From b879fc70e3926daa51386c73ee20cb2d597511ed Mon Sep 17 00:00:00 2001 From: Michael Weiss Date: Thu, 10 Sep 2026 22:01:35 +0200 Subject: [PATCH] =?UTF-8?q?=F0=9F=93=9D=20Clarify=20expectations=20for=20A?= =?UTF-8?q?I-generated=20PRs=20and=20issues?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Adds guidance on bug report quality, low-impact LLM-found issues, discussing scope before larger PRs, and how we handle large auto-generated contributions. Co-Authored-By: Claude Sonnet 5 --- CONTRIBUTING.md | 7 +++++++ 1 file changed, 7 insertions(+) diff --git a/CONTRIBUTING.md b/CONTRIBUTING.md index d95e2a9..ee736ed 100644 --- a/CONTRIBUTING.md +++ b/CONTRIBUTING.md @@ -14,6 +14,13 @@ We are always looking for people to improve the library. Contributions include, 2. Add sufficient tests to your PRs. 3. Document your code. 4. Don't hesitate to ask questions. +5. For bug reports, a minimal reproduction example plus a short note on real-world impact goes a long way. +6. If an issue was found by an LLM rather than something you actually hit, and isn't security-critical, please only submit it if you can clearly justify a real practical impact — theoretical issues with no real-world relevance add review burden without much benefit. +7. For anything beyond a small fix, open an issue to discuss scope and approach before investing time in an implementation. + +### A note on AI-generated PRs + +Using AI tools is fine, and encouraged — including to self-review your PR before opening it. What doesn't work well is large, auto-generated PRs or issues with no clear human benefit behind them: they're inefficient to review. We reserve the right to close these without merging, even when technically valid, since at that point it's faster for us to just generate a fix ourselves. ### Version 1 vs version 2