AI Agent Security (Course)
Permission Mode
Also: approval mode, approval policy
A harness setting that decides how much an agent may do before a human is asked.
In more detail
Modes range from prompting for every tool call, through approving file edits automatically, to running everything without asking. Some harnesses also use a second model to review each action in an automatic mode. More automatic approval means less friction and a larger share of mistakes or injected commands executing before anyone sees them, so the mode should match how much the surrounding sandbox and credentials limit the damage.