How to Implement the Investment-Checklist with Buffett's Screening Process in AI-Berkshire

The investment-checklist skill in AI-Berkshire implements Warren Buffett's six-gate pre-buy screening methodology through a parallelized, tool-driven workflow that evaluates business understanding, economic characteristics, competitive moats, and margin of safety before generating auditable markdown reports.

The AI-Berkshire repository provides a sophisticated /investment-checklist slash-command that transforms Buffett's qualitative investment philosophy into a repeatable, automated screening process. This skill functions as both a Claude Code command and a Codex skill, leveraging parallel Task-based agents and deterministic Python tools to evaluate stocks against six rigorous gates inspired by Warren Buffett's "circle of competence" and margin-of-safety principles.

Architectural Overview of the Investment-Checklist

The skill workflow begins with input processing and proceeds through parallelized data collection before executing the six-gate evaluation. According to skills/investment-checklist.md, the architecture follows a layered orchestration designed to minimize latency while maximizing analytical rigor.

Input Normalisation and Bias Assessment

The process starts at lines 5-13 of skills/investment-checklist.md, where the skill receives free-form company identifiers (e.g., "Tencent, 茅台, NVDA"). It parses each entry to extract full names, tickers, and exchanges, flagging unlisted firms for bypass. Immediately following (lines 15-24), an AI-driven "information-richness" rating (A/B/C) is assigned to determine how strictly the subsequent checklist applies to each candidate.

Parallel Data Harvesting

For every validated company, the system spawns a Task-based background agent (lines 29-40). These agents concurrently pull eight data groups: profitability, valuation, growth, financial health, competitive landscape, moat evidence, management quality, and recent news. This parallelism dramatically reduces latency compared to sequential processing.

The Six-Gate Buffett Screening Process

The core evaluation occurs within six gates that mirror Buffett's buying criteria, as implemented in skills/investment-checklist.md between lines 44-172. Each gate yields a ★-rating (1-5) and can trigger a hard-stop if mandatory conditions fail.

Gate 1 – Business Understanding (能力圈)

Lines 44-72 verify that the business model can be described succinctly and remains viable a decade ahead. This gate enforces Buffett's "circle of competence" rule—a hard-stop occurs if the system cannot articulate how the company makes money.

Gate 2 – Economic Characteristics (经济特征)

Lines 84-92 invoke tools/financial_rigor.py to compute quantitative metrics including ROE, margin, free-cash-flow, asset intensity, and debt ratios. Unlike manual calculations, this Python script ensures reproducible financial rigor through deterministic algorithms.

Gate 3 – Moat Depth Assessment

Lines 95-112 evaluate sustainable competitive advantages through brand strength, switching costs, network effects, scale advantages, and IP barriers. The gate includes a "competitor replication" thought experiment to test moat durability.

Gate 4 – Management Trustworthiness

Lines 116-133 assess qualitative factors including capital-allocation skill, shareholder orientation, founder versus professional management, and governance structures. Honesty and transparency in communications carry significant weight in this evaluation.

Gate 5 – Margin of Safety Valuation

Lines 136-162 execute a three-scenario valuation using financial_rigor.py to obtain PE, forward-PE, PB, dividend yield, and FCF-Yield bands. The system compares current price against intrinsic value estimates, requiring a sufficient safety margin for passage.

Gate 6 – Position-Sizing Discipline

Lines 165-172 screen for emotional triggers including FOMO and hype indicators. This final gate ensures the investment thesis can be expressed in ≤200 words, enforcing clarity and conviction before capital allocation.

Mirror Test and Final Evaluation

Following the six gates, the skill generates a "mirror-test" paragraph (lines 75-85) and runs a quick-reject checklist (lines 90-102) examining red-flag items such as negative cash-flow trends, management scandals, or eroding moats. Any triggered item immediately marks the company as Reject, preventing further analysis.

For multi-company runs, the system builds a comparison table (lines 105-124) summarizing gate results, safety-margin scores, and final verdicts. The complete analysis writes to ~/巴菲特Checklist-{公司名}.md (lines 125-140), creating an auditable record with full citations.

Invoking the Skill and Tool Commands

You can trigger the Buffett screening process through the Claude Code interface or invoke underlying tools directly.

Run the complete checklist for multiple tickers:

claude-code run /investment-checklist "Tencent, Moutai, NVDA"

Execute direct valuation verification (used in Gate 2):

python3 tools/financial_rigor.py verify-valuation \
  --price 215.30 --eps 8.24 --bvps 45.12 --fcf-per-share 6.58 --dividend 2.10

Run three-scenario valuation analysis (Gate 5):

python3 tools/financial_rigor.py three-scenario \
  --price 215.30 --eps 8.24 --shares 1.2 \
  --growth 0.12 0.08 0.04 \
  --pe 15 20 30 \
  --currency CNY

Access generated reports programmatically:

from pathlib import Path
report_path = Path.home() / "巴菲特Checklist-多公司对比.md"

# The skill automatically writes compiled markdown to this location

File Structure and Sync Automation

The investment-checklist implementation spans several coordinated files:

• skills/investment-checklist.md – Canonical markdown definition of the six-gate process and data requirements (lines 5-172)
• codex-skills/investment-checklist/SKILL.md – Auto-generated Codex wrapper mapping slash commands to skill logic
• codex-prompts/investment-checklist.md – Prompt declarations for downstream agent loading
• tools/financial_rigor.py – Core financial-analysis utilities invoked by Gates 2 and 5
• scripts/sync-codex-skills.py – Automation script that regenerates Codex artifacts when source markdown changes

According to the AI-Berkshire source code, maintaining synchronization between the human-readable skill definition and the Codex runtime requires running scripts/sync-codex-skills.py after any modifications to skills/investment-checklist.md.

Summary

• The investment-checklist skill implements Buffett's six-gate screening via parallel Task-based agents and deterministic Python tools.
• Input normalisation (lines 5-13) and bias pre-checks (lines 15-24) filter candidates before parallel data harvesting begins.
• Six quantitative and qualitative gates evaluate business understanding, financial rigor, moat depth, management quality, margin of safety, and emotional discipline.
• Hard stops enforce Buffett's "don't buy what you don't understand" rule, while the mirror test and quick-reject list catch red flags post-evaluation.
• Financial rigor relies on tools/financial_rigor.py for reproducible valuation calculations rather than AI estimation.
• Auditable outputs write to ~/巴菲特Checklist-{公司名}.md, enabling transparent review of every rating and data source.

Frequently Asked Questions

How does the investment-checklist handle multiple companies simultaneously?

The skill spawns a Task-based background agent for each company (lines 29-40), enabling concurrent data harvesting across profitability, valuation, growth, and competitive metrics. This parallel architecture reduces total execution time significantly compared to sequential processing, then aggregates results into a single comparison table written to ~/巴菲特Checklist-多公司对比.md.

What triggers a hard-stop rejection in the Buffett screening gates?

Hard-stops occur when mandatory conditions fail, such as the inability to articulate how a company makes money (Gate 1), or when the mirror-test quick-reject list (lines 90-102) identifies red flags including negative cash-flow trends, management scandals, or eroding competitive moats. These stops enforce Buffett's discipline of avoiding investments outside the "circle of competence" or exhibiting fundamental deterioration.

Which Python tools perform the quantitative calculations in Gates 2 and 5?

The tools/financial_rigor.py script handles all deterministic financial calculations. Gate 2 invokes verification functions for ROE, margin, and debt ratios, while Gate 5 executes the three-scenario command to generate PE, PB, and FCF-Yield bands across optimistic, baseline, and pessimistic growth assumptions, ensuring reproducible valuation metrics rather than AI hallucinations.

How do I synchronize the skill definition after making edits?

Run python3 scripts/sync-codex-skills.py after modifying skills/investment-checklist.md. This script regenerates the auto-generated wrappers in codex-skills/investment-checklist/SKILL.md and updates prompt declarations in codex-prompts/investment-checklist.md, ensuring the Claude Code and Codex runtimes reflect your latest screening logic.

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