An approval prompt is a user confirmation step that appears before a tool performs a sensitive action, such as running a command, reading secrets, or changing files. It is a security control that forces the tool to stop and ask before crossing a trust boundary.
In cyber security, approval prompts matter because they reduce the damage from untrusted input. If a repository, document, or agent instruction tries to trigger command execution, the prompt gives the user a chance to block it before the action reaches the host system. This is especially important for AI coding agents, where background tasks like Git inspection or shell calls can happen early in a workflow. Without a prompt, repository-controlled settings or helper commands may run outside the sandbox and with the user’s privileges. Defenders use approval prompts together with sandboxing and least privilege, but prompts only help if they appear before the risky action and cover all sensitive operations.