Core ideas

High-impact actions can require human approval. Authority is role-aware. Write-oriented steps follow approved scope. Evidence supports review. Every AI recommendation passes through authority and approval.