atai-anomaly-discovery-agent-data-prep
Pass
Audited by Gen Agent Trust Hub on Nov 1, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill implements local data transformation logic for sensor recordings. It uses standard data science libraries (pandas, numpy, scipy) to resample data onto a uniform grid and validate its suitability for anomaly detection fitting. No suspicious instructions or override attempts were found.
- [SAFE]: All dependencies listed in requirements.txt are well-known, versioned packages from the official Python Package Index (PyPI). The skill does not perform any dynamic package installation or remote script execution.
- [SAFE]: No network operations or credential-handling patterns were detected. The scripts operate exclusively on local files provided via command-line arguments.
- [INDIRECT_PROMPT_INJECTION]: The skill processes external sensor data.
- Ingestion points: references/ad_common.py (read_recording), references/prepare.py (prepare_one).
- Boundary markers: Data is forced into a structured grid with fixed columns.
- Capability inventory: Local file read/write (pandas/pyarrow) and process parallelism (ProcessPoolExecutor).
- Sanitization: Inputs are checked for numeric finiteness and specific label strings (normal/anomaly). The risk is minimal as the output is structured data for a downstream agent.
Audit Metadata