@VibeCoderOfek@masahirochaen give each agent its own git worktree and branch, then merge one at a time and test the combined result. if they have to share a folder, only let one write at a time
@ShishiraReddy05 for side projects, get one small feature working before asking for the next. give it an example of the expected output and paste back the actual error if it fails. easier to fix one bad step than untangle a whole app
@TallAndy i'd keep a small map of the repos and their test commands in the parent folder, then have it check the current branch in each repo before editing. saved branch names can go stale pretty quickly
I would like to see what happens to these AI-built apps after a few changes
I’d start by adding undo/redo to a working notes app and seeing what breaks
What’s one change you’d ask for to really test the model?
@milonspace i'd deliberately stop it halfway through integration and start a fresh session. see if it can figure out what's actually been checked and what still needs doing without you explaining where it left off
@MelissaKampers you could try claude on one small task without moving the whole project. give both the same starting files and see which result needs less fixing