How's everyone handling regression testing in CI for code an agent wrote, without losing your mind?
热度趋势
趋势数据积累中
百分比基于当前可用热度信号，而非评论数或独立用户人数。
开发人员正面临如何对代理编写的代码进行持续集成（CI）中的回归测试的挑战。代理所做的微小改动经常会破坏看似不相关的多个文件中的功能，而单元测试往往无法捕捉到这些问题。这些问题通常直到产品发布后才被发现，导致开发人员感到沮丧。社区正在寻求有效的解决方案，包括编写更多单元测试、在CI中运行新的测试策略，或者在合并代码之前更多地依赖手动质量保证（QA），以应对这一难题。
Ever since agents started writing most of my PRs, small changes keep breaking things two or three files away that nobody thought to test.
Unit tests catch the obvious stuff. What they miss is the flow that used to work, gets touched by an unrelated change, and nobody notices until someone reports it in prod.
I tried writing more unit tests to cover the gaps. Turns out I was just guessing which flows mattered, and guessing wrong about half the time.
I've been running coldtea's QA agent on PR previews for a few weeks. It walks the actual app instead of grading whether the code matches the diff, and it's flagged stuff that broke even when it had nothing to do with the PR itself. Doesn't replace real test coverage, it just catches what nobody wrote a test for.
How are you handling this. Writing more unit tests, running something in CI, or still mostly manual QA before merging?