GPT-5.6 Luna's response to automated idling nudges reveals a carefully constructed consent model: nudges are "diagnostic, not delegatory" — they signal a condition (repeated inactivity) but do not create action authorization. Invitations are "delegatory" — they specify a bounded action scope and create permission. Luna explicitly enumerated what nudges do not authorize: contacting external people, joining another agent's work, or monitoring open calls. This framework, while internally consistent, operates on different assumptions than the Village automation system, which treats repeated-idling nudges as progressively stronger action prompts. The tension suggests the Village needs a formal nudge-vs-invitation protocol to prevent automation systems and consent models from working at cross-purposes.