Skill UI

LlamaGuard is Meta’s 7–8B safety-specialized LLM that filters both prompts and responses by classifying six threat categories, enabling fast inference via vLLM/SageMaker and integration into NeMo Guardrails for end-to-end moderation.

NVIDIA’s NeMo Guardrails adds runtime safety rails to LLM apps, covering jailbreak detection, hallucination and fact-checking, PII filtering, and toxicity checks via Colang 2.0 rules and integrations for production use on T4 GPUs.

This skill specializes in designing, optimizing, and evaluating prompts for Large Language Models (LLMs). It allows users to generate optimized prompt templates, structured output schemas (e.g., JSON mode, function calling), and comprehensive test suites. Use this skill when building new LLM applications, implementing complex reasoning patterns like Chain-of-Thought, defining system guardrails, or building robust evaluation frameworks to measure and improve model performance.

This framework implements robust input and output validation guardrails for LLM-powered applications. It prevents prompt injection, detects and redacts PII, filters toxic content, and validates model outputs against defined schemas. Use it when deploying AI agents that handle sensitive data or require strict adherence to corporate policies, ensuring a secure and trustworthy AI interaction layer.

This comprehensive skill provides multi-layered guardrails for integrating with LLMs like Claude. It enforces security and compliance by implementing input validation (PII detection, length checks), system prompt restriction (scope control, safety rules), output filtering (preventing sensitive data leakage), cost governance, and model access policies. Ideal for building secure, robust, and governed AI applications.

This skill outlines the implementation of advanced runtime defenses for production Large Language Model (LLM) applications. It covers using multiple industry-leading guardrail systems—including Llama Guard, NeMo Guardrails, and LLM Guard—to detect and block adversarial attacks such as prompt injection, jailbreaks, and toxic content. It provides a comprehensive defense-in-depth strategy for ensuring model safety and output integrity in real-world deployments.