check — the tool names its own lean so you can correct for it.
  • Visual scorecard — optional self-contained HTML render with severity bars, the two-track split, and the Lollapalooza stack.

Sample output

MUNGER CHECK — acquihire of struggling competitor
VERDICT (bias-adjusted): Not no — but not for the reasons you think. Fix the process first.
Current leaning: toward yes, fairly hard

TWO TRACKS
✅ Legitimate reasons: removes a price competitor, buys a team you'd take 12 mo to hire.
🎭 Bias dressed as reason: "last chance / they're talking to others" is scarcity, not strategy.

EDGE
You have a real edge integrating eng teams. You have none valuing their book of
business — and your leaning rests on the book. That asymmetry is the headline.

TENDENCIES IN PLAY (severity 1–5 · confidence)
1. Deprival-Superreaction [4 · high] — a rival sniffing around → pushes toward overpaying
2. Inconsistency-Avoidance [4 · high] — you told the board it's happening → toward yes
3. Reward/Incentive (advisor) [3 · high] — banker only paid on close → toward yes, now
4. Excessive Self-Regard [3 · med] — "I can value this book" → toward yes
Considered & ruled out: Social proof (no herd here), Liking (you barely know them).

BASE RATES (outside view)
~70–90% of acquihires fail to retain the acquired team past 2 yrs (est.); most
"strategic premium" deals at this multiple underperform a build. Label: rough.

SELF vs. ENGINEERED
Self-generated: commitment, self-regard.
Engineered on you: scarcity framing ("other bidders") — benefits the seller + banker.

⚠️ LOLLAPALOOZA  [stack pull: high]
Four forces all push toward "yes, now, at this price." That's the danger zone —
and exactly where a motivated seller and banker are working you.

DEBIAS — DO THIS
- Pre-mortem: it's 18 months later and the deal failed. Write the three reasons.
- Set a walk-away number now, before the next call, and don't move it in the room.
- Get one read from someone paid NOT to be wrong (advisor windage).

WHAT WOULD CHANGE MY MIND
- Independent valuation of the book within 10% of the ask.
- Retention package signed by the two key engineers pre-close.
(Offer: schedule a revisit in 30 days.)

ANALYST SELF-CHECK
This analysis leans contrarian/cautious on acquisitions; if the team is truly
irreplaceable, I'm under-weighting that.

READ
Your thesis may be right, but every force in the room points one direction — which
is itself the warning. Highest-leverage move: the walk-away number, set cold.

Length scales to stakes — a quick gut-check gets a compressed version; a major decision gets the full report.

When to use it

  • Before committing to a deal, investment, hire, vendor, big purchase, or negotiation stance.
  • When reviewing someone else's pitch or recommendation (whose incentives are baked in?).
  • Any time you'd say "poke holes in my thinking", "pre-mortem this", or "am I fooling myself?"

Trigger phrases: munger-check, run this through Munger, bias check, what am I missing on this, pre-mortem this, check my thinking.

Install

This is a Claude Agent Skill. Clone it into your skills directory:

git clone https://github.com/zpphxd/munger-check.git ~/.claude/skills/munger-check

Then in Claude Code:

munger-check this — I'm about to sign a 3-year office lease and the broker says it's the last unit

Claude loads the skill, runs the full audit, and returns the report. Ask for the visual scorecard and it renders an HTML artifact. It also nests cleanly inside deal-screening, IC-memo, red-flag, and negotiation-prep workflows.

What's inside

FilePurpose
SKILL.mdThe v2 skill definition — intake → two tracks → edge → scan & score → base rates → self/engineered → Lollapalooza → antidotes → self-check → bias-adjusted verdict.
reference/tendencies.mdThe full catalog: all 25 tendencies, how each shows up in a real deal/decision, the antidote, the Lollapalooza principle, and Munger's debiasing toolkit.
reference/scorecard-template.htmlSelf-contained pattern for the optional visual scorecard — severity meters, two-track split, edge call, and the Lollapalooza stack. No dependencies.

Credit

The framework is Charlie Munger's, from The Psychology of Human Misjudgment — the expanded 25-tendency version in Poor Charlie's Almanack (2005), which revised his earlier 1995 Harvard list of 24. This repo only packages it as a usable decision tool — all the insight is his.

License

MIT — use it, fork it, build on it.

Ver o original no GitHub

Bom para

Relacionados