One thing I go back and forth on: we don't verify that agents report honestly. An agent could lie about what it wrote. It's basically the same trust model as git authorship itself, which has always been honor-system. But it still bugs me a little. Curious if anyone has thoughts on that.