guardrails-safety-filter-builder

What problem does it solve?

This skill enables teams to implement robust safety filters for LLMs, including PII redaction, policy-driven constraints, prompt-injection detection, and safe refusal templates to reduce risk and protect user data.

Core Features & Use Cases

• Input filtering to block malicious prompts and reduce attack surface.
• Output redaction to mask sensitive information in responses.
• Policy constraints to enforce allowed content and provide safe refusals.
• PII detection and redaction to protect personal data in user interactions.
• Prompt-injection detection to identify and block attempts to manipulate system prompts or dynamics.

Quick Start

Configure the guardrails builder in your ML pipeline to automatically apply safety filters during request handling.