Jason Zou
04Stock screenerClaude skill · public repository · v5.0

The same dashboard for any stock or ETF

Screening stocks by hand is slow and inconsistent. This Claude skill, a folder of instructions and scripts, fills one fixed dashboard with six tabs for a stock and five for an ETF. I've run it on LEU, CRDO, PSTG, STLD, MTSI, XRX, CLSK and GE.

Output
One React dashboard (a Claude Artifact) with six tabs and all the data held in the file
Data
Yahoo Finance through yfinance by default, or LSEG Refinitiv, Alpha Vantage or another connected data source when one is set up. Web search supplies the qualitative context.
Versions
The public repository is v5.0 and defaults to yfinance. The version I run locally is v6.3, which uses LSEG Refinitiv where a licence is available and yfinance otherwise.
Repository
Stock-investment-screener-dashboard-for-Claude
Live example

The dashboard

The window runs the public repository's template on synthetic data. The site changes only the colours, two typefaces and four chart heights, and opens the revenue chart on its first slice. You can switch tabs, click value-chain stages, open metrics, pick catalysts and filter articles.

Synthetic data for a fictional company
dashboard-template.jsx XMPL · Example Corp (fictional) · synthetic data Template sourceSource

XMPL · six tabs · Overview, Analysis, MACD Strategy, Street Consensus, News & Sentiment, Export & Citations

About 780 KB of JavaScript, the template with its chart library and React. It loads on its own as this window comes into view.

Press Expand to see the window at full height. The firms, publishers and products in the data are all made up, named Example, Sample or Demo.

The six tabs

What the skill fills in

The template is the same for every run, and only the data changes. For each tab, the notes below say what the skill has to supply, as set out in the repository's instructions and reference files.

01 · Overview

The tab shows the price, six headline metrics, about 250 weekly bars with volume, and returns against the S&P 500 from 1M to 5Y. Then come the business and industry, revenue streams, regions, value chain, threats, opportunities and key drivers.

02 · Analysis

The tab charts drawdowns and lists any deeper than 10% (7% for broad ETFs). Moving averages and returns give a five-level momentum label, and golden or death crosses from the past year are flagged. Fundamentals are rated in five sections.

03 · MACD Strategy

The tab plots monthly candles, a 10-month moving average and the MACD. The stock gets one of six regime labels and a rule-based conviction score, which is not a forecast. Moves over 8% or two standard deviations mark catalyst months.

04 · Street Consensus

The tab shows the rating mix, a box plot of targets against the current price, and ten to fifteen analyst actions from the past 60 days, with theses and sources. ETFs have no consensus tab, so they get five tabs.

05 · News & Sentiment

The target is twenty to thirty articles, and fewer than eight is labelled limited coverage. Each article is scored on fundamental impact, market sentiment, urgency and confidence, weighted by source tier and recency. A keyword panel filters the article reader.

06 · Export & Citations

The tab lists the sources in a fixed order, starting with filings. A source with no reliable link is cited by name. Expandable tables of key metrics, drawdowns and analyst actions follow, then the data attribution and disclaimer.

04Definitions and checksstock-metric-definitions.md · SKILL.md step 5

Metric definitions

The skill's metric reference file fixes one definition for each ratio, so every run uses the same one. It also says when a ratio is not meaningful, and the dashboard then shows N/M instead of a negative multiple.

stock-metric-definitions.md · the variant the skill uses
P/E (TTM) = price ÷ trailing diluted EPS
Forward P/E = price ÷ consensus EPS for the next 12 months (three or more analysts)
PEG = P/E ÷ consensus long-term EPS growth
EV = market cap + total debt − cash
EV/EBITDA = EV ÷ trailing EBITDA
ROE = net income ÷ average equity
Debt/Equity = total liabilities ÷ equity (the skill's definition, closer to liabilities to equity)
Beta = 5-year monthly, against the S&P 500 total return
Avg volume = mean daily shares over the last 63 sessions
// N/M when a ratio is not meaningful: P/E if EPS ≤ 0, PEG if growth ≤ 0, P/B if book value < 0, EV/EBITDA if EBITDA < 0, ROE if average equity < 0
// returns use adjusted prices; the quote and the 52-week range are unadjusted
// no dividend shows as a dash, never as 0%

How the screener checks itself

  • Claude only supplies the data. The dashboard is a fixed template, and the data sits in constants at the top of the file. The skill fills those constants and leaves the rendering code alone, so the layout is the same for every ticker.
  • The indicators are computed in code. The moving averages, MACD (12, 26, 9), drawdowns and returns all come from the price history. The MACD's exponential averages start from a simple average. The template only draws what it is given.
  • Missing data is not filled in. Early months before the MACD is defined keep null values. Gaps are stated in words, such as "limited coverage". A source without a reliable link is cited by name, with no invented URL.
  • A ten-point checklist runs last. It confirms that every constant is filled and that about 250 weekly and 96 monthly points exist. Consensus counts must match analyst actions, every article and citation needs a link, and the rendering code must stay untouched.
  • Judgement calls are labelled. The reference files describe the sentiment weighting as "judgment, not formula". Cost-cutting news is marked Mixed, because deciding whether a cut is strategic or forced "requires human judgment".

The tool

  • Pulling the numbers from Yahoo Finance, LSEG Refinitiv, Alpha Vantage or another connected source, and the context from web search
  • Computing returns, drawdowns, momentum, the monthly MACD regime and the fundamentals
  • Running the checklist, showing the warnings and filling the fixed template

You

  • Reading the output and deciding whether the name deserves real work
  • Judging ambiguous news, such as whether a cost cut is strategic or forced (the tool scores it Mixed)
  • Forming any buy or sell view, which the screener does not give
04Notes

Notes

The fictional ticker

The screens of LEU, CRDO and the others used LSEG Refinitiv and yfinance data that cannot be republished, so this page shows the repository's template with synthetic figures for a company that does not exist. The numbers are not a view on any security.

Judgement and data

Prices, returns, drawdowns, moving averages, the MACD, the regime label and fundamentals come from code. Claude writes value-chain text, threats and opportunities, each metric's likely cause, sentiment scores and catalyst notes. The reference files flag judgement calls.

The synthetic data

A fixed-seed random generator makes 260 weekly bars ending at $50.00 on 18 September 2026. Monthly candles are built from those weeks, so the Overview and MACD tabs match. Company figures come from a small internal model, so ratios tie.