# AI Berkshire industry-research vs industry-funnel: Complete Skill Comparison

> Compare AI Berkshire's industry-research and industry-funnel skills. Understand how industry-research maps sectors and industry-funnel identifies top stocks for investment.

- Repository: [Xbt Lin/ai-berkshire](https://github.com/xbtlin/ai-berkshire)
- Tags: comparison
- Published: 2026-07-02

---

**`industry-research` generates a comprehensive value-chain panorama of an entire sector, while `industry-funnel` executes a disciplined six-stage filter to identify the top three individual stocks for investment.**

The `xbtlin/ai-berkshire` repository provides two distinct **industry screening skills** that serve sequential stages of the investment workflow. Understanding when to deploy `industry-research` versus `industry-funnel` determines whether you receive a macro-level sector map or a micro-level actionable shortlist. Both skills are defined as canonical markdown files under `skills/` and leverage parallel agent execution, but they differ fundamentally in granularity, output structure, and analytical depth.

## Primary Goals and Output Focus

### Industry-Research: Value-Chain Mapping

The `industry-research` skill produces a **full-panorama of an industry's value-chain**, mapping every investment opportunity across upstream, mid-stream, downstream, and auxiliary segments. According to [`skills/industry-research.md`](https://github.com/xbtlin/ai-berkshire/blob/main/skills/industry-research.md), this skill executes a five-step workflow: build and validate a logic chain, draw the value-chain diagram, scan all global listed companies, perform four-master analysis on head-companies per segment, and deliver portfolio-allocation recommendations.

**Output granularity** remains at the **segment level**. You receive tiered company classifications (Tier 1-4), logic-chain diagrams, and tables grouping companies by their position in the value chain. This serves investors who need to understand every foothold an industry offers before selecting a focus area.

### Industry-Funnel: Discriminating Stock Selection

The `industry-funnel` skill, as defined in [`skills/industry-funnel.md`](https://github.com/xbtlin/ai-berkshire/blob/main/skills/industry-funnel.md), performs a **layer-by-layer "funnel" selection** that narrows an entire market down to exactly three final target stocks. The workflow emphasizes hard elimination: full-market scan by activity and price-rise, application of five hard-value-investing criteria, rough filtering to ten or fewer companies, detailed 300-500 word analysis, and final deep dives of 800-1200 words on the three survivors.

**Output granularity** operates at the **company level**. The deliverable includes a scanned-company table with entry classes (A/B/C), a filter-by-criteria table documenting pass/fail statuses, and a final recommendation matrix categorizing selections as core, satellite, or option positions.

## Technical Implementation in the Codebase

### Skill Definition and Agent Orchestration

Both skills are implemented as canonical markdown specifications under [`skills/industry-research.md`](https://github.com/xbtlin/ai-berkshire/blob/main/skills/industry-research.md) and [`skills/industry-funnel.md`](https://github.com/xbtlin/ai-berkshire/blob/main/skills/industry-funnel.md). The repository generates Codex-compatible wrappers at [`codex-skills/industry-research/SKILL.md`](https://github.com/xbtlin/ai-berkshire/blob/main/codex-skills/industry-research/SKILL.md) and [`codex-skills/industry-funnel/SKILL.md`](https://github.com/xbtlin/ai-berkshire/blob/main/codex-skills/industry-funnel/SKILL.md), ensuring a single source of truth for both Claude-Code and Codex clients.

Each skill spawns multiple **Agents** that execute sub-steps in parallel. For `industry-research`, agents handle "全市场扫描" (full-market scanning) and segment-level four-master analysis. For `industry-funnel`, agents manage the "价值投资 5 条硬指标" (five hard-value criteria) filtering and the sequential elimination pipeline. These agents return structured JSON that the skill's orchestration script stitches into the final markdown report.

### Audit and Validation Tools

Both skills invoke **[`tools/report_audit.py`](https://github.com/xbtlin/ai-berkshire/blob/main/tools/report_audit.py)** for data-quality checks and **[`tools/financial_rigor.py`](https://github.com/xbtlin/ai-berkshire/blob/main/tools/financial_rigor.py)** for precise valuation mathematics. These tools execute the "准出" (audit-pass) validation before publishing any output. The [`report_audit.py`](https://github.com/xbtlin/ai-berkshire/blob/main/report_audit.py) script verifies that segment tables in `industry-research` contain complete global coverage, while [`financial_rigor.py`](https://github.com/xbtlin/ai-berkshire/blob/main/financial_rigor.py) confirms market-cap calculations and multi-scenario valuations in `industry-funnel` meet Buffett-style rigor standards.

## Practical Usage Examples

### Running a Sector-Wide Analysis

Execute the value-chain scan to understand all opportunities within a sector:

```text
/industry-research 核电

```

This generates `核电-industry-20260409.md`, containing the logic chain, segment-level company tables, four-master analyses per segment, and a portfolio allocation matrix.

### Executing a Targeted Selection

Run the funnel to distill a known sector into three actionable equities:

```text
/industry-funnel AI算力

```

This produces `AI算力-funnel-20260509.md`, documenting the initial 30-60 candidates, the filtered shortlist of ten or fewer companies, and the final three-stock recommendation with risk/return profiles.

### Integrated Workflow

Combine both skills for comprehensive coverage:

```text

# Step 1: Map the entire sector structure

/industry-research AI算力

# Step 2: Select the optimal individual equities

/industry-funnel AI算力

```

The second command can reuse segment tables generated by the first, ensuring consistency between the macro sector view and the micro stock selection.

## Summary

- **`industry-research`** operates at the **segment level**, producing exhaustive value-chain maps and tiered company classifications for entire global sectors.
- **`industry-funnel`** operates at the **company level**, applying five hard-value criteria to eliminate candidates and deliver exactly three deeply-analyzed stock recommendations.
- Both skills utilize **parallel agent execution** and mandatory **audit validation** via [`tools/report_audit.py`](https://github.com/xbtlin/ai-berkshire/blob/main/tools/report_audit.py) and [`tools/financial_rigor.py`](https://github.com/xbtlin/ai-berkshire/blob/main/tools/financial_rigor.py).
- Use **`industry-research`** first to reveal sector structure, then **`industry-funnel`** to distill the most attractive individual equities.

## Frequently Asked Questions

### Can I use industry-funnel without first running industry-research?

Yes, but you lose the segment-level context. The `industry-funnel` skill performs its own full-market scan via the "全市场扫描" agent, so it can operate independently. However, as noted in the "与 `industry-research` 的区别" section of [`skills/industry-funnel.md`](https://github.com/xbtlin/ai-berkshire/blob/main/skills/industry-funnel.md), running `industry-research` first provides the value-chain logic that ensures your funnel analysis targets the correct industry segments.

### How do the five hard-value criteria in industry-funnel differ from the four-master analysis in industry-research?

The **five hard-value criteria** in `industry-funnel` serve as binary elimination filters (pass/fail) designed to quickly reduce 30-60 candidates to 10 or fewer. The **four-master analysis** in `industry-research` provides qualitative, narrative depth on segment leaders without eliminating candidates. The former focuses on investment viability; the latter on competitive positioning and industry structure.

### Why does industry-funnel limit output to exactly three stocks?

The constraint targets **portfolio construction discipline**. According to the workflow defined in [`skills/industry-funnel.md`](https://github.com/xbtlin/ai-berkshire/blob/main/skills/industry-funnel.md), the final stage categorizes selections into core, satellite, and option positions—three distinct risk/return profiles that provide sufficient diversification without diluting analysis quality. The 800-1200 word deep-dive format requires this focus to maintain the "准出" audit standards enforced by [`tools/report_audit.py`](https://github.com/xbtlin/ai-berkshire/blob/main/tools/report_audit.py).

### Where are the skill definitions stored in the repository?

Canonical definitions reside in [`skills/industry-research.md`](https://github.com/xbtlin/ai-berkshire/blob/main/skills/industry-research.md) and [`skills/industry-funnel.md`](https://github.com/xbtlin/ai-berkshire/blob/main/skills/industry-funnel.md). Auto-generated Codex wrappers exist at [`codex-skills/industry-research/SKILL.md`](https://github.com/xbtlin/ai-berkshire/blob/main/codex-skills/industry-research/SKILL.md) and [`codex-skills/industry-funnel/SKILL.md`](https://github.com/xbtlin/ai-berkshire/blob/main/codex-skills/industry-funnel/SKILL.md). The high-level comparison appears in [`README.md`](https://github.com/xbtlin/ai-berkshire/blob/main/README.md) lines 456-510 under the "行业筛选类" section.