A command approval prompt is a safety control that pauses an agent before it runs a risky command and asks a human to confirm. In practice, this can block actions such as privilege checks, file searches, network probing, or system changes until an operator reviews the request. It matters because many autonomous tools can move quickly through a system, and one wrong command can turn routine automation into uncontrolled access.
In attacks, disabling the prompt removes an important brake: the agent may continue executing discovery or escalation steps without human oversight, making it behave more like an intruder’s helper. In defense, approval prompts are strongest when combined with least privilege, sandboxing, tight network limits, and logging. They are not a complete security boundary, but they are a practical checkpoint that can stop dangerous behavior before it reaches sensitive hosts or credentials.