pai-load-policy
Pass
Audited by Gen Agent Trust Hub on Oct 29, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill functions as a documentation guide for the Physical AI Runtime library. No malicious code, command execution, or obfuscation patterns were detected.\n- [EXTERNAL_DOWNLOADS]: The skill describes loading models from remote Hub snapshots using the InferenceModel.from_pretrained method. These downloads originate from the openvinotoolkit organization, a well-known provider of AI inference technologies.\n- [INDIRECT_PROMPT_INJECTION]: The skill involves processing model artifacts which constitutes a potential attack surface.\n
- Ingestion points: Reads manifest.json, model files, and Hub snapshots (SKILL.md).\n
- Boundary markers: Explicit boundary markers for untrusted content are absent in the prompt interpolation.\n
- Capability inventory: InferenceModel performs local file system reads and remote network downloads for model artifacts (SKILL.md).\n
- Sanitization: The skill recommends validating artifacts against the manifest-schema.md contract, which provides structural integrity for the ingested data.
Audit Metadata