How to Perform Industry Research with AI Berkshire's `/industry-research` Skill: The Complete 8-Step Workflow

AI Berkshire's /industry-research skill executes a systematic eight-phase workflow that transforms a plain industry name into a comprehensive investment report, covering logic-chain construction, value-chain mapping, and portfolio allocation.

The AI Berkshire repository provides an agent-driven framework for institutional-grade investment analysis. The /industry-research skill, defined in skills/industry-research.md, automates deep fundamental research by orchestrating multiple AI agents through a rigorous pipeline that evaluates sectors from macro trends down to individual security selection.

The 8-Step Industry Research Workflow

The skill processes any industry name provided as $ARGUMENTS through eight distinct phases defined in the canonical skill file. Each phase produces structured intermediate outputs that culminate in a final investment memorandum.

Step 1: Investment Logic-Chain Construction

According to skills/industry-research.md (lines 5-30), the workflow begins by sketching the investment logic chain from macro trend to ultimate industry beneficiary. This phase maps the causal pathway: macro trend → demand generation → supply bottleneck → beneficiary identification. Every link requires validation through evidence searches, case studies, and contract analysis. This foundational step ensures the investment thesis rests on verifiable causality rather than narrative speculation.

Step 2: Full Value-Chain Mapping

As defined in lines 41-58 of the skill definition, the system deconstructs the sector into a complete value-chain diagram spanning upstream, midstream, downstream, and ancillary nodes. Each node receives detailed annotation covering business model mechanics, margin range estimates, competitive landscape dynamics, barrier types (regulatory, capital, or technological), and cyclicality characteristics. This mapping provides the structural foundation for subsequent security selection.

Step 3: Global Company Scan

Lines 82-94 specify that a background Task agent executes a comprehensive global securities scan across US exchanges, A-shares, Hong Kong listings, and other international markets, plus relevant ETFs and notable private players. For each identified firm, the agent extracts the ticker symbol, market capitalization, one-sentence business role description, and precise placement within the value-chain segment identified in Step 2.

Step 4: "Four Masters" Deep-Dive

For each Tier 1 and Tier 2 company within every value-chain node, the skill applies the Four Masters framework (段永平, 巴菲特, 芒格) as detailed in lines 112-158. This evaluation covers:

• Business essence (段永平): Core profitability and simplicity
• Economic moat (巴菲特): Sustainable competitive advantages
• Risk assessment (芒格): Potential failure modes and downside scenarios
• Management quality (段永平 + 巴菲特): Capital allocation and integrity

Each company receives valuation snapshots and a star rating (★) based on this multi-dimensional analysis.

Step 5: Industry-Level Risk Assessment

Lines 161-174 mandate the assembly of a systematic risk matrix categorizing policy risks, technological disruption, cyclical exposure, and valuation extremes. This phase incorporates historical analogue comparison and explicit bias checks for narrative fallacy, anchoring, and herd behavior, ensuring the analysis accounts for behavioral vulnerabilities.

Step 6: Civilisation-Trend Judgement

Applying the Li Lu (李录) framework specified in lines 187-194, the skill determines whether the sector represents a genuine long-run civilisation shift or temporary market hype. This judgement predicts the ultimate industry outcome (consolidation, obsolescence, or structural growth) based on alignment with fundamental human and economic needs.

Step 7: Portfolio Allocation Blueprint

Lines 197-209 require the generation of a tiered allocation table specifying:

• Core positions (high conviction, large allocations)
• Satellite positions (tactical exposure)
• Optionality positions (asymmetric upside)
• ETF substitutes (diversified exposure)

Each tier includes specific position sizing guidelines, entry and exit signals, and overall theme-level exposure caps to manage concentration risk.

Step 8: Final Decision Memo

The pipeline concludes with a decision memorandum (lines 221-239) summarizing logic-chain confidence levels, the highest-conviction value-chain segment, aggregate moat strength assessment, primary risk factors, trend positioning, and valuation metrics. The report embeds simulated commentary from the four masters, providing qualitative perspectives on the investment thesis.

Output Requirements and Report Structure

The skill enforces strict output requirements defined in lines 40-48 of skills/industry-research.md to ensure institutional reproducibility. Every factual claim must cite specific data sources, all tables render in Markdown format, and the value-chain diagram appears as a plain-text block. The final report automatically saves to ~/[行业名]产业链投资研究报告.md, where [行业名] represents the input industry name.

Before publication, the tools/report_audit.py script performs post-generation audits to verify data provenance and citation accuracy.

How to Invoke the Industry Research Skill

You can trigger the workflow through multiple interfaces depending on your execution environment.

Claude-Code Slash Command

For interactive Claude-Code sessions, use the slash command syntax:

/industry-research 半导体

This command initiates the full eight-step workflow for the semiconductor industry.

Codex Skill CLI

For command-line execution via the generated Codex skill package:

python -m codex-skills.industry-research "新能源汽车"

Python Programmatic Call

Import the skill directly for integration with existing Python workflows:

from ai_berkshire.skills import industry_research

report_path = industry_research.run("云计算")
print("Report saved to:", report_path)

The function returns the absolute path to the generated Markdown report.

Batch Execution via Task

Schedule the skill as a background task in AI Berkshire's orchestrator:

task run industry-research "可再生能源"

Each interface executes the identical eight-phase pipeline, producing the standardized Markdown report at ~/[行业名]产业链投资研究报告.md.

Summary

• AI Berkshire's /industry-research skill implements an eight-phase agent-driven workflow defined in skills/industry-research.md that transforms industry names into comprehensive investment reports.
• The process covers logic-chain construction, value-chain mapping, global company scans, and Four Masters analysis (段永平, 巴菲特, 芒格) before generating portfolio allocation recommendations.
• Output standards require cited data sources, Markdown tables, and specific file naming conventions (~/[行业名]产业链投资研究报告.md), with post-generation validation via tools/report_audit.py.
• Users can invoke the skill via Claude-Code slash commands, Codex CLI, Python imports, or Task orchestration, with each method producing identical analytical depth.

Frequently Asked Questions

What exchanges does the global company scan cover?

The global company scan in Step 3 covers US listings, A-shares (China), Hong Kong exchanges, and other international markets, plus relevant ETFs and significant private companies. The background Task agent scrapes ticker symbols, market capitalizations, and business descriptions for each entity to build a complete investment universe.

How does the Four Masters framework evaluate companies?

The Four Masters framework applies four distinct investment lenses: 段永平's focus on business essence and simplicity, Buffett's moat analysis for competitive durability, Munger's risk assessment for downside identification, and the combined judgment on management quality and capital allocation. Each company receives a star rating (★) based on this multi-dimensional evaluation.

Can I run industry research on multiple sectors simultaneously?

While the /industry-research skill processes one industry per invocation, you can execute multiple instances simultaneously through the Task orchestrator (task run) or by calling the Python API asynchronously. Each execution operates independently and generates separate report files named according to the input industry.

What distinguishes the Civilisation-Trend Judgement from regular trend analysis?

The Civilisation-Trend Judgement (Step 6) specifically applies Li Lu's framework to distinguish between structural, multi-decade civilisation shifts (such as energy transitions or computing paradigms) and temporary market narratives or cyclical hype. This determination predicts ultimate industry outcomes—whether the sector will consolidate, become obsolete, or experience structural growth—rather than just short-term price movements.

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