The Agent Skills Directory

[SAFE]: The skill provides standard operational scripts for vLLM, a high-throughput LLM serving engine. The provided shell script start_vllm.sh correctly uses environment variables with safe defaults to configure the server, avoiding hardcoded sensitive values.
[SAFE]: The Python client code and readiness auditor utilize standard libraries and follow best practices for secret management, specifically recommending the use of HF_TOKEN from environment variables rather than hardcoding credentials.
[SAFE]: The diagnostic directives focus on infrastructure health (GPU memory, NCCL topology, and OOM troubleshooting) using legitimate tools like nvidia-smi and health endpoints.
[INDIRECT_PROMPT_INJECTION]: The skill includes code to process user-supplied prompts via an LLM client. While this represents an attack surface for indirect prompt injection, it is the primary intended purpose of the skill and follows standard implementation patterns using structured roles (e.g., system vs. user messages), which mitigates risk.

vllm