Conversation
|
|
||
| ## Maintainer and Release Context | ||
|
|
||
| Most agents should not perform maintainer-only operations unless explicitly asked. |
There was a problem hiding this comment.
Should this explicitly list the maintainer-only operations that are forbidden? I'm assuming the intent is to keep agents from submitting PRs without human oversight?
There was a problem hiding this comment.
First of all, the current version of the AGENTS.md file was generated entirely by Codex, so some wording may not be ideal.
I actually think it's fine to let agents make whatever changes they consider appropriate. The responsibility is then on us, as maintainers, to carefully review the resulting PRs and merge them manually.
There was a problem hiding this comment.
My concern is that agents will begin sending in excessive PRs that aren't necessarily constructive; I obviously don't think it will be as bad as on really popular packages like curl, but I think that agent-submitted PRs could lead to extra work for volunteer maintainers.
In my own experience with using AI to suggest code improvements, I find that it makes a lot of style suggestions that don't actually fix bugs, add features, or meaningfully improve documentation without more specific direction. My concern is that there would be PRs that would simply suggest renaming variables/functions or other similar changes that would take away from reviewers' time.
Also, as someone who came into open-source development by finding the low-hanging fruit on a PyGMT, my concern is that agents will consume all of the fixes that used to be tagged as "good first issue" and limit the chances for newcomers to the project to meaningfully contribute without extensive experience.
2 participants
Initial draft done by codex.
Address #4677