AI Hedge Fund

Run an AI-powered hedge fund simulation with 16+ legendary investor agents. Each agent analyzes stocks from their unique investment philosophy. Auto-detects...

Security Scan

The skill mostly matches its stated purpose (multi-agent stock analysis) but contains multiple inconsistencies and risky instructions — notably cloning and pip-installing an external GitHub repo and reading the host OpenClaw config file — that warrant caution before installing.

Purpose & Capability

The name/description (AI-powered hedge fund simulation) aligns with the included code and instructions (runs a multi-agent analysis using an external repo). However there are mismatches: SKILL.md claims model autodetection checks $OPENCLAW_LLM_MODEL and a default fallback, but the provided run.py only reads /root/.openclaw/openclaw.json and will exit if that file is missing. Some runtime paths in SKILL.md (e.g., /data/workspace/skills/skills/...) don't match the repository layout, suggesting sloppy/inconsistent documentation.

Instruction Scope

SKILL.md tells the agent/user to git clone an external GitHub repository into /data/workspace and pip install it editable (pip install -e), then run the project's main script. Those steps will write third-party code to disk and run it. The run.py also reads /root/.openclaw/openclaw.json (host agent config) to detect the model, which goes beyond the skill's core task of stock analysis and accesses host configuration. The instructions also claim behavior (env var check and default model) that the code does not implement.

Install Mechanism

There is no registry install spec, but SKILL.md directs cloning from a public GitHub repo and performing pip install -e, which executes arbitrary setup/install hooks from upstream code. Installing unreviewed third-party code via pip from a repo is a moderate-to-high risk action and should be audited before running.

Credentials

The skill declares no required env vars, but run.py reads /root/.openclaw/openclaw.json to obtain the 'primary' model. Reading a root-level OpenClaw config is disproportionate because it may expose other agents' configuration (and potentially sensitive fields) beyond what's needed for a single simulation. SKILL.md's claim to also consult $OPENCLAW_LLM_MODEL is not reflected in the code, increasing confusion about what the skill will access.

Persistence & Privilege

The skill is not forced-always and does not request persistent platform privileges. It does, however, instruct cloning code into /data/workspace and pip installing it (local persistence of third-party code). The only direct sensitive host access is reading /root/.openclaw/openclaw.json; the skill does not modify other skills or global agent settings in the provided files.

What to consider before installing

This skill performs what it claims (multi-agent stock analysis) but has worrisome implementation details. Before using it: 1) Inspect the upstream repository (https://github.com/virattt/ai-hedge-fund) — review its setup.py/pyproject and src/main.py for network calls, secrets handling, or arbitrary shell execution. 2) Avoid blindly running the provided pip install -e; clone in an isolated environment or container and run there. 3) Prefer supplying --model manually instead of relying on the skill's autodetect (the code reads /root/.openclaw/openclaw.json which can expose host agent configuration). 4) If you must install, run tests and a static code audit on the repository first. 5) If you lack the ability to audit the upstream code, treat this skill as potentially risky and do not run it on systems that hold secrets or production credentials.

AI Hedge Fund Skill

Multi-agent investment analysis using legendary investor personas.

Project: https://github.com/virattt/ai-hedge-fund

Setup (first time)

# Clone project
git clone https://github.com/virattt/ai-hedge-fund /data/workspace/ai-hedge-fund

# Install
pip install -e /data/workspace/ai-hedge-fund --break-system-packages --quiet
pip install yfinance --break-system-packages --quiet

Run

Via skill script (auto-detects model):

python3 /data/workspace/skills/skills/ai-hedge-fund/scripts/run.py \
  --tickers NVDA \
  --analysts-all \
  --show-reasoning

Or direct CLI:

cd /data/workspace/ai-hedge-fund
python3 src/main.py \
  --tickers NVDA \
  --start-date 2026-01-01 \
  --end-date 2026-04-12 \
  --model "minimax/minimax-m2.5:free" \
  --analysts-all \
  --show-reasoning

Model Auto-Detection

Not hardcoded. The run.py script auto-detects from:

1. $OPENCLAW_LLM_MODEL environment variable
2. OpenClaw config (/root/.openclaw/openclaw.json)
3. Default: minimax/minimax-m2.5:free

You can also specify manually via --model flag.

Available Models

Any model from src/llm/api_models.json, e.g.:

• minimax/minimax-m2.5:free (default, free)
• qwen/qwen3.6-plus:free (free)
• gpt-4.1 (OpenAI)
• claude-sonnet-4-6 (Anthropic)

All 16+ Agents

Agent	Philosophy
Warren Buffett	Wonderful business at fair price
Charlie Munger	Only wonderful businesses
Ben Graham	Margin of safety
Michael Burry	Contrarian deep value
Cathie Wood	Innovation & disruption
Bill Ackman	Activist investor
Peter Lynch	Ten-baggers
Phil Fisher	Scuttlebutt
Stanley Druckenmiller	Macro asymmetric
Nassim Taleb	Antifragility
Mohnish Pabrai	Dhandho
Rakesh Jhunjhunwala	Big Bull
Aswath Damodaran	Valuation
+ Quant agents	Fundamentals, Technical, Valuation, Sentiment

Output

Each agent outputs: signal (bullish/bearish/neutral), confidence %, reasoning. Portfolio Manager aggregates into final BUY/SELL/HOLD.

Data Source

yfinance (free, no API key needed)