How can I test a coding-agent interface in smart glasses?
Start with a harmless sample project and a repeatable sequence of events: progress, a request for input, failure and completion. Then interrupt the connection and change tasks while a request is pending. This tests whether the interface preserves the meaning of an action under less convenient conditions.
Verify what happened on the host
A successful animation only shows that the glasses responded to an input. The test also needs to show which request the host received and what it did. Record task and request identifiers so duplicate messages or a late response can be traced to the original event.
Include a deliberately expired approval and require the host to reject it. That is a more revealing reliability check than repeatedly approving a single fresh request while the phone and glasses remain connected on a desk.