01 · Edge Condition (conditional edges)

What this lesson demonstrates

This is the first 03-workflows lesson. A workflow is a graph of executors — plain Go functions wrapped as nodes — joined by edges. Most edges are unconditional, but AddDirectEdge lets you attach a condition: a func(any) bool over the message flowing along that edge. When two edges leave the same executor with complementary conditions, the graph forks.

The scenario is email triage. A SpamDetectionExecutor classifies an incoming email string, then the message is routed either down the “assistant → send” path or to the “handle spam” terminal — depending only on a boolean in the emitted DetectionResult.

The real code

The whole branch lives in the builder. The condition is the fourth argument to AddDirectEdge:

return workflow.NewBuilder(detect).
    AddDirectEdge(detect, assistant, false, func(msg any) bool { return !msg.(DetectionResult).IsSpam }).
    AddEdge(assistant, send). // unconditional
    AddDirectEdge(detect, spam, false, func(msg any) bool { return msg.(DetectionResult).IsSpam }).
    WithOutputFrom(send, spam). // either terminal is a workflow output
    Build()

Both edges leave detect. The predicate on each decides which one the DetectionResult is actually delivered on, so exactly one branch runs.

What to notice

• Executors are just functions. workflow.NewExecutor(id, fn).Bind() wraps a func(In) Out as a node; the framework routes a message to an executor when its input type matches. DetectionResult flows to the two branches, string flows on to send.
• Conditions live on the edge, not the node. This keeps the routing logic separate from the work each executor does. AddEdge(src, dst) is the unconditional shorthand.
• Type-assert inside the predicate. The condition receives any; here it does msg.(DetectionResult).IsSpam. The gotcha: the two conditions must be exact negations, or you can drop a message (no edge accepts it) or double-deliver it.
• WithOutputFrom(send, spam) marks both terminals as outputs; the run surfaces them as workflow.OutputEvents you drain from run.NewEvents().

Construction is factored into buildWorkflow() so the offline test builds the identical graph, asserts both edges leaving detect carry a condition, and runs the workflow on two inputs to prove the conditions route correctly.

How it maps to the Agent Framework

In the Microsoft Agent Framework Go SDK, workflow is the graph-orchestration layer that sits above single agents. Conditional edges are the primitive every higher-level construct — switch cases, multi-selection fan-out, group-chat routing — is built on. Getting comfortable here means the rest of 03-workflows reads as sugar over this one idea. Nothing calls a model, so this lesson is a clean way to learn the graph API before wiring Azure AI Foundry agents into it.

Run it

go run ./tutorial/03-workflows/conditional-edges/01_edge_condition

Fully offline — every executor is a pure function, no credential needed. The built-in input contains both “prize” and “wire transfer”, so it is flagged as spam and yields Email marked as spam: matched suspicious wording.

Next: 02 · Switch Case — multi-way conditional routing