What problem does it solve? Writing reliable long-running serverless workflows is error-prone: state is lost on interruption, retries cause duplicate side effects, and non-deterministic code silently corrupts execution on replay. This Skill guides the correct implementation of AWS Lambda durable functions so workflows survive interruptions and run for up to one year. ## Core Features & Use Cases - Replay-Safe Workflow Construction: Enforces the critical replay model rules, including keeping non-deterministic code inside steps, avoiding nested durable operations, and returning values instead of mutating closures. - Orchestration Patterns: Covers steps, waits, callbacks for human-in-the-loop approval, parallel/map concurrency with completion policies, saga compensating transactions, and circuit breakers. - Testing and Deployment: Provides LocalDurableTestRunner and DurableFunctionTestRunner testing patterns plus CloudFormation, CDK, and SAM deployment templates with correct IAM policies and qualified ARN invocation. - Use Case: Build an order-processing workflow that reserves inventory, charges payment, waits up to 24 hours for manager approval via callback, and automatically runs compensating refunds if any step fails. ## Quick Start Ask the assistant to create a TypeScript durable Lambda function that fetches a user in a step, waits five seconds, processes the result, and deploys it with CDK.