The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads the NV-Generate-CTMR repository and model weights from official NVIDIA GitHub and Hugging Face repositories. These resources are from a trusted vendor and are handled according to standard machine learning development practices.
[COMMAND_EXECUTION]: The script scripts/run_mr.py uses subprocess.run to execute the image generation engine and Git commands. The command arguments are strictly controlled, and input configurations undergo schema and value validation (checking dimensions, spacing, and seeds) before being passed to the underlying process.
[PROMPT_INJECTION]: The skill instructions and evaluation datasets focus on technical execution and do not contain any patterns attempting to override agent behavior or bypass safety guardrails.
[DATA_EXFILTRATION]: No sensitive file access or unauthorized network communications were detected. Network activity is limited to established model hosting services for the purpose of downloading necessary weights.
[SAFE]: The skill demonstrates good security practices, such as providing a cost-preview confirmation for expensive runs and isolating model configurations into staged JSON files.

nv-generate-mr