magic-data-transformation

Warn

Audited by Gen Agent Trust Hub on Jul 3, 2026

Risk Level: MEDIUMCOMMAND_EXECUTION
Full Analysis
  • [COMMAND_EXECUTION]: The script scripts/derive_columns.py uses the Python eval() function to process calculated column expressions. While it attempts to implement a 'safe' namespace and blocks certain keywords like 'import' and 'exec', eval() can often be bypassed via Python introspection to achieve arbitrary code execution.
  • [SAFE]: The skill implements mandatory 'PAUSE gates' for all database writes in scripts/deliver_to_db.py and HuggingFace Hub uploads in scripts/deliver_to_hf.py, ensuring human-in-the-loop confirmation for sensitive operations.
  • [SAFE]: The HuggingFace delivery script includes a built-in credential scanner that checks files for common secret patterns (e.g., HF tokens, AWS keys) before they are uploaded to the Hub.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jul 3, 2026, 07:00 AM
Security Audit — agent-trust-hub — magic-data-transformation