Post

open source projects may need to lock PRs to trusted folks. careless slop is sucking the life out of maintainers. we need better tools to talk to and onboard new contributors while interacting on issues. more friction could actually help new folks who care stand out, and let the community breathe

9 reposts118 likes

2 questions 1) How would you like to see new people get involved in open source in a helpful, positive, and productive manner? 2) What tools would you like to see?

new folks are helpful to a project when they put real effort on whatever they want to bring to the table. It can be helping others in the community, sharing their experience to improve things, and if they want to fix issues or implement features, understand that generating code isn't the bottleneck

patak@patak.cat

The changed code is a small fraction of an open source contribution. Your commitment to understand the issue, how your proposed solution fits with the project, and be ready to own and push the review process forward is the biggest chunk of the work. Your effort is the contribution, not the code.

If PRs needs to be locked, there should be new tools to let people into the project in an easy way, and that let maintainers interact with these potential contributors (chat, calls, more conversational focus on issues) so they're able to get to know them and move them to a trusted party

I just started hosting my own projects on my own private infrastructure, and now embed a source.zip in my distributed packages, for the people who want to extend/hack on the code itself. Maximum friction, maximum FOSS.

Haven’t tried it, but I'd like to: community reports bugs via issues and feature requests via discussions/issues, in their own simple words/ideas (AI-generated reports get closed). Maintainers review/accept ideas, then use AI to turn them into detailed issues. Only maintainers create PRs.

I fully agree, but after having done this myself, I have found people still try to provide AI slop via an issue with a link to the diff of their fork and the main repo, and mentioning they can't create a PR. This means the noise is still there, sadly. Have you found a way to prevent this?

That's a bummer. I don't know how we fight that. We need better tools, an issues sandbox for a start. Better comms tools to talk to these contributors about what they're trying to do and not their generated code.

Better tools are definitely needed. I'm starting to think GitHub might not be the right platform anymore, because they are too invested in AI to really provide tooling for this kind of mindset. I have strongly considered Codeberg, but the migration is very time-consuming.

The flue project has an interesting take. It doesn't accept PRs (except for a few). PRs are transformed into discussions and then iterated by an agent and supervised by the owner of the project.

The motivation: "If nowadays people just send automated PRs, at this point I prefer to automate them myself so I can overview things"

3 times now I've seen an author (very obviously an agent acting on behalf of the account owner) close its own PR after it realized it had violated the AI policy. Overall those PRs were a waste of time, but it does make me think that writing docs for agents can possibly help stop low-effort spam PRs.

No, I haven't seen that, but that site certainly seems helpful! I've also had a bit of luck writing an agents.md file that tells agents to add a comment that a human should later remove. It's simple, but so far it's helped to quickly identify code that has not been reviewed by the human user.

the vitest team decided to lock PRs

Vitest@vitest.dev

We are updating our AI Policy to reduce maintenance stress. All pull requests are now automatically closed, unless they are created by maintainers or users from an approved list of contributors. The PR is not lost. If the team decides the change is worth pursuing, we will reopen it.

didn’t lock it, it’s more of an approve-first approach rather than just leaving all PRs open for an eventual review PRs are still there, they are just closed

It is conceptually locking it to me. It would be nice to have a PR sandbox with all these. I imagine you do that already with a label.

ya, but I understand them here, hopefully tooling evolves to allow for better flows in the future

we were talking about this with @vladimir.berlin and @bluwy.me. i've always pushed back against locking PRs. i feared it would make the barrier higher for new folks to join open source communities. but if their work is covered by noise, other avenues than PRs to gain trust may end up helping them

My current hatred is slop accounts just targeting any "good first issue" or "help wanted" labelled issue. They usually have zero interest in actually working with the project. I'm not sure what the grift is yet though

I think the main grift is still generating activity to use GitHub when looking for a job. Some of it may be people that think they are actually helping, but I don't think they would continue after so many maintainers tell them to stop, though.

I'm wondering whether it's actually to gain CI access on a broad range of repositories? I don't think anyone cares about GitHub activity in the age of AI?

maintainers are starting to ignore PRs altogether as the chance to find a real caring human on the other side is diminishing. being ignored is worse for new folks than having to go through a longer process to be able to contribute. communities will adapt, here hoping towards more human connection

By now I'd prefer a well written issue-description without PR. Let my own coding-agent write the implementation. Mixing pre- and post- agentic-AI workflows can cause a lot of friction.

I've rarely found an agent that can't assimilate good ideas where it finds them. Start yours without looking at the other solution! But double back, have some tokens to compare, see what to take forward.

yes, the value of a PR isn't in the generated code

patak@patak.cat

The changed code is a small fraction of an open source contribution. Your commitment to understand the issue, how your proposed solution fits with the project, and be ready to own and push the review process forward is the biggest chunk of the work. Your effort is the contribution, not the code.