Your agent writes the code. Can you explain what happens when a request is retried, or a worker crashes halfway through?

Sphinx pauses a Git push and has a separate model read the change and ask you about its behavior. A missed answer leads to an explanation and another scenario. The reviewer doesn't see the coding conversation that produced the patch.

It's a bypassable learning aid, and its examiner can be wrong. Using it requires a reviewer model; installation configures Git hooks across your machine.

Code and documentation:
github.com/parthh01/sphinx