deidentify

Pass

Audited by Gen Agent Trust Hub on Nov 1, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill operates as a local-only tool. The included deidentify.py script does not utilize any network libraries and is designed to run in a standalone environment without AI or network calls.
  • [SAFE]: The skill implements strict data isolation. Instructions in SKILL.md and skill.yml explicitly forbid the AI agent from reading or processing raw Protected Health Information (PHI). The agent is only permitted to read metadata reports and de-identified outputs after human verification.
  • [SAFE]: Cryptographic best practices are followed. The script uses secrets.SystemRandom for non-deterministic date shifting and HMAC-SHA256 with per-run keys for audit logging. This prevents brute-force re-identification of sensitive values (like dates) from the logs.
  • [SAFE]: The skill enforces secure file management. Sensitive mapping files containing re-identification data are automatically restricted to owner-only permissions (0600) via os.chmod.
  • [SAFE]: The skill includes robust verification mechanisms. The data_fingerprint function uses salted PBKDF2 to ensure the input data has not been modified between the scan and apply phases, preventing race conditions or data injection attacks.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 1, 2026, 03:18 PM