maia-skill

Tododeia — Investment Analysis Skill for Claude Code

by @quebert · modCarlos/maia-skill

A Claude Code skill that pre-fetches real market data for 60+ tickers using Python tools, then spawns a single MegaAgent to produce a risk-adjusted portfolio with investment theses, carry-forward of active positions, and a Next.js dashboard.

Leer en Español

---

How It Works

  "run tododeia" / "analyze markets"
           │
           ▼
  ┌─────────────────────┐
  │  Risk profile prompt │  conservative / moderate / aggressive
  └────────┬────────────┘
           │
           ▼  Python tools (no LLM needed)
  ┌─────────────────────────────────────────────────────┐
  │  pre_fetch.py       — yfinance data for 60 tickers  │
  │  news_fetch.py  ┐                                   │
  │  sec_risk_fetch ┘ parallel — headlines + 10-K risks │
  │  accuracy_windows.py — 1d/5d/30d pick accuracy      │
  │  update_stops.py     — ATR trailing stops           │
  │  compress_context.py — compact ~7,500 char context  │
  └────────────────────────┬────────────────────────────┘
                           │  /tmp/mega_context.txt
                           ▼
              ┌────────────────────────┐
              │       MegaAgent        │
              │   (Claude Sonnet 4.x)  │
              │                        │
              │  • Applies CORRELATION │
              │    LIMITS (hard rules) │
              │  • Carry-forward of    │
              │    active positions    │
              │  • 13 risk-adj. picks  │
              │    with thesis +       │
              │    invalidators        │
              └────────────┬───────────┘
                           │
           ┌───────────────┼───────────────┐
           ▼               ▼               ▼
   output/history/   active_positions  dashboard/
   YYYY-MM-DD.json       .json         public/data/
                                           │
                                           ▼
                                  Next.js localhost:3420

What Changed from the Original

This fork diverged significantly from the upstream project:

Feature	Original	This fork
Architecture	5 parallel AI agents (4 sector + 1 strategy)	Pre-fetch tools + 1 MegaAgent
Asset universe	Dynamic discovery by agents	Fixed 60-ticker screened list
Data source	Agent web searches	yfinance (real market data, no LLM searches)
Sectors	Crypto, Stocks, Forex, Commodities	Stocks + commodities/gold integrated
Active positions	None	Carry-forward between sessions
Split detection	None	Hard exclusion when split detected (±60 days)
Accuracy tracking	Basic	1d / 5d / 30d windows with outlier filtering
Trailing stops	None	ATR-based, updated every run
Context to LLM	Raw per-sector JSON	Compressed 7,500-char structured block

Features

• Pre-fetched real data: RSI, trend, support/resistance, ATR, Altman Z, Piotroski score, insider signals — all from yfinance, zero LLM web searches for technicals
• Risk profiles: Conservative, moderate, aggressive — position sizes and allocation change accordingly
• Carry-forward: Active positions persist between sessions; excluded only when thesis explicitly invalidated
• Split detection: Tickers with recent stock splits are automatically excluded from the screener
• Multi-window accuracy: 1-day, 5-day, 30-day accuracy tracked per session — fed back to the MegaAgent
• Trailing stops: ATR-14 based stops that only move up, updated every run
• Interactive dashboard: Next.js on localhost:3420 with bilingual support (EN/ES)
• History: Every session saved to output/history/YYYY-MM-DD.json for backtesting and accuracy tracking

Requirements

• Claude Code
• Python 3.9+ with yfinance (pip install yfinance)
• Node.js 18+ (for the dashboard)
• Internet connection (yfinance + news fetch need it)

Installation

Option A: One-liner

curl -sL https://raw.githubusercontent.com/modCarlos/maia-skill/main/install.sh | bash

Clones the repo, symlinks the skill into Claude Code, and installs dashboard dependencies.

Option B: Manual

# 1. Clone
git clone https://github.com/modCarlos/maia-skill.git

# 2. Symlink into Claude Code
mkdir -p ~/.claude/plugins
ln -s "$(pwd)/maia-skill" ~/.claude/plugins/maia-skill

# 3. Install Python dependencies
pip install yfinance pandas numpy requests

# 4. Install dashboard dependencies
npm install --prefix maia-skill/dashboard

Usage

In any Claude Code conversation:

"run tododeia"
"run tododeia moderate"
"analyze markets aggressive"
"investment analysis conservative"

You'll be asked your risk profile if not provided, then the pipeline runs. The report opens at http://localhost:3420.