Overview
Safe mode restricts automation based on confidence. Low-confidence issues may only get recommendations; high-confidence issues may be auto-resolved when policies allow.
Confidence Thresholds
Admins set thresholds (e.g., 80% confidence for auto-execute). Below the threshold, actions may be:
- Recommendation only
- Require approval
- Simulation only
Safe Mode Levels
| Level | Behavior | |-------|----------| | Recommend only | No execution; humans act | | Simulate | Dry-run; no write-back | | Approve then execute | Human approval required | | Auto-execute | Execute when confidence and policy allow |