creating-online-evaluations

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

• Third-party content exposure detected (high risk: 0.70). The required runtime workflow for the eval is to run posthog:llma-evaluation-* against existing $ai_generation events; those events include the model output (assistant text) and may originate from outside-authored user/traffic content, which the agent/LLM context would indirectly ingest to judge/evaluate.