A team wants additional checks on agent input and output content. Which feature should be configured and evaluated?

1. Removal of all traces
2. Only a reassuring agent name
3. Only a higher response token budget
4. Agent guardrails at the appropriate supported nodes