How to Add Tool Approval Gates to an AI Agent Using Nuxt and AI SDK 7

A developer building a superhero-themed Nuxt app connected to an Anthropic model via Amazon Bedrock accidentally gave an AI agent the ability to delete files without any confirmation step, resulting in unintended deletions. The incident prompted an investigation into tool approval mechanisms, leading to the discovery that AI SDK 7 supports approval gates at the model-call level, pausing execution until a user confirms or rejects a pending action. The tutorial walks through setting up a Nuxt 4 project with AI SDK 7, AWS Bedrock credentials, and sandboxed file tools that restrict path access to a fixtures directory. An additional section integrates the Kiro CLI behind the same approval interface using the Agent Client Protocol (ACP). The guide requires Node.js 22 or later, npm 11, valid AWS credentials, and an authenticated Kiro CLI for the optional ACP portion.