About
The Guardrails & Safety skill provides a robust framework for instrumenting safety checks at both the input and output stages of AI agent workflows. It enables developers to detect prompt injections, filter harmful content, redact PII (Personally Identifiable Information), and perform hallucination checks using LLM-as-a-judge or NLI models. By integrating with observability platforms like Langfuse, it ensures that every safety intervention is logged, allowing teams to monitor latency overhead, track false positive rates, and maintain a high-quality user experience without compromising security.
