In the old days, we built things like lintian-brush and debcraft and called them "expert systems". They are "rule based" good old fashioned AI (GOFAI). And as we can see from all the little issues, it's a pretty brittle way to build systems of this sort. Basically you end up playing a lot of whack-a-mole with no end in sight. And for some sociological reason people get into disagreements and build parallel systems based on what seem externally like minor technical issues. Like, should debcraft and lintian-brush *really* be separate programs? I know the authors can make a case, but come on.
Times have changed, and now we use machine learning and agentic AI and fancy LLMs and all this cool stuff. All the functionality of lintian-brush and debcraft can be subsumed into some instructions for an LLM-based coding system. And you can tell it when what you want isn't its default and it will listen. Push your repo onto github and tell github copilot "update the debian packaging information according to the most modern policies and up-to-date standards and address any and all issues, then fetch the bugs.debian.org page for this source package and address each bug one-by-one, being sure to mark bugs closed in debian/changelog, then fetch the debian package tracker page for this package and address any issues listed there". It's not perfect but it's pretty darn remarkable. Will it do everything our lovingly hand-crafted tools can? Give it a shot; try it on your debcraft and lintian-brush test suites.
