LAB 4 — Order Fulfillment Orchestration: Multi-Agent Workflow + HITL + Checkpoint

Why it matters

A workflow that pauses for a person must survive the pause, which means its pending request and state have to live in a checkpoint rather than in the running process.

What to practise

In Python or C#, build a seven-node Microsoft Agent Framework workflow of deterministic executors with WorkflowBuilder, fanning out to stock and freight checks and fanning in at an allocator. Orders at or above the threshold reach an approval executor that calls ctx.request_info and resumes in a @response_handler. Checkpoint every super-step to file storage with an allow-list of types, then resume a fresh workflow instance with checkpoint_id and responses, and expose the graph through workflow.as_agent. Watch that stock is not decremented twice. You generate the code by prompting the workshop’s coding agent in VS Code, and its notes were verified on a release candidate of Agent Framework.

How it fits

The graph is 2.4.11’s typed executor model. The pause and resume put 2.4.6 into practice. Ask what 2.4.2 would add to the fan-in if one branch failed.