The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes a diagnostic script scripts/vllm-health and associated tests tests/test_vllm_health.py. The test_vllm_health.py file uses subprocess.run to execute the local health script as part of its deterministic testing suite. This is a standard practice for verifying CLI behavior and is contained within the test environment.
[EXTERNAL_DOWNLOADS]: The documentation and templates provide docker run and pip install commands for vLLM. These reference official sources, including the vllm/vllm-openai Docker image and vllm package on PyPI. The vllm-health script performs HTTP GET requests to the user-supplied server URL to collect status metrics, which is consistent with its primary purpose as a health probe.
[DATA_EXFILTRATION]: The vllm-health script is verified to be read-only, using method="GET" for all network requests. It does not write files or transmit collected metrics to any external or third-party domains. The skill explicitly instructs agents not to print or commit sensitive environment variables or Hugging Face tokens to logs or chat sessions.
[CREDENTIALS_UNSAFE]: The skill correctly follows security best practices by recommending the use of environment variables (HF_TOKEN) and Kubernetes secrets for authentication tokens, rather than hardcoding them in scripts or configuration files.

vllm