ai-permission-mode
ai-permission-mode: the popup answering an agent's permission request — CR selects the option under the cursor, Esc or q defers it.
The popup an agent's permission request opens: the agent wants to do something (write a file, run a command), and this is where you say yes, no, or which variant.
Chords
| Chord | Action |
|---|---|
<CR> | Select the option under the cursor |
<Esc> or q | Defer — leave the request pending |
Deferring is not denying
<Esc> and q dismiss the popup without answering. The request stays pending and the agent stays blocked on it; you can come back and answer later.
That's deliberate: a permission prompt appearing while you're mid-thought shouldn't force a decision, and dismissing a dialog is a reflex. Making dismissal mean "not now" rather than "no" keeps the reflex from silently denying something you meant to allow.
To answer, pick an option and press <CR>.
The option list is the agent's
The options come from the request itself, not from a fixed yes/no set — an agent may offer "allow once", "allow for this session", "deny", or something specific to what it's asking. The popup renders whatever the agent sent, so read them rather than pressing by position.
Avoiding the prompt entirely
<C-t> in the conversation buffer toggles trust mode. In auto-accept, requests are granted without surfacing here. Worth it for a long mechanical task you're supervising; not worth it otherwise.
See also
ai-conversation-mode— the conversation and the trust toggle.