Let’s talk about AI slop | Blog

The End of Open Source as We Know It

AI bots arrived and blew up the issue, taking it to 253 comments total, poisoning the conversation with pointless “implementation plans” and even pure aggression toward the maintainers!

AI accounts started flooding not just this issue — but the entire repo. Every sloppy comment triggered a notification for every team member watching the repo. Our GitHub notifications became a wall of noise. Real conversations from contributors like @ethanwater, @developerfred, and @Geetk172 — people actively working on bounties — were getting buried.

One of our team members had to spend half a day every week cleaning AI garbage out of the repo, removing untested PRs and closing hallucinated issues. When we forgot to do so, our repo quickly became a place completely unfriendly to legitimate contributors.

Fighting Back

We’ve decided that we need to fight back and insist on making our repo a comfortable and safe space for legitimate contributors, responsible AI users, newbies, and seasoned engineers.

Today we’re blocking the ability to create issues, open PRs, and leave comments for those who didn’t go through the onboarding.

It’s a nuclear option, yes. It’s especially sensitive for a VC-backed startup that is measured thoroughly by GitHub activity, but we have to pull the trigger: we value quality over quantity. We don’t value metrics pumped by AI slop.

We want Archestra to be a great piece of software that everyone can contribute to, without it being swallowed by AI bots.

Doing It in GitHub

There is no straightforward way to whitelist those who can comment or create PRs on an open source repo, so we had to hack around.

There is a setting called “Limit to prior contributors.” Simple rule: if you haven’t previously committed to main, you can’t comment on issues or PRs.

The setting can’t tell the difference between an AI bot and a real developer who signed up to work on a bounty. Both are “not prior contributors.” Both get locked out.

GitHub defines “prior contributor” as someone whose GitHub account is the author of a commit on main. Git commits have two identity fields — author and committer — and they can be different people.

You can create a commit attributed to someone else using Git’s --author flag. If the email matches their GitHub account, GitHub links the commit to their profile and grants them contributor status.

Every GitHub account has a noreply email: <id>+<username>@users.noreply.github.com. Look up the ID via the API and commit:

gh api users/their-username --jq '.id'

git commit 
  --author="their-username <ID+their-username@users.noreply.github.com>" 
  -m "chore: add their-username to external contributors"

Push to main, and they can comment immediately.

The external user shows up as the author, our account as the committer. That’s all GitHub needs to consider them a prior contributor.

The full flow:

1. Onboarding on our website with ethical AI rules and a CAPTCHA: https://archestra.ai/contributor-onboard
2. A GitHub Action that fires on submission, looks up the user’s GitHub ID, adds their handle to an EXTERNAL_CONTRIBUTORS.md file, and pushes a commit to main authored under their account.
3. The user becomes whitelisted and gets access to the repo.

Final Words

While GitHub reports massive metric growth — a substantial part of which is AI-generated — we as an open source project team have to do the heavy lifting of cleaning up AI slop from our repository and come up with esoteric workarounds to keep the level of legitimacy of our open source audience.

Dear community, it’s time to have a serious talk about the effect AI has on open source.