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
Risk Level
SAFE
Analyzed
Jul 15, 2026, 03:25 AM
Security Audit — agent-trust-hub — scientific-data-preprocessing