vllm

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 0.90). The skill contains examples that embed API keys and show passing bearer tokens (e.g., --api-key sk-local-xyz, api_key="sk-local-xyz", Authorization: Bearer ...) which encourages including secret values verbatim in commands/code and thus requires the LLM to handle/output secrets directly.

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (medium risk: 0.30). The runtime workflow is a self-hosted inference server (vllm serve) that ingests outsider-authored user prompt text via its OpenAI-compatible HTTP endpoints (e.g., POST /v1/chat/completions and POST /v1/completions) without first selecting a specific item.