langsmith-online-eval-engineering

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

• Third-party content exposure detected (medium risk: 0.30). The runtime workflow fetches “recent root-level traces” from the user-specified LangSmith project (via client.list_runs(...)), and those traces include user-authored input/output text that is later fed into the LLM-as-judge prompt or passed into code evaluators via run["inputs"]/run["outputs"].