tail-risk-hedging
Pass
Audited by Gen Agent Trust Hub on Sep 3, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from CSV files, creating a surface for potential indirect prompt injection attacks.\n
- Ingestion points: The Python script
scripts/tail_risk_hedging_diagnostics.pyreads user-provided CSV files.\n - Boundary markers: There are no explicit markers or instructions to isolate the diagnostic output when it is returned to the agent's context.\n
- Capability inventory: The skill is capable of reading files and writing JSON artifacts to the local file system but lacks network connectivity or arbitrary shell command execution.\n
- Sanitization: While numeric values are coerced for calculation, the CSV column names are extracted and included directly in the output summary without sanitization, allowing malicious instructions in headers to enter the agent's context.
Audit Metadata