Responses get mixed into think blocks with Z.AI GLM 4.5

[academo]

I am using Z.AI GLM 4.5 via openrouter

At some point after several tools usages the respoens from the model get hidden behind its think blocks and crush simply will hide the think blocks

Is there anyway to expand think blocks? or a config where I can tell crush to always show the content of think blocks?

[meowgorithm]

Hey! There's no way to expand the blocks, but this is a UX change we should make. We initially designed it to operate similar to the way ChatGPT thinks out loud, but in practice I think makes more sense to keep the blocks expanded.

I'll review the behavior to double check, and if it makes sense will convert this to an issue for tracking.

[academo]

thanks! that'd be very nice. I am sure there's something wrong with Z.AI that somehow starts messing up its think blocks, because this doesn't happen every time but it happens often and unless I read very fast I can't see the final output of its response.

Something simple like a shortcut to expand the think blocks would be great but even a config to always expand think blocks would be enough for this kind of edge cases.