01Access Control (ACL) with per-agent, per-backend, per-tool rules
02Two-Level Rate Limiting for both individual agents and global cross-agent activity
03Human Approval Workflows for destructive operations via webhook notifications
04FIFO Execution Queue with randomized delays to mimic human interaction patterns
05Comprehensive Audit Logging of every tool call, verdict, and execution result
061 GitHub stars