ai-agents-bedrock-guardrail-ha-when-disable-exception-on-block-is-set-to-true-for-affe1121

THEN Modify the LiteLLM guardrail handling to skip invoking the model when a pre-call guardrail blocks content and disable_exception_on_block is true. Instead, return the guardrail's blocked response as the final output to the client, avoiding unnecessary token usage and preserving the intended blocking behavior.

if guardrail_mode == 'pre_call' and guardrail_response.blocked:
    if disable_exception_on_block:
        # Return guardrail response directly, skip model call
        return guardrail_response.output
    else:
        raise HTTPException(status_code=400, detail=guardrail_response.output)