What is Automated Reasoning in Bedrock Guardrails?

Automated Reasoning provides mathematical verification against formal policy rules, offering up to 99% accuracy in ensuring that model responses strictly adhere to defined safety and compliance protocols.

Does this skill only work with AWS Bedrock models?

No. While it uses the Amazon Bedrock service, the ApplyGuardrail API is model-agnostic and can be used to validate inputs and outputs for any LLM, including OpenAI, Google Gemini, or self-hosted models.

How does Contextual Grounding prevent hallucinations?

It validates the factual accuracy of a model's response by comparing it against provided source documents, ensuring the AI doesn't 'hallucinate' information that isn't present in your reference data.

Can I define custom sensitive data patterns?

Yes, the skill supports custom regex patterns, allowing you to identify and block or anonymize domain-specific information like internal employee IDs, API keys, or proprietary project codes.

Bedrock AI Safety & Guardrails

Name: Bedrock AI Safety & Guardrails
Author: adaptationio

byadaptationio

0•

安全与测试

Implements comprehensive Amazon Bedrock Guardrails to secure generative AI applications with content filtering, PII protection, and hallucination detection.

This skill provides a standardized implementation for Amazon Bedrock Guardrails, enabling developers to enforce strict safety policies across any foundation model. It automates the setup of six critical safeguard layers—including PII redaction, topic denial, and mathematical verification via automated reasoning—ensuring that AI-driven interactions remain compliant, secure, and contextually accurate. By leveraging the ApplyGuardrail API, it allows for consistent policy enforcement regardless of the underlying LLM provider, making it an essential tool for enterprise-grade AI deployment, risk management, and RAG application validation.

主要功能

01Contextual Grounding to detect and prevent AI hallucinations

02Customizable Topic Denial for financial, medical, and legal boundaries

03Model-agnostic enforcement via the ApplyGuardrail API

040 GitHub stars

05Automated Reasoning for mathematical verification of policy rules

06Six safeguard policies including Content Filtering and PII Redaction

使用场景

01Securing enterprise chatbots against jailbreaks and prompt injection attacks

02Validating RAG application responses for factual accuracy and source relevance

03Automating the redaction of sensitive PII from model inputs and outputs

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add adaptationio/skrillz bedrock-guardrails

For use in Claude.ai and ChatGPT

Download Skill