context-ramp
Pass
Audited by Gen Agent Trust Hub on Aug 12, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill downloads text-to-speech models and runtimes from NVIDIA's official repositories on Hugging Face and GitHub. These references use pinned revisions for stability and security.
- [COMMAND_EXECUTION]: Provides local scripts for building and serving the TTS engine. The processes are restricted to the local loopback interface (127.0.0.1), preventing external exposure of the service.
- [SAFE]: An eval() call detected in the Python narration script is an invocation of torch.nn.Module.eval(), a standard PyTorch method for setting a model to inference mode. It does not execute arbitrary code and is not a security risk.
- [SAFE]: The skill processes external repository content and documentation. It includes strong boundary markers, requiring the agent to 'obey evidence and authority rules' and use 'exact source anchors' for all claims. These constraints, combined with local audio processing, mitigate the risk of indirect instruction following.
Audit Metadata