rtvi-cv-customize-model

Pass

Audited by Gen Agent Trust Hub on Aug 16, 2026

Risk Level: SAFE
Full Analysis
  • [PRIVILEGE_ESCALATION]: The skill instructs the user to use sudo for creating data directories and modifying file ownership on the host. These operations are standard practices for configuring persistent storage in Docker-based deployments.\n- [DYNAMIC_EXECUTION]: The skill documents the compilation of a C++ shared library for DeepStream and the runtime generation of a TensorRT engine from an ONNX model. These processes are required for the skill's primary purpose of model customization and do not involve executing untrusted code.\n- [INDIRECT_PROMPT_INJECTION]: The architecture includes ingestion points for external data via video streams and Kafka messaging. The mandatory evidence chain identifies these ingestion points, notes the absence of explicit boundary markers or sanitization within the skill's scope, and inventories capabilities such as Docker and shell access. This constitutes a low-risk surface for indirect prompt injection at runtime.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 16, 2026, 05:53 PM
Security Audit — agent-trust-hub — rtvi-cv-customize-model