Add task: happy-dog-latex

[Leolty]

A complex and interesting task requires GUI, visual inspection, and complex coding

[zzn-nzz]

Hi @Leolty , this is a creative and challenging task with an interesting idea. It requires many crucial abilities in realistic scenarios, including web interaction, visual perception, and code implementation and execution.

I have two small concerns:

• Note-head counts problem: I believe the answer makes sense, but if the number of correct models is not specified, I would assume that GPT-5.2-Thinking (Extended) has five note-heads based on the confusing definition of note-head. You might consider explicitly specifying the number of models that satisfy the requirement that “note-head counts match the original image.”

• The agent output example in evaluation.md seems to be missing.

Once these two issues are resolved, I’d be happy to include this amazing task. Thank you!