This Claude Skill Runs a Billionaire's Brain Every Day

This Claude Skill Runs a Billionaire's Brain Every Day

What if you could ask Stanley Druckenmiller what he thinks about today's market — and get an answer in under 3 seconds?

Not a book summary. Not a podcast recap. His actual framework, running on live data, weighted exactly how his decades of 30%+ annual returns proved it should be weighted. The secret most retail investors miss isn't stock-picking skill. It's that billion-dollar managers operate on systems, not gut feelings. They don't read more news — they read better signals. And until now, those systems lived behind fortress walls at Duquesne Capital, accessible only to the ultra-wealthy.

That just changed. druckenmiller.skill is a Claude Code skill that weaponizes one of history's greatest macro investors for your daily market decisions. It doesn't predict. It doesn't hedge. It delivers weighted conviction scores based on four battle-tested signals — liquidity regime, forward earnings, market breadth, and price action — then responds in Druckenmiller's unmistakable voice: direct, unsparing, brutally honest.

Ready to stop guessing and start knowing? Here's everything you need to deploy this in the next 10 minutes.

What Is druckenmiller.skill?

druckenmiller.skill is a Claude Code skill built by developer xingpt88 that operationalizes Stanley Druckenmiller's legendary investment framework into an automated daily analysis pipeline. For the uninitiated: Druckenmiller managed George Soros's Quantum Fund, founded Duquesne Capital, and generated 30% annualized returns over 30 years — one of the greatest track records in financial history.

The skill isn't a chatbot costume. It's a structured decision architecture that replicates how Druckenmiller actually thinks:

• Liquidity-first hierarchy: Fed policy and credit conditions drive everything else
• No-prediction discipline: Score today's conditions, don't forecast tomorrow's prices
• Fat-pitch selectivity: Swing hard only when all signals align (85-100 score)
• Capital preservation reflex: When conviction collapses, cash is a position

What makes this trending now? Three converging forces: Claude Code's skill system (launched 2024) finally enables portable, reproducible AI workflows; retail traders are desperate for systematic edges after the 2022-2024 volatility roller coaster; and macro uncertainty — Fed pivot timing, AI bubble dynamics, geopolitical fragmentation — has made Druckenmiller's liquidity-centric framework more relevant than ever. The repo's minimalist design (just two files: SKILL.md and PERSONA.md) belies its sophistication — this is institutional-grade thinking compressed into developer-friendly infrastructure.

The Four Signals That Actually Move Markets

Druckenmiller famously said: "It's liquidity that moves markets, not earnings." Most investors do the opposite — they obsess over earnings reports while missing the credit tide going out. This skill corrects that fatal inversion with four weighted signals:

1. Liquidity Regime (35% Weight)

The kingmaker signal. Measures Federal Reserve direction, credit conditions, and systemic money flow. When liquidity contracts, even great companies get revalued downward. When it expands, mediocre companies float higher. This 35% weighting reflects Druckenmiller's core insight from the 1987 crash, the dot-com bust, and the 2022 bear market: macro liquidity trumps micro fundamentals.

2. Forward Earnings (25% Weight)

Not trailing earnings — revision trends. What are analysts quietly upgrading or downgrading before it hits headlines? This captures consensus mispricing. If 60% of S&P 500 companies see estimate cuts but prices haven't adjusted, that's actionable divergence.

3. Market Breadth (25% Weight)

Is the rally broad-based or concentrated in 7 mega-caps? Narrow leadership (think Q1-Q2 2024) is a classic warning sign. Healthy breadth means participation, durability, and lower drawdown risk. This signal caught the 2021 ARKK collapse and the 2023 regional banking stress before they exploded.

4. Price Signal (15% Weight)

The contrarian canary: earnings beats that sell off. When NVDA, MSFT, AMD report strong numbers and drop anyway, that's not "profit-taking." That's informed capital exiting. Druckenmiller treats this as a 6-month leading indicator — the market's way of whispering that the cycle is turning.

These weights aren't arbitrary. They're extracted from Druckenmiller's documented decisions, interview transcripts, and Soros-era position structures. The skill doesn't just calculate — it explains the divergence between what you're reading and what the market is actually saying.

Where This Skill Destroys Traditional Analysis

Scenario 1: The "Everything Is Fine" Trap

Your Bloomberg terminal shows green across the board. Earnings season beat rate: 78%. But druckenmiller.skill returns score 47, Low Conviction. Why? Liquidity regime turned neutral-to-negative, breadth collapsed to 30% of stocks above 50-day moving averages, and three mega-caps sold off post-earnings. You reduce exposure before the correction that wipes out momentum traders.

Scenario 2: The Fat Pitch Nobody Sees

October 2023: rates peak, credit spreads stabilize, earnings revisions inflect positive, breadth expands to 75% participation. Score hits 91. The skill screams "swing hard" while Twitter debates recession probabilities. You deploy 95% equity — catching the Q4 rally that leaves cash-holders demoralized.

Scenario 3: The Earnings Beat Headfake

Your favorite AI semiconductor crushes estimates. Stock drops 4% after hours. Traditional analysis: "buy the dip." Druckenmiller skill: "Price signal triggered — this is 6-month warning, not entry." You wait. Three months later, inventory glut confirmed, stock down 35%.

Scenario 4: The Bilingual Edge

Global macro doesn't sleep when US markets close. The skill's multilingual support (English, 简体中文) lets you query conviction in Mandarin during Asian session, then switch to English for London open — maintaining consistent framework application across time zones and language barriers.

Step-by-Step Installation & Setup Guide

Deploying druckenmiller.skill takes under 5 minutes. You need Claude Code installed (available via Anthropic's console or API access).

Method 1: Direct Download (Recommended)

From your project root, create the skill directory and fetch both required files:

# Create the Claude Code skills directory structure
mkdir -p .claude/skills/druckenmiller

# Download the core skill definition — contains signal weights, scoring logic, zone mappings
curl -o .claude/skills/druckenmiller/SKILL.md \
  https://raw.githubusercontent.com/a0981456759/Druckenmiller/main/SKILL.md

# Download the persona layer — governs voice, personality, decision rules under pressure
curl -o .claude/skills/druckenmiller/PERSONA.md \
  https://raw.githubusercontent.com/a0981456759/Druckenmiller/main/PERSONA.md

Method 2: Clone and Symlink (For Contributors)

# Clone the full repository for inspection, modification, or contribution
git clone https://github.com/a0981456759/Druckenmiller.git

# Manually copy both files into your Claude Code skills folder
# This preserves your ability to git pull updates without reconfiguring paths
cp Druckenmiller/SKILL.md .claude/skills/druckenmiller/
cp Druckenmiller/PERSONA.md .claude/skills/druckenmiller/

Verification

Launch Claude Code and test activation with any of these prompts:

今天市场怎样？
What's the conviction today?
Should I add to my position?
Druckenmiller, what do you see?

The skill auto-detects and fetches today's conviction JSON. If the daily report isn't generated yet, Claude responds transparently — no hallucinated scores, no fabricated data. This integrity layer is critical: the skill refuses to guess when signal integrity is compromised.

REAL Code Examples: Inside the Machine

Let's dissect the actual architecture powering this skill. These patterns from the repository reveal how institutional-grade decision-making gets encoded into AI-native infrastructure.

Example 1: The Five-Layer Persona Architecture

Before any number is crunched, the persona layer enforces cognitive discipline. From PERSONA.md:

## How It Responds

The skill runs in two layers:

**1. Persona first** — before any analysis, Druckenmiller's five-layer personality structure governs the response:
- Layer 0: Core rules (liquidity first, no hedging, no predictions)
- Layer 1: Identity (who he is, how he sees his job)
- Layer 2: Expression style (short sentences, conclusion before reasoning)
- Layer 3: Decision logic (fat pitch threshold, exit rules)
- Layer 4: Behavior under pressure (losing streaks, disagreement, euphoria)

Why this matters: Most AI "personas" are superficial — a few adjectives in a system prompt. This is behavioral architecture. Layer 0 prevents the AI from doing what every amateur does: predicting price targets. Layer 2 enforces cognitive efficiency — Druckenmiller famously said "I make decisions in 30 seconds"; the skill mirrors that brevity. Layer 4 is the secret weapon: it governs responses during drawdowns when most systems (and humans) panic. This isn't decoration — it's drawdown control encoded into response generation.

Example 2: Signal Weighting and Scoring Logic

The core scoring matrix from SKILL.md:

| Signal | Weight | What It Measures |
|--------|--------|-----------------|
| Liquidity Regime | 35% | Fed direction, credit conditions — the only thing that moves markets |
| Forward Earnings | 25% | Analyst revision trend — what consensus hasn't priced in yet |
| Market Breadth | 25% | Participation health — broad rally vs. dangerous mega-cap concentration |
| Price Signal | 15% | Earnings beats that sell off — almost certainly a 6-month warning |

Implementation insight: These weights sum to 100% and create a non-linear scoring function. A liquidity score of 20 (crisis) with perfect earnings (100) and breadth (100) still yields: 0.35×20 + 0.25×100 + 0.25×100 + 0.15×100 = 62. That's "Moderate" zone — you don't get bailed out by one strong signal if the kingmaker fails. This prevents the classic investor error of averaging good news while ignoring the fatal factor.

Example 3: Conviction Zone Execution Table

The actionable output framework:

| Score | Zone | Equity | Action |
|-------|------|--------|--------|
| 85–100 | Fat Pitch | 90–100% | Swing hard — once or twice a year |
| 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 |

Critical implementation detail: Notice the non-linear equity allocation. The jump from "Moderate" to "High Conviction" adds 20 percentage points; from "High Conviction" to "Fat Pitch" adds another 20. But the drop from "Moderate" to "Low Conviction" is 30 points. This asymmetric risk response mirrors Druckenmiller's actual behavior: he builds positions slowly but exits fast. The skill encodes this convexity — slow to enter, quick to preserve capital.

Example 4: Real Response Output (Score 61)

Here's the actual output format when querying the skill:

> 流动性中性——转向迹象有，但没确认。这不是 fat pitch。
> 61.4，标准仓位，50–69%。不加码，等催化剂。
> 广度健康，盈利上调，这两个不错。但看价格——NVDA、MSFT、AMD 全都盈利好但股价跌。这是 6 个月后的坏消息预警，不是买点。
> *"When you don't see it, don't swing."*

Parse this response structure: First sentence — liquidity assessment with explicit non-confirmation (no false precision). Second — score, zone, exact equity range, and action boundary ("不加码" = don't add). Third — signal-by-signal breakdown, highlighting the divergence (breadth/earnings good, price signal warning). Fourth — behavioral quote reinforcing discipline. Every element serves a decision function. No filler, no hedging, no probabilistic waffle ("might go up, might go down").

Advanced Usage & Best Practices

Pro Tip 1: Batch Morning Routine

Set a cron job or GitHub Action to query conviction at 6:30 AM ET, before pre-market. Pipe output to Slack/Discord for team alignment. The skill's structured output (score, zone, equity%, key divergences) is machine-parseable for automated position sizing.

Pro Tip 2: Divergence Logging

When the skill flags "price signal" warnings (earnings beats selling off), log these in a tracking spreadsheet. Backtest: how many 6-month warnings materialized? This builds your personal validation dataset and calibrates trust in the framework.

Pro Tip 3: Layer 4 Stress Testing

Deliberately query during market crashes: "Druckenmiller, we're down 15% this month." The persona's Layer 4 (behavior under pressure) activates — watch how the response structure changes. Use this to pre-commit to discipline before emotional decisions arise.

Pro Tip 4: Multi-Language Arbitrage

Asian markets often price US macro signals with lag. Query in 简体中文 during Shanghai/Hong Kong hours for local context, then English for US open. The consistent framework across languages prevents cognitive regime switching.

druckenmiller.skill vs. The Competition

The verdict: Bloomberg has better data. Robo-advisors have better UX. But no commercial product operationalizes a verified billionaire framework with this transparency and zero cost. The skill's edge isn't information — it's cognitive architecture that prevents expensive mistakes.

FAQ: What Developers Ask Before Installing

Is this actually Stanley Druckenmiller's official tool?

No — it's an independent open-source project by xingpt88 that codifies Druckenmiller's publicly documented framework from interviews, speeches, and the Soros/Druckenmiller era. The persona layer is reconstructed from primary sources, not authorized. Treat as research infrastructure, not endorsed product.

Where does the daily conviction data come from?

The repository author updates conviction reports using Yahoo Finance, FRED (Federal Reserve Economic Data), and Financial Modeling Prep. The skill fetches this pre-computed JSON. You cannot self-calculate without rebuilding the data pipeline — but the skill architecture is extensible for custom data feeds.

Can I modify the signal weights?

Absolutely — SKILL.md is plaintext. Want to test Ray Dalio's all-weather weighting? Or your own tactical overlay? Fork the repo, adjust weights, redeploy. The MIT license permits commercial use with attribution.

What happens if the daily report isn't ready?

The skill explicitly tells you — no hallucination, no stale data. This integrity check is hard-coded: the persona refuses to generate scores when signal freshness is compromised.

Does this work with Claude API or only Claude Code?

Currently optimized for Claude Code's skill system. API deployment requires manual prompt injection of SKILL.md and PERSONA.md content. Community adaptations for LangChain, AutoGen, or OpenAI assistants are welcome contributions.

Is this investment advice?

Explicitly not. The README states: "Research only, not investment advice." This is a decision-support framework, not a signal service. Your capital, your responsibility, your due diligence.

How do I contribute or report issues?

Visit github.com/xingpt88/Druckenmiller — open issues for data quality concerns, PRs for framework improvements, or fork for private modifications.

Conclusion: Stop Reading, Start Asking

Here's the uncomfortable truth: you already have access to more market information than Stanley Druckenmiller did in 1985. What you lack is his filter — the weighted signal hierarchy that discards 90% of noise, the persona discipline that prevents emotional override, the convex risk response that preserves capital for asymmetric opportunities.

druckenmiller.skill isn't a crystal ball. It's a cognitive exoskeleton — the closest most developers will get to installing a billionaire's decision architecture on their own machine. For zero cost, five minutes of setup, and the price of Claude Code access, you get a daily interlocutor that won't flatter you, won't hedge for social comfort, and won't predict prices to make you feel smart.

It will tell you when to swing hard. When to hold. When cash is the position. And when the market is whispering warnings that earnings headlines drown out.

Install it now. Query it tomorrow morning. Compare its discipline to your last three trades. The gap you discover might be the most expensive thing you've ever measured — or the cheapest edge you'll ever acquire.

→ Get druckenmiller.skill on GitHub ←

Remember: "When you don't see it, don't swing." But when you do? Swing hard.