Unskip the listing-create suite, updating tests to current command behavior

[richardhjtan]

Turns the listing-create live suite back on. The suite was skipped wholesale by the host→catalog migration and its tests had drifted behind three deliberate behavior changes — they were asserting stale contracts, not catching bugs:

1. Auto-patch is background work now. The command returns before autoPatchName/Summary/autoLink/linkSpecs run; they land via the backgroundWork promise on the result (the create-listing-modal awaits it the same way). Tests now await it before asserting those fields.
2. Only the listed export gets a spec. linkSpecs deliberately specs just the listed export of the main module, so a module defining both Author and AuthorCompany yields one spec.
3. SearchAndChoose speaks numbered options. Candidates go to the LLM as an id-less numbered list (anti-hallucination) and the reply is option numbers. The mock now parses the wanted entry's number out of the prompt, and link assertions compare card names instead of realm-specific ids — the candidate pool spans every readable realm (mock + real catalog), so ids aren't deterministic.

Two tests remain skipped with reasons in the file: they're written against the pre-storefront isolated-view DOM (data-test-catalog-listing-embedded-*) and the retired open-on-stack flow — the create-listing-modal now navigates to code mode with an isolated preview after the command runs. They need a rewrite against that flow (same bucket as the browse-suite DOM rewrite).

Test plan

• listing create suite locally: 2 pass / 2 skip / 0 fail
• recognised imports test passes unchanged (guards spec-creation realm boundaries)

Part of the Eliminate Skipped Tests in Catalog CI project (CS-11462).

🤖 Generated with Claude Code

[github-actions]

Staging Submissions Preview

This PR's content is pushed to the staging submissions realm: https://realms-staging.stack.cards/submissions/

Changed folders:

• tests/

Updated at 2026-08-03 04:14:09 UTC for commit 65933eb. Shared realm: only this PR's changed files are pushed; files touched by multiple PRs reflect whichever pushed last, and deleted files are not removed.

[Copilot]

Pull request overview

Re-enables the listing create live acceptance suite and updates the tests to match current ListingCreateCommand behavior (notably: async background auto-patch/link work, updated spec-linking semantics, and SearchAndChoose’s numbered-option prompting).

Changes:

• Unskips the listing create suite and converts key cases from skip to test.
• Updates the proxy/LLM mock to handle SearchAndChoose’s numbered candidate list and asserts linked entities by stable names instead of realm-specific ids.
• Awaits result.backgroundWork before asserting auto-patched/link-populated fields and updates spec assertions to expect only the listed export to be spec’d.

Suppressed comments (1)

tests/live/catalog-app/listing-create.test.gts:345

• This test is now un-skipped, but it never awaits the command’s backgroundWork. Since linkSpecs runs in the background and backgroundWork uses allSettled (doesn’t throw), the later listing.specs.every(...) assertion can pass vacuously when specs is still empty, reducing the test’s ability to catch regressions. Await backgroundWork and assert that specs were actually generated.

          test('listing will only create specs with recognised imports from realms it can read from', async function (assert) {
            const cardId = mockCatalogURL + 'UnrecognisedImports/example';
            const commandService = getService('tool-service');
            const command = new ListingCreateCommand(
              commandService.commandContext,

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[richardhjtan]

[Claude Code 🤖] Two updates in d423931, following review discussion:

• Deleted the open-on-stack test instead of leaving it skipped — it asserted navigation the command no longer performs (the create-listing-modal owns the post-create flow now, navigating to code mode with an isolated preview), so there's nothing left for it to guard at the command level.
• Added supportingCards coverage, which the suite previously had none of: open cards that aren't instances of the listed type get linked as supportingCards, explicit supportingCardIds are honored verbatim, and a card classified as an example is never duplicated into supportingCards. The assertions read the persisted document's relationships rather than the in-memory instance — the doc is what the command writes, and linksToMany entries on the live instance resolve lazily (the raw doc showed supportingCards.1 present while the instance still reported it as unresolved).

Suite now: 3 pass / 1 skip (the app-listing test, pending the storefront-DOM rewrite) / 0 fail.