Approval
Persetujuan
The moment a person says yes before the agent does something it cannot undo.
Approval is a deliberate checkpoint where a person reviews a proposed action and allows it. It exists because some actions are hard or impossible to reverse, and because a system that can act without asking is a system that can act wrongly without asking.
The design problem is deciding where to put the checkpoints. Too many and the user stops reading them; too few and the irreversible action happens unattended. The usual answer is to gate on reversibility and blast radius rather than on the kind of action: reading a file needs no approval, deleting a directory does, and sending an email to a customer probably does even though it is technically reversible.
An approval prompt has to carry the information needed to decide. "Run rm?" is not enough. "Delete 240 files in src/generated, not recoverable" is.
Related terms
Human in the Loop
A person who reviews, approves, or corrects the agent's work at key points.
Guardrails
Checks that stop or constrain an action before it happens.
Agent Mode
A preset that bundles permissions and behaviour, like a profile for the agent.
Permissions
The rules that decide what the agent is allowed to touch.
Tool Call
The model's request naming a tool and its arguments. Structured output, not a side effect.
Want this working inside your team?
It starts with one conversation about your workflow.