nv-segment-ctmr
Pass
Audited by Gen Agent Trust Hub on Jul 8, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill fetches model assets and configuration from trusted external sources during setup. It downloads model weights from HuggingFace using the nvidia/NV-Segment-CTMR repository and clones the upstream code bundle from NVIDIA-Medtech's official GitHub repository.- [COMMAND_EXECUTION]: The skill performs command execution to run medical imaging inference. The script scripts/run_ctmr.py invokes python -m monai.bundle run using subprocess.run to execute the segmentation model. The execution environment is configured with specific CUDA allocation settings to manage GPU resources.- [PROMPT_INJECTION]: The skill exposes an attack surface for indirect prompt injection by processing external data. Ingestion points: Reads NIfTI imaging volumes and JSON configuration files (label_dict.json, metadata.json) from the upstream repository or user-provided paths. Boundary markers: The skill provides instructions for wrapping the tool and validating output JSON, but lacks explicit delimiters for the ingested medical data. Capability inventory: Uses subprocess.run for model execution and performs file system writes to a caller-provided output directory. Sanitization: Implements geometry checks (spacing, affine, shape) and label ID validation in scripts/run_ctmr.py to ensure output consistency.
Audit Metadata