cuda-kernels
Pass
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: SAFE
Full Analysis
- [EXTERNAL_DOWNLOADS] (SAFE): The scripts 'scripts/ltx_kernel_injection_example.py' and 'scripts/transformers_injection_example.py' download pre-trained models from HuggingFace (e.g., 'Lightricks/LTX-Video'). This is a standard and expected operation within the AI development ecosystem using a trusted source.
- [COMMAND_EXECUTION] (SAFE): Documentation in 'references/a100-optimization-guide.md' and 'references/troubleshooting.md' includes instructions for using standard GPU development tools such as nvcc, nsys, and ncu. These are presented as user-executed commands for development and profiling.
- [DATA_EXFILTRATION] (SAFE): No patterns of unauthorized network communication or sensitive data access were found. The scripts focus on local model optimization and performance measurement.
- [REMOTE_CODE_EXECUTION] (SAFE): The skill utilizes standard PyTorch extension loading mechanisms for local CUDA kernels in 'scripts/benchmark_rmsnorm.py'. No remote execution of untrusted scripts was observed.
Audit Metadata