explore-ml-data

Pass

Audited by Gen Agent Trust Hub on Jun 13, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes a Python script (data/eda.py) using a shared runner located at .agents/skills/audit-ml-pipeline/scripts/run_cells.py. This relies on an external script for core execution logic.
  • [REMOTE_CODE_EXECUTION]: The skill generates and executes Python code at runtime. It populates data/eda.py from a template, substituting user-defined paths and data-loading logic, and executes it to generate analysis artifacts. This behavior is restricted to the generated script and intended for its primary function.
  • [EXTERNAL_DOWNLOADS]: The skill may trigger the installation of the ipython package via the python-env-manager skill if it is not present in the environment, which is required for the execution runner to function.
  • [PROMPT_INJECTION]: The skill has an indirect prompt injection surface as it ingests untrusted user data files and uses the analysis results to generate a narrative report. Evidence chain: 1. Ingestion points: Raw data files located via user-provided paths or workspace scanning. 2. Boundary markers: No explicit boundary markers for raw data content; execution output is parsed via JSON. 3. Capability inventory: File-write (project deliverables), file-read, and shell execution (bash/pixi). 4. Sanitization: No explicit sanitization of data values before processing through skrub's profiling tools.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 13, 2026, 03:18 PM
Security Audit — agent-trust-hub — explore-ml-data