opik-optimizer

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

• Third-party content exposure detected (high risk: 0.80). The skill explicitly ingests dataset items from opik_optimizer.datasets (e.g., hotpot, cnn_dailymail, driving_hazard) and uses those public dataset rows as inputs to optimize_prompt/metric evaluations, which means the agent will read and interpret open/public (potentially user-generated) third‑party content.