Difference Between industry-research and industry-funnel Skills in AI Berkshire: Complete Workflow Guide

The industry-research skill produces a comprehensive value-chain panorama mapping all investment opportunities across industry segments, while the industry-funnel skill executes a disciplined, multi-stage filtration process to narrow the market down to three final target stocks.

Understanding the difference between industry-research and industry-funnel skills in the xbtlin/ai-berkshire repository is essential for executing a complete investment workflow. Both skills belong to the "行业筛选类" (industry screening) category, but they operate at different analytical altitudes and serve distinct stages of the research pipeline. The source code defines these workflows in skills/industry-research.md and skills/industry-funnel.md, with the explicit comparison documented in the README under the "行业筛选类" section.

Core Functional Differences

The primary distinction lies in their analytical scope and end objectives. While both skills scan global markets and analyze listed companies, their outputs target fundamentally different investment needs.

industry-research operates at the segment level, dissecting every link in an industry's value chain—upstream, midstream, downstream, and auxiliary segments—to generate a complete opportunity map. This skill answers the question: What are all the investment footholds available in this industry?

industry-funnel operates at the company level, applying hard-value-investing criteria to progressively eliminate candidates until only three stocks remain. This skill answers the question: Which three specific equities offer the best risk-adjusted returns right now?

Workflow Comparison

The step-by-step execution paths differ significantly in granularity and filtering logic.

Industry-Research Workflow

As defined in skills/industry-research.md, the workflow follows five sequential stages:

1. Build and validate a logic chain establishing the industry's economic model
2. Draw the value-chain diagram visualizing upstream-to-downstream relationships
3. Scan all listed companies globally to identify every player in each segment
4. Apply four-master analysis to head companies per segment (Tier 1-4 classification)
5. Generate portfolio-allocation recommendations across the entire industry spectrum

The output includes logic-chain diagrams, segment-grouped company tables, and a full industry-level risk checklist.

Industry-Funnel Workflow

According to skills/industry-funnel.md, this skill implements a six-stage funnel architecture:

1. Full-market scan filtering by activity level, price-rise metrics, and market capitalization
2. Apply 5 hard-value-investing criteria (specific quantitative thresholds)
3. Rough filter reducing the universe to ≤10 companies
4. Detailed 300-500 word analysis of remaining candidates
5. Final 3-company deep dive producing 800-1200 word reports per stock
6. Portfolio recommendation classifying selections as core, satellite, or option positions

The output includes scanned-company tables with entry class ratings (A/B/C), filter-by-criteria pass/fail tables with elimination reasons, and a final recommendation matrix.

Technical Implementation and Architecture

Both skills are implemented as canonical markdown files under the skills/ directory, with auto-generated Codex wrappers exposing them to CLI clients.

Skill and Agent Layers

The markdown definitions in skills/industry-research.md and skills/industry-funnel.md serve as the single source of truth. Each skill spawns multiple Agents that execute sub-steps in parallel (e.g., "全市场扫描" for market scanning, "价值投资 5 条硬指标" for criteria application). These agents return structured JSON that the orchestration script stitches into the final markdown report.

Data Quality Assurance

Both skills invoke tools/report_audit.py to execute the "准出" (audit-pass) data-quality checks before publishing. They also utilize tools/financial_rigor.py for precise valuation mathematics, including market-cap verification and multi-scenario valuation models. This ensures that whether producing a broad industry map or a narrow three-stock recommendation, all outputs meet the repository's financial rigor standards.

Practical Usage Examples

Execute these commands within the AI Berkshire environment to trigger the respective workflows.

Running Industry Research

/industry-research 核电

This generates 核电‑industry‑20260409.md, containing the complete value-chain logic, segment tables with Tier 1-4 classifications, and four-master analyses for head companies in each segment.

Running Industry Funnel

/industry-funnel AI算力

This produces AI算力‑funnel‑20260509.md, starting with 30-60 initial candidates, showing the elimination table through the 5 hard criteria, and ending with the final three-stock recommendation matrix.

Combining Both Skills

# First, map the entire sector structure

/industry-research AI算力

# Then, select the top three equities from the mapped universe

/industry-funnel AI算力

The second command can ingest segment tables generated by the first, ensuring consistency across the macro-to-micro workflow.

Summary

• industry-research provides macro-level, value-chain mapping with exhaustive company scanning across all industry segments
• industry-funnel provides micro-level, disciplined shortlisting that narrows the market to exactly three target stocks through quantitative criteria
• File locations: Core definitions reside in skills/industry-research.md and skills/industry-funnel.md, with the explicit comparison in README.md (lines 456-510)
• Quality gates: Both skills rely on tools/report_audit.py and tools/financial_rigor.py to validate outputs before publication
• Typical workflow: Use industry-research first to understand sector structure, then industry-funnel to select specific equities

Frequently Asked Questions

When should I use industry-research versus industry-funnel?

Use industry-research when you need to understand every available investment opportunity in a sector before committing capital (e.g., "What are all the plays in nuclear power?"). Use industry-funnel when you already understand the industry and need a quick, quantitative shortlist of the best individual stocks (e.g., "Select the top 3 AI computing stocks now"). The README explicitly positions them for these distinct scenarios.

Can industry-research and industry-funnel be used together?

Yes, they are designed to be complementary. According to the source code in skills/industry-funnel.md (lines 13-15), the industry-research output often feeds the funnel's market-scan input. Run the research skill first to generate the segment tables, then use the funnel skill to drill down into the most attractive segments identified.

What are the specific output formats for each skill?

industry-research outputs logic-chain diagrams, tables of companies grouped by value-chain segment (upstream/midstream/downstream), and tiered company classifications (Tier 1-4). industry-funnel outputs scanned-company tables with entry classes (A/B/C), filter-by-criteria tables with elimination reasons, and a final three-stock recommendation matrix designating positions as core, satellite, or option.

How do the tools ensure data quality in both skills?

Both skills invoke tools/report_audit.py to execute mandatory "准出" audits that verify data completeness and accuracy. They also call tools/financial_rigor.py to perform precise valuation calculations and market-cap verifications. This dual-tool validation ensures that whether the skill produces a broad industry map or a narrow stock selection, the underlying financial data meets rigorous standards.

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