physicalai-train-working-with-datasets

Pass

Audited by Gen Agent Trust Hub on Jul 6, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill fetches machine learning datasets from HuggingFace, a well-known service, using the repo_id parameter (e.g., lerobot/pusht) within the LeRobotDataModule configuration. These downloads are standard for the skill's documented purpose of training AI policies.
  • [COMMAND_EXECUTION]: The skill uses the physicalai CLI and uv package runner to execute training tasks and run unit tests. These tools are project-specific utilities intended for development and training workflows.
  • [INDIRECT_PROMPT_INJECTION]: The skill defines a surface for ingesting external data via dataset repositories. While datasets are untrusted external inputs, the skill's primary function is to facilitate this data flow for training, and no exploitable capabilities (like unauthorized file writes or exfiltration) are exposed through this ingestion.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 6, 2026, 07:19 PM
Security Audit — agent-trust-hub — physicalai-train-working-with-datasets