How Management-Deep-Dive Evaluates Executives Using the "Buy Stock Is Buy People" Framework

The management-deep-dive skill in the ai-berkshire repository implements a 9-step systematic workflow that quantifies Warren Buffett's "buy stock = buy people" principle into a data-driven executive rating system with integrity as a hard-stop veto.

The open-source xbtlin/ai-berkshire project translates qualitative investment philosophy into reproducible analysis workflows. At its core, the management-deep-dive skill—defined in skills/management-deep-dive.md—evaluates corporate leadership through the lens that purchasing equity means betting on specific individuals, applying criteria popularized by Duan Yongping and Li Lu.

The 9-Step Evaluation Architecture

The skill orchestrates parallel agents to execute a comprehensive assessment across eight analytical dimensions before applying a weighted scoring algorithm.

1. Identify Key Executives

The workflow begins by distinguishing formal titles from actual influence. Using web search capabilities, the system constructs a table of CEOs, CFOs, founders, actual controllers, and other senior leaders who hold decision-making power.

2. CEO Capability Circle

This step evaluates whether the CEO can simultaneously see the future and execute. The system aggregates public statements from shareholder letters, earnings calls, interviews, and social media, then measures:

• Prediction accuracy over time
• Strategic vision clarity
• Execution track record
• Crisis handling effectiveness
• Iteration speed

3. Integrity Assessment (The Veto Item)

Integrity serves as a binary gate in the framework. The skill tracks concrete commitments made over the past three years, calculating a commitment-fulfillment rate. It evaluates behavior during crises and attitudes toward all stakeholders—shareholders, employees, customers, suppliers, and regulators. According to the "段永平 '买人' 标准" implemented in the scoring rules, if integrity is absent, the final rating defaults to ★ (do not invest) regardless of other factors.

4. Capital-Allocation Ability

The system catalogues major capital-allocation decisions including M&A, share repurchases, dividends, and new-business investments. Each event receives ratings on price paid, integration success, timing, and sustainability, aggregating into a weighted capital-allocation score. This step leverages tools/financial_rigor.py to validate buy-back valuations against intrinsic value calculations.

5. Governance Structure

This dimension examines whether ownership aligns with management through:

• Equity structure analysis (A-shares, super-voting rights, VIEs)
• Executive compensation relative to profit
• Related-party transaction patterns

6. Side-Channel Validation

To cross-check management narratives, the skill pulls employee sentiment from Glassdoor and Zhihu, customer/merchant feedback from App Store reviews and forums, and industry reputation data.

7. Succession Resilience

The evaluation analyzes whether the business can survive leadership changes by reviewing past transitions, executive bench depth, and whether competitive moats rely on a single person versus systemic processes.

8. Structured Reporting

The system generates a markdown report at reports/{Company}-management-{YYYYMMDD}.md containing tables for each dimension, supporting evidence, and preliminary scores.

9. Final Scoring Rules

The composite score weights four dimensions: integrity (35%), strategic-execution capability (25%), capital-allocation ability (25%), and governance structure (15%). The final output assigns 3-5 star ratings, with integrity acting as a hard-stop filter.

Technical Implementation

The skill utilizes the internal Task tool to orchestrate parallel agents, enabling concurrent data fetching across multiple sources without manual searching. The codex-skills/management-deep-dive/SKILL.md file provides a Codex-compatible wrapper that mirrors the Claude Code skill definition for cross-platform compatibility.

For capital-allocation validation, the system invokes tools/financial_rigor.py to perform rigorous valuation checks on buy-back events, ensuring that repurchase decisions meet Buffett-style criteria of buying below intrinsic value.

How to Invoke the Skill

Users trigger the evaluation through simple command patterns in the AI-Berkshire environment:

# Simple company query

/management-deep-dive 美团

# Specify a person when the CEO differs from the brand

/management-deep-dive 王兴 美团

Understanding the Output Report

The generated report follows a standardized structure. A typical output (reports/美团-management-20260409.md) includes:

一、关键人物速览
| 角色 | 姓名 | 任期 | 背景 | 持股/期权 |
|------|------|------|------|----------|
| CEO/董事长 | 王兴 | 2004‑至今 | 创始人 | 22% |
...

二、诚信度评估
- 承诺兑现率：87% → 优秀
- 困难时期表现：积极沟通、主动承担
- 利益相关方态度：股东‑尊重，员工‑激励...

六、综合评分与结论
| 维度 | 权重 | 评分(1‑5) | 加权 |
|------|:----:|:---------:|:----:|
| 诚信度 | 35% | 4 | 1.4 |
| 战略与执行能力 | 25% | 4 | 1.0 |
| 资本配置能力 | 25% | 5 | 1.25 |
| 治理结构 | 15% | 3 | 0.45 |
**综合评分** = 4.1 ★★★★★ → 推荐持有

Summary

• The management-deep-dive skill transforms the qualitative "buy stock = buy people" framework into a quantifiable 9-step evaluation system.
• Integrity functions as a veto item: any integrity deficit automatically triggers a "do not invest" recommendation regardless of other capabilities.
• The workflow evaluates capital-allocation decisions using tools/financial_rigor.py to validate buy-back economics.
• Parallel agent architecture enables comprehensive data gathering across executive statements, employee sentiment, and governance filings.
• Final reports provide weighted scores (integrity 35%, strategy 25%, capital allocation 25%, governance 15%) with clear 3-5 star investment ratings.

Frequently Asked Questions

What makes integrity a "hard stop" in the evaluation?

The framework implements Duan Yongping's investment philosophy where character is non-negotiable. As specified in skills/management-deep-dive.md lines 158-169, if the commitment-fulfillment rate or stakeholder treatment analysis reveals integrity gaps, the system automatically assigns a ★ rating (do not invest), overriding any strong performance in strategic vision or capital allocation.

How does the skill validate capital-allocation decisions?

The system catalogs historical M&A, dividends, and share repurchases, then applies price-vs-value analysis using tools/financial_rigor.py. Each decision receives ratings on timing, price paid, and integration success. Buy-backs specifically trigger valuation checks to ensure management purchased shares below intrinsic value rather than destroying shareholder capital through overpriced repurchases.

Can the evaluation work for non-Chinese companies?

While the default configuration targets Chinese equities (using sources like Zhihu and local app stores), the architecture is source-agnostic. Users can invoke the skill on any public company (/management-deep-dive Apple or /management-deep-dive Microsoft), with the web search agents automatically adjusting to available English-language sources including Glassdoor, LinkedIn, and SEC filings.

How is the "succession resilience" score calculated?

The skill analyzes historical leadership transitions documented in public records, evaluates the depth of the current executive bench below the CEO, and determines whether operational processes or personal relationships drive the business. Companies where the moat depends on a single "key person" receive lower longevity scores, while those with institutionalized processes score higher regardless of the current leader's brilliance.

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