convez.ai
AgenteFinanceiroAvançado
Pruthavirajsingh

Good Decisions Only

Pruthavirajsingh/good-decisions-only

Parar de decidir por intuição. Começar a decidir por matemática.

Ver no GitHub
InicialPluginScore · de 0 a 100

Um agent de análise de decisões baseado em matemática que avalia escolhas de carreira, relacionamento, investimento e oportunidades usando 6 frameworks cognitivos: Expected Value, Base Rates, Teorema de Bayes, Sunk Cost, Survivorship Bias e Kelly Criterion. Transforma decisões emocionais em recomendações estruturadas e quantificadas.

Instalação

$npx convez-ai Pruthavirajsingh/good-decisions-only

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
1
Forks
0
Acompanhando
0
Saúde
42%
Último commit
há 8 meses
Versão
—
Issues abertas
0
Contribuidores
1

Casos de uso

  • Avaliação de oportunidades de carreira com análise de risco-retorno
  • Decisões de investimento e apostas financeiras estruturadas
  • Análise de escolhas pessoais (relacionamentos, saúde, mudanças de vida)
  • Identificação e correção de vieses cognitivos em decisões importantes

Veredito

Agent robusto que democratiza análise quantitativa em decisões de alto impacto. Forte em frameworks matemáticos, mas sucesso real depende de inputs honestos e contexto adequado do usuário.

README

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

🧠 good-decisions-only

Stop deciding on vibes. Start deciding on math.

A specialized AI agent that runs every real decision through 6 battle-tested
cognitive frameworks — and delivers a clear, structured recommendation.

License: MIT Claude Code Compatible Agent Type PRs Welcome


🤔 Why This Exists

You make roughly 35,000 decisions a day.

For small stuff — what to eat, which show to watch — gut instinct is fine.

But for the decisions that actually shape your life? Career moves, investments, relationships, health? Vibes are a disaster.

The problem isn't intelligence. It's that the human brain is wired to:

  • Feel losses 2x more than equivalent gains (loss aversion)
  • Ignore base rates and chase inspiring anecdotes (base rate neglect)
  • Let past investment contaminate forward decisions (sunk cost fallacy)
  • Only see the winners and miss the graveyard (survivorship bias)
  • Either ignore new evidence or wildly overreact to it (bad Bayesian updating)
  • Go all-in or nothing, when the math says something in between (Kelly violations)

This agent fixes all of that. Not with motivation. With math.


⚙️ The 6-Framework Engine

Every decision gets run through all 6. No shortcuts.

┌─────────────────────────────────────────────────────────────┐
│                    YOUR DECISION INPUT                       │
└─────────────────────┬───────────────────────────────────────┘
                      │
          ┌───────────▼───────────┐
          │   1. REFRAME FIRST    │  Strip emotion & sunk costs.
          │                       │  What are you ACTUALLY deciding?
          └───────────┬───────────┘
                      │
          ┌───────────▼───────────┐
          │  2. EXPECTED VALUE    │  EV = Σ(probability × payoff)
          │                       │  Show the math. Make it visible.
          └───────────┬───────────┘
                      │
          ┌───────────▼───────────┐
          │   3. BASE RATE CHECK  │  What's the historical success rate
          │                       │  for this category of decision?
          └───────────┬───────────┘
                      │
          ┌───────────▼───────────┐
          │  4. SUNK COST AUDIT   │  Past investment = zero votes.
          │                       │  Only the future counts.
          └───────────┬───────────┘
                      │
          ┌───────────▼───────────┐
          │  5. BAYESIAN UPDATE   │  What does your evidence actually
          │                       │  justify believing? Update proportionally.
          └───────────┬───────────┘
                      │
          ┌───────────▼───────────┐
          │ 6. SURVIVORSHIP CHECK │  Find the graveyard.
          │                       │  Who tried this and failed silently?
          └───────────┬───────────┘
                      │
          ┌───────────▼───────────┐
          │  7. KELLY SIZING      │  How much to commit? Quarter-Kelly
          │                       │  to Half-Kelly for real humans.
          └───────────┬───────────┘
                      │
          ┌───────────▼───────────┐
          │   CLEAR RECOMMENDATION│  A verdict. Not a list of considerations.
          └───────────────────────┘

🚀 How to Use

Option 1 — With Claude Code (Recommended)

# Clone and install
git clone https://github.com/Pruthavirajsingh/good-decisions-only.git
cp good-decisions-only/decision-analyst.md ~/.claude/agents/

Then in any Claude Code session:

"Activate decision analyst — I need to think through [your decision]"

Option 2 — As a System Prompt (Claude, ChatGPT, Gemini, etc.)

  1. Open decision-analyst.md
  2. Copy the entire contents
  3. Paste it as the system prompt in any AI chatbot
  4. Describe your decision and let it run the full framework

💡 Example Decisions It Handles

Decision TypeExample Input
💼 Career"Stable $120K job vs. startup offer with equity. Switch?"
💰 Investment"Down 50% on this stock. Sell or hold?"
❤️ Relationship"Together 6 years but unhappy. Stay or leave?"
🚀 Opportunity"Everyone I know is making money in X. Should I jump in?"
🎓 Education"Drop out to build my startup or finish the degree?"
🏠 Life"Move to a new city for a job I'm not sure about?"

📊 Real Example Output

Input: "I have a stable $120K job. A startup offered me $90K base + equity. Should I switch?"

Agent Output (condensed):

Reframe: This is a forward-only EV question. Your tenure at your current job is a sunk cost — it doesn't vote.

EV Calculation:

  • Option A (Stay): EV = $120K (certain)
  • Option B (Startup): 60% success → $250K total comp | 40% failure → $70K EV = (0.60 × $250K) + (0.40 × $70K) = $178K

Base Rate: Startup success rate ~20-30%. Equity actually vesting: ~40% of that. Adjusted EV drops to ~$145K — still higher than staying.

Sunk Cost flagged: "I've invested 4 years here" — removed from calculation.

Survivorship Bias: You're hearing about startup wins on LinkedIn. The failures don't post. Denominator matters.

Kelly Sizing: Don't quit cold. Negotiate a 3-month notice. Build runway. That's your quarter-Kelly move.

Verdict: The math favors the switch IF you have 6+ months of savings as runway. If you don't — build it first, then switch. The opportunity isn't going anywhere.


📚 The Intellectual Foundation

These 6 frameworks aren't invented here — they're sourced from the best thinkers in decision science:

FrameworkSource
Expected ValueStandard probability theory
Base Rate NeglectDaniel Kahneman — Thinking, Fast and Slow
Sunk Cost FallacyRichard Thaler — Behavioral Economics
Bayesian UpdatingThomas Bayes + Philip Tetlock — Superforecasting
Survivorship BiasNassim Taleb — Fooled by Randomness
Kelly CriterionJohn Kelly + William Poundstone — Fortune's Formula

Recommended reading if you want to go deeper:

  • Thinking, Fast and Slow — Daniel Kahneman
  • Superforecasting — Philip Tetlock
  • Fooled by Randomness — Nassim Taleb
  • Fortune's Formula — William Poundstone
  • Thinking in Bets — Annie Duke
  • The Signal and the Noise — Nate Silver

🗂️ Repo Structure

good-decisions-only/
├── decision-analyst.md   ← The full agent prompt (paste this as system prompt)
├── README.md             ← You are here
├── CHANGELOG.md          ← Version history
└── LICENSE               ← MIT

🤝 Contributing

Found a framework that should be added? A bias that's missing? PRs are welcome.

  1. Fork the repo
  2. Create a branch: git checkout -b improve-framework
  3. Make your changes
  4. Submit a PR with a clear explanation of what you added and why

🏗️ Built With

Agent structure inspired by msitarzewski/agency-agents. Frameworks sourced from Kahneman, Taleb, Tetlock, Thorp, Silver, and Duke.


📜 License

MIT — use freely, commercially or personally.


Ver o original no GitHub

Bom para

Relacionados