Financial Data Acquisition & Cross-Validation
Multi-market financial data acquisition and cross-validation spec ensuring each key metric comes from two independent sources with >1% error flagged.
The skill mandates two-source cross-validation and clear labeling of data sources and discrepancies, enhancing data transparency; however, no user confirmation is required, least privilege or sandbox isolation is not declared, and tools like twstock_data.py may auto-access external APIs and handle sensitive tokens, posing privacy risks. Deducted for lack of user confirmation, permission management, and full data-flow disclosure from max 25.
The skill's instructions are coherent, but key dependencies (external websites and APIs) are unverified, and tests only cover tool functions, not the skill itself, so reproducibility cannot be confirmed statically. Per static calibration cap, scored 1 (some instructional consistency but no execution evidence).
Audience (value investors) and scenarios are clear, with explicit data source priorities and boundaries (e.g., treatment of unlisted companies); however, non-fit scenarios are not defined, and external services like macrotrends and stockanalysis may be inaccessible from mainland China, impacting environment fit. Scored 12 due to partially clear boundaries and triggers.
MIT license present, repository has README, security policy, and evidence of changelog via test comments, but the skill itself lacks versioning, changelog, and FAQ; maintenance responsibility is unclear (no maintainer specified). Scored 10 due to clear information architecture with gaps.
The skill describes a complete workflow and cross-validation methodology, but no example outputs or real-case evidence show task completion, and reliance on external services may render it non-functional. Scored 5 per static calibration cap, as the core idea is plausible but unverified.
Repository includes test suites and popularity signals (Trendshift badge), but tests cover only tool functions, not the skill workflow, and there is no independent execution record, so claimed reliability cannot be cross-corroborated. Scored 1 per static calibration cap, as no reproducible third-party evidence exists.
- Several external data sources (macrotrends, stockanalysis, aastocks, eastmoney, etc.) may be inaccessible from mainland China due to network restrictions; consider providing alternatives or verifying reachability.
- While the FINMIND token is required to be stored locally, the skill does not explicitly enforce sandboxing or least privilege; ensure it is not run in untrusted environments.
- The skill lacks user confirmation mechanisms and may automatically perform external calls and data processing; evaluate its impact on your environment.
What does this skill do, and when should you use it?
Part of the AI Berkshire investment research framework, this skill defines a rigorous financial data acquisition and cross-validation process. It specifies primary and backup data sources for US, HK, A-share, and Taiwan stocks, and requires each key financial metric to be pulled from two independent sources, flagging discrepancies over 1% and mandating a check of original filings for discrepancies over 5%. It includes a Taiwan stock data tool (FinMind API) and financial rigor utilities, plus rules for historical price adjustment (adjusted prices) to avoid distortion in trend analysis.
The skill executes the following: 1) Selects primary/backup data sources based on target market (e.g., macrotrends, stockanalysis, aastocks, Eastmoney, FinMind API); 2) Fetches each financial metric from two independent sources; 3) Calculates error rate and grades it (≤1% consistent, 1-5% discrepancy, >5% major discrepancy requiring original filing check); 4) Uses tools/financial_rigor.py for market cap verification, valuation verification, cross-validation, three-scenario valuation, and Benford's law detection; 5) For Taiwan stocks, calls tools/twstock_data.py to fetch quotes, valuations, financials, monthly revenue, dividends; 6) States data cutoff date in report header and labels all sources and errors.
- An individual investor analyzing US-listed companies like PDD or Tencent ADR can invoke this skill to cross-verify data from macrotrends and stockanalysis.
- A researcher covering HK-listed companies (e.g., Tencent 0700, NetEase 9999) can contrast aastocks with macrotrends (TCEHY/NTES) and verify against HKEX original annual reports.
- A-share investors analyzing stocks like 37 Interactive Entertainment or G-bits can pull data from Eastmoney and CNinfo and cross-validate.
- Taiwan stock analysis (e.g., TSMC, MediaTek) uses the FinMind tool for monthly revenue and dividend data, cross-checked with Goodinfo or ADR (TSM).
- Any analysis involving historical prices, valuation percentiles, or long-term returns must follow the adjusted-price rules to ensure consistency.
What are this skill's strengths and limitations?
- Clear priority tables for data sources across US, HK, A-share, and Taiwan markets, highly practical.
- Enforces cross-validation and error grading, significantly reducing data error risk.
- Includes a Taiwan stock data script that directly fetches high-frequency monthly revenue, a unique advantage.
- Provides financial rigor tools for market cap verification, three-scenario valuation, etc.
- Detailed rules for adjusted prices in historical analysis avoid common pitfalls.
- Relies on external websites (macrotrends, eastmoney, aastocks) for availability and stability; no fallback strategy for automated fetching failures is mentioned.
- No test suite or automated test evidence; cross-validation relies on manual confirmation.
- FinMind API has hourly rate limits; unregistered use may affect large-scale analysis.
- Does not cover all markets (e.g., Europe, Japan), only four specific ones.
How do you install this skill?
The skill is part of the ai-berkshire repository at codex-skills/financial-data/SKILL.md. Install the entire repo, then copy the codex-skills/financial-data folder to ~/.codex/skills (Codex) or the appropriate Claude Code commands directory. Commands: git clone https://github.com/xbtlin/ai-berkshire.git; cd ai-berkshire; ./scripts/install-codex-skills.sh (macOS/Linux) or .\scripts\install-codex-skills.bat (Windows).
How do you use this skill?
After restarting Codex, instruct it with a prompt like 'Use financial-data to get financial data for Tencent.' The skill will first run the 'date' command to confirm today, then select data sources based on the market, fetch and cross-verify each metric. For Taiwan stocks, you can manually run python3 tools/twstock_data.py quote 2330 or describe the task to the agent.