Programmable Tool Confirmation

[Stebalien]

It would be nice if tool confirmation were a bit more programmable.

In my specific case:

1. I'd like the tool to be able to provide a description of what the tool is doing rather than displaying the raw call description. This especially matters when some of the call arguments are large.
2. I'd like to be able to ask for confirmation for risky operations and/or the first time some operation is performed on a buffer. E.g., when the LLM tries to edit a buffer in its context the first time, I want to confirm that tool call to give the LLM permission to edit the buffer, but I don't want to confirm every single call.

The simplest solution, IMO, is to allow the tool's :confirm parameter to be a synchronous function that:

1. Takes the same parameters as the tool.
2. Returns one of allow, deny (maybe with an optional reason?), ask, or a string (to be displayed in the buffer as the tool confirmation).

A more flexible alternative would be to have the tool call some kind of gptel-ask-confirmation function, but that starts turning into callback hell pretty quickly.

[karthink]

See #968. This is on the docket, but I need to set some time aside to figure out the best way to do it.

[Stebalien]

Ah, awesome! I searched the open issues but not the open PRs.