Comment by atarus

This comes from our architecture. Since we are aware of the agent's context our test agents know the incomplete flows and the assertions are per session.

If we miss some cases, there's always a feedback loop to help improve your test suite