scikit-learn
Pass
Audited by Gen Agent Trust Hub on Oct 12, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill instructions demonstrate processing external data sources, such as loading data via
pd.read_csv('data.csv'). This creates a potential attack surface where an agent processing this data could be influenced by malicious instructions embedded within the dataset. - Ingestion points: The skill examples include loading tabular data from CSV files and generic data variables (
SKILL.md). - Boundary markers: There are no explicit instructions or delimiters provided to ensure the agent ignores natural language instructions that might be present within the input data.
- Capability inventory: The skill utilizes standard Python libraries for machine learning tasks. No evidence of unauthorized network exfiltration, sensitive file access, or arbitrary command execution was found in the analyzed instructions.
- Sanitization: No specific data sanitization, schema validation, or input filtering techniques are mentioned or required for the data ingestion steps.
Audit Metadata