TL;DR
I took the excellent ai-hedge-fund project — a multi-agent system where AI versions of famous investors analyze stocks — and rebuilt it as a pure-prompt Claude Code skill.
No Python. No API keys. No financial data subscription. Just:
/sages NVDA
And 13 legendary investors — Buffett, Munger, Graham, Burry, Taleb, Cathie Wood, Druckenmiller, Ackman, Fisher, Lynch, Pabrai, Damodaran, Jhunjhunwala — walk into the room and argue about your stock.
Repo: https://github.com/hyhmrright/market-sages
Why rebuild it as a skill?
ai-hedge-fund is brilliant, but to run it you need a Python environment, multiple LLM API keys, a paid financial data API, and some patience with setup. That's a high bar for the "I just want to hear what Buffett thinks about NVDA on a Tuesday night" crowd — which is most of us.
Claude Code skills are basically a single markdown file that teaches Claude a new trick. That made me wonder: how much of ai-hedge-fund's value is in the orchestration code, and how much is in the personas and frameworks?
Turns out: a lot of the magic is in the frameworks. And frameworks compress beautifully into prompts.
What it looks like
╔══════════════════════════════════╗
║ 🧠 Warren Buffett ║
║ Signal: NEUTRAL ║
║ Confidence: 55% ║
║ Reasoning: Exceptional moat ║
║ (gaming ecosystem), but current ║
║ P/E leaves no margin of safety. ║
╚══════════════════════════════════╝
╔══════════════════════════════════╗
║ 🧠 Michael Burry ║
║ Signal: BEARISH ║
║ Confidence: 72% ║
║ Reasoning: FCF yield ~1.8%, ║
║ EV/EBIT ~80x — no deep value ║
║ here, this is momentum not math. ║
╚══════════════════════════════════╝
... (11 more sages)
╔══════════════════════════════════════════════╗
║ 🏦 PORTFOLIO MANAGER — FINAL VERDICT ║
║ Action: WATCH ║
║ Conviction: MEDIUM ║
║ Entry Strategy: Wait for 20-25% pullback ║
╚══════════════════════════════════════════════╝
Each sage has their own signal rules, valuation lens, and voice. Buffett will talk about moats and owner earnings. Burry will quote EV/EBIT and FCF yield. Taleb will refuse to give you a price target and lecture you about tail risk instead.
The design choices that mattered
1. Signal rules per investor, not one shared rubric.
Buffett's BULLISH is not Cathie Wood's BULLISH. If you give every agent the same "how to decide" rubric, they all sound the same. Each sage has their own thresholds (e.g., Pabrai wants downside <10% before even considering an idea).
2. Voice style is part of the spec.
Munger should be blunt. Fisher should be meticulous. Damodaran should sound like a professor. Specifying tone explicitly makes the output way more readable — you can scan 13 opinions because they don't blur together.
3. A "Portfolio Manager" synthesis agent at the end.
13 opinions is noise unless someone reconciles them. The final agent weighs confidence, flags disagreement clusters, and outputs an actionable verdict.
4. Let users paste fresh data.
LLMs have stale training data. So the skill explicitly tells Claude: if the user pastes financial figures or news, trust those over training knowledge.
/sages AAPL
Revenue: $391B (+2% YoY)
Net Income: $94B
FCF: $107B
P/E: 28x, EV/EBITDA: 22x
5. Cross-CLI portability.
The same skill.md works in Claude Code. I also wrote thin adapters (AGENTS.md, GEMINI.md) so it runs in Codex CLI and Gemini CLI. One prompt, three ecosystems.
Install (30 seconds)
Claude Code (recommended):
/plugin marketplace add hyhmrright/market-sages
/plugin install market-sages@market-sages-marketplace
Then: /sages AAPL
Codex CLI:
python3 ~/.codex/skills/.system/skill-installer/scripts/install-skill-from-github.py \
--repo hyhmrright/market-sages --path skills --name market-sages
Gemini CLI:
/extensions install https://github.com/hyhmrright/market-sages
Any other LLM: copy skill.md as your system prompt.
Fun things to try
/sages compare AAPL MSFT GOOG META AMZN
/sages TSLA --value
/sages NVDA @buffett @burry @taleb
Review my portfolio: 40% AAPL, 30% NVDA, 20% TSLA, 10% BTC
Caveats (real ones)
- Not financial advice. This is an educational/entertainment tool.
- LLMs hallucinate numbers. Paste real data for anything serious.
- These are AI impressions of how these investors think, not the actual investors.
Credit
Huge credit to @virattt and ai-hedge-fund for the original concept. Market Sages is a lighter, prompt-only sibling — if you want full multi-agent orchestration with real data pipelines, use ai-hedge-fund.
Repo + docs (中文 / 日本語 / 한국어 included): https://github.com/hyhmrright/market-sages
If it helps you think more clearly about a stock, a GitHub star is appreciated 🌟
Which sage would you add? I'm thinking about Soros (reflexivity) or Howard Marks (cycles) next — curious what you'd want to see in the council.
Top comments (1)
Thinking about adding Soros (reflexivity) or Howard Marks (cycles) next — curious which one people would find more useful. Also: anyone else running multi-persona prompt skills in Claude Code? Would love to hear what worked.