fix: run the two test-apps that turbo has never seen

[gitKrystan]

custom-root-url and markdown-and-gjs-md-app name their browser-test script _test:browser, so pnpm turbo test:browser has never matched either package. At first, this appears to have been intentional, but they appear to have real tests in them now that aren't running in CI. This PR renames the scripts and fixes what that turns up.

Why now

8333e36 noted "Omit not implemented tests from running yet", at the end of #264. Both apps were new in that PR, and custom-root-url's crawl really was a stub: a commented-out body behind an assert.ok(true). Both have had real tests for a while.

The gap costs something. For example, #369 rewrote markdown-and-gjs-md-app's crawl snapshot, and three of its six entries are markdown-only's URLs rather than this app's extensionless .gjs.md ones, with /my-folder-name/baz missing.

Worth a look

• markdown-and-gjs-md-app also spelled its build build:test, which is not a turbo task. test:browser dependsOn build:tests, so renaming just the test script may have handed testem an unbuilt dist/. Both are renamed.
  • That app never wired handlePotentialIndexVisit, so every run logged Page not found for path "/Docs" ten times and passed regardless. It has app/routes/page.ts now and a test that asserts the redirect. The crawl cannot catch this by itself: it only records paths, and /Docs's redirect target still starts with /Docs, which is the whole of what the crawler compares.
• markdown-only has the same missing hook and logs the same error. I left it alone, since it is green and running today, but it might need a fix in follow-up.

Claude spew: audit of each failure and what fixed it

Five failures, one silent pass, and one CI-only flake. Only one was a bug in the code under test.

Stale assertions: all five failures. Every one traces to 0dab708, which moved the Home group's nav link to the app root. The behavior that commit changed was the bug, and these tests still asserted the old href. Updating them lines up with the assertion 0dab708 already added to multiple-docs-routes.

Test	Asserted	Now	Same property?
nav-active, Documentation page	.../Home not active	/my-github-project/ not active	yes
nav-active, Home page	.../Home is active	/my-github-project/ is active	yes, plus a new /Home doesNotExist guard
nav-active, casing	.../Home not active	/my-github-project/ not active	yes
custom-root-url crawl	/Home reachable	/my-github-project/ reachable	yes
markdown-and-gjs-md-app crawl	/Home reachable, ten times	absent	yes

Nothing was loosened to make these pass. The one edit that looks like it might have been is deleting the KNOWN_REDIRECTS entry for /my-github-project/Home, which used to pin that visit to bar.md. That entry was already dead: the link is /my-github-project/ now, so the lookup missed and expected fell back to the href either way. The redirect itself is still pinned by index-redirect-test.ts, which visits /Home and asserts it lands on /my-folder-name/bar.md.

One real bug, which was not among the failures. markdown-and-gjs-md-app's /Docs group root errored on every run while the suite reported green. visitAllLinks compares URLs rather than checking that a page rendered, and an erroring page keeps its URL, so current.startsWith(expected) held. Wiring handlePotentialIndexVisit is the only change here that touches behavior rather than assertions.

One upstream bug, visible only on CI. The first CI run failed on Must pass an element, selector, or descriptor to click, the revisit flake 0d52f50 documented, in test-support rather than in kolay. The workaround for it is reverted now that 0.9.0 carries the real fix.

Claude spew: Why the two crawl snapshots are written differently

Both crawls now assert deepEqual(visited.sort(), [...]), per @NullVoxPopuli's review.

markdown-and-gjs-md-app is sorted rather than re-recording its visit order. Its old list disagreed with reality by more than the ten /Home visits, and pinning a fresh order would pin exactly the timing artifacts 0d52f50 called out. That app's runtime-compiled .gjs.md pages are the loosest in the repo on ordering.

test-support 0.9.0, which arrived with #369 partway through this branch, dedupes by target itself, so neither crawl needs its own Set any more.

An earlier push here skipped custom-root-url's crawl, because CI hit the revisit flake 0d52f50 left open. 0.9.0 fixed that too: it visits by URL instead of clicking an anchor unless asked for mode: 'click', and the flake lived entirely in the click path. The workaround is reverted and the crawl runs, in 57ms rather than 440ms. docs-app may no longer need its own skipAllLinks either, which I have not tested.

Testing

CI should be running all of these tests now. Locally, one app at a time:

pnpm build                                   # once, at the root
cd test-apps/<app> && pnpm build:tests && pnpm test:browser

Run them one at a time. pnpm turbo test:ember across every app in parallel starves them into 120s browser timeouts.

To see the bug this fixes, check out main and try pnpm test:ember in either app.

_{AI attribution: 🤖 Claude using Opus 5. Used skills including pr-description and humanizer. The code changes have been reviewed by the author.}

[bolt-new-by-stackblitz]

Run & review this pull request in StackBlitz Codeflow.

[vercel]

@gitKrystan is attempting to deploy a commit to the universal-ember Team on Vercel.

A member of the Team first needs to authorize it.

[NullVoxPopuli]

this is a little silly, just do assert.deepEqual(visit.sort(), [...]);

[gitKrystan]

deepEqual done in 177a98d; custom-root-url's crawl matches now too (0d9884a).

On the comment about module order you dropped: I'd asserted that by analogy from custom-root-url and never checked it here. Ran this app with the crawl first and all three tests still pass, so the claim was wrong.

Drafted by Claude, reviewed before posting.

[gitKrystan]

What does "wedges" mean here?

[gitKrystan]

keeps its own what?

[NullVoxPopuli]

tyty!