Guide · Agent skill
How to use the bottleneck-hunter agent skill for supply chains
Updated · written with AI from the project's own documentation and checked against our catalogue data
The bottleneck-hunter skill is a set of written instructions that tells an AI coding agent, such as Claude Code or Codex, to work as an investment researcher. It looks for hard-to-replace parts of a supply chain that could limit a major trend, such as AI data-center buildout, and then lists and ranks the listed companies that supply them. This guide explains what the skill does, how to install it, how to phrase a request, and what it needs and risks, so you can run it with your eyes open.
Supply Chain Bottleneck Hunterxbtlin/ai-berkshire · MIT Download the skill Quick start
- 1Download the ZIP. Download the bottleneck-hunter skill ZIP from the download link on this page.
- 2Unzip into a skills folder. Unzip it into ~/.claude/skills/ for all projects, or into a project's .claude/skills/ folder for one project.
- 3Or upload it in the Claude apps. In the Claude apps, go to Settings → Capabilities → Skills and upload the ZIP.
- 4Check out the repository. Clone or check out xbtlin/ai-berkshire so that tools/ and AGENTS.md are present, and start your agent from the repository root.
- 5Install Python 3. Make sure python3 is available in the shell your agent uses, because the bundled calculation tools need it.
- 6Make a request. Ask for a named trend, such as AI infrastructure, and request a bottleneck map or opportunity board.
- 7Cross-check the results. Verify the cutoff date, the financial figures and the valuation flags against primary sources before acting.
What the skill makes the agent do
An agent skill is a folder with a SKILL.md file that gives an AI agent a repeatable workflow. This skill is an adapter: its own text points the agent to a canonical workflow in the same repository, skills/bottleneck-hunter.md, which is written in Chinese.
The workflow starts from the question of which link in a physical supply chain would run short first if a trend keeps growing. It confirms the trend, maps the supply chain in layers, rates each choke point against six criteria, lists listed companies that supply it, and checks valuation (P/S, P/E and market cap against market size) before assigning a signal-strength score from one to five stars.
- Trend check: the trend must look durable, physical, large (global capex above 500 億 USD a year is one stated threshold) and supply-constrained.
- Layer map: the chain is broken down from end products down to materials, equipment and infrastructure.
- Bottleneck rating: six criteria (supplier concentration, expansion time, substitutability, utilisation, demand growth, customer qualification time) give S, A or B ratings.
- Valuation check: red, yellow and green flags are applied to each candidate before ranking.
- Output: a ranked opportunity table, a one-page summary per candidate, and suggested actions such as watchlist or further research.
Install the skill in Claude Code or Codex
Download the skill ZIP from the download link on this page. Then choose one of two install locations, depending on whether you want the skill for all your projects or just one.
For a personal install that works in every project, unzip the file into ~/.claude/skills/. For a single project, unzip it into that project's .claude/skills/ folder. You can also upload the ZIP in the Claude apps under Settings → Capabilities → Skills.
- Personal install: ~/.claude/skills/
- Project install: the .claude/skills/ folder inside your project
- Claude apps: Settings → Capabilities → Skills, then upload the ZIP
~/.claude/skills/Set up the repository the skill depends on
The skill file alone is not enough. Its instructions call shared scripts in tools/ and research rules in AGENTS.md, both of which live in the xbtlin/ai-berkshire repository. Check out that repository and start your agent from its root, so the paths in the instructions resolve.
The skill tells the agent to run the bundled calculators with Python 3, for example financial_rigor.py for exact valuation and maths, and twstock_data.py for Taiwan stock figures. Taiwan figures come from the FinMind data source, described in skills/financial-data.md; the source does not state the API key requirements, so check that file before running Taiwan-stock queries.
- Repository: xbtlin/ai-berkshire, with tools/ and AGENTS.md present
- Python 3 for the calculation tools
- FinMind access for Taiwan stock data (see skills/financial-data.md)
- Web search capability for market research
python3 tools/financial_rigor.py
python3 tools/twstock_data.pyHow to trigger the skill with a natural request
The skill reads your request as the trend to analyse. Name a trend from the tracked list or describe one, and the agent narrows its scan to that trend. If you name no trend, the agent works from its default list.
Ask for the output you want. The workflow is built for ranked candidates, so asking for a bottleneck map or an opportunity board matches its outputs better than a general question about stocks.
- Scan the AI infrastructure trend and show the S-grade and A-grade bottlenecks.
- Find listed suppliers of optical transceiver lasers and check their valuations.
- Map the bottlenecks in semiconductor reshoring and update the master map.
- Run the energy transition trend and add anything new to the watchlist.
What the output looks like and how to read it
The report opens with a data cutoff date. The skill tells the agent to run the date command first and to treat that date as the baseline for 'latest' data. Check that date before trusting any figure.
Read the bottleneck map first, then the ranking table. An S-grade bottleneck means four or more of the six criteria are red, which the skill calls a single-point-of-failure level. A star rating reflects both cross-checks and valuation: a candidate with a red valuation flag is capped at two stars and labelled as overvalued.
- Ranking table: market cap, annual revenue, P/S, P/E, bottleneck grade, market share, revenue growth, signal strength and valuation verdict.
- Each candidate needs market cap, revenue, P/S and P/E. If financial data cannot be found, the skill caps the rating at two stars.
- Cross-check status lines mark verified items and unverified items separately.
Safety notes: scripts, network and financial risk
This skill runs scripts and uses the network. Its instructions have the agent run shell commands against the repository's tools/, and perform web searches and data lookups. Read the commands the agent proposes before approving them, and only run the skill in a repository you trust.
The output is an analytical framework, not investment advice. The skill itself says results need cross-checking, and its figures depend on web searches that may be stale or wrong. Verify every price, market cap and financial figure from a primary source before acting on it.
- Review shell commands before approval, especially those that run scripts in tools/.
- Expect network calls for web search and FinMind data.
- Verify all financial figures independently.
- Keep your own risk limits; the skill does not provide personalised advice.
What people use it for
- Building a bottleneck map for the AI infrastructure trend, from GPUs down to optical components and substrates.
- Screening listed suppliers of a narrow component, then comparing their valuations on P/S and P/E.
- Maintaining a watchlist of candidates over time in the reports/bottleneck-map/ folder.
- Studying how an agent can be instructed to run a multi-step financial workflow with shell tools.
Frequently asked questions
How do I install the bottleneck-hunter skill?
Download the skill ZIP from this page and unzip it into ~/.claude/skills/ for all projects, or into a project's .claude/skills/ folder for one project. You can also upload the ZIP in the Claude apps under Settings → Capabilities → Skills. Then check out the xbtlin/ai-berkshire repository so the tools it calls are present.
What do I need to run the bottleneck-hunter skill?
You need Codex or Claude Code with shell access, the xbtlin/ai-berkshire repository including tools/ and AGENTS.md, and Python 3 for the bundled tools. Taiwan stock figures use FinMind, and the skill also needs web search for market research. The source does not state FinMind key requirements, so check skills/financial-data.md.
Is the bottleneck-hunter skill free and what is its licence?
The item is listed under the MIT licence. The source does not state any price, so this page makes no claim about cost beyond the licence. The skill depends on tools and data sources that you may need to set up yourself.
Can the skill give me guaranteed investment advice?
No. The skill produces an analytical framework and its own instructions say results need cross-checking. Figures come from web searches and may be stale or incorrect, so you should verify them before making any decision.
Why does the skill fail or say a tool is missing?
The most likely cause is that the repository is not checked out, or your agent is not running from its root. The instructions call scripts such as tools/financial_rigor.py and tools/twstock_data.py using paths relative to the repository root. Check that tools/ and AGENTS.md exist and that python3 is available.
How does this compare with other stock screening tools?
The source does not compare this skill with other tools, so this page makes no comparison. Its distinctive approach is to start from physical supply-chain choke points and rate them against six criteria, then apply valuation checks before ranking.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
This skill suits investors or analysts who already research stocks and read Chinese, and who want a structured bottleneck framework run by an AI agent. It is a poor fit for anyone wanting personalised advice or who cannot verify financial data, and it will not work without the repository's tools/ folder.
Download the skill