Automate value investing research workflows with AI agents
This framework enables individual developers to perform analysis comparable to that of a professional investment research team. By installing it in Claude Code or Codex, you gain access to 22 research commands that combine the investment philosophies of Buffett, Munger, Dan Yuming, and Li Lu. Rather than simply listing information, the system clarifies the basis for investment decisions by having AI agents with different perspectives collide and verify each other.
Structure that overcomes the limits of single AI analysis
Requesting corporate analysis from a general AI often yields ambiguous answers. This repository is designed to force clear conclusions of 'pass', 'fail', or 'gray area' along with specific price ranges. It also runs four AI agents with distinct investment perspectives simultaneously to synthesize their independent judgments. To ensure data accuracy, the system uses Python's precise calculation modules and cross-verifies key financial metrics from at least two independent sources. This structure reduces potential AI biases and calculation errors, leading to reproducible research results.
Command set for 22 research scenarios
The repository provides 22 skills tailored to various stages of investment research.
- In-depth corporate analysis Performs comprehensive analysis of a single company.
/investment-researchexecutes seven modules sequentially, while/investment-teamruns four agents in parallel to generate a comprehensive report at the fastest speed. - Industry and portfolio management Offers features to scan an entire industry's value chain (
/industry-research) or filter promising companies from the whole market (/industry-funnel). It also supports rebalancing existing portfolios (/portfolio-review) and tracking investment logic (/thesis-tracker). - Real-time event response Includes the
/news-pulseskill, which identifies the cause of sharp stock price fluctuations within 10 minutes. Instead of merely listing news, it analyzes the factors contributing to price changes through causal relationships and suggests response strategies.
Installation and execution environment
The framework is designed to work with both Claude Code and Codex AI clients. Users run the installation script suitable for their environment to register skills locally. Claude Code users install the client with npm install -g @anthropic-ai/claude-code and then execute the repository's install-claude-commands.sh script. Codex users prepare the client via npm install -g @openai/codex or the official installation script, then run install-codex-skills.sh. After installation, entering the relevant command in the AI client immediately starts the research process. Since token consumption can be high, it is recommended to use lighter skills for initial screening and more capable models for in-depth analysis.
Practical verification and precautions
The repository records actual investment returns of +69.29% and +66.38% for 2024 and 2025, respectively, indicating high excess returns compared to major indices. However, these returns are past results and do not guarantee future performance. The tool assumes that humans remain the primary decision-makers for investments. AI-generated reports are for reference only, and users bear the final responsibility for buy/sell decisions. Specifically, using the --dangerously-skip-permissions mode increases security risks, so it is advisable to use it only in trusted environments and with limitations. License information is not specified in the README, so verification is required before commercial use.
