Outside contributors are being refused CreatePullRequest

[rodriguescarson]

Is there an interaction limit on this repo at the moment? PR creation is being refused from my account:

GraphQL: rodriguescarson does not have the correct permissions to
execute `CreatePullRequest` (createPullRequest)

I can push branches to my fork, open issues, and comment — only PR creation fails. It has been failing since 29 August. #1644, #1645 and #1646 were created from the same fork on 29 August, so it worked before then, and the fork's parent link is intact.

I am not the only one. @ge-al reported the same thing on #1482 today: "GitHub is refusing PR creation from my account on this repo, so linking here until that clears."

What it looks like from outside. Everyone I can see who opened a PR here on 8 September (#1737, #1736, #1735, #1733, #1732, #1731, #1729) is a prior merged contributor. That is the shape a "limit to prior contributors" interaction setting would produce. My own account is not restricted GitHub-wide — I opened and merged 12 PRs on another repo on 8 September.

Why it is worth checking. If it is deliberate, that is completely fine and I will keep posting branches on issues instead — it would just help to know. If it is not deliberate, it is currently stopping first-time outside contributors from opening pull requests at all, which seems contrary to what the progress prizes are trying to encourage. There are at least two of us sitting on finished branches we cannot submit normally:

• ink_detection: native patch retry can enter deterministic cycles and hang workers #1482 — @ge-al's ge-al:fix/native-patch-retry-bounded, and mine at rodriguescarson:fix/native-retry-terminates
• a docs page for vc_grow_seg_from_seed params at rodriguescarson:docs/grow-seg-from-seed-params

Happy to be told it is intentional and move on.

AI-assisted (Claude Code), human-directed.

[rodriguescarson]

This is resolved as of today. CreatePullRequest now succeeds for this account against ScrollPrize/villa: #1771, #1772 and #1773 were all created a few minutes ago from the same fork that was refused for the previous two weeks. Whatever changed, thank you.

One note in case it is useful to whoever looks at this: the three PRs I could not reopen are the same three that the staleness bot closed this morning under the 14-day rule (#1644, #1645, #1646). Reopening a bot-closed PR was refused for this account even after creation started working, so they had to be recreated rather than reopened. That is a small trap for an outside contributor, since the branch and the discussion history end up split across two PR numbers.

Closing this.

[rodriguescarson]

Spoke too soon. CreatePullRequest is refusing my account again today for a docs-only PR from the same fork (branch docs/grow-seg-from-seed-params, one file), with the identical error. It accepted #1771, #1772 and #1773 yesterday. So whatever this is, it's intermittent rather than fixed. Reopening so it's visible.

[rodriguescarson]

One concrete consequence, in case it helps prioritise this: #1776 (the silent normal_grid_level fallback) ended with "happy to send this as a PR", and I couldn't, because CreatePullRequest was refusing the account again that day. #1798 opened the fix instead, which is great for the codebase and exactly how it should work, but it means the block is now costing the repo fixes from the people who found the bugs. The three PRs I could open (#1771, #1772, #1773) only exist because the refusal lifted for one day.

If someone with admin on the repo could check the interaction limits or whatever is gating outside contributors, I'd be grateful. Happy to test again on request.

[CVasilopoulos]

I think this is GitHub's open pull request limit, not a permissions problem.

I hit the same does not have the correct permissions to execute CreatePullRequest error today from gh pr create. Opening the compare page in the browser for the same branch shows the actual reason:

You've reached the open pull request limit for this repository
Users without write access can have up to 3 pull requests open at a time. Merge or close your open pull requests to submit a new one. Drafts don't count toward the limit.

That matches everyone in this thread. Right now @rodriguescarson, @ge-al, @Bullo27, @mithilp, @jonmarrs and I each have exactly 3 open PRs here. It also explains the pattern above: PR creation came back after the bot closed #1644, #1645 and #1646, allowed exactly three new PRs (#1771, #1772, #1773), and then stopped again.

Two things that help while it's in place:

• gh pr create --draft (or "Create draft pull request" in the web UI) works, because drafts don't count. I opened vesuvius.train: generate configured auxiliary targets instead of training them against nothing #1799 that way. It can be marked ready once one of the other three is merged or closed.
• The web compare page shows the limit message. gh only reports it as a permissions error.

If the limit is intended, it might be worth a line in CONTRIBUTING.md so people don't lose time on it.

AI-assisted (Claude Code), human-directed.

[pmh47]

Yes this is to avoid people spamming too many random AI-generated PRs (which typically come from asking claude to hunt for issues, rather than real users).