[devin-ai-integration]

What and why

The develop ruleset requires the checks Changed areas and Build and test, both produced by .github/workflows/pr.yml. With a merge queue on the branch, GitHub builds a commit on refs/heads/gh-readonly-queue/develop/pr-<n>-<base-sha> for each enqueued pull request and asks for the required checks on it through the merge_group event — but pr.yml only listened for pull_request, so nothing ran on the queue commit, nothing reported, and the entry sat until the status-check timeout dequeued it as failed (PR #1580: queue commit created, no workflow run).

pr.yml now triggers on merge_group (checks_requested) as well, and every place that read the pull-request payload has a merge-group branch:

• the concurrency group is merge-group-<ref> for a queue commit (each entry has its own ref) and pr-<number> for a pull request;
• Changed areas treats a queue commit as a ready pull request — DRAFT and the ci:full label are both false, so the full tier runs — and computes the areas from the merge group's own diff, merge_group.base_sha..merge_group.head_sha, through the same scripts/ci-changed-areas.sh, so a queue commit gets exactly the checks a ready pull request with the same diff gets;
• the protobuf wire-compatibility baseline is the merge group's base_ref (refs/heads/develop, stripped to the branch name) where a pull request uses github.base_ref;
• the binaries are stamped VERSION=merge-queue (with the commit) instead of pr-<number>.

Build and test stays the single aggregate (if: always(), same needs) and reports on the queue commit.

The other workflows were audited: the ruleset's required checks are exactly the two pr.yml emits, and no other workflow gates pull requests, so none of them changes.

Also: tests/hygiene/merge_group_workflow_test.go holds the workflow to the trigger, the merge-group branches of Changed areas (diff range and draft flag), the two concurrency-group forms and the unconditional aggregate; CONTRIBUTING.md gains a paragraph on the queue next to the quick/full tier description; changelog fragment changes/unreleased/merge-group-ci.changed.md.

The merge queue is currently switched off in the develop ruleset. The change is inert while it is off and removes the stall if the queue is re-enabled.

How it was verified

• actionlint .github/workflows/pr.yml: no finding introduced. The two it reports (shellcheck SC2129 style, SC2034 warning) are in untouched run: blocks and are reported identically on develop.
• go test ./tests/hygiene -run TestPRWorkflow → ok. Against develop's pr.yml (the test file alone checked out over origin/develop) the five tests fail: no merge_group trigger, pull-request-only concurrency key, unguarded DRAFT/label reads, no merge-group diff range, unguarded payload reads.
• go build ./... && go vet ./... && gofmt -l . → clean, gofmt -l . empty. python3 scripts/changelog.py check passes.
• Dry evaluation of the changes job's shell for the merge_group case: the tier step with DRAFT=false CI_FULL_LABEL=false yields tier=full; a queue-shaped commit (origin/develop merged with this branch in a detached worktree) run through scripts/ci-changed-areas.sh <base_sha> <head_sha> with CI_TIER=full produced the same area outputs (go_shards=["runtime","runtime-corpus","model","export","graph","codegen","rest","migrate","cli"], …) as the pull-request form scripts/ci-changed-areas.sh origin/develop HEAD on this branch — byte-identical.

What this PR could not test

A real merge_group run cannot be triggered from this pull request: the queue runs the workflow that is on develop, and this is the change that makes develop's workflow listen. After merging, and once the queue is on again, enqueue a trivial pull request and confirm a PR workflow run appears for the gh-readonly-queue/develop/pr-<n>-... ref with Changed areas and Build and test reporting on the queue commit.

Ruleset settings to make when re-enabling the queue (GitHub UI, no CLI): repository Settings → Rules → Rulesets → develop → Require merge queue. Keep the required status checks exactly Changed areas and Build and test, and set Status check timeout to at least 120 minutes — of the last 40 green pr.yml runs, 10 exceeded 60 minutes (longest 92), so the previous 60-minute timeout would dequeue slow-but-passing entries as failed.

Checklist

• make test and make lint pass locally (go build, go vet, gofmt -l, the hygiene suite; CI runs the rest)
• Tests added or updated for the change
• Documentation extended where it already covers the surface (see CONTRIBUTING.md)
• Changelog entry added as changes/unreleased/<slug>.<section>.md, not as an edit to CHANGELOG.md
• No internal work-item labels in the body, docs, or changelog

Link to Devin session: https://nasa-jpl-demo.devinenterprise.com/sessions/41bd27e6714f4b5ca3c4fac9f92bb1fb
Open in Devin Desktop: https://nasa-jpl-demo.devinenterprise.com/desktop/session/41bd27e6714f4b5ca3c4fac9f92bb1fb?variant=devin
Requested by: @HuiJun

[devin-ai-integration]

I'll fix CI failures and address comments from users with write access. I'll skip comments containing "(aside)".

• Disable automatic comment, CI, and merge conflict monitoring