Thanks for putting this up Ash, I think the brown m&m is a cheap check to start with.
To add on to that idea, what if we introduce a soft workflow on pull_request_target that just posts a comment when the brown m&m test fails. We may / may not fail the build, and just nudge the author to re-read the diff, walk our code-review checklist, and that probably would indicate them to fix the brown m&m test. If we wanna be a bit more cautious to reduce the bad quality PRs from unknown actors (thus far), we can consider having a first-time contributor gate. A lot of the slop is first PRs, so trigger the heavier checks only there and leave known contributors alone. Thanks & Regards, Amogh Desai On Tue, Jun 9, 2026 at 10:15 AM 陳弘逸 <[email protected]> wrote: > I think the core issue isn't that AI is automatically creating PRs, > but rather that the human submitters behind them aren't actually > reviewing the diffs. If a human has thoroughly reviewed all the > changes, then even if the PR itself was submitted by an AI agent, it > doesn't necessarily mean it's a "slop" PR. > > However, I do agree with Shahar's suggestion—perhaps we could > implement a simple AI review step in our CI workflow to quickly filter > out the completely useless PRs. > > Thanks, > Henry > > --------------------------------------------------------------------- > To unsubscribe, e-mail: [email protected] > For additional commands, e-mail: [email protected] > >
