arize-trace

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

• Third-party content exposure detected (high risk: 0.80). The skill explicitly instructs the agent to run ax spans/traces export to fetch Arize trace/span data (e.g., fields like attributes.llm.input_messages, attributes.input.value, attributes.retrieval.documents.contents) which is user-generated, untrusted content that the agent is expected to read and use to guide filtering/export decisions, so it can carry indirect prompt-injection payloads.