bigquery-agent-analytics-sdk

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

• Third-party content exposure detected (high risk: 0.85). The skill’s runtime path for LLM-as-Judge/AI integration reads BigQuery event text fields like agent_end_events.input_text / output_text (e.g., via judge.evaluate_traces() and AIMLIntegration.generate_*), which are outsider-authored user/agent messages stored in the telemetry and then fed as free text into the LLM context.