Agent Ops / Standard term
Approval gate
A policy checkpoint that requires authorization from a designated actor or system before a prepared operation can execute.
A gate can require a named account owner, role, quorum, policy service, or separate signing system. It is appropriate when authorization itself cannot be delegated, when evidence remains ambiguous, or when policy requires separation of duties. It is not the only way to contain risk: hard limits, narrow tools, simulations, staged rollout, and automatic verification may provide stronger controls for repeatable cases.
Builder example
A gate is useful only if the authorizer receives the decisive evidence and has authority to decide. Requiring a person to click every routine action can create queues and approval fatigue without detecting errors. Place gates by authorization and residual risk after other controls, not by a rule that every irreversible action is manual.
Common confusion: Approval and guardrails can coexist. A guardrail can reject an operation, cap it, transform it, or route it to an approval state; the gate specifically represents authorization still required before execution.

