How to use the Industry Funnel Stock Screener agent skill

What an agent skill is and what this one does

An agent skill is a folder containing a SKILL.md file, which gives an AI agent a reusable set of instructions for a specific kind of task. When a request matches the skill, the agent follows its workflow instead of improvising.

The industry-funnel skill (skill name industry-funnel) narrows a sector down in four layers. First it builds a candidate pool of 30 to 60 companies from actively traded stocks, large gainers and the largest companies in the industry, across A-shares, Hong Kong, US and some international markets, plus a section for unlisted candidates. Second it applies five hard value-investing tests: PE valuation, ROE, operating cash flow relative to net profit, debt-to-asset ratio and a qualitative moat rating, keeping no more than about 10 companies. Third it writes a structured analysis of each survivor. Fourth it produces a detailed report on three final picks, covering a combined portfolio table, position-size suggestions and a data-accuracy check.

Every company dropped at each layer must be recorded with a reason, so the process can be reviewed later.

Install the skill from the ZIP download

Download the ZIP for the industry-funnel skill from this page and unzip it. The skill is a folder that contains SKILL.md. Put that folder in one of these locations, depending on how you want to use it.

For all your projects, place it in your personal skills folder. For a single project, place it in that project's .claude/skills/ folder so it applies only there.

Alternatively, you can upload the skill in the Claude apps under Settings, then Capabilities, then Skills.

• Personal install: ~/.claude/skills/industry-funnel/
• Project install: .claude/skills/industry-funnel/ inside your project folder
• Claude apps: Settings → Capabilities → Skills, then upload the skill

Set up the repository tools the workflow expects

The SKILL.md file refers to Python scripts in a tools/ folder of the source repository, including financial_rigor.py for exact arithmetic on valuation figures and report_audit.py for the data-accuracy check. The skill tells the agent to run commands from the repository root, such as python3 tools/financial_rigor.py, and to locate the real checkout path if the session starts outside the repository.

The source does not bundle these scripts with the ZIP. If you install only the SKILL.md folder, the audit and math steps will not have their tools available, so you need access to the repository's tools/ folder as well.

Trigger the skill with a natural request

You do not need special syntax. Name an industry or investment theme in plain language and ask for a funnel-style screen. The skill's examples include themes such as AI computing power, innovative drugs and robotics. The agent will recognise the request and follow the workflow.

Good requests name the sector and say what you want back. For example, ask for a funnel screen of the robotics sector that ends with three finalists and a log of rejected companies. The skill also expects you to say whether you want the report in Chinese; the source sets Chinese as the report language.

The source also describes a Codex form in which the request is treated as the arguments of the current thread. The skill works the same way: the name of the industry in your request is what the agent screens.

• Screen the AI computing power industry with the industry-funnel skill and list every rejected company with a reason.
• Run an industry funnel on innovative drugs and give me three finalists with position-size suggestions.
• Use industry-funnel on humanoid robots. Include unlisted candidates and tell me the data cutoff date.

Read the output and the finalist tables

A complete run produces a report saved under reports/ with a name in the form {industry}-funnel-{date}.md. Expect a funnel overview, the candidate table with market and category tags, the five-test screening table with pass, fail and rejection reasons, per-company analysis, the three-finalist portfolio table, an industry-level ETF alternative, a self-rated information-sufficiency table and a list of data points to update.

Read the rejection log as carefully as the picks. The skill is designed so that you can see why each company was dropped and check whether you agree with the reasoning. Also check the data-cutoff date, because the source says the agent should confirm today's date from the system and use it as the baseline for 'latest' data.

• Check the data-cutoff date in the report header
• Review the rejection reasons, not just the final three
• Check items marked as estimates and the data points flagged for later verification

Know what the skill needs and what it can do on your machine

The skill needs web search so the agent can gather financial data. It also needs access to the repository's tools/ folder. Financial data sources such as company filings, Wind, Tonghuashun or Yahoo Finance are referenced in the workflow but are not bundled, so you need to be able to reach them yourself.

Safety notes: this skill runs scripts and uses the network. The workflow runs shell commands such as python3 for the tools, and it searches the web for financial data. Review the scripts in the tools/ folder before running the skill on a machine with sensitive data, and be aware that web queries are sent out during research.

The skill does not provide real-time market data by itself. Its accuracy depends on the sources the agent reaches and on your own checks.

• Requires web search capability
• Requires access to the repository tools/ folder (financial_rigor.py, report_audit.py)
• Requires financial data sources you can access, which are not bundled
• Runs scripts and uses the network

Tips for getting useful results

Name a specific sector rather than a broad one. A narrow theme gives the five-test screen a coherent candidate pool. Ask the agent to include unlisted candidates and Asian listings, because the skill warns that English-language sources tend to under-represent A-shares and Hong Kong stocks.

Treat the moat rating and the 'story versus real business' distinction as judgements to check, not facts. The skill itself notes that AI research has biases such as favouring large companies and high-media-attention stocks.

• Name a narrow sector for a tighter candidate pool
• Ask for both Chinese and English sources where relevant
• Verify figures against the filings listed in the source list section
• Use the skill as the entry point, then run the related follow-up skills on each finalist if you want deeper work