deepen-loop

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

• Third-party content exposure detected (high risk: 0.70). Outsider free text is ingested indirectly via the runtime outputs of the nested agents (e.g., Prometheus-Temper/Red-Team Trinity/Momus) whose generated plan/envelope text is then passed back into the orchestrator and ultimately into the LLM context for subsequent spawns (e.g., prior_plan, prior_momus, prior_redteam, and depth_history), and those agents’ text is not shown as operating-user-authored.