Clojurians Log v2
emacs 2026-09-21
hey there, I am doing some experimenting to improve PR reviewing in the AI age with my fork of code-review https://github.com/ag91/code-review I was wondering if I there are some users here that would like to share ideas. Some of the things I am experimenting: 1. difftastic support 2. tree-sitter support to highlight the most valuable bits of the code (trying constants, variable&fun names) 3. declutter view from AI bot reviews 4. reorder files to show tests first to get an early sense of how the change was evaluated 5. focus mode to hide large AI generated readme 6. review both local and remote changes (for users of AI agents, usually even your own code needs review) 7. clone and open in git worktree a change for code navigation (also tried to bind xref in review) I have seen people delegating review to AI agents as well, but the aim of my changes is to have the user absorbing the most important information to keep building the theory of the program sustainably.
maybe @ag are you still using it?
Hey Andrea! Apologies, I'm not ghosting you - I've been swamped with work and stuff. I saw your PR earlier, about a fix I complained at some point, but I never got time to respond. Yes, I still use the package, your fork specifically. Thank you for taking a good care of it. Although tbh, reviewing code by manually going through the diff happens less and less these days - some LLM-based PRs are far too big to go through each line.
same. I am trying to refine code-review so that reviewing those large PRs can still bring value without wasting too much time into it. At least in my experience devs are losing touch with how software systems actually work, introducing all sort of unexpected behavior. These models generate mostly boiler plate, so hopefully I manage to declutter and prioritize, so on review I can get the most important bits.