Guide · Agent skill
How to use the Quality Screen (7-Metric Filter) agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The quality-screen skill is a checklist that an AI agent such as Claude or Codex follows when you ask it to screen companies. It applies seven financial-quality tests to each company and marks the ones that clearly fall short. This guide explains what the skill does, how to install it, how to phrase requests so it runs, and where its limits are.
Quality Screen (7-Metric Filter)xbtlin/ai-berkshire · MIT Download the skill Quick start
- 1Download the skill. Download the quality-screen ZIP from this page and unzip it on your computer.
- 2Place or upload the folder. Put the folder in ~/.claude/skills/ or a project's .claude/skills/, or upload it under Settings, Capabilities, Skills in the Claude apps.
- 3Prepare the repository. Clone or locate a checkout of the xbtlin/ai-berkshire repository so the agent can run tools/financial_rigor.py and read AGENTS.md.
- 4Make sure web search and shell are available. The skill needs web search to collect figures and shell access to run repository commands from the repo root.
- 5Ask for a screen. Name one or more companies, an industry, an index, or a theme in a plain request.
- 6Review borderline results. Check the warning-marked companies and the data-insufficient labels before accepting any exclusion.
What the skill makes the agent do
An agent skill is a folder with a SKILL.md file that tells the agent how to handle a certain kind of task. This one, named quality-screen, is a first-pass filter for ruling out non-first-class companies. It is generated from the AI Berkshire project's skills/quality-screen.md so that Claude Code and Codex users follow the same workflow.
The skill tests each company against seven hard metrics: 10-year average ROE, five-year cumulative free cash flow, interest coverage (EBIT divided by interest expense), long-term gross margin, the five-year average of operating cash flow to net profit, long-term net margin, and five-year share-count expansion not caused by acquisitions. It also defines three exemption rules for companies that miss a metric but meet specific conditions.
- Test 1: 10-year average ROE below 8% is excluded
- Test 2: negative five-year cumulative free cash flow is excluded
- Test 3: interest coverage below 2x is excluded (not applied to banks and insurers)
- Test 4: long-term gross margin below 15% is excluded
- Test 5: operating cash flow to net profit (five-year average) below 0.7 is excluded
- Test 6: long-term net margin below 5% is excluded
- Test 7: share count growth above 20% over five years, not from acquisitions, is excluded
Install the quality-screen skill in Claude
You can install the skill in one of three ways. Download the ZIP file from the skill's page on this site and unzip it. Then place the unzipped folder in your personal skills directory, ~/.claude/skills/, or in a project's .claude/skills/ folder if you want the skill available only inside that project.
Alternatively, if you use the Claude apps, you can upload the skill through Settings, then Capabilities, then Skills. The skill is a single file, SKILL.md, so the folder should contain that file at minimum.
- Personal install: unzip into ~/.claude/skills/
- Project install: unzip into the project's .claude/skills/
- Claude apps: Settings → Capabilities → Skills, then upload
~/.claude/skills/quality-screen/SKILL.mdSet up what the skill needs
The skill needs three things to work well. It needs web search access, because it gathers financial figures from the web. It needs shell access, because the skill refers to repository tools such as python3 tools/financial_rigor.py. Those tools are referenced by the skill but are not included in the file you download, so you need a checkout of the xbtlin/ai-berkshire repository to run them. The repository also holds the AGENTS.md rules the skill tells the agent to follow.
If you only want the agent to read figures and reason through the checklist, the skill can still be useful with web search alone, but the source does not say how it behaves without the shared tools, so check the output carefully.
- Web search access for financial data
- Shell access to run repository tools from the repo root
- A checkout of the xbtlin/ai-berkshire repository for shared tools and AGENTS.md
python3 tools/financial_rigor.pyTrigger the skill with natural requests
Ask for a screen by naming the companies, an industry, an index, or a theme. The skill accepts four input types: individual stocks, an industry, a market or index, and a theme. For industry, market, and theme requests, the agent first searches for the main listed companies (industry mode covers roughly the 15 to 20 largest by market value, theme mode covers 15 to 30), then screens each one. If there are more than 30 companies, the agent splits the work into batches.
Before researching, the skill tells the agent to run the date command to confirm today's date and to state the data cutoff date in the report header. Expect the agent to ask you to confirm the company list when it is long.
- Single companies: 'Screen Tencent, Meituan and NVIDIA with the quality screen skill.'
- Industry: 'Run the quality screen on the global cloud computing industry.'
- Index: 'Quality screen the Hang Seng Index constituents.'
- Theme: 'Use quality-screen on the global AI compute chain theme.'
Read the results table
The output starts with a summary table showing each company's result on the seven tests, using a check mark for a pass, a cross for a fail, and a warning sign for a borderline value with an explanation. Each company is then listed as passed, excluded, or exempted under one of the three rules. Companies with missing data are labelled 'data insufficient' rather than passed or failed, and companies with short listing histories are labelled as having an insufficient data window.
For industry, market, or theme runs, the report adds a pass rate, a sector summary, and a three-tier ranking: first-class, acceptable, and eliminated. Read borderline cases in the dedicated section before drawing conclusions.
- Pass / excluded / exempted lists with reasons
- Borderline cases with the actual numbers
- Pass rate and sector ranking (batch runs only)
Understand the three exemption rules
Exemption A covers companies listed fewer than 10 years that have gross margin above 30% and positive operating cash flow in each of the last two years. Exemption B covers companies with gross margin above 30% whose net margin has recovered to 5% or more in the last two years, or is clearly rising. Exemption C covers high-turnover, low-margin models such as membership, platform commission, or high-volume retail, when ROE is above 20%, operating cash flow to net profit is above 1.0, and the business model fits that category.
The skill gives examples: Meituan for Exemption A, Amazon for Exemption B, and Costco for Exemption C. These are examples from the skill's reasoning, not verdicts on those companies.
- Exemption A: ROE shortfall during an investment phase
- Exemption B: low net margin by strategic choice
- Exemption C: thin margins with high capital returns
What people use it for
- Screening a list of listed companies to remove clearly weak businesses before deeper research
- Comparing peers within one industry by seeing which pass the seven tests and which need exemptions
- Running a quick, consistent check across a stock index's constituents
- Learning value investing by seeing a structured checklist applied to real companies
Frequently asked questions
Is the quality-screen skill free to use?
The skill is released under the MIT licence, as listed for the xbtlin/ai-berkshire repository. Check the licence file in the repository for the full terms before reusing or redistributing it.
What do I need to run the quality screen skill?
You need web search access to collect financial data, shell access to run repository tools such as python3 tools/financial_rigor.py, and a checkout of the xbtlin/ai-berkshire repository for the shared tools and AGENTS.md rules. The tools are referenced but not included in the downloaded file.
Where do I install an agent skill like this?
Unzip the folder into ~/.claude/skills/ for personal use or into a project's .claude/skills/ folder. In the Claude apps, you can upload it under Settings, then Capabilities, then Skills.
Does a company passing the screen mean it is a good investment?
No. The skill states that passing the screen only means the company was not ruled out by the seven tests. It does not assess valuation, management, business model durability, or competition, so further research is still required.
Does the skill work for banks and insurers?
The interest coverage test does not apply to banks and insurers, because their business is built on interest spreads. The skill says to skip test 3 for them.
Are there alternatives or does this compare to other stock screeners?
The source does not compare this skill with other screeners or tools, so this guide cannot make a claim about how it ranks against them. It is a fixed seven-test checklist run by an agent that relies on web search, so its results depend on the figures the agent finds.
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
Use this skill if you want a quick, consistent first-pass filter across many listed companies, or if you are learning value investing and want to see a structured checklist applied to real firms. Pick something else if you need a buy or sell decision, a valuation, a judgement on management, or reliable data for companies where public figures are thin, since the skill itself says it does not provide those.
Download the skill