Current behavior: Super Agents are shared with all members and admins at creation. Privacy must be manually toggled from the agent's profile afterward. The only exception is when a private location or personal connection is added, which auto-sets privacy.

Risk: Super Agents can be designed to cross-reference meeting transcripts with task activity to surface employee behavioral patterns: escalation frequency, deliverable follow-through, conflict indicators. This makes them powerful coaching tools for managers, but the underlying data is sensitive. Between creation and manual privacy configuration, any workspace member or admin can view the agent's instructions (which may reference specific employees and their patterns) and trigger the agent. A single @mention in a shared channel could expose coaching observations in an unintended context.

Proposed solution: A workspace-level default that sets new Super Agents to private at creation. Users opt collaborators in explicitly. This eliminates the exposure window and aligns agent privacy with the sensitivity of the data they're designed to handle.