How the Management-Deep-Dive Skill Evaluates Corporate Leadership: An 8-Step Technical Breakdown

The management-deep-dive skill evaluates corporate leadership through an eight-step data-driven workflow that assesses executive integrity, strategic vision, capital allocation discipline, and governance structure, culminating in a weighted composite score and star rating.

The management-deep-dive skill in the xbtlin/ai-berkshire repository transforms raw public data into rigorous leadership assessments using a structured, parallelized workflow. Defined in skills/management-deep-dive.md, this tool mirrors institutional investment research processes to evaluate corporate management across multiple quantitative and qualitative dimensions.

Architecture of the Management-Deep-Dive Skill

The skill operates as an orchestrated multi-agent system that gathers disparate data streams concurrently. According to the source code in skills/management-deep-dive.md, the workflow spans lines 25–170 and implements a data-driven methodology where every qualitative judgment maps to a table of sourced facts. The architecture supports parallel execution, allowing multiple agents to harvest executive statements, financial records, and side-channel feedback simultaneously. The repository’s overall agent architecture is documented in AGENTS.md, which explains how these skills sync with the Codex execution environment.

The 8-Step Leadership Evaluation Workflow

Step 1: Identify Key Executives and Parallel Data Collection

The workflow initiates with a WebSearch to identify CEOs, CFOs, founders, actual controllers, and senior managers. The system then spawns parallel agents to gather public statements, capital-allocation histories, governance disclosures, and side-channel intelligence (lines 25–44).

Step 2: CEO Ability-Circle Evaluation

This phase splits into two critical components:

• Strategic Vision: Validates past forecasts against actual outcomes to assess predictive accuracy.
• Execution Capability: Measures delivery metrics, talent attraction success, crisis handling, and iteration speed (lines 45–68).

Step 3: Integrity Assessment

The most critical evaluation tracks promises versus delivery through a commitment-fulfilment table. The system examines behavior during crises and evaluates attitudes toward shareholders, employees, customers, suppliers, and regulators (lines 70–112).

Step 4: Capital-Allocation Ability

The skill catalogs M&A activity, share repurchases, dividends, and new-business investments. Each entry receives a 1–5 score and cross-checks against tools/financial_rigor.py verify-valuation for financial rigor (lines 116–154).

Step 5: Governance Structure Review

This inspection covers equity structure (A-shares, super-voting rights, VIEs), board independence metrics, large-shareholder activity, and senior-executive compensation relative to profit performance (lines 159–176).

Step 6: Side-Channel Validation

To triangulate the leadership narrative, the system pulls employee sentiment from Glassdoor and Zhihu, customer/merchant reviews from App Store and forums, and industry reputation data (lines 83–110).

Step 7: Post-CEO Continuity Analysis

The evaluation asks whether the firm survives the CEO’s departure. It assesses management team depth, succession plan quality, and historic hand-over success (lines 113–122).

Step 8: Report Generation and Final Scoring

The skill generates a markdown report at reports/{company_name}-management-{YYYYMMDD}.md applying weighted scores:

• Integrity: 35%
• Ability: 25%
• Capital-Allocation: 25%
• Governance: 15%

The final "buy-person" rule translates the composite score into a ★-rating (lines 124–170).

Implementation: Invoking the Skill

Developers can trigger the workflow programmatically or via command-line interfaces.

# Example: invoking the skill from a Python script using the Codex CLI

from codex import run_skill

# Evaluate the leadership of “美团”

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

print(result)   # prints the generated markdown report

Alternatively, use the slash command defined in codex-prompts/management-deep-dive.md:

# Example: using the Claude-compatible slash command in a chat interface

/management-deep-dive 美团

Both methods forward arguments to the underlying workflow defined in the markdown skill file.

Summary

• The management-deep-dive skill in xbtlin/ai-berkshire implements an eight-step parallel workflow to evaluate corporate leadership.
• Integrity assessment carries the highest weight (35%) and tracks promises versus delivery through sourced fact tables.
• Capital-allocation decisions are scored 1–5 and validated using tools/financial_rigor.py verify-valuation.
• The workflow generates timestamped markdown reports with weighted composite scores and star ratings.
• Invocations are supported via Python run_skill() API or the /management-deep-dive slash command.

Frequently Asked Questions

What is the primary source file for the management-deep-dive skill?

The complete workflow specification resides in skills/management-deep-dive.md within the xbtlin/ai-berkshire repository. This file defines all eight evaluation steps, scoring weights, and output formats. A minimal wrapper exposing the skill as a slash command exists in codex-prompts/management-deep-dive.md.

How does the skill weight different leadership criteria?

The final composite score applies specific weights: integrity at 35%, ability at 25%, capital-allocation at 25%, and governance at 15%. These weightings reflect the AI Berkshire team's emphasis on management integrity as the primary determinant of long-term investment quality.

What tool validates the financial assumptions in capital-allocation analysis?

The skill calls tools/financial_rigor.py verify-valuation to cross-check and validate financial assumptions underlying M&A decisions, share repurchases, and investment moves. This ensures quantitative rigor in the 1–5 scoring of capital allocation decisions.

Can the workflow evaluate private companies or only public entities?

While the source code demonstrates evaluation of public companies like "美团" (Meituan), the architecture relies on WebSearch and public data sources (Glassdoor, App Store, regulatory filings). The framework can theoretically assess any entity with sufficient public digital footprint, though data availability determines output quality.

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