I've been working on a side project where I created a build wrapper that automatically diagnoses CI build failures, applies fixes, and retries without any human intervention. Rather than waiting until someone wakes up at 2 AM to fix a failed build, my tool could potentially handle it. For example, if a build fails due to missing dependencies or linker errors, it attempts to fix those issues automatically. However, I have a few concerns and would appreciate feedback from those familiar with CI/CD pipelines:
1. Am I solving a real problem here, or do teams just usually fix CI configs once and forget about it?
2. What are the security concerns with having a tool that can auto-install packages in CI?
3. Is there a risk of adding more complexity by trying to support different build systems?
I think it could be useful for new projects that struggle with initial CI setup, open-source projects where contributors have varied environments, and to mitigate late-night pipeline failures. But I'm not looking to replace thorough CI management or make decisions over a human expert's knowledge of the codebase. What do you think? Would you use this, and how could I make it trustworthy in a real pipeline?
5 Answers
It’s interesting, but it kinda feels like a solution looking for a problem. CI should ideally mirror production setups, and automatic fixes might lead to non-reproducible builds or incorrect fixes. While your idea seems useful for initial setups, the risk of misuse might outweigh benefits in ongoing scenarios. Just some concerns to consider!
For me, this isn't really necessary; CI build failures are pretty rare in my experience. Most of the time, if something goes wrong, I can fix it in just a few minutes myself. It could depend on the programming language and build tools you're using, though.
This sounds like it could be helpful, but I personally don't trust tools that make changes to code automatically, especially if they're using AI to generate fixes. A better approach might be generating a pull request for any suggested fixes instead of applying them right away.
This isn't something I would use personally.
I kind of like this idea! Many CI failures aren’t real bugs—they're often just setup issues. While teams usually implement tools for build speed, having something that can heal itself might be a fresh take.
Totally agree! I've seen AI-generated code work in some cases, but it often doesn't pass linting or security checks, so keeping a human in the loop is essential.