How to Use the Management-Deep-Dive Skill to Evaluate Company Leadership

The management-deep-dive skill executes an eight-step, data-driven workflow that transforms raw public information into a comprehensive leadership assessment with weighted scoring across integrity, ability, capital allocation, and governance.

The management-deep-dive skill in the xbtlin/ai-berkshire repository provides a systematic framework for assessing corporate leadership through automated, parallel data collection. Defined in skills/management-deep-dive.md, this open-source tool orchestrates multiple agents to gather executive profiles, track capital deployment decisions, and validate management integrity against historical commitments. Investors and analysts can invoke this skill to generate rigorous, evidence-based leadership ratings that mirror institutional due diligence processes.

What Is the Management-Deep-Dive Skill?

The management-deep-dive skill is a structured evaluation workflow designed to assess company leadership through eight distinct analytical phases. According to the source code in skills/management-deep-dive.md, the skill employs parallel agent architecture to harvest disparate data streams simultaneously, then synthesizes results into a scored markdown report. This approach ensures every qualitative judgment is backed by sourced facts, creating an audit trail for investment decisions.

The Eight-Step Leadership Evaluation Process

The skill follows a rigorous sequence that evaluates leadership from multiple angles:

1. Identify Key Executives and Parallel Data Collection

The workflow begins by using WebSearch to identify CEOs, CFOs, founders, actual controllers, and other senior managers. The system then spawns parallel agents to simultaneously gather executive statements, capital-allocation histories, governance documents, and side-channel feedback. This parallel architecture accelerates data collection while ensuring comprehensive coverage.

2. CEO "Ability-Circle" Evaluation

This step bifurcates into Strategic Vision and Execution Capability. The system checks past management forecasts against actual outcomes to assess foresight accuracy. It also measures delivery metrics, talent attraction capabilities, crisis handling responses, and organizational iteration speed to gauge operational excellence.

3. Integrity Assessment

Considered the most critical component, this phase tracks promises versus actual delivery through a commitment-fulfilment table. The skill examines behavior during crises and evaluates attitudes toward shareholders, employees, customers, suppliers, and regulators.

4. Capital-Allocation Ability

The skill catalogues historical M&A activity, share repurchases, dividend policies, and new-business investments. Each capital deployment decision receives a 1-5 score and is cross-checked using the tools/financial_rigor.py script with the verify-valuation sub-command to ensure financial rigor in the assessment.

5. Governance Structure Review

This phase inspects equity structures including A-shares, super-voting rights, and VIE arrangements. The evaluation covers board independence metrics, large-shareholder activity monitoring, and senior-executive compensation relative to company profitability.

6. Side-Channel Validation

To triangulate the leadership narrative, the skill pulls employee sentiment data from Glassdoor and Zhihu, customer and merchant reviews from App Stores and forums, and industry reputation indicators. This validates whether public statements align with ground-level reality.

7. Post-CEO Continuity Analysis

The workflow evaluates organizational resilience by asking whether the firm would survive the CEO's departure. It assesses management team depth, succession plan clarity, and historical hand-over success rates to determine dependence on key individuals.

8. Report Generation and Scoring

The final step produces a markdown report at reports/{company_name}-management-{YYYYMMDD}.md containing weighted scores across four dimensions. The composite score translates into a star rating (★) through the "buy-person" rule, providing an immediate visual indicator of leadership quality.

Scoring Weights and Output Format

The management-deep-dive skill applies a weighted scoring system that prioritizes integrity while balancing operational competence:

• Integrity: 35%
• Ability (Strategic and Execution): 25%
• Capital Allocation: 25%
• Governance: 15%

Each dimension receives a numerical score based on documented evidence tables, including commitment-fulfilment matrices and crisis-response logs. The output file follows the naming convention reports/{公司名}-management-{YYYYMMDD}.md, creating a dated audit trail for each evaluation.

How to Invoke the Skill

You can trigger the evaluation workflow through Python scripts or CLI commands:

Python Implementation

from codex import run_skill

# Evaluate the leadership of Meituan

result = run_skill(
    skill="management-deep-dive",
    arguments="美团"
)

print(result)   # prints the generated markdown report

Command Line Interface

# Using the Claude-compatible slash command

/management-deep-dive 美团

Both methods ultimately call the same workflow defined in skills/management-deep-dive.md, forwarding the company argument to the underlying orchestration layer.

Core Architecture and Dependencies

The management-deep-dive skill relies on several key files within the xbtlin/ai-berkshire repository:

• skills/management-deep-dive.md: Contains the complete eight-step workflow specification and evaluation logic.
• codex-prompts/management-deep-dive.md: A minimal wrapper that exposes the skill as a slash command for Codex users.
• tools/financial_rigor.py: Provides the verify-valuation sub-command used to validate capital-allocation decisions.
• AGENTS.md: Documents the repository's overall architecture and the synchronization process for Codex skills.

Summary

• The management-deep-dive skill provides an eight-step structured workflow for evaluating corporate leadership through skills/management-deep-dive.md.
• Parallel agent architecture enables simultaneous data collection from web searches, employee sentiment platforms, and financial databases.
• Weighted scoring prioritizes integrity (35%) while assessing ability (25%), capital allocation (25%), and governance (15%).
• The verify-valuation tool in tools/financial_rigor.py validates financial decisions with quantitative rigor.
• Output generates as dated markdown files in reports/{company}-management-{YYYYMMDD}.md with star ratings.

Frequently Asked Questions

What data sources does the management-deep-dive skill use?

The skill aggregates public data through WebSearch for executive biographies, Glassdoor and Zhihu for employee sentiment, App Store reviews and industry forums for customer feedback, and financial filings for capital allocation histories. This multi-source approach ensures triangulation of leadership narratives against ground-level reality.

How is the final leadership score calculated?

The composite score weights four dimensions: integrity at 35%, ability (strategic and execution) at 25%, capital allocation at 25%, and governance at 15%. Each component receives individual scores based on documented evidence tables, which then feed into the final "buy-person" rule that translates the percentage into a star rating system.

Can I evaluate private companies or only public entities?

The current implementation in skills/management-deep-dive.md is optimized for public companies with available disclosure data, executive statements, and tracked capital allocation histories. While the parallel agent architecture could theoretically apply to private firms, the verification mechanisms (particularly the verify-valuation cross-checks) require public financial transparency to function effectively.

What is the difference between the skill file and the Codex prompt wrapper?

The skills/management-deep-dive.md file contains the complete evaluation logic, step definitions, and scoring methodology. The codex-prompts/management-deep-dive.md file serves as a minimal interface that forwards user arguments to the underlying skill, enabling slash-command invocation (/management-deep-dive) within chat interfaces without exposing the full workflow complexity to end users.

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