tensorrt-llm
Pass
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: SAFE
Full Analysis
- [EXTERNAL_DOWNLOADS] (SAFE): Installation instructions utilize standard package managers (pip, docker) and reference models from trusted organizations like Meta-Llama.
- [COMMAND_EXECUTION] (SAFE): Shell commands are restricted to deployment orchestration (Kubernetes, Ray), local server management, and GPU monitoring using nvidia-smi.
- [CREDENTIALS_UNSAFE] (SAFE): The documentation follows secure practices by suggesting the use of environment variables for API keys rather than hardcoding credentials.
- [DATA_EXFILTRATION] (SAFE): Network communication is confined to local service requests and Prometheus monitoring on whitelisted domains like localhost.
- [PROMPT_INJECTION] (SAFE): No malicious instructions or behavioral overrides were found. The skill also manages the indirect prompt injection surface safely: 1. Ingestion points: llm.generate() and trtllm-serve API endpoints. 2. Boundary markers: Absent in snippets, which is typical for library-level documentation. 3. Capability inventory: GPU model inference. 4. Sanitization: Handled by the underlying inference library logic.
Audit Metadata