Archestra blocks AI slop on GitHub by abusing Git's --author flag

Archestra’s open source repo was overrun by AI-generated noise: a $900 bounty issue ballooned to 253 comments of hallucinated implementation plans, a single feature request drew 27 mostly-untested PRs, and a team member was losing half a day each week pruning garbage. Homegrown defenses — a reputation bot and an ‘AI sheriff’ that mistakenly closed legitimate PRs — failed to stem the flow, burying real contributors under notification spam.

The team flipped GitHub’s ‘Limit to prior contributors’ setting, which blocks anyone who hasn’t committed to main from opening issues, PRs, or comments. To onboard real humans through that wall, they exploit Git’s separation of author and committer identities: after a contributor clears a CAPTCHA-gated onboarding form, a GitHub Action looks up their numeric ID, then pushes a commit to an EXTERNAL_CONTRIBUTORS.md file with —author set to the contributor’s noreply email. GitHub treats them as a prior contributor and unlocks the repo.

The author frames this as a deliberate quality-over-metrics stance — painful for a VC-backed startup judged on GitHub activity — and warns the problem is more than annoyance. He cites the LiteLLM incident where attackers used AI bots to steer repo conversations, arguing AI slop is now a supply-chain risk to open source, not just a moderation headache.