How to Use the Investment Checklist Feature in AI Berkshire: A Complete Guide

The investment checklist feature in AI Berkshire evaluates stocks through a six-gate framework defined in skills/investment-checklist.md, spawning parallel agents to collect financial data and output deterministic Markdown reports with Pass/Fail/Gray verdicts.

The AI Berkshire repository provides an autonomous investment analysis framework that implements Warren Buffett's investment principles through structured AI agents. The investment checklist feature serves as the primary pre-buy screening tool, applying rigorous financial rigor to determine whether a stock meets fundamental quality standards. This guide explains how to invoke the checklist, interpret its six-gate evaluation, and leverage the parallel agent architecture for efficient multi-stock analysis.

How the Investment Checklist Architecture Works

The investment checklist operates across three distinct layers: Skill, Agent, and Tool.

Skill Definition Layer

The core logic resides in [skills/investment-checklist.md](https://github.com/xbtlin/ai-berkshire/blob/main/skills/investment-checklist.md). This markdown file defines the six-gate "pre-buy" process, information-richness ratings (A/B/C), and required data-collection steps. The skill parses the $ARGUMENTS string by commas, spaces, or Chinese punctuation to extract company symbols, then orchestrates the entire evaluation workflow.

Agent and Tool Integration

When invoked, the skill spawns specialized agents that delegate precise calculations to [tools/financial_rigor.py](https://github.com/xbtlin/ai-berkshire/blob/main/tools/financial_rigor.py). The checklist calls the verify-valuation and three-scenario sub-commands to compute ROE, margin-of-safety, and scenario-based target prices using exact decimal arithmetic.

Parallel Processing

For each ticker, the skill launches independent background agents using the generic Task utility. This parallelism allows concurrent data collection for profitability metrics, valuation multiples, growth trends, balance-sheet health, moat evidence, and management records.

Running the Investment Checklist

Claude Code CLI Method

After installing the AI Berkshire commands, invoke the checklist directly:

investment-checklist Moutai, NVIDIA, Apple

The command spawns parallel agents for each company and prints a summary table to the console while generating detailed Markdown reports in your home directory.

Codex Slash Prompt Method

For Codex users, use the slash-prompt syntax:

/prompts:investment-checklist Tesla, Alibaba

Alternatively, after installing the skills package:

./scripts/install-codex-skills.sh
./scripts/install-codex-prompts.sh

You can then use /investment-checklist followed by your tickers.

Direct Tool Invocation

For custom scripts or manual verification of specific calculations, call the financial rigor tool directly:

python3 tools/financial_rigor.py verify-valuation \
  --price 650 --eps 4.2 --bvps 35 --fcf-per-share 2.5 --dividend 0.8

This outputs exact decimal results for the Margin of Safety gate calculations.

The Six-Gate Evaluation Framework

The checklist sequentially evaluates companies through six deterministic gates:

1. Circle of Competence – Determines if the business model is understandable
2. Good Business – Assesses ROE, margins, and capital efficiency
3. Moat – Evaluates competitive advantages and durability
4. Management – Reviews capital allocation and integrity records
5. Margin of Safety – Calculates valuation gaps using verify-valuation and three-scenario outputs
6. Decision Discipline – Applies final bias checks and information-richness ratings (A/B/C)

Each gate receives a ★1-5 rating. Unlisted companies are marked "N/A" and skipped automatically.

Understanding the Output Reports

After evaluation, the skill generates timestamped Markdown reports at ~/巴菲特Checklist-<company>.md containing:

• Score tables – Six-gate ratings (★1-5) for each evaluation criterion
• Core financial tables – ROE, FCF, and valuation multiples
• Mirror Test – A mandatory 5-sentence sanity check paragraph; failure to articulate coherent logic results in an automatic "Fail"
• Final verdict – Pass, Fail, or Gray classification with core rationale

For multi-company runs, the system creates ~/巴菲特Checklist-多公司对比.md with side-by-side comparison tables.

Installation and Setup

Clone the repository and install the appropriate command layer for your AI coding tool:

git clone https://github.com/xbtlin/ai-berkshire.git
cd ai-berkshire

# For Claude Code users

./scripts/install-claude-commands.sh

# For Codex users

./scripts/install-codex-skills.sh
./scripts/install-codex-prompts.sh

Once installed, the investment-checklist command becomes available immediately.

Summary

• The investment checklist feature lives in skills/investment-checklist.md and implements a six-gate Buffett-style screening process
• Parallel agents collect data concurrently for each ticker, calling tools/financial_rigor.py for exact decimal calculations
• Invoke via investment-checklist <tickers> in Claude Code or /prompts:investment-checklist in Codex
• Reports output to ~/巴菲特Checklist-<company>.md with ★1-5 gate scores and Pass/Fail/Gray verdicts
• The Mirror Test requirement ensures qualitative reasoning backs every quantitative result

Frequently Asked Questions

How does AI Berkshire handle unlisted companies in the checklist?

Unlisted companies are automatically detected during the company metadata resolution phase and marked "N/A", causing the checklist to skip evaluation for those symbols while processing valid tickers normally.

What is the information-richness rating (A/B/C) in the checklist output?

The information-richness rating categorizes data quality: A indicates comprehensive, reliable data; B indicates moderate data gaps requiring caution; C signals poor data quality that triggers quick-kill red-lines and typically results in a "Gray" or "Fail" verdict regardless of other metrics.

Can I customize the six-gate evaluation criteria?

The gate logic is defined in skills/investment-checklist.md and operates deterministically. While you can modify the markdown skill file directly, the framework is designed for consistent, disciplined application of Buffett-style criteria rather than ad-hoc customization.

Where are the checklist reports saved and how long do they persist?

Reports are written to your home directory (~/巴菲特Checklist-<company>.md) as timestamped Markdown files. They persist indefinitely as standard text files, allowing you to build historical comparison archives and track rating changes over time.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}