atai-anomaly-discovery-agent
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill and its associated scripts follow secure coding practices. No malicious command execution, data exfiltration to unauthorized domains, or obfuscation techniques were identified.
- [EXTERNAL_DOWNLOADS]: The skill utilizes the 'archetypeai' Python package, which is the official client for the vendor's platform and is considered a trusted resource.
- [INDIRECT_PROMPT_INJECTION]: The skill processes numerical sensor data. While this creates an ingestion point for external content, the risk is negligible due to the data's non-textual nature.
- [INDIRECT_PROMPT_INJECTION] Ingestion points: Sensor data CSV files (e.g., 'bearing_eval.csv') in 'references/run_ad_agent.py'.
- [INDIRECT_PROMPT_INJECTION] Boundary markers: Delimiters are not explicitly defined in the ingestion pipeline.
- [INDIRECT_PROMPT_INJECTION] Capability inventory: The skill can upload/download files and run remote agents via the Archetype AI Agents API.
- [INDIRECT_PROMPT_INJECTION] Sanitization: Content is parsed as CSV data without specific sanitization for instruction-like patterns.
Audit Metadata