You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
🌱 Fork PRs that touch the pipeline wait on a maintainer push before Canary can pass (6 open now) #369
Six open PRs can't pass the required Canary check until a maintainer pushes their branch into this repo, and most of them come from Hive agent forks that will keep hitting this.
@hanthor, raising this with you since you built the canary. I think the guard is doing what it was designed to do. What I'm asking about is the queue it creates.
What a contributor sees today
A fork PR that changes a pipeline path gets a red Canary. The job log ends with:
##[error]a fork pull request cannot publish the canary image; a maintainer must push this branch to the repository
Canary is a required status check on main:
$ gh api repos/projectbluefin/utah-packages/rules/branches/main --jq '.[]|select(.type=="required_status_checks").parameters.required_status_checks'
[{"context":"Canary","integration_id":15368}]
So these PRs cannot merge until someone with write access pushes the branch here. A fork PR that does not touch the pipeline passes. For example, #368's canary logged "no pipeline path changed; nothing to prove".
Which PRs are waiting (checked 2026-10-04 ~02:00Z)
All six also have a change request from you, so the canary is the second gate on each, not the only one.
Why I think it will keep happening
The Hive contributor lanes work from forks with pull-only tokens, so every lane PR that touches the pipeline will land in this state. The guard is right that a fork can't get packages: write. The cost is that each of these PRs needs a maintainer to push its branch by hand, once per head.
Options I can see (untested; you know the constraints better than I do)
Keep it as is. A maintainer pushes the branch for a PR once the other review is done. This is fine if the volume stays low.
Add a maintainer-run workflow_dispatch that copies a PR head into an in-repo branch (for example canary/pr-<N>) and dispatches the canary there. That would turn a manual push into one button. I haven't checked how the result would attach to the fork PR's head, so this needs your read.
Let a fork PR's canary run the stages that need no publish, and pass if they pass. The publishing stages would be left to the in-repo run at merge time. This weakens what "Canary green" means on a PR, so it may not be what you want.
Open question
Is option 1 the intended long-term flow? If it is, I'd suggest saying so in the error message or the contributor docs, so lanes stop asking for a push on each PR.
Current review reproduces this blocker for #432 and #433: Canary runs 38030238130 and 38048828235 fail with the fork-publication guard before package builds. Static validations are green, but no new recipe build/publish evidence exists. Those PRs remain unmerged; an admin bypass would remove a real validation boundary here rather than establish readiness. A trusted same-repository run is still needed.
@hanthor update with today's numbers, since the volume this issue worried about has arrived.
On 2026-10-25 the 38 open parity issues you had queued in 3-clanker-queue were also labelled triage/accepted so the hive's contributor agents can work them. They're now producing PRs, and they all come from forks. As of 16:50Z, 13 fork PRs are open and every one is blocked by the Canary fork-publication guard ("a fork pull request cannot publish the canary image; a maintainer must push this branch to the repository"):
The hive sends these back to the contributor as "failing checks" (072), but no contributor can fix this from a fork, so the PRs just sit there and the queue grows with every parity issue picked up.
Option 1 (a maintainer pushes each branch by hand) now means a dozen-plus pushes, and again for every new head.
Could you pick one of the three options above? Option 2 (a maintainer-run workflow that copies a PR head to an in-repo canary/pr-<N> branch) looked like the least risky. Until then, if you'd rather not have the agents produce more PRs that can't pass, I can pull triage/accepted off the remaining queued parity issues; just say so.
Decision: option 2. Opened #619: canary-fork-pr.yml, a maintainer-run workflow (Actions tab -> "Canary for a fork pull request", inputs pr and the full head_sha that was reviewed). It copies exactly that commit to canary/pr-<N> and dispatches the real Canary there, so the Canary check is created on the fork PR's own head commit. It refuses if the head moved since review, if the PR is not open against main, or if it changes .github/ (workflow changes stay a hand push). Nothing from the fork is run in the copying workflow. The dispatch names no pr input so the PR's own (older) canary.yml takes the same-repository path; that means it works for #600-#618 as they stand.
Why not the others: option 1 does not scale to the 13 PRs above, and option 3 would make "Canary green" on a PR mean less than it does on main.
What a maintainer still has to do per PR: review the diff, then press the button. The canary run builds the fork's code with packages: write, so that review is the boundary; I have not pressed it on any of the 13.
Not yet verified (needs #619 merged, because workflow_dispatch only runs from the default branch): that the dispatched green Canary supersedes the fork's own red pull_request Canary run for the required-check rule. First real use will show it, and #619's docs say so. For the hive: the guard's message now tells authors this is expected and not theirs to fix, so "returned to contributor (072) as failing checks" on this specific error should be treated as waiting-on-maintainer, not a contributor action.
Done: option 2 shipped in #619 (merged 2026-10-25), a maintainer-run "Canary for a fork pull request" workflow. Fork PRs that need Canary (e.g. #753) now go through that workflow.
Activity
hivecommons-hive commentedon Oct 4, 2026
What to reply
/hive approve: go with the recommendation/hive decision <your instructions>: go ahead, following your instructions/hive help: show these againIf this issue is waiting only for reporter confirmation after a merged fix, the reporter or a maintainer can reply
/fixedinstead.Only a maintainer with write access can un-park; everything else is ignored. Full command reference: https://github.com/hivecommons/hive/blob/v5/src/docs/maintainer-commands.md.
hanthor commentedon Oct 10, 2026
Current review reproduces this blocker for #432 and #433: Canary runs 38030238130 and 38048828235 fail with the fork-publication guard before package builds. Static validations are green, but no new recipe build/publish evidence exists. Those PRs remain unmerged; an admin bypass would remove a real validation boundary here rather than establish readiness. A trusted same-repository run is still needed.
Reviewed with Codex.
Danathar commentedon Oct 25, 2026
@hanthor update with today's numbers, since the volume this issue worried about has arrived.
On 2026-10-25 the 38 open parity issues you had queued in
3-clanker-queuewere also labelledtriage/acceptedso the hive's contributor agents can work them. They're now producing PRs, and they all come from forks. As of 16:50Z, 13 fork PRs are open and every one is blocked by the Canary fork-publication guard ("a fork pull request cannot publish the canary image; a maintainer must push this branch to the repository"):072): feat: add support for camera RAW image files #618, feat: add the remaining braille document converters #616, feat: add Bison for building language parsers #615, feat: add missing compatibility libraries for development tools #614, feat: add Python libraries for dates and older software #613, feat: add a tool for automated remote logins #611, fix: honor the decision against bulk font additions #610 (mendezr), import: codecs-media batch (libultrahdr, svt-vp9-libs) #609 (kylerankin)074): feat(long-tail): import 6 source packages, resolve hyperv subpackages (#500) #607, fix(sources): re-point 27 Fedora-primary locks at their upstream release #608085): fix: bootstrap tool no longer wipes the source list on a rejected import #600, fix: keep package update history current #601; stalled in review: feat: add libraries for isolated containers on Hummingbird #602Two side effects worth knowing:
072), but no contributor can fix this from a fork, so the PRs just sit there and the queue grows with every parity issue picked up.Could you pick one of the three options above? Option 2 (a maintainer-run workflow that copies a PR head to an in-repo
canary/pr-<N>branch) looked like the least risky. Until then, if you'd rather not have the agents produce more PRs that can't pass, I can pulltriage/acceptedoff the remaining queued parity issues; just say so.hanthor commentedon Oct 25, 2026
Decision: option 2. Opened #619:
canary-fork-pr.yml, a maintainer-run workflow (Actions tab -> "Canary for a fork pull request", inputsprand the fullhead_shathat was reviewed). It copies exactly that commit tocanary/pr-<N>and dispatches the real Canary there, so theCanarycheck is created on the fork PR's own head commit. It refuses if the head moved since review, if the PR is not open against main, or if it changes.github/(workflow changes stay a hand push). Nothing from the fork is run in the copying workflow. The dispatch names noprinput so the PR's own (older)canary.ymltakes the same-repository path; that means it works for #600-#618 as they stand.Why not the others: option 1 does not scale to the 13 PRs above, and option 3 would make "Canary green" on a PR mean less than it does on main.
What a maintainer still has to do per PR: review the diff, then press the button. The canary run builds the fork's code with
packages: write, so that review is the boundary; I have not pressed it on any of the 13.Not yet verified (needs #619 merged, because
workflow_dispatchonly runs from the default branch): that the dispatched greenCanarysupersedes the fork's own redpull_requestCanary run for the required-check rule. First real use will show it, and #619's docs say so. For the hive: the guard's message now tells authors this is expected and not theirs to fix, so "returned to contributor (072) as failing checks" on this specific error should be treated as waiting-on-maintainer, not a contributor action.ci(canary): let a maintainer run Canary on a reviewed fork pull reque…
fix(canary): a pull request that only adds recipes and inventory pass…
Danathar commentedon Oct 28, 2026
Done: option 2 shipped in #619 (merged 2026-10-25), a maintainer-run "Canary for a fork pull request" workflow. Fork PRs that need Canary (e.g. #753) now go through that workflow.