Skip to content
Get packs

Guide · Agent skill

How to use the Financial Data Cross-Check agent skill

Updated · written with AI from the project's own documentation and checked against our catalogue data

The financial-data skill is a set of written rules that tells an AI agent how to collect company financial figures such as revenue, net profit, margins and cash flow. Instead of accepting one website's number, the agent fetches each key figure from two independent sources, compares them, and labels the result as consistent, showing a discrepancy, or needing a check against the original filing. This guide explains what the skill does, how to install it, how to trigger it, and what it needs to run.

Financial Data Cross-Checkxbtlin/ai-berkshire · MIT

Quick start

  1. 1Download and unzip. Download the ZIP from this page and unzip it to get the financial-data folder containing SKILL.md.
  2. 2Place the skill. Move the folder to ~/.claude/skills/ for personal use, or to .claude/skills/ in your project. Alternatively, upload the ZIP under Settings → Capabilities → Skills in the Claude apps.
  3. 3Check the tools. Make sure tools/financial_rigor.py and tools/twstock_data.py exist in your repository, since the skill references them but they are not in the reviewed file. Confirm Python 3 is installed.
  4. 4Optional token. If you use Taiwan-stock data at scale, set FINMIND_TOKEN or save the token in local/finmind_token.txt, which stays out of git.
  5. 5Ask a specific question. Name the company, ticker, metric and period, for example TSMC's 2330 revenue for last year.
  6. 6Review the report. Check the data cutoff date and each figure's consistency label: consistent, showing a discrepancy, or major discrepancy.

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

An agent skill is a folder containing a SKILL.md file. The file gives an AI agent a specific workflow it can follow when a request matches the skill's description. This skill is named financial-data and is described as the AI Berkshire skill for financial data acquisition and cross-verification.

The workflow requires that every key figure come from two independent sites. The agent calculates the error rate between them and marks each figure as consistent (error of 1% or less), showing a discrepancy (1% to 5%, with both values and a likely cause), or a major discrepancy (over 5%, which must be checked against the original filing before use). Figures for private companies, which have only one source, are marked as estimates.

  • Checks the current date first and states the data cutoff date in the report header.
  • Lists source priority by market: US stocks, Hong Kong, A-shares and Taiwan-listed companies.
  • Explains common reasons numbers differ, such as GAAP vs non-GAAP, exchange rates, fiscal year definitions and consolidation scope.
  • Sets rules for stock price adjustment: forward-adjusted prices for historical comparisons, and no mixing of adjusted and unadjusted data in one analysis.
  • For Taiwan-listed stocks, points the agent to a bundled FinMind script (tools/twstock_data.py) for quotes, valuation, financials, monthly revenue and dividends.

Install the skill from the ZIP file

Download the ZIP from this page and unzip it. You can install the skill for yourself across all projects, or for one project only.

For a personal install, place the unzipped folder in ~/.claude/skills/. For a project install, place it in the project's .claude/skills/ folder. You can also upload the ZIP in the Claude apps under Settings → Capabilities → Skills.

  • Personal: ~/.claude/skills/ (available everywhere on your machine).
  • Project: .claude/skills/ inside your project folder (available only in that project).
  • Claude apps: Settings → Capabilities → Skills, then upload the ZIP.
~/.claude/skills/financial-data/SKILL.md

How do you trigger the skill with a request?

You do not need a special command. The skill is meant to be used when you ask for company financial figures that you want checked. Describe the company, the metric and the purpose in plain language. Naming the market or ticker helps the agent pick the right sources from the priority tables.

If you want the agent to use the checking workflow rather than a quick lookup, say so, for example by asking for cross-checked figures or a sourced report.

  • "Get Tencent's 2023 revenue and net profit, cross-checked across two sources, and tell me if they disagree."
  • "Pull TSMC's monthly revenue for the last year and check it against a second source."
  • "Write a short report on NetEase's margins and free cash flow, and state the data cutoff date."

What can you ask it to do? Examples by market

The source gives specific source lists for each market. For US-listed companies such as PDD, it names macrotrends and stockanalysis as the two main sites, with SEC EDGAR filings as the original reference. For Hong Kong stocks such as Tencent (0700), it names aastocks and macrotrends, and for A-shares it names Eastmoney and cninfo.com.cn.

For Taiwan-listed companies, the skill bundles a FinMind-based script. Its subcommands include quote, valuation, financials, revenue, dividend and search. The source notes that the figures are in New Taiwan dollars (TWD) and must be labelled when compared with other currencies.

  • US-listed: PDD, Tencent ADR, NetEase ADR, Nintendo, Capcom (macrotrends and stockanalysis, checked against SEC EDGAR).
  • Hong Kong: Tencent 0700, NetEase 9999, Meituan 3690 (aastocks and macrotrends ADR codes, checked against HKEX filings).
  • A-shares: 三七互娱 (002555), 吉比特 (603444) (Eastmoney and cninfo.com.cn).
  • Taiwan: TSMC 2330, MediaTek 2454 (FinMind script, Goodinfo, MOPS filings).
python3 tools/twstock_data.py financials 2330

What do you need before you start?

The skill itself is a text file, but its workflow relies on tools and internet access. The tools named in the source are tools/financial_rigor.py and tools/twstock_data.py. These files are referenced by the skill but are not included in the file reviewed, so you must have them in your repository before the scripts can run.

The source says the Taiwan script is dependency-free and needs Python 3. A FinMind API token is optional. It can be set as the FINMIND_TOKEN environment variable or saved in a local file, local/finmind_token.txt, which is git-ignored. Without a token, FinMind works but has lower rate limits.

  • Python 3 to run the bundled tools.
  • Internet access to the financial sites named in the workflow.
  • Optional: a FinMind API token, via the FINMIND_TOKEN environment variable or local/finmind_token.txt.
  • The skill is designed to run inside the AI Berkshire repository, where tools/ and AGENTS.md are present.

Safety notes: scripts and network access

This skill runs scripts and uses the network. The agent may execute the bundled Python tools and fetch pages from the financial sites named in the workflow. Read the scripts in tools/ before running them, especially if you did not write them yourself, because the reviewed SKILL.md file does not include them.

Keep your FinMind token out of reports, skills and git commits. The source says the token should be stored only on your own machine.

  • Review tools/financial_rigor.py and tools/twstock_data.py before first use.
  • Never commit the FinMind token; keep it in the environment variable or the git-ignored local file.
  • The network calls go to the sites named in the source, such as macrotrends, stockanalysis, aastocks, Eastmoney, cninfo, FinMind and Goodinfo.

Tips for getting reliable results

Use the skill for the figures that matter to your decision, not every number on a page. Check the data cutoff date in the report header, because some sites lag behind the latest filings. When a discrepancy appears, read the explanation column first: a gap between GAAP and non-GAAP profit is common and is not necessarily an error.

  • Prefer annual figures for cross-checks; the source notes quarterly data may lag on some sites.
  • When a number differs by more than 5%, check the original annual report or 10-K before using it.
  • For Taiwan stocks, use monthly revenue; the source describes it as the fastest public signal of changes in fundamentals.
  • For ADRs, remember the ratio: the source notes 1 TSM ADR equals 5 shares of 2330.

What people use it for

  • Comparing a US-listed Chinese internet company's revenue and net profit across macrotrends and stockanalysis before writing a report.
  • Tracking TSMC's monthly revenue trend using the FinMind-based script and checking it against Goodinfo.
  • Checking a Hong Kong company's margins across aastocks and HKEX filings to see whether the gap is a GAAP vs non-GAAP difference.
  • Building a repeatable research workflow in the AI Berkshire repository so each report follows the same verification rules.

Frequently asked questions

How do I install the financial-data agent skill?

Download the ZIP from this page and unzip it. Place the folder in ~/.claude/skills/ for personal use or in a project's .claude/skills/ folder, or upload the ZIP in the Claude apps under Settings → Capabilities → Skills.

What do I need to run the financial-data skill?

You need Python 3 for the bundled tools, internet access to the financial sites the workflow names, and optionally a FinMind API token for higher rate limits. The tools/ scripts it references are not in the reviewed file, so they must be present in your repository.

Is the financial-data skill free?

The source lists the license as MIT, which is a permissive open-source license. The page does not state any price, so check the license file in the repository for the exact terms.

Does the skill give investment advice?

No. The skill is a research process for gathering and checking figures. It is not financial advice, and the sources it names can be wrong or out of date, so independent review is needed before any investment decision.

Which markets and companies does it support?

It covers US-listed, Hong Kong, A-share and Taiwan-listed companies, with named sources for each market. For private companies or sources not on its list, it gives no method beyond marking figures as estimates.

Why do two sites show different revenue or profit numbers?

The most common reason is GAAP vs non-GAAP accounting. Other causes include exchange-rate timing, fiscal year definitions and whether minority shareholder interests are included. The skill asks the agent to explain the likely cause when figures differ by more than 1%.

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 an AI agent to gather company financials with a documented two-source check, date stamp and clear labels for discrepancies, and you are comfortable reviewing the scripts it runs. Pick something else if you need real-time trading data, figures for private companies with no public source, or a finished investment recommendation.

Download the skill