agent-backend-adapter

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

• Third-party content exposure detected (medium risk: 0.30). templates/vllm_adapter_template.py ingests outsider-provided free text at runtime via the CLI argument --prompt (and optionally chat messages -> prompt text) and passes it into vllm.LLM.generate(prompts, ...), so an outsider can supply arbitrary text that the LLM reads.