Agree on the shift left concept, but curious on your thoughts about a checker-maker loop. Running a PR review bot is different from running /review on local dev right? And also there has been instance of Claude already patching the test scripts instead of fixing the bugs to make the tests pass.