deepstream-sop

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

• Third-party content exposure detected (high risk: 0.95). The skill ingests OUTSIDER-authored free text from the user’s request body (/v1/chat/completions → SOPProcessManager.create_video_processor(..., prompt=text_content)), and that prompt is then passed into the embedded vLLM (VLLMInference.inference(..., prompt=engine_prompt)), so arbitrary user text becomes LLM context (indirect prompt-injection risk).

MEDIUM W013: Attempt to modify system services in skill instructions.

• Attempt to modify system services in skill instructions detected (high risk: 1.00). The skill includes instructions that modify system-level files (e.g., the "DeepStream lib hide" rename trick and writing to /opt paths) and build/deploy steps that implicitly require elevated privileges, which can change the host system state and thus pose a compromise risk.