Prompt injection detection API: screen text bound for an agent and get a safe, suspicious or malicious verdict with the patterns that matched. Rule-based and deterministic.