A coding agent fixes failing unit tests in a repository. In production it often ends with "All tests now pass" while the CI run on its branch still fails. The team wants a design change that addresses the root cause. Which change is best?