statistical-arbitrage
Pass
Audited by Gen Agent Trust Hub on Sep 3, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill includes an ingestion surface for untrusted external data via CSV files. \n
- Ingestion points: External data is read from
input.csvinto a pandas DataFrame inscripts/statistical_arbitrage_diagnostics.py. \n - Boundary markers: None defined. \n
- Capability inventory: Local file reading, writing of JSON results to a user-specified path, and standard console output. \n
- Sanitization: The script performs strict schema validation and coerces data into numeric and datetime types using
pd.to_numericandpd.to_datetimewitherrors='coerce', which effectively neutralizes non-numeric injection payloads before they are processed for statistical calculations. \n- [COMMAND_EXECUTION]: The skill instructions include the execution of a local Python script (scripts/statistical_arbitrage_diagnostics.py). This script is a deterministic utility for computing financial metrics and operates entirely within the local environment without network access.
Audit Metadata