This project provides an automated, agent-driven stock analysis engine for A-shares, Hong Kong, and US markets. It aims to democratize in-depth financial analysis by simulating the insights of 51 diverse investment personalities and applying 17 institutional methodologies, delivering comprehensive reports using only free data sources. The tool is designed for investors and power users seeking a sophisticated, multi-dimensional view of stocks without the cost of traditional terminals or proprietary data.

How It Works

The UZI-Skill employs a two-stage pipeline orchestrated by AI agents. Initially, scripts gather data across 22 dimensions from free sources and set up quantitative models. An agent then intervenes, role-playing as 51 distinct investors (from value to growth to macro hedge fund styles), each with unique quantitative rules. The agent synthesizes these simulated perspectives, providing qualitative justifications that can override mechanical scores, focusing on real-world investment logic like portfolio fit and industry affinity. A mandatory 13-point mechanical self-review gate (v2.9+) ensures critical data integrity and consistency before the final report is generated.

Quick Start & Requirements

• Primary Install: Integrate via agent marketplaces (e.g., Claude Code /plugin marketplace add wbh604/UZI-Skill). Specific instructions are provided for various agent frameworks like Cursor, Gemini CLI, and Codex.
• Prerequisites: Python 3.9+. No API keys are strictly required, as it relies on free data sources. Optional configuration (e.g., MX_APIKEY for Eastmoney API) can enhance performance in restricted network environments.
• Resource Footprint: Full analysis takes approximately 5-8 minutes per stock, primarily due to data acquisition.
• Links: Installation guides are available within the README and linked external files (e.g., .codex/INSTALL.md, AGENTS.md).

Highlighted Details

• Simulated Investor Panel: Features 51 simulated investors with distinct quantitative rule sets (180 rules total) and investment styles, offering diverse perspectives.
• 17 Institutional Analysis Methods: Incorporates methodologies like DCF, Comps, LBO, and institutional report formats (JPM/GS style).
• Free Data Aggregation: Leverages multiple free sources (e.g., Eastmoney, Xueqiu, akshare, yfinance) with fallback mechanisms, eliminating API key requirements.
• Mandatory Self-Review Gate (v2.9+): A critical 13-point automated check prevents report generation if significant issues (e.g., incorrect industry classification, missing data) are detected, enhancing reliability.
• Multi-Format Output: Generates HTML reports, shareable social media images, and concise text summaries.

Maintenance & Community

The project is under active development with frequent updates (e.g., v2.9.0 released April 17, 2026). The README notes that the current version is "not very stable" and encourages user testing and feedback via a WeChat group or GitHub Issues. The primary developer is listed as FloatFu-true.

Licensing & Compatibility

• License: MIT License.
• Compatibility: Designed for integration with various AI agent frameworks. No explicit restrictions on commercial use are noted beyond standard MIT terms.

Limitations & Caveats

The project is described as having stability issues and a significant number of reported bugs, requiring active user testing. Data acquisition can be slow, and some data sources may require opt-in login or specific network configurations (e.g., using MX_APIKEY for mainland China users) to function reliably. Web search results, used for macro data, can be inconsistent.