deepstream-sop

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

• Third-party content exposure detected (medium risk: 0.30). The runtime workflow reads outsider-authored free text from the POST body to /v1/chat/completions (the messages[].content text item becomes the prompt passed into SOPVideoProcessor, which forwards it to the VLM via VLLMInference.inference).

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

• Attempt to modify system services in skill instructions detected (high risk: 0.90). The skill includes the DEEPSTREAM_LIB_HIDE rule that explicitly instructs renaming DeepStream library files ("rename lib → lib.tmp") during a gst-plugin build, which entails modifying system library files (likely under system paths such as /opt or /usr) and therefore risks requiring sudo or otherwise changing the host system state.