The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external financial data from sources including Sina Finance, EastMoney, Yahoo Finance, and CNINFO. This establishes a theoretical attack surface for indirect prompt injection via malicious content in financial filings or announcements.
Ingestion points: Financial reports, income statements, balance sheets, cash flow statements, and corporate announcements fetched via tools such as sina_income_statement, em_global_income, yahoo_financials, and cninfo_announcements.
Boundary markers: The instructions do not explicitly specify delimiters for external content, but they enforce a highly structured output format (Facts Table, Derived Table, Inference Section) which provides structural isolation.
Capability inventory: The agent is restricted to data retrieval tools and specialized calculation functions (quarterize, ttm_sum, ratio, etc.). It is explicitly instructed not to perform mental math or manual file writing, instead using a controlled --run-dir for logging calculations.
Sanitization: No specific sanitization methods for external text are mentioned, although the focus is primarily on numerical extraction and structured data processing.
[SAFE]: The skill follows security best practices by discouraging the agent from performing manual arithmetic or file system operations, instead routing all logic through specific calculation tools (calc) and designated directories.
[SAFE]: All external data sources identified (Sina, EastMoney, Yahoo Finance, CNINFO) are well-known and standard providers in the financial industry.
[SAFE]: The skill explicitly forbids the agent from providing investment advice, reducing the risk of being used for unauthorized financial guidance.

earnings-analysis