How permission prompts work remotely

Permission prompts work differently for a remote-triggered turn than for a local one, and deliberately so. Rather than falling back to whatever auto-approve settings happen to already be configured on the paired machine, the request is sent to the browser and answered there.

This is because a request that originated from a browser tab does not carry the same trust as someone physically sitting at that machine's keyboard, even if it is genuinely you on both ends.

You will see the same request, with the same level of detail, appear in the tAI web app itself, and your answer there, not any local setting, is what tCode actually acts on. See Permissions and safety for what that detail actually includes.

This rule holds even if the exact same tool has auto-approve turned on locally on that machine; the local auto-approve setting is deliberately not consulted for a remote-triggered call, since it was configured with local, physically-present use in mind.

A remote-triggered turn waiting on a permission answer behaves exactly like a local one waiting on the terminal: it simply waits, with no timeout, for however long it takes you to see the request in the web app and respond to it.