discussion Guidelines/best practices for /quality/ agent-assisted development?

I'd like to invite a conversation about code quality and agentic coding.

So I have been using claude heavily for shaping up some emacs packages and managing my config. If I'm attentive and involved, I can produce great code this way; if I'm lazy I can produce vibecoded slop. While I use agents myself, every time I see a published package that is LLM coded, I nope out.

I wonder if a standard set of norms and practices about how to use agents well, a sort of opt-in code of conduct, could strike an interesting balance? Obviously some will be disinterested in any agent-assisted code, but for those whose biggest concern is drive-by vibe coded junk that might not be reliable or last, opening a repo marked as "Coded with LLM norm set X" could be really helpful.

What sorts of norms would you want in something like this? Potential categories could be architecture, planning, pre-write pseudocode, code review and testing. What would you want in such categories? Are there other categories you would want?



https://redd.it/1wz7g8g
@r_emacs