vllm-deploy-simple
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [SAFE]: The skill provides a legitimate utility for deploying the vLLM inference engine.
- [COMMAND_EXECUTION]: The skill executes shell commands to install packages, manage server processes, and perform hardware detection. All actions are consistent with the tool's stated purpose of deploying and testing an LLM server.
- [EXTERNAL_DOWNLOADS]: The skill downloads the 'vllm' and 'openai' packages from standard Python registries using pip or uv. These are established packages relevant to the skill's functionality.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests user-provided arguments for model names, ports, and paths. 1. Ingestion points: Command-line arguments in scripts/quickstart.sh. 2. Boundary markers: None. 3. Capability inventory: Package installation, subprocess execution, and local network requests. 4. Sanitization: Arguments are double-quoted but not otherwise sanitized against command injection sequences. This represents a standard shell script design and is considered safe for the intended use case.
- [DATA_EXFILTRATION]: Network operations are limited to downloading official packages and performing health checks against the local server.
Audit Metadata