Why this matters
Human-led fundamental research scales poorly: it's slow, brittle, and prone to bias. This project uses multiple LLM-powered agents to simulate a small investment research team and turn qualitative frameworks (Buffett, Munger, Duan Yongping, Li Lu) into repeatable, auditable Skills. The core insight is not “more AI” but “structured adversarial thinking + guaranteed data rigor” so outputs can be treated as inputs to decisions rather than vague essays.
What Sets It Apart
- Four-master adversarial framework: research is produced from four distinct thinking lenses (business essence, moat/valuation, inversion/risks, long-term civilizational trends) to surface real tensions rather than a single balanced narrative — outputs include forceful pass/fail/gray conclusions and price ranges.
- Multi-agent parallelism: each Skill runs multiple independent agents in parallel (e.g., commercial, financial, industry, management) and then a Team Lead aggregates findings, increasing coverage and reducing single-agent blind spots.
- Financial rigor and reproducibility: built-in tools perform precise decimal-based valuation checks, multi-source cross-validation, Benford tests and three-scenario valuations, and the workflow enforces consistent output formats for longitudinal comparison and backtesting.
- Skill-first, Claude Code / Codex integration: delivered as a set of command/skill files to run inside Claude Code or Codex environments, enabling scripted workflows (investment-research, investment-team, earnings-review, industry-funnel, etc.).
Who it's for — and tradeoffs
Great fit if you are an individual investor or small research team wanting disciplined, auditable research templates that produce actionable recommendations (buy/sell/hold with price bands) and you already use or can run Claude Code/Codex. The repo suits users who accept some engineering setup to get deterministic, repeatable outputs.
Look elsewhere if you need turnkey portfolio automation, full regulatory-grade compliance, or purely quantitative systematic strategies — this is a qualitative/quantitative hybrid focused on research discipline and decision hygiene rather than automated execution.
Where it fits
Positions between ad-hoc LLM prompting and full prod-grade investment platforms: it converts investment heuristics into agentized Skills with explicit anti-bias checks and calculation tooling. Use it to scale analyst workflows, enforce decision rules, and produce reproducible research that can be re-run and compared over time.
