financial-computing-python

Pass

Audited by Gen Agent Trust Hub on Sep 3, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes external CSV data using a Python diagnostic script (scripts/financial_computing_python_diagnostics.py) and returns a JSON summary to the agent context. This creates a potential attack surface where malicious instructions embedded in data fields could influence the agent's subsequent reasoning.
  • Ingestion points: scripts/financial_computing_python_diagnostics.py reads user-provided CSV files via pd.read_csv(arguments.input_csv).
  • Boundary markers: The output is formatted as a structured JSON object (payload = json.dumps(...)), which provides some delimiter-based isolation, but no explicit 'ignore embedded instructions' warnings are present in the prompt instructions.
  • Capability inventory: The skill possesses file-write capabilities (open(arguments.output, 'w').write()) and relies on standard data science libraries.
  • Sanitization: The script performs type conversion (pd.to_numeric, pd.to_datetime) and drops invalid timestamps, but does not sanitize the content of the instrument or other non-numeric columns before they are included in the columns list of the JSON output.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 3, 2026, 08:46 AM
Security Audit — agent-trust-hub — financial-computing-python