Governed persistent AI workers with default-deny authority, human approvals, dry-runs, and an authenticated MCP.
Scoring was not performed