memory-user-edits

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

• Third-party content exposure detected (medium risk: 0.65). The required workflow uses memory_user_edits with user-provided “control” text (e.g., the user’s “remember X” / “add a memory edit …” content) as the “Memory Edits (Your Explicit Instructions)” that are ingested into the agent’s context; if that “X” includes text from an outsider (or is derived from outsider-authored sources), it becomes free-form text in the LLM context.