adverse-event-narrative
Pass
Audited by Gen Agent Trust Hub on Sep 23, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied adverse event data in JSON format, creating a surface where malicious instructions embedded in clinical descriptions could attempt to influence the agent's output.
- Ingestion points: Adverse event case data is read from JSON files by
scripts/main.py. - Boundary markers: The skill does not employ explicit delimiters or instructions to ignore embedded commands within the medical data fields.
- Capability inventory: The
scripts/main.pytool possesses file-write capabilities via the--outputcommand-line argument. - Sanitization: Input is parsed as structured JSON, though the resulting narrative is generated through direct string interpolation of the field values.
- [COMMAND_EXECUTION]: The skill provides standard command-line instructions for script validation and help usage. These are restricted to local operations and are appropriate for a developer-oriented medical tool.
- Evidence:
python -m py_compile scripts/main.pyis used for syntax verification, andpython scripts/main.py --helpis used for CLI documentation access.
Audit Metadata