magic-data-loading

Pass

Audited by Gen Agent Trust Hub on Jul 3, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: Database connections are managed securely through environment variables and utilize dialect-specific read-only enforcement (such as PRAGMA query_only for SQLite and session read-only modes for PostgreSQL/MySQL) to prevent accidental data modification.
  • [SAFE]: SQL query execution in extract_data.py uses parameterized queries via SQLAlchemy and pandas, effectively mitigating SQL injection risks. The script also automatically injects LIMIT clauses to prevent resource exhaustion from large data fetches.
  • [SAFE]: External dataset downloads are restricted to HuggingFace Hub, a recognized and well-known service, using official client libraries.
  • [SAFE]: The generate_dataset_card.py tool includes proactive credential scrubbing using regex patterns to identify and redact API keys and tokens from metadata before documentation is generated.
  • [SAFE]: The use of subprocess.run is limited to the test suite (tests/test_text_parser.py) for validating the skill's own CLI tools and does not present a command execution vulnerability in production scripts.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 3, 2026, 07:00 AM
Security Audit — agent-trust-hub — magic-data-loading