arize-trace

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

• Third-party content exposure detected (high risk: 0.80). The skill's required workflow instructs running ax spans export / ax traces export to ingest and read exported span data (see SKILL.md "Export Spans/Traces" and "Where to Find Prompts and LLM I/O"), which explicitly contains untrusted, user-generated fields like attributes.llm.input_messages, attributes.input.value, attributes.output.value and attributes.retrieval.documents.contents that the agent is expected to read and could influence decisions—so it clearly consumes untrusted third-party content.