vllm-ascend
Pass
Audited by Gen Agent Trust Hub on Oct 18, 2026
Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The service deployment script (scripts/deploy_service.sh) includes a command to terminate Python 3 processes (killall -9 python3) as a method for stopping the server. This is a common management task in development and staging environments.
- [DYNAMIC_EXECUTION]: The skill documentation and benchmarking scripts (scripts/benchmark.py) utilize the trust_remote_code flag. This is a native feature of the vLLM and Hugging Face ecosystems required to load specific model architectures that include custom logic. While this allows execution of code provided by the model repository, it is a standard requirement for LLM inference serving.
- [SAFE]: Automated scan flags regarding piped execution to python3 were verified as false positives. The diagnostic commands in scripts/deploy_service.sh use python3 -m json.tool specifically to format and validate JSON output from a local health check endpoint.
Audit Metadata