ConsensusAI

investment analysis that aggregates signals from 21 analyst personas to generate consensus-based trading recommendations.

Overview

ConsensusAI uses Claude Code Skills to simulate multiple investment analysts, each with their own philosophy and methodology. The system aggregates their signals to produce consensus recommendations.

Analyst Personas (12)

Analyst	Style
Warren Buffett	Value investing, intrinsic value, competitive moats
Charlie Munger	Mental models, quality-focused, multidisciplinary
Ben Graham	Deep value, Graham Number, margin of safety
Peter Lynch	GARP, PEG ratios, 10-baggers
Phil Fisher	Scuttlebutt method, 15-point checklist, R&D focus
Michael Burry	Contrarian deep value, distressed, short positions
Mohnish Pabrai	Dhandho framework, asymmetric payoffs
Bill Ackman	Activist investing, turnarounds, governance
Cathie Wood	Disruptive innovation, AI, high-growth
Stanley Druckenmiller	Macro-driven, sector rotation, trend-following
Rakesh Jhunjhunwala	Emerging markets, macro-aware growth
Aswath Damodaran	Rigorous DCF, valuation models, WACC

Analysis Skills (9)

Skill	Function
Fundamentals	Financial statements, ratios, profitability
Technicals	Chart patterns, indicators, momentum
Valuation	Multi-model DCF, comparables, asset-based
Sentiment	Insider trades, institutional ownership
News Sentiment	News flow, media coverage, event-driven
Growth Analyst	Revenue acceleration, margin expansion
Risk Manager	Position sizing, portfolio constraints
Portfolio Manager	Aggregate signals, final trading decisions
Financial Data	Fetch metrics, prices, news, insider trades

Installation

# Clone repository
git clone https://github.com/ancs21/ai-sub-invest.git
cd ai-sub-invest

# Install dependencies (requires Python 3.11+)
uv sync

# Set up API key
cp .env.example .env
# Edit .env and add: FINANCIAL_DATASETS_API_KEY=your-key

# Optional: Install SDK for programmatic access
uv add claude-agent-sdk

Usage

Via Claude Code Chat

Simply ask Claude to analyze stocks:

Analyze AAPL like Warren Buffett

Run all analysts on TSLA and give me a trading decision

What's the valuation for MSFT?

Example Prompts

Task	Prompt
Single analyst	Analyze NVDA using Charlie Munger's approach
Multiple analysts	Run Buffett, Graham, and Lynch on META
Full analysis	Aggregate all signals for AMZN
Specific focus	What's the DCF valuation for GOOGL?
Comparison	Compare AAPL and MSFT using fundamentals
Technical	Show me the technical analysis for SPY
Sentiment	What's the insider sentiment on TSLA?

Via SDK

# Single ticker
uv run python src/sdk_main.py AAPL 2025-12-06

# Multiple tickers
uv run python src/sdk_main.py AAPL,MSFT,NVDA 2025-12-06

# With custom portfolio cash
uv run python src/sdk_main.py AAPL,GOOGL 2025-12-06 --cash 50000

# Specific analysts only
uv run python src/sdk_main.py NVDA 2025-12-06 --analysts warren-buffett,cathie-wood,technicals

# Verbose output (see intermediate steps)
uv run python src/sdk_main.py TSLA 2025-12-06 -v

# Save results to file
uv run python src/sdk_main.py AAPL,MSFT 2025-12-06 -o results.json

# Full example with all options
uv run python src/sdk_main.py AAPL,MSFT,GOOGL 2025-12-06 \
  --start-date 2025-06-01 \
  --cash 100000 \
  --analysts warren-buffett,ben-graham,technicals,valuation \
  --verbose \
  --output analysis.json

SDK Options

Flag	Description	Default
tickers	Comma-separated tickers	Required
end_date	Analysis date (YYYY-MM-DD)	Required
--start-date	Historical data start	6 months before end
--cash	Initial portfolio cash	100,000
--analysts	Specific analysts to use	Core 6 analysts
--verbose, -v	Show intermediate output	False
--output, -o	Save JSON to file	stdout

Run Individual Skills

# Warren Buffett analysis
uv run python .claude/skills/warren-buffett/scripts/analyze.py AAPL 2025-12-06

# Technical analysis (requires date range)
uv run python .claude/skills/technicals/scripts/analyze.py AAPL 2024-06-01 2025-12-06

# Risk calculation
uv run python .claude/skills/risk-manager/scripts/calculate.py AAPL 2025-12-06 100000

# Aggregate signals
uv run python .claude/skills/portfolio-manager/scripts/aggregate.py '{
  "warren_buffett": {"signal": "bullish", "confidence": 85},
  "ben_graham": {"signal": "neutral", "confidence": 52}
}'

Output Format

All analysts return JSON with:

{
  "ticker": "AAPL",
  "signal": "bullish",
  "confidence": 85,
  "score": 18,
  "max_score": 22,
  "reasoning": "...",
  "market_cap": 3500000000000
}

Portfolio manager aggregation returns:

{
  "action": "buy",
  "confidence": 75,
  "reasoning": "Bullish consensus (5/8 analysts) with 75% avg confidence",
  "consensus_details": {
    "bullish_count": 5,
    "bearish_count": 1,
    "neutral_count": 2,
    "weighted_score": 0.45
  }
}

Project Structure

consensusai/
├── .claude/
│   └── skills/           # 21 analyst skills
│       ├── warren-buffett/
│       ├── ben-graham/
│       ├── portfolio-manager/
│       └── ...
├── src/
│   ├── data/
│   │   ├── cache.py      # API response caching
│   │   └── models.py     # Pydantic models
│   ├── tools/
│   │   └── api.py        # Financial data API
│   ├── utils/
│   │   └── api_key.py    # API key validation
│   └── sdk_main.py       # SDK entry point
├── pyproject.toml
└── README.md

Disclaimer

This project is for educational and research purposes only.

• Not intended for real trading or investment
• No investment advice or guarantees provided
• Consult a financial advisor for investment decisions
• Past performance does not indicate future results

Inspiration

Inspired by ai-hedge-fund.

License

MIT License