Output guardrail check

output-guardrail-check

Decorate a function with @output_guardrail and attach it to Agent.output_guardrails. If the guardrail detects a problem, tripwire_triggered=True raises OutputGuardrailTripwireTriggered.

When to apply

Preventing sensitive data leakage (PII, phone numbers, credentials) in agent replies, compliance filtering, or any post-generation validation.

Core snippet

from pydantic import BaseModel, Field
from agents import (
    Agent, GuardrailFunctionOutput, OutputGuardrailTripwireTriggered,
    RunContextWrapper, Runner, output_guardrail,
)

class MessageOutput(BaseModel):
    reasoning: str = Field(description="Thoughts on how to respond")
    response: str = Field(description="The response to the user")
    user_name: str | None = Field(description="User name if known")

@output_guardrail
async def sensitive_data_check(
    context: RunContextWrapper, agent: Agent, output: MessageOutput
) -> GuardrailFunctionOutput:
    phone_number_in_response = "650" in output.response
    return GuardrailFunctionOutput(
        output_info={"phone_in_response": phone_number_in_response},
        tripwire_triggered=phone_number_in_response,
    )

agent = Agent(
    name="Customer service",
    instructions="You are a helpful customer service agent.",
    output_type=MessageOutput,
    output_guardrails=[sensitive_data_check],
)

async def main():
    try:
        result = await Runner.run(agent, "What is the support phone number?")
        print(result.final_output)
    except OutputGuardrailTripwireTriggered as e:
        print(f"Output blocked: {e.output.output_info}")

Key notes

• Output guardrails run only for the last (final) agent in the chain
• Output guardrails always run after the agent completes (no parallel option)
• output parameter type must match the agent's output_type
• e.output.output_info contains the diagnostic info set in the guardrail