scientific-data-preprocessing
Pass
Audited by Gen Agent Trust Hub on Jul 15, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill defines functions that process external data and print column names or statistics directly to the agent context, creating a potential surface for indirect prompt injection.
- Ingestion points: Data frames processed in the data_quality_report and preprocessing_pipeline functions in SKILL.md.
- Boundary markers: No delimiters or explicit instructions are provided to the agent to ignore instructions embedded within the data.
- Capability inventory: The pipeline performs file writing (to_csv) to the local results directory and outputs text to the agent's active context.
- Sanitization: Input data, such as column names or categorical values, is not sanitized or escaped before being displayed to the agent or saved.
Audit Metadata