data-analysis

Pass

Audited by Gen Agent Trust Hub on Jun 23, 2026

Risk Level: SAFECOMMAND_EXECUTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill generates Python code files (analysis.py) and executes them using a system interpreter (defaulting to python3) to perform data computations. This is the primary intended behavior for the data analysis loop.\n- [COMMAND_EXECUTION]: The skill provides instructions for the agent to potentially install necessary Python dependencies such as pandas or numpy via uv pip install with user consent.\n- [SAFE]: The skill instructions explicitly restrict the agent to read-only access for the input dataset and require all generated files and logs to be stored within a restricted <sandbox_root> directory to prevent accidental system modifications.\n- [PROMPT_INJECTION]: The skill processes external data from a user-provided dataset, creating an indirect prompt injection surface. Evidence chain:\n
  • Ingestion points: The <dataset> binding in SKILL.md identifies an external file (CSV, Parquet, etc.) as the source of data.\n
  • Boundary markers: The skill instructions explicitly state the dataset is "read-only ground truth" and that the agent should not modify it.\n
  • Capability inventory: The skill possesses the ability to execute generated scripts via analysis_cmd and write to the filesystem within <sandbox_root>.\n
  • Sanitization: No explicit sanitization or filtering of the dataset's column names or content is described before using them in code generation.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 23, 2026, 04:42 PM
Security Audit — agent-trust-hub — data-analysis