The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads model weights and source code from trusted platforms, including NVIDIA's official GitHub repository and Hugging Face. These operations are part of the documented setup process and target well-known, reputable services.
[COMMAND_EXECUTION]: The entry point script scripts/run_ctmr.py uses subprocess.run to execute the MONAI bundle inference engine. The command is constructed using a list of arguments rather than a shell string, and user-provided inputs such as modalities and label IDs are validated or coerced to safe types before execution.
[CREDENTIALS_UNSAFE]: The skill implements a security-conscious environment handler in scripts/run_ctmr.py that only forwards a whitelist of necessary variables to the segmentation process. It explicitly drops potentially sensitive environment variables like AWS_SECRET_ACCESS_KEY, a behavior verified by the included unit tests.
[DATA_EXPOSURE]: File operations are restricted to reading input medical images (NIfTI format) and writing results to a caller-specified output directory. No unauthorized file system access or unexpected network data transmission was detected.

nv-segment-ctmr