How to Analyze Any Stock with Claude (Free Buffett Skill)
A free Claude skill that scores any stock out of 100 with Buffett's rules and gives you a watch price. Setup steps, how it works, and a real Netflix run.
17 min read
You can analyze any stock with Claude by giving it a fixed process instead of a loose question. The Buffett Business Analyzer is a free Claude skill that pulls current financials, scores a company out of 100, and ends with a watch price: the price at which the stock would start to look like a buy.
A loose question gets a loose answer. "Is this stock a good buy?" invites a confident paragraph and nothing you can check. This skill makes Claude run the same checks on every company, show the numbers behind each score, and say no when the evidence is thin. It works on one company at a time, and it is built to look for reasons not to buy.
Written 7 October 2026. The example run further down is from 20 July 2026.
Educational, not financial advice
The skill produces an AI-generated analysis built from web data. It can contain errors (I found one in my own example, covered below), and it knows nothing about your finances or the rest of your portfolio. Check the numbers against the company's filings before you act on anything.
What is the Buffett Business Analyzer skill?
The Buffett Business Analyzer is a free Claude skill that runs one listed company through a Buffett-style checklist. It returns a score out of 100, an estimate of what the business is worth, a watch price, a bear case and a position-size ceiling. It covers Indian (NSE and BSE), US and Australian listings.
A Claude skill is a folder of instructions that Claude loads when your request matches it. This one is a SKILL.md file plus eight reference files, such as scoring tables, valuation methods and sector playbooks. Anthropic explains how skills work in its guide to using skills in Claude.
Under the hood, a normal company is valued with a ten-year owner-earnings DCF, and a bank with a price-to-adjusted-book model. You do not need to know either term to use it. The ones that matter are explained below.
How do you analyze a stock with it?
There are two ways to set it up: upload it to claude.ai, or drop it into Claude Code. Both take a few minutes.
Option 1: claude.ai, no terminal
- Download the skill. Get buffett-business-analyzer.zip. It holds one folder,
buffett-business-analyzer, withSKILL.mdand areferencesfolder. The same files are on GitHub. - Upload it. In claude.ai open Customize, then Skills. Click the plus button, choose Create skill, then Upload a skill, and pick the ZIP. Menu names change, so check Anthropic's skills guide if yours looks different.
- Switch on what it needs. Skills need code execution (Settings, then Capabilities). Keep web search on as well, because the skill tells Claude to pull live numbers and never to score from memory.
- Ask for a company. Type something like the prompts below.
Option 2: Claude Code
git clone https://github.com/gkv856/buffett-ai-stock-skills.git
cd buffett-ai-stock-skills
./install.shThe script links every skill in the repo (this one and a stock screener) into ~/.claude/skills. Restart Claude Code afterward. To do it by hand, copy the whole buffett-business-analyzer folder there. Copy the folder, not just SKILL.md: the skill loads files from references/ as it works and does worse without them. You can also call it by name: /buffett-business-analyzer NSE: DIVISLAB.
What do you type?
The skill responds to plain requests. These work:
Analyse NSE: COROMANDEL
Should I buy Asian Paints at current prices?
What's the margin of safety on SBI?
Run HDFC Bank through the framework
Score Microsoft with the Buffett business analyzerThree habits make the output more useful:
- Name the exchange for non-US stocks. "NSE: COROMANDEL" removes any doubt about which company you mean.
- Push back. Ask "but it is down 40% from its high" or "what would change the verdict?" and the skill re-runs the logic rather than repeating itself.
- Run it again later. A score is true for one date and one price. When the price moves, the answer can move.
What do you get back?
The skill builds a one-page scorecard and gives you a short written summary in the chat. These are the parts:
| Part of the scorecard | What it tells you |
|---|---|
| Verdict and score | A number out of 100 and a band: buy, watchlist, speculative or avoid |
| Hard-gate results | Whether the company can be analyzed at all |
| Stage-by-stage scores | Where the business is strong or weak, with the metric behind each score |
| Reverse DCF | The growth the current price assumes, next to what the company has delivered |
| Bear case | What you lose if you are wrong, and whether the loss is permanent |
| Watch price | The price at which the stock becomes interesting |
| Position-size ceiling | The most the framework would put in one stock |
| Sector risks | At least three risks the scoring does not capture |
How does the 100-point score work?
Four stages share the 100 points. Penalties can take away up to 25, and a catalyst bonus can add up to 10.
| Stage | Points | What it measures |
|---|---|---|
| Business quality | 30 | Earnings predictability, gross margin and its trend, pricing power, owner earnings yield |
| Capital efficiency | 25 | Return on invested capital, return on retained earnings, record on capital allocation |
| Balance sheet | 15 | Debt to equity against the sector, interest coverage (banks use capital adequacy and bad-loan ratios) |
| Valuation | 30 | Margin of safety from a conservative ten-year DCF, and today's multiple against the company's own history |
The pass mark is 70 after penalties and before any bonus. The verdict depends on the band:
| Score | Verdict | What it means |
|---|---|---|
| 70 or more | Buy | Only if the margin of safety is at least 25% |
| 60 to 69 | Watchlist | A good business; wait for a better price |
| 50 to 59 | Speculative | A significant, specific risk; only for informed conviction |
| Under 50 | Avoid | The evidence does not support the price |
Valuation gets 30 points because a great business bought at too high a price is still a bad purchase. In the skill's own rules, a company that scores 70 out of 70 on quality but has no margin of safety is still a fail.
What does the skill check before it scores anything?
Four hard gates come first. If any one fails, the skill stops, shows only the gate results and gives no score.
- The Graham investment test. Benjamin Graham defined an investment as an operation that, after thorough analysis, promises safety of principal and an adequate return. The skill checks all three parts: enough verified data, a concrete floor under the price, and an expected return that beats the risk-free rate. If one fails, the verdict is SPECULATION and scoring stops.
- Circle of competence. Claude must be able to explain the business model, the unit economics and the main competitive threat in plain terms.
- Management integrity. Claude looks for restatements, regulator enforcement, guidance manipulation and self-dealing. One confirmed intentional misrepresentation fails the gate.
- Moat identification. Claude has to name one specific moat (brand pricing power, switching costs, network effects, a cost advantage, a regulatory licence or a sovereign backstop), explain how it stops competitors copying it, and say how long it should last. "Market leader" does not count.
The Graham gate exists to close a specific hole. Without it, a company with missing financials could still end up with a number built from guesses, and that looks like analysis without being analysis. If the data is not there, the honest output is "this cannot be analyzed".
How do you read the result?
Five terms carry most of a scorecard.
Intrinsic value: the skill's estimate of what the business is worth, built from its owner earnings (net income plus depreciation, minus the capital spending needed to stay in business). It uses a conservative case, so the number is meant to be a floor and not a hope.
Margin of safety: the gap between the price you pay and that estimate of value. Benjamin Graham's idea, and one Buffett repeats. In his 2008 letter to Berkshire Hathaway shareholders he wrote that Graham taught him "Price is what you pay; value is what you get." The skill asks for a 25% margin on quality businesses and 40% or more on speculative ones.
Reverse DCF: instead of asking what a stock is worth, it asks what growth today's price assumes. Claude solves for the yearly growth rate that justifies the price, then sets it beside what the company has actually delivered over five and ten years. A price that needs 25% growth from a company that has grown 8% is an expensive story, whatever the chart looks like.
Bear case: the same valuation re-run with lower earnings (85% of the base), growth cut to 3% and then 0%, an 8 times exit multiple and a discount rate one point higher. The skill then separates volatility (the price falls and later recovers) from permanent loss (the business is worth structurally less than you paid), because only the second one ruins a long-term owner.
Watch price and position size: the watch price is the conservative value multiplied by one minus the required margin of safety. The position ceiling is the lower of two numbers: a cap set by the score (from 10% of a portfolio at 85 or more, down to 2% for speculative stocks) and a cap set by the downside to the bear-case value. It is a ceiling, not a suggestion to buy that much, and it cannot see the rest of your portfolio.
What does a real run look like?
To show the output, I ran Netflix through the skill on 20 July 2026, when the stock was 44% below its 52-week high. Treat it as a demo of what comes back, not a view on Netflix. The date matters: the price has moved since, and every figure below is what the scorecard reported that day.
| Stage | Score | Out of | What drove it |
|---|---|---|---|
| Business quality | 18 | 30 | Gross margin up more than 7 points in three years, but an owner earnings yield of 3 to 4% against a 4.55% ten-year Treasury |
| Capital efficiency | 16 | 25 | Return on invested capital of 34.7%, offset by buybacks of about 89% of free cash flow at high multiples |
| Balance sheet | 13 | 15 | Debt to equity of 0.47 and interest coverage of 16.9 times |
| Valuation | 8 | 30 | Conservative value of $49.01 against a price of $67.57 |
| Penalties | 0 | up to -25 | None of the six triggers fired |
| Total (before any bonus) | 55 | 100 | Speculative |
The scorecard also noted a +5 catalyst bonus for Netflix's ad tier, but it is a memo line only: the bonus cannot lift a score under the pass mark, and with it the total would be 60. The shape of the table is the point. A business can be strong on quality and still score low overall, because the price asks too much of it. That is the pattern the skill is built to surface.
The reverse DCF shows the other half of the story. At $67.57 the scorecard said the market was still pricing in roughly 16 to 17% yearly owner-earnings growth for five years, then about 10% for five more, while viewing hours were growing 2%.
Then the watch price. For a speculative score of 50 to 59, the skill asks for a margin of safety of at least 40%, so the entry price on the $49.01 value was $29.41. The scorecard also lists $36.76 (a 25% margin), but only for the case where the score first improves to 70 or more.
I also pushed back, because the stock was down 40% from its high. Claude ran the reverse DCF at the high as well. At $121.59 the price had assumed 26.1% growth for five years, then 15.7%, against 16.1% and 9.6% at $67.57. A falling price on its own was not a valuation.
What are the limits of using Claude to analyze stocks?
- Claude can make mistakes. The numbers come from web searches and arithmetic done by a language model. Verify anything you rely on in the company's own filings, such as SEC EDGAR for US companies or the exchange filings for NSE, BSE and ASX stocks.
- It looks at one stock at a time. It has no view of your portfolio, so the sector-concentration check is yours to make.
- It is biased toward saying no, deliberately. A missed opportunity costs you a return you never had. A permanent loss costs capital you did have, plus everything it would have compounded into. The two errors are not equal, so the framework leans toward rejection and will pass on many good companies.
- A score is a snapshot. It reflects one date and one price. It is not a forecast.
- My example used an earlier version. The Netflix run used version 2 of the framework. The current skill is version 3, which adds the Graham test, a mandatory reverse DCF and bear case, mandatory position sizing, and a check that a scorecard's arithmetic closes. The Netflix scorecard therefore has no Graham test and no position-size line.
A mistake I caught in my own example
The scorecard's closing line says Netflix scores "62/70" on the three quality stages. Its own stage scores are 18 of 30, 16 of 25 and 13 of 15, which add up to 47 of 70. The total of 55 is right and the 62 is a slip in the summary text. It is a small error, and exactly the kind you only find by adding the numbers up yourself. Always check that a scorecard's arithmetic closes.
Frequently asked questions
Can Claude analyze stocks?
Yes, if it has current data and a clear method. Claude can search the web, read filings and do the math, but it can also get numbers wrong. Give it a fixed framework like this skill, make it show its working, and check the key figures yourself.
Is the Buffett Business Analyzer free?
Yes. It is open source under the MIT license on GitHub. You need a Claude account that supports skills, with code execution and web search switched on.
Does it work for Indian stocks?
Yes. It covers NSE and BSE listings as well as US and Australian ones. It includes an India market reference file, and for banks and NBFCs it swaps debt-to-equity for capital adequacy and bad-loan ratios.
What is a good score?
A score of 70 or more after penalties, with at least a 25% margin of safety, gives a buy verdict. A score from 60 to 69 is a watchlist: a good business at a price that is still too high. Below 50 is an avoid.
What is a margin of safety?
It is the discount between the price you pay and a conservative estimate of what the business is worth. A bigger margin means more room for your estimate to be wrong. The skill requires 25% or more for quality businesses and 40% or more for speculative ones.
What is a reverse DCF?
A reverse DCF works backward from today's price to the growth the market is assuming. You then compare that growth with what the company has actually delivered. If the price needs far more than the company has ever managed, the stock is expensive.
Does it tell me how much to buy?
It gives a position-size ceiling: the lower of a cap based on the score and a cap based on the downside to the bear case. It is a limit for one position, and it does not know the rest of your portfolio.
Can I use it with ChatGPT or Gemini?
The repository README includes a copy-paste setup prompt for other assistants. I built and use it with Claude, so that is where I can vouch for it.
Is this financial advice?
No. It is an educational framework that produces an AI-generated analysis. It does not know your goals, your risk tolerance or your other holdings, so check everything and decide for yourself.
Get the skill
Download buffett-business-analyzer.zip or take it from GitHub. Then pick a company you already understand, run it, and compare the verdict with your own instinct. The most useful part is usually the reverse DCF, because it shows what the price is asking you to believe.
Comments