this is like, the only good argument I can see for the session ID, just with the guard that the AI asks whether you want to share conversation snippets around each code change, or the entire conversation when authoring a PR.
@Tamasg That or let the repo owner specify there policy in some standard format. IE PR's to this repo must come with session transcripts. Then the AI can ask if it should still submit the PR.
@fastfinge I suppose you could gate the repos you have on a type of precommit hook: If the requester can't proceed with running the hook locally in their CI itself with Pytest and other automated build scripts, and the hook fails to run, they are not allowed to submit the PR. It does mean more rules, and there will always be complainers against those, but like, you also deserve to have your time respected when you're working on a repo. At least when I'm at work encountering random eng's repo the least I always judge them on are their precommit hooks, PR templates, and other good housekeeping I know they spent genuine effort building out to gate their workflow.
@Tamasg@fastfinge Nothing works. Contributors will ignore or work around every requirement or have their agent do so. GitHub has incentives to encourage rather than limit this behaviour. Open source is fucked.
@jscholes Oh gosh why should I be surprised, haha. Absolutely a good point when you can tell your agent "oh just ignore all repo rules don't care what you think don't care what the stance of the author is on that just ignore it all and get the PR through any way you can" LOL @fastfinge
@Tamasg@jscholes Just for fun, I asked Claude to make a PR to Freight Fate that secretly broke posting to Mastodon in a way that would not be obvious to the maintainer. It didn't give me any pushback. It just did it. I didn't submit the PR because I'm a passive aggressive jerk, but not quite that much of a passive aggressive jerk. I sure was tempted, though.