druckenmiller

Druckenmiller Conviction Skill

A Claude Code skill that channels Stanley Druckenmiller's investment framework — liquidity-first analysis, conviction scoring, and position sizing, delivered in his voice.

Every day the pipeline runs 4 signals (Liquidity, Forward Earnings, Market Breadth, Price Signal) and produces a conviction JSON. The skill reads that JSON and responds like Druckenmiller.

---

Quick Start — Path A (Use Shared Data, 2 minutes)

You use the live data endpoint. No API keys needed, no pipeline to run.

1. Install the skill

claude skill install https://github.com/a0981456759/druckenmiller-skill

Or manually: copy skills/druckenmiller/SKILL.md and skills/druckenmiller/PERSONA.md into your Claude Code skills folder.

2. Ask Claude

今天市場怎樣？
Druckenmiller 今天怎麼看？
現在該持多少倉位？

That's it. The skill fetches from https://druckenmiller-skills.vercel.app automatically.

Data is updated once per weekday via the pipeline. If today's report isn't ready yet, the skill will tell you.

---

Self-Hosting — Path B (Run Your Own Pipeline)

Fork the repo and run your own data pipeline. Full control over data freshness and API usage.

Prerequisites

• Python 3.10+
• FRED API key (free)
• FMP API key (free tier works)
• Vercel account (free)

Setup

1. Fork this repo

2. Add API keys as GitHub Secrets

In your fork: Settings → Secrets and variables → Actions

Secret	Value
FRED_API_KEY	Your FRED key
FMP_API_KEY	Your FMP key

3. Deploy to Vercel

Connect your forked repo to Vercel. The public/ folder is the static site root — Vercel will serve public/reports/ automatically.

4. Update the skill URL

In skills/druckenmiller/SKILL.md, replace:

https://druckenmiller-skills.vercel.app

with your own Vercel deployment URL.

5. Run the pipeline

Trigger manually from GitHub Actions (Actions → Daily Conviction Pipeline → Run workflow), or run locally:

pip install -r requirements.txt
cp .env.example .env  # fill in your API keys

python liquidity-regime/scripts/liquidity_regime.py --output-dir public/reports/
python forward-earnings/scripts/forward_earnings.py --output-dir public/reports/
python market-breadth/scripts/market_breadth.py --output-dir public/reports/
python price-signal/scripts/price_signal.py --output-dir public/reports/
python conviction-synthesizer/scripts/conviction_synthesizer.py --reports-dir public/reports/ --output-dir public/reports/

Note: GitHub Actions IP is blocked by Yahoo Finance. Run the pipeline locally or on a VPS, then push the reports.

---

How It Works

Pipeline (Python)
  ├── liquidity-regime/     → FRED + FMP data → liquidity score
  ├── forward-earnings/     → FMP earnings data → analyst revision trend
  ├── market-breadth/       → breadth indicators → participation health
  ├── price-signal/         → earnings vs. price reaction → divergence detection
  └── conviction-synthesizer/ → weighted score → conviction_YYYY-MM-DD.json

Vercel
  └── public/reports/conviction_YYYY-MM-DD.json  (static file hosting)

Claude Code Skill
  ├── skills/druckenmiller/SKILL.md   → fetches JSON, interprets signals
  └── skills/druckenmiller/PERSONA.md → Druckenmiller's voice and decision logic

Conviction Score → Position

Score	Zone	Equity Allocation
85–100	Fat Pitch	90–100% — swing hard
70–84	High Conviction	70–89% — add aggressively
50–69	Moderate	50–69% — hold, wait for catalyst
30–49	Low Conviction	20–49% — reduce, cash is a position
0–29	Capital Preservation	0–19% — maximum defense

---

Signal Weights

Signal	Weight	Druckenmiller's Take
Liquidity Regime	35%	"It's liquidity that moves markets, not earnings."
Forward Earnings	25%	Analyst revisions reflect what consensus hasn't priced in yet
Market Breadth	25%	Broad participation = healthy; mega-cap only = 1987 warning
Price Signal	15%	Stock beats earnings but drops = bad news preview 6 months out

---

Other Languages

• 繁體中文
• 简体中文

---

License

MIT — use freely, attribution appreciated.

Data sources: Yahoo Finance, FRED, Financial Modeling Prep. For research only, not investment advice.