Skip to content

Permission Prompts

Some coding agents pause to ask for permission before taking an action — editing a settings file, running a command, calling a tool. If those prompts aren’t answered, a run can stall.

Orbit includes an auto-responder for an agent’s “Do you want to proceed?” style prompts, so routine permission questions don’t freeze a run. A parked question that genuinely needs your input surfaces as an amber pause-for-input card you can answer or dismiss.

Separately from permission prompts, Orbit installs a Destructive Command Guard that intercepts genuinely destructive shell commands before a supported agent CLI can execute them, regardless of whether the agent itself would have paused to ask. An agent session reports an intercepted command as blocked rather than silently failing. If the guard didn’t get installed during first-run onboarding, install or retry it from Settings → Coding Agent.

  • A run that exits very quickly with no output can indicate a permission prompt that wasn’t satisfied in the spawn environment, even when the same prompt runs fine when launched manually.
  • Some agent versions invalidate a local permissions file when an agent edits it. If permissions seem to “reset,” reloading the agent’s environment (for example, restarting the shell window that holds its permission cache) restores them.