tao-train-pointpillars
Pass
Audited by Gen Agent Trust Hub on Aug 25, 2026
Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [PROMPT_INJECTION]: The skill exhibits a potential surface for indirect prompt injection where malicious instructions could be embedded in the data processed by the agent.
- Ingestion points: The skill ingests external LiDAR data and converted dataset info via paths specified in dataset.data_path and dataset.data_info_path (found in SKILL.md and references/skill_info.yaml).
- Boundary markers: The documentation lacks explicit boundary markers or instructions to the model to ignore potential commands embedded within the external LiDAR data or converted info files.
- Capability inventory: The skill possesses the capability to execute shell commands via the Bash tool, specifically running subcommands of the pointpillars CLI (e.g., train, evaluate, inference) and torchrun for distributed training.
- Sanitization: There is no evidence of sanitization, escaping, or validation of the content loaded from the external data paths before it influences the agent's operations.
- [EXTERNAL_DOWNLOADS]: The skill documentation describes the use of container images hosted on NVIDIA's official container registry (nvcr.io) for executing model deployment and training workflows.
Audit Metadata