This pull request is part of a Mergify stack:

Merge Protections

🟢 All 6 merge protections satisfied — ready to merge.

This stack answers three of the four docs asks filed on 2026-08-17 (Mergifyio/ci-bot#369, #370, #372). The fourth, Mergifyio/ci-bot#366, turned out not to be a docs change and is Mergifyio/monorepo#38775 instead.

#366 asked whether this repo should grow a section on how a ruleset bypass mode interacts with condition injection, taking the #bypass-actors-and-injection anchor that five engine messages already link to, or stay silent and have those links repointed at the section that exists.

I read the flag that decides it. FORCE_EXEMPTED_RULESET_CONDITIONS_INJECTION landed six days ago, is opt-in per org with no grandfathering, and turns off the skip one org at a time; its own note says it is temporary and disappears once every org is enabled. So the destination is known but almost nobody is there yet. Writing "exempt no longer skips injection" would be wrong for nearly every reader today, and writing the current behavior would push customers onto something that reverses under them without warning. The reversible half is to point the links at the section that does exist, which is also the section those messages are actually about, so that is what Mergifyio/monorepo#38775 does. When the rollout finishes, the docs get a section on injection and the links can move again.

One thing that fix deliberately does not settle, noted on that PR: the branch_protection_injection_mode field description still tells customers that exempt is how you skip injection for one ruleset, which is the behavior being removed.

Revision history

Restacked only — c604f0e → 10ece46 (compare). The content of this commit is unchanged; it moved because #12415 at the bottom of the stack was amended to address a review comment.

Restacked only — 10ece46 → 0ec2b78 (compare). The content of this commit is unchanged; it moved because the base commit (#12415) was amended and the stack rebased on main.

@jd this pull request is now in conflict 😩

Rebased onto main and resolved the conflict 0ec2b78 → 3d4d468 (compare).

#12439 landed the id: tests half of this change across the same recipes while this sat open, which is what conflicted. I kept the two continue-on-error: true lines it did not add — junit.mdx (Gradle) and minitest.mdx (minitest-reporters) — because the quarantine partial these recipes include needs the test step to keep the job alive, otherwise a failing test run ends before the upload step and produces no report at all.

What is left in this PR is therefore the dashboard link fix (all eleven recipes, via the new _review-in-test-insights.mdx partial) plus those two lines. The commit message also credited #12411 for the overlap; it was #12439, and that is corrected.

Rebased onto main and rewrote this one, because #12490 landed the same link fix while this sat here — 3d4d468 → a2f4cc7 (compare).

#12490 corrected all eleven closing links in place, to the same page this PR was pointing them at (/test-insights/detection). So the dead link is fixed on main already and I dropped the Fixes keyword — I'll close the ask against #12490 rather than against this.

What is left here is the part #12490 could not do, and I think it is still worth landing:

• The sentence becomes one partial. _review-in-test-insights.mdx, included by the recipes. The rendered text is byte-identical to what docs(agent): point the Test Insights dashboard link at a page that exists #12490 shipped — I resolved every conflict to the partial and checked the built HTML. The reason to keep it is what just happened: one stale URL became ten broken links precisely because the paragraph existed eleven times, and an eleventh copy had already drifted to having no link at all. Next time that link moves it is one edit. This is also how these files already handle the Buildkite quarantine setup and the upload steps, so it is not a new pattern.
• Two continue-on-error: true lines that are still missing on main — the Gradle example in the JUnit recipe and the rake test example in the Minitest one. Both include the quarantine partial, which needs it: without it a failing test step ends the job before the upload runs, so the run that most needs a report produces none. In both files the sibling example directly above already has the line, which is what makes me read it as drift rather than intent.

If you would rather not carry the partial, the two continue-on-error lines are the only part that has to land and I'll cut it down to those.

Merge Queue Status

• ✅ Entered queue — 2026-08-24 11:52 UTC · Rule: default · triggered by merge protections
• ✅ Checks passed · on draft merge queue: checking main (bc9c489) and #12417 together #12502
• ✅ Merged — 2026-08-24 11:55 UTC · at a2f4cc7ad58ccd915400230cfb1837397a711486 · squash

This pull request spent 3 minutes 24 seconds in the queue, including 2 minutes 51 seconds running CI.