AI Coding: Workflow Engine for Delayed Delivery Refunds + AI-generated Tests

Problem: Build a Parsable and Executable Workflow Engine for Delayed Delivery Refunds Implement a simplified workflow engine to automate refund handling for delayed deliveries. Users provide a text-based workflow description where each step is a node. Your system must parse the text, build a workflow graph, and execute it to produce a refund decision (full/partial/no refund, follow-ups). Inputs A workflow text description (you define the syntax, but must specify it clearly) supporting at least: Sequential execution Conditional branching (if/else) Node parameters (e.g., delay thresholds, refund percentages) An Order object (promised ETA, actual delivery time/current time, order amount, delivered flag, etc.). Output Execution result: action: NO_REFUND / PARTIAL_REFUND / FULL_REFUND amount: refund amount (0 if none) reason: explanation of which rule matched Example capabilities Delivered & delay > 30 min => full refund Delivered & delay 10~30 min => partial refund (e.g., 20%) Not delivered & past ETA by 20 min => partial credit or trigger human follow-up Constraints Parser must handle invalid input with readable errors. Extensible design: easy to add new node types. Testing: provide runnable tests, including hand-written unit tests an "AI-generated test case" strategy (property-based / random generation / mutation), and show at least 5 generated test examples. Scale Workflow nodes: 1 ~ 200 Execution time target: < 100ms per run (excluding external IO) Deliverables Syntax definition (EBNF or examples) Core parsing + execution code Test code and sample AI-generated cases Example Input workflow: IF delivered AND delay_min > 30 THEN FULL_REFUND order: delivered=true, delay_min=45, amount=30.00 Output action=FULL_REFUND, amount=30.00