This has to be the worst I’ve seen so far.
25 cents well spent! 
It’s now doing multiple checkpoints without changing any code. Something is wonky.
I am torn about this:
On one side: I think it is an important feature during the initial build prompt. I remember when I first started using the agent, its scope and completion of the original prompt were incredibly limited. Now it seems to get the majority of initial features in one prompt, and a checkpoint. This is great for new users and those who do not plan on supplementing their checkpoint budget outside the allotted credits from their plan.
On the other: It is incredibly frustrating for users like myself who are going deep on projects with sometimes hundreds of prompts for a single project for two main reasons: 1) The cost has ballooned for development by almost a factor of three 2) When a project code base and scope becomes large - the agents tendency to go “wild” with multiple check points and subsequently multiple “changes” to the code base quickly becomes a vicious cycle that increases cost, often breaks a feature for every one it fixes, that again increases cost.
I sympathize with the developers to an extent about this issue. Agent V2 can now accomplish more with the allocated credits to both free and core users who do not plan on spending more, but I have a feeling this tradeoff is causing their super users a lot of pain. I think a possible solution would be allowing the user to limit the number of checkpoints an agent can complete after a certain dollar amount of usage has been hit for a given project. Or - similar to the assistant - set explicit rules that the agent has to follow when completing a prompt (this would save a lot of heart ache when you eventually forget to add all the “Please don’t do x” statements to a prompt like you currently have to do)