aws-cloudformation-bedrock_skill

Overview

This skill provides AWS CloudFormation patterns to provision and manage Amazon Bedrock resources for production-ready AI infrastructure. It covers agents, knowledge bases, data sources, guardrails, prompts, flows, and inference profiles. Use it to codify RAG, agent actions, moderation, and multi-model routing as repeatable CloudFormation templates.

How this skill works

The skill supplies reusable CloudFormation constructs and parameter patterns that define Bedrock Agents, KnowledgeBases, Guardrails, VectorStore configurations, Flows, and InferenceProfiles. Templates include IAM roles, parameter validation, mappings for environment tiers, outputs for cross-stack exports, and examples for common vector stores (OpenSearch, Pinecone, pgvector). Deploying the templates creates the Bedrock resources and exports IDs/ARNs for application stacks to import.

When to use it

• Creating Bedrock agents with action groups, Lambda integrations, and function definitions
• Implementing Retrieval-Augmented Generation (RAG) using managed knowledge bases and vector stores
• Connecting data sources like S3, web crawls, or custom connectors to a knowledge base
• Applying content moderation guardrails to enforce safe AI outputs
• Orchestrating multi-step workflows with Bedrock Flows and versioned prompts
• Configuring inference profiles for optimized model routing and multi-model access

Best practices

• Use CloudFormation parameter types (SSM, ARNs, resource IDs) to validate model identifiers and environment-specific values
• Segment infrastructure with nested stacks or separate stacks (infrastructure vs application) and export/import critical IDs
• Define least-privilege IAM roles for Bedrock agents and Lambda functions, including explicit bedrock:InvokeModel actions
• Parameterize vector store type and mapping fields to support OpenSearch, Pinecone, pgvector, or Redis without template changes
• Use Mappings and Conditions to adjust capacity and inference units across dev/staging/production
• Version prompts, guardrails, and flows; export aliases and IDs to avoid breaking application deployments

Example use cases

• Deploy a customer-support Bedrock Agent with an OpenSearch-backed knowledge base for RAG
• Provision a guardrail that blocks sensitive PII and denies regulated advice topics
• Create an application stack that imports AgentId and invokes the agent from a Lambda
• Switch vector store provider by changing a parameter (OPENSEARCH_SERVERLESS, PINECONE, PGVECTOR)
• Define inference profiles to route requests to different foundation models based on workload