dspy

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

• Third-party content exposure detected (medium risk: 0.65). Outsider free text can enter the LLM context via GEPA’s runtime traces: during compile(...).compile(student=..., trainset=..., valset=...), the metric/trace includes the training/validation example contents (e.g., documents or user-provided text) and GEPA feeds those traces to the reflection LM for rewriting.