ai-observability-and-debugging

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

• Third-party content exposure detected (high risk: 0.75). The skill’s required workflow includes “Replay CLI + back-office UI” that “pulls the full audit row from ai_requests + S3 payload” and “reconstructs the exact gateway input,” which can include outsider-authored free text (e.g., customer/support ticket content or user prompts) stored in the audit/S3 payload and then ingested into the agent’s LLM context during replay.