Guardrails

Example	Description
Custom Guardrail	Subclass BaseGuardrail and raise InputCheckError to block security-abuse prompts as a pre-hook.
OpenAI Moderation	Block flagged input with OpenAIModerationGuardrail as a pre-hook, including custom category lists and image moderation.
Output Guardrail	Reject responses under 20 characters with a post-hook that raises OutputCheckError.
PII Detection	Block SSNs, credit cards, emails, and phone numbers with PIIDetectionGuardrail, or mask them with mask_pii=True.
Prompt Injection	Block prompt injection and jailbreak attempts with PromptInjectionGuardrail as a pre-hook.
Mixed Hooks and Guardrails	Combine a logging pre-hook with PIIDetectionGuardrail; blocked runs return RunStatus.error instead of raising.

⌘I