Deep Research Skills in AI Berkshire: Management Deep-Dives and Company Analysis

AI Berkshire provides two specialized deep research skills—the Management Deep-Dive and the Deep-Company Series—that generate long-form, citation-rich reports for multi-day investment investigations.

AI Berkshire is an open-source investment research framework designed for rigorous company analysis. Unlike standard screening tools, its deep research skills conduct comprehensive investigations into management quality and business models, outputting structured markdown reports suitable for institutional-grade due diligence.

What Are Deep Research Skills in AI Berkshire?

Deep research skills differ fundamentally from quick-look screening capabilities. According to the canonical definitions in skills/management-deep-dive.md and skills/deep-company-series.md, these skills execute multi-day investigation workflows with explicit scoring methodologies and strict output formats. They require specific slash-command triggers and produce deliverables including composite scoring models, cross-chapter consistency checks, and series indexes.

Available Deep Research Skills

Management Deep-Dive Skill

The Management Deep-Dive skill performs a longitudinal assessment of company leadership based on the principle that "buying stocks means buying people" (买股票就是买人).

Triggered via the command /management-deep-dive <company> [<person> <company>], this skill produces a multi-section markdown report saved to reports/[Company]-management-[YYYYMMDD].md. The output includes:

• Key management personnel tables
• Promise-versus-fulfillment tracking matrices
• Capital allocation records (M&A, buybacks, dividends, new ventures)
• Governance structure assessments (equity structure, compensation, related-party transactions)
• Side-channel verification (Glassdoor ratings, customer reviews)
• Composite scoring model weighted as: Integrity (35%), Strategy (25%), Capital Allocation (25%), Governance (15%)

The skill internally calls tools/financial_rigor.py verify-valuation to validate valuation assumptions during the analysis.

Deep-Company Series Skill

The Deep-Company Series skill generates a comprehensive 3-8 article series titled "Understanding X" (看懂X), covering company panoramas, economic moats, profit engines, hidden assets, AI variables, financials, management, and investment decisions.

Triggered via /deep-company-series <company>, the skill creates a dedicated folder structure at reports/[Company]/《看懂[Company]》/ containing sequentially numbered markdown files:

• 01-开篇-认识[Company].md – Opening chapter establishing company context
• 07-管理层-买人三问.md – Management evaluation using the "Three Questions" framework
• 08-决策-估值三情景.md – Investment decision scenarios with three valuation contexts

Each article follows a strict editorial skeleton requiring head-quote blocks, narrative hooks, key-point recaps, next-chapter previews, and legal disclaimers. The skill generates a series index file 00-系列说明.md (not published) and enforces cross-chapter consistency checks for terminology, figures, and citations before final release.

Source Files and Implementation

The deep research capabilities are defined across multiple canonical locations in the xbtlin/ai-berkshire repository:

• skills/management-deep-dive.md – Canonical workflow definition including investigation steps, scoring tables, and governance evaluation criteria
• skills/deep-company-series.md – Canonical definition for the article series workflow, chapter templates, and writing standards
• codex-skills/management-deep-dive/SKILL.md – Auto-generated Codex wrapper for GitHub Copilot integration
• codex-skills/deep-company-series/SKILL.md – Auto-generated Codex wrapper mirroring the series workflow
• codex-prompts/management-deep-dive.md and codex-prompts/deep-company-series.md – Slash-command prompt layer definitions

The public command reference in README.md lists both /management-deep-dive and /deep-company-series as available deep-research triggers.

Summary

• AI Berkshire offers two distinct deep research skills: Management Deep-Dive and Deep-Company Series.
• Management Deep-Dive evaluates leadership across four weighted dimensions, generating timestamped markdown reports with composite governance scores and capital allocation histories.
• Deep-Company Series produces structured 3-8 article investigations with enforced consistency checks, standardized editorial frameworks, and non-published series indexes.
• Both skills are defined in skills/ with Codex-compatible wrappers in codex-skills/ and prompt integrations in codex-prompts/.
• These tools target multi-day research workflows rather than rapid screening, as implemented in the xbtlin/ai-berkshire framework.

Frequently Asked Questions

What is the difference between deep research skills and standard skills in AI Berkshire?

Standard skills perform quick screening and checklist tasks, while deep research skills conduct multi-day investigations generating long-form, citation-rich output. The deep research capabilities specifically target management assessment and comprehensive business model analysis through structured methodologies defined in skills/management-deep-dive.md and skills/deep-company-series.md.

How do I run a management deep-dive analysis?

Execute the slash command /management-deep-dive <company-name> to trigger the skill. The system generates a report at reports/[Company]-management-[Date].md containing management scoring tables, capital allocation histories, and governance assessments using the composite weighting model (Integrity 35%, Strategy 25%, Capital Allocation 25%, Governance 15%).

What outputs does the deep-company series skill create?

The skill creates a folder structure at reports/[Company]/《看懂[Company]》/ containing 3-8 markdown files following the "Understanding X" series format. Each article follows a strict skeleton with head-quotes, narrative hooks, and cross-references, plus a non-published index file 00-系列说明.md that maintains series continuity and enforces terminology consistency across chapters.

Where are the deep research skill definitions stored in the repository?

Canonical definitions reside in skills/management-deep-dive.md and skills/deep-company-series.md. Auto-generated Codex-compatible wrappers exist in codex-skills/management-deep-dive/SKILL.md and codex-skills/deep-company-series/SKILL.md, with prompt integrations located in codex-prompts/management-deep-dive.md and codex-prompts/deep-company-series.md.

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