bio-clinical-biostatistics-cdisc-data
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is architected to ingest and process external data files (e.g., DM.csv, AE.csv, and .xpt domains) which serves as a vector for indirect prompt injection. A malicious actor could embed instructions within clinical datasets to influence the agent's behavior during data derivation or analysis.
- Ingestion points: Data ingestion occurs in
SKILL.mdandexamples/cdisc_data_preparation.pyviapd.read_csv(),pd.read_sas(), andpyreadstat.read_xport(). - Boundary markers: There are no explicit delimiters or system instructions provided to the agent to distinguish between data and potential commands within the files.
- Capability inventory: The skill demonstrates capabilities for local file system operations, including reading multiple domain files and writing the final analysis dataset to disk via
to_csv(). - Sanitization: The provided code and instructions do not include mechanisms to sanitize, validate, or escape the content of the external files to prevent the agent from interpreting data values as instructions.
Audit Metadata