How to use the Bottleneck Hunter agent skill (supply chain)

What is an agent skill, and what does this one do?

An agent skill is a folder containing a SKILL.md file that gives an AI agent a set of instructions for a specific kind of task. When a request matches the skill, the agent follows its method instead of answering from general knowledge alone.

Bottleneck Hunter starts from a super-trend you name, such as AI infrastructure, energy transition, defense, semiconductor reshoring, or the space economy. It decomposes that trend into physical layers, identifies Layer 2 and Layer 3 choke points such as optics, lasers, InP and SOI substrates, IC substrates, probe cards and specialty fiberglass, and looks for listed companies tied to those links.

• Confirms the trend against durability, physicality, scale and acceleration criteria using web search.
• Scores each supply-chain link on six scarcity criteria and grades it.
• Screens listed companies tied to S- and A-grade bottlenecks.
• Runs valuation gates through the financial_rigor tool and a bear-case reverse check.
• Outputs a ranked bottleneck opportunity board.

Install the Bottleneck Hunter skill

Download the ZIP file for the skill from this page and unzip it. The skill folder is named bottleneck-hunter and contains SKILL.md. Place it in one of two locations depending on how you use the agent.

For a personal setup, put the folder in ~/.claude/skills/. For a single project, put it in that project's .claude/skills/ folder so it applies only there.

In the Claude apps, you can instead upload the skill under Settings → Capabilities → Skills.

• Personal install: ~/.claude/skills/bottleneck-hunter/
• Project install: the project's .claude/skills/bottleneck-hunter/ folder
• Claude apps: Settings → Capabilities → Skills, then upload the skill

How do you trigger the skill with a natural request?

You do not need a special command. The skill's description says to use it when you want hidden beneficiaries of a structural trend rather than already-priced leaders, so ask for that directly. Name the trend and say you want under-the-radar companies tied to its bottlenecks.

Specific requests work better than general ones. Naming the trend, the layers you care about, and the market or listing venues you want covered helps the agent stay on the method. The skill is written around listed companies in markets such as A, HK, US, JP, TW and EU.

• "Use the bottleneck hunter on AI infrastructure and list under-the-radar suppliers of optical components."
• "Find hidden beneficiaries of the semiconductor reshoring trend, starting from Layer 2 choke points."
• "Run a supply-chain bottleneck screen on the energy transition, focusing on grid and storage components."

What output should you expect?

The main deliverable is a ranked bottleneck opportunity board. Each row lists the company, ticker, market cap, revenue, PS and PE, the bottleneck link, its grade, revenue share from that link, growth, a signal strength rating and a valuation note. Market cap, revenue, PS and PE are required fields in the source, and if financials cannot be obtained the signal strength is capped at two stars.

The grading is mechanical. Each link is scored red, yellow or green on six criteria, and four or more reds make an S-grade single-point risk, three reds make an A-grade, one or two make a B-grade, and no reds means the link is skipped.

• Bottleneck grades: S (four or more red criteria), A (three), B (one or two), skipped (none).
• Signal strength runs from one to five stars and depends on cross-validation and valuation colour.
• Valuation gates can cap a signal at two stars when the valuation is red.

Read the valuation gate and the bear case

The valuation gate is mandatory in the method. For each company the agent should compute PE, PB, ROE and FCF yield with financial_rigor (command=verify_valuation) and run scenario valuation with financial_rigor (command=three_scenario). Red-light flags include a market cap above 20% of TAM and a PS above 30x with revenue growth under 100%. Yellow flags include loss-making companies with PS above 15x.

The skill also asks a sanity question: if the most optimistic scenario fully plays out and you exit at 25x PE in ten years, what is the annualised return? An answer under 10% per year should be flagged as having no margin of safety at the current price.

The reverse checks ask why smart investors are not already buying, whether the bottleneck can be bypassed, whether competitors can add capacity quickly, and what happens to the company if end demand falls by half.

• Red light: caps signal strength at two stars and flags valuation stretched.
• Yellow light: needs extra justification or is downgraded.
• Green light: a bonus, for example PS under 10x with growing revenue.

Safety, data limits and what the skill needs

The skill needs a web search tool, the financial_rigor tool, and screen_market plus get_financial_statements or get_stock_news for market data. It does not run scripts, but it does use the network to fetch information. Treat its outputs as research notes that you verify yourself.

The source states that the skill does not verify data against primary filings, that its outputs depend on web search results and model judgment, and that errors are likely. Its output can also encourage speculative trading. Check every company against its own filings and do not treat a ranking as a recommendation.

The method is also weak for trends with little physical supply chain, such as pure software, because its core approach assumes hardware or material bottlenecks.

• Required tools: web search, financial_rigor (verify_valuation and three_scenario), screen_market, get_financial_statements, get_stock_news.
• No scripts are run; network access is used.
• Verify figures against primary filings before acting on them.