DM reactions use the easy-task model settings
Contextual emoji selection now reuses the existing easy-task model policy. The reply body keeps the Agent's configured model.
Improved
- Live Luna calls returned appropriate gratitude, celebration and empathy reactions. This small comparison did not show a latency improvement.
Fixed
- Workers select an available easy-task model, retaining Agent settings when no candidate qualifies. Selection failure or timeout keeps the existing neutral reaction fallback.