The Agent Skills Directory

[SAFE]: The skill provides a methodology for performance benchmarking using standard infrastructure tools like chalk container and kubectl for their intended monitoring purposes.
[CREDENTIALS_UNSAFE]: Sensitive tokens, including Hugging Face and API keys, are handled securely through environment variables and CLI arguments rather than being hardcoded in the skill.
[EXTERNAL_DOWNLOADS]: The skill utilizes uv for managing the httpx dependency and pulls container images from official vLLM repositories, which are well-known and reputable sources.
[COMMAND_EXECUTION]: Shell commands are used appropriately for container deployment, local port-forwarding for telemetry metrics, and cost calculation queries without evidence of malicious intent.

benchmark-vllm-model