physicalai-train-training-a-policy

Pass

Audited by Gen Agent Trust Hub on Sep 6, 2026

Risk Level: SAFE
Full Analysis
  • [COMMAND_EXECUTION]: The skill instructs the agent to execute project-specific CLI commands such as physicalai fit, physicalai validate, and pytest. These are standard tools for the defined ML training workflow and operate on local configurations.
  • [INDIRECT_PROMPT_INJECTION]: The skill describes an ingestion point for external datasets via the LeRobotDataModule using a repo_id. While this involves fetching data from remote repositories (such as the Hugging Face ecosystem), it is a standard machine learning practice and the skill does not expose dangerous capabilities to the content of those datasets.
  • [SAFE]: The skill adheres to legitimate development practices, using standard library imports and well-known ML frameworks. No obfuscation, credential exposure, or unauthorized persistence mechanisms were found.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 6, 2026, 01:43 PM
Security Audit — agent-trust-hub — physicalai-train-training-a-policy