# Bottleneck-Hunter Skill: A Systematic Method to Identify Supply Chain Investment Opportunities

> Discover the Bottleneck-Hunter skill a Claude Code automation that uncovers supply chain investment opportunities by identifying scarce components through six quantitative criteria.

- Repository: [Xbt Lin/ai-berkshire](https://github.com/xbtlin/ai-berkshire)
- Tags: how-to-guide
- Published: 2026-07-25

---

**The Bottleneck-Hunter skill is a Claude Code automation that decomposes macro trends into physical supply chain layers and identifies scarce components using six quantitative criteria to generate ranked investment targets.**

The `bottleneck-hunter` skill in the **xbtlin/ai-berkshire** repository provides a structured, data-driven framework for scanning global supply chains to uncover physical scarcity bottlenecks. Unlike traditional stock screeners that rely on financial metrics alone, this skill evaluates **supply chain opportunities** through a rigorous seven-stage pipeline that maps "super-trends" (such as AI infrastructure or semiconductor re-industrialization) down to specific constrained components and their public market suppliers.

## How the Bottleneck-Hunter Skill Works

The skill operates as an end-to-end analytical pipeline defined in [`skills/bottleneck-hunter.md`](https://github.com/xbtlin/ai-berkshire/blob/main/skills/bottleneck-hunter.md). It transforms high-level macro narratives into concrete investment candidates by enforcing physical-world constraints throughout the analysis.

The methodology rests on six quantitative bottleneck criteria defined in **Section 3.1** of the source documentation: **集中度** (market concentration), **扩产周期** (capacity expansion cycle), **替代难度** (substitution difficulty), **利用率** (utilization rate), **需求增速** (demand growth velocity), and **客户验证周期** (customer validation cycle). Each candidate component receives a score against these standards to determine if it represents a genuine supply constraint.

## The Seven-Stage Pipeline Architecture

### Stage 1: Trend Confirmation

The process begins by selecting a "super-trend" that satisfies four threshold conditions: longevity, physicality, scale, and acceleration. According to the *Trend筛选标准* section (lines 21-27) in [`skills/bottleneck-hunter.md`](https://github.com/xbtlin/ai-berkshire/blob/main/skills/bottleneck-hunter.md), only trends with multi-year duration and tangible hardware requirements proceed to decomposition.

### Stage 2: Physical Supply-Chain Decomposition

The skill breaks the validated trend into five distinct layers, mapping from end terminals (终端) down to foundational infrastructure. As illustrated in lines 58-100 of the source file, an AI infrastructure trend decomposes into specific sub-components—such as high-bandwidth memory substrates, liquid cooling manifolds, or advanced packaging substrates—each assigned to its appropriate layer in the supply hierarchy.

### Stage 3: Bottleneck Identification

Every Layer-2 and Layer-3 element undergoes scoring against the six bottleneck standards. Components achieving **S-level** or **A-level** bottleneck status exhibit high concentration ratios, long capacity expansion cycles (typically 18-36 months), high utilization rates (>85%), and rapid demand acceleration. These characteristics indicate genuine physical scarcity rather than temporary market noise.

### Stage 4: Company Screening

For each verified bottleneck, the skill queries publicly listed companies supplying the constrained component. The initial filter matrix (**4.2 初筛标准**, lines 68-74) applies strict eligibility rules: active listing status, minimum business segment revenue percentage (typically >30%), adequate market capitalization, and sufficient liquidity thresholds.

### Stage 5: Deep Valuation Check

This stage implements a "red/yellow/green" valuation guardrail system (**4.2.1 估值检查**, lines 77-95). The skill calculates current PS ratios, PE multiples normalized against sector growth rates, and performs a "10-year 25× PE exit" back-test. Candidates with implied annual returns below the hurdle rate receive automatic disqualification, preventing entry into overvalued scarcity plays.

### Stage 6: Cross-Verification

Every bottleneck and candidate company requires corroboration from at least two independent data sources. The cross-validation tables (**5.1 正向验证** and **5.2 反向验证**, lines 45-64) mandate evidence triangulation through industry reports, customer disclosure filings, or supply chain news before final inclusion.

### Stage 7: Output Generation

The pipeline culminates in a ranked "bottleneck opportunity board" (**6.1 瓶颈机会排名表**) and standardized one-page company summaries (**6.2 一页纸摘要**). These outputs aggregate in [`reports/bottleneck-map/master-map.md`](https://github.com/xbtlin/ai-berkshire/blob/main/reports/bottleneck-map/master-map.md), which serves as the persistent knowledge base for ongoing trend monitoring.

## Key Implementation Files

Understanding the repository structure reveals how the skill maintains consistency across Claude Code and Codex environments:

- **[`skills/bottleneck-hunter.md`](https://github.com/xbtlin/ai-berkshire/blob/main/skills/bottleneck-hunter.md)**: The canonical skill definition containing methodology, scoring tables, and valuation red-lines (lines 14-100).
- **[`codex-prompts/bottleneck-hunter.md`](https://github.com/xbtlin/ai-berkshire/blob/main/codex-prompts/bottleneck-hunter.md)**: Auto-generated mirror prompt enabling Codex CLI usage through the `sync-codex-skills` utility.
- **[`reports/bottleneck-map/master-map.md`](https://github.com/xbtlin/ai-berkshire/blob/main/reports/bottleneck-map/master-map.md)**: The central aggregation point for daily scan results and persistent bottleneck tracking.
- **[`tools/twstock_data.py`](https://github.com/xbtlin/ai-berkshire/blob/main/tools/twstock_data.py)**: Data retrieval helper for Taiwanese equity markets, referenced in Stage 4 screening and Stage 5 valuation calculations.

## How to Invoke the Bottleneck-Hunter Skill

### Direct CLI Invocation (Claude Code)

Execute the skill by passing a validated super-trend as the argument:

```bash
/bottleneck-hunter AI基础设施

```

This command triggers the full seven-stage pipeline and returns a markdown report containing the opportunity board and company summaries.

### Programmatic Integration (Python)

For automated workflows or scheduled scans, invoke the skill via subprocess:

```python
import subprocess, json, pathlib

def run_bottleneck_hunter(trend: str) -> str:
    """Execute the skill via the local Claude Code entry point."""
    cmd = ["claude-code", "run", "bottleneck-hunter", trend]
    result = subprocess.run(cmd, capture_output=True, text=True, check=True)
    return result.stdout   # markdown report

report_md = run_bottleneck_hunter("AI基础设施")
print(report_md[:500])  # preview first 500 chars

```

The script calls the `claude-code` binary installed by the repository's tooling chain and captures the generated analysis.

### Codex CLI Usage

After running `sync-codex-skills`, the prompt becomes available as a first-class Codex command:

```text
> bottleneck-hunter AI基础设施

```

## Summary

- **The Bottleneck-Hunter skill** automates supply chain scarcity detection through a rigorous seven-stage pipeline defined in [`skills/bottleneck-hunter.md`](https://github.com/xbtlin/ai-berkshire/blob/main/skills/bottleneck-hunter.md).
- **Six quantitative criteria** (concentration, expansion cycle, substitution difficulty, utilization, demand growth, and validation cycle) determine whether a component represents a genuine bottleneck.
- **Physical decomposition** maps abstract trends into five concrete supply layers, ensuring analysis targets tangible hardware constraints rather than narrative themes.
- **Valuation guardrails** filter candidates using PS/PE ratios and 10-year exit back-tests to prevent overpayment for scarcity assets.
- **Cross-verification requirements** mandate dual-source confirmation for all bottlenecks and suppliers, maintaining data integrity throughout the pipeline.

## Frequently Asked Questions

### What distinguishes the Bottleneck-Hunter skill from traditional stock screeners?

Traditional screeners filter by financial metrics like PE or ROE in isolation. The Bottleneck-Hunter skill starts with **physical supply chain constraints**—evaluating whether a component is genuinely scarce based on capacity utilization, expansion lead times, and demand velocity—then maps those constraints to financially viable public companies. This "physical-first" approach identifies opportunities arising from real-world manufacturing limitations rather than accounting anomalies.

### How does the skill define and score bottleneck severity?

According to **Section 3.1** of [`skills/bottleneck-hunter.md`](https://github.com/xbtlin/ai-berkshire/blob/main/skills/bottleneck-hunter.md), the skill evaluates six standards: market concentration (集中度), capacity expansion cycle length (扩产周期), substitution difficulty (替代难度), facility utilization rates (利用率), demand growth velocity (需求增速), and customer validation timelines (客户验证周期). Components scoring highly on concentration and expansion cycle metrics while showing high utilization and demand growth receive **S-level** or **A-level** bottleneck classifications, indicating severe supply constraints.

### What valuation safeguards prevent investing in overpriced bottleneck plays?

Stage 5 implements a strict **red/yellow/green** valuation system (**4.2.1 估值检查**). The skill calculates current Price-to-Sales ratios, normalized PE multiples against sector growth rates, and performs a "10-year 25× PE exit" back-test. If the implied compound annual growth rate falls below the required hurdle rate (typically 15-20%), the candidate receives a red light and exclusion from the final opportunity board, regardless of scarcity severity.

### Can this skill operate outside Claude Code environments?

Yes. The repository includes a [`codex-prompts/bottleneck-hunter.md`](https://github.com/xbtlin/ai-berkshire/blob/main/codex-prompts/bottleneck-hunter.md) file that mirrors the full skill definition for OpenAI Codex usage. After executing the `sync-codex-skills` command, the identical seven-stage pipeline becomes available as a native Codex command. Additionally, the Python subprocess example demonstrates how external automation scripts can trigger the skill via the `claude-code` CLI binary.