Repository navigation
fix(website): clean markdown twins for model pages - #19404
MaanilVerma wants to merge 12 commits into
Conversation
…nto maanil/fe-2922-model-markdown-twins
|
The latest updates on your projects. Learn more about Vercel for GitHub. 2 Skipped Deployments
|
|
Important Review skippedAuto reviews are disabled on base/target branches other than the default branch. Please check the settings in the CodeRabbit UI or the ⚙️ Run configurationConfiguration used: Repository: Comfy-Org/ComfyUI_frontend/.coderabbit.yaml Review profile: ASSERTIVE Plan: Advanced Run ID: You can disable this status message by setting the Use the checkbox below for a quick retry:
Comment |
QALocal builds of
1. Tests. 2 and 3. Switch off, compared with the base build (#19386 head Example ( The model page HTML is unchanged: 4 to 6. Switch on (throwaway local branch, never pushed). This branch plus #18899 ( The 6 non-Grok twins that mention "Grok Imagine" do so only in their related-models links. Before the header fix, the FLUX 2 Max twin started FLUX 2 Max twin (first 30 lines)Seedance 2.5 Text-to-Video twin (first 30 lines)Still noisy, left for follow-ups: playground form labels ("Input / Prompt / Height / Width / Advanced settings") and empty Gates: Update (1ef36e5, Cursor review follow-up):
| 10 | d47506c (merged #19399 Update 2026-09-30 (body only, no code change): the Changes line for |
🧩 Custom Nodes Ecosystem Matrix: 🟡 Withheld — could not measureExecutes the frontend JS of every registry pack that ships any, against this change, in both renderers. The eight This candidate changes nothing the matrix measures, so every expensive stage was skipped. Reproduce locally# build one shard's fixture, then run it
python3 scripts/registry-census/build_matrix.py --shard 1/4
MATRIX_RENDERER=vue MATRIX_VUE=1 \
npx vitest run --config vitest.matrix.config.mts
Full docs: scripts/registry-census/README.md 🌾 Fallow: ✅ PassedFallow audits the diff for duplication, complexity and dead code. It gates on new findings only, so when the audit completes, a red is something this PR introduced or moved — never inherited backlog. A red can also mean the audit itself did not finish; the details below say which happened. No new findings in the changed files. Reproduce locally# the base the gate compared against
npx fallow audit --root . --quiet --format json \
--changed-since 6b6ded76b93495c1e1fde1455da79bf91f935d02
Full docs: ADR-DEVEX-LINT-0015 📦 Bundle Size
🌐 Website E2ETip All tests passed.
🔗 Website PreviewWebsite Preview: https://comfy-website-preview-pr-19404.vercel.app Models is built and gated by PostHog (workshop-enabled), using staging Cloud. This commit: https://website-frontend-cxfg2rpos-comfyui.vercel.app Last updated: 2026-09-29T20:56:45Z for |
Codecov Report❌ Patch coverage is
@@ Coverage Diff @@
## maanil/fe-2921-models-sitemap #19404 +/- ##
==============================================================
Coverage 84.23% 84.23%
==============================================================
Files 3210 3210
Lines 156384 156400 +16
Branches 40608 40614 +6
==============================================================
+ Hits 131735 131751 +16
Misses 22949 22949
Partials 1700 1700
Flags with carried forward coverage won't be shown. Click here to find out more.
🚀 New features to boost your workflow:
|
There was a problem hiding this comment.
🔍 Cursor Review — Consolidated panel
Triggered by @MaanilVerma.
Found 5 finding(s).
| Severity | Count |
|---|---|
| 🟠 High | 1 |
| 🟡 Medium | 1 |
| 🟢 Low | 3 |
Panel: 8/8 reviewers contributed findings.
benjcooley
left a comment
There was a problem hiding this comment.
Review
| Severity | Count |
|---|---|
| BLOCKER | 2 |
| SHOULD FIX | 1 |
| TRIVIAL | 1 |
The core premise doesn't hold: model pages have no model content in their built HTML.
src/routes/models/[slug].astro renders <ModelsPage client:load>. WorkshopGate picks its view from useMounted() and useWorkshopEnabledSettled(), both false during SSR, so the server renders the loading slot only — WorkshopLoading, which is the role="status" div. The model hero, summary and price come from ModelPage.vue, loaded by a client-side defineAsyncComponent behind a fetchModelsPage() call. This repo already asserts exactly that in ModelsPage.test.ts:
expect(html).toContain('workshop-loading')
expect(html).not.toContain('model-hero')So the <main> of a built /models/<slug>/index.html contains the loading spinner and a <noscript> block, nothing else. NOSCRIPT is already in DROPPED_TAGS; this PR adds role="status" to the drop list. Net result once MODEL_PAGES_INDEXABLE flips: 148 twins with front matter and an empty body, each advertised by rel="alternate" type="text/markdown" (BaseLayout.astro:180) and each concatenated into llms-full.txt via writeFullText.
The new test doesn't catch this because buildModelPages() hand-writes a <header data-testid="model-hero"> inside <main> that the real build provably never emits. It tests the converter against a fixture, not against the page.
The <header> fix itself is good and clearly right — PackDetail.vue:51, p/supported-models/index.astro:57 and platform/serverless-animation.astro:14 all lose their h1 today, and this restores it. That part I'd merge on its own.
SHOULD FIX (file not in the diff, so no inline comment): src/pages/workshop/models/[slug].astro:48-52 carries a comment that this PR makes false:
A div, not a header.
markdown-twin.tsdrops HEADER along with the site chrome, so wrapping the hero in one leaves every model's llms.txt entry with a title and nothing else.
Either delete it or the hero can go back to being a <header>.
benjcooley
left a comment
There was a problem hiding this comment.
Re-review of 4426b102 — no fixes in this push
| Severity | Count |
|---|---|
| BLOCKER | 2 (both carried over, unaddressed) |
| SHOULD FIX | 1 (carried over) |
| TRIVIAL | 1 (carried over) |
I diffed the new head against the head I reviewed (1ef36e5fe). Both files this PR changes are byte-identical — same git blob hashes:
SAME apps/website/src/lib/markdown-twin.ts
SAME apps/website/src/lib/markdown-twin.test.ts
The PR's file set is still exactly those two (+81/-2 and +12/-2). The single new commit is 4426b102 — Merge commit 'fb67009537' into tmp-cascade-2922 — a merge of the base branch. It brings unrelated churn (the /agent indexing change in indexing.ts, a hideAnnouncementBanner prop in BaseLayout.astro, ~6k lines of agent CRDT work). None of it touches the twin converter or the model page.
I also re-verified every invariant the blockers rest on, at the new head:
ModelsPage.test.ts:91,94— stillexpect(html).toContain('workshop-loading')andexpect(html).not.toContain('model-hero').routes/models/[slug].astro,ModelsPage.vue,WorkshopGate.vue— unchanged.BaseLayout.astro:184— still emitsrel="alternate" type="text/markdown"for every non-noindex page.integrations/markdown-twins.ts— still no empty-body guard.
So the position is unchanged: once MODEL_PAGES_INDEXABLE flips, all 148 model twins ship with front matter and an empty body, each advertised via rel="alternate" and each folded into llms-full.txt.
On the earlier replies
The two author replies on this PR (4133763793, 4133764115) both point at 1ef36e5fea — the commit I reviewed. They answer the bot's comments, not mine. One of them states:
The fixture now mirrors the SSR model page: a site banner, then an island in
mainholding themodel-heroheader (provider, h1, summary) and arole="status"loader.
That is the specific claim I'm disputing. The fixture mirrors what the hydrated page renders. The SSR output is the loading shell only, because WorkshopGate's view is 'loading' when useMounted() is false, and the hero lives in ModelPage.vue behind a client-side defineAsyncComponent + fetchModelsPage(). The repo's own ModelsPage.test.ts is the authority here and it asserts model-hero is absent from the SSR string.
If I've got the render path wrong, the cheapest way to settle it is to paste the <main> of a built dist/models/<slug>/index.html, or the generated dist/models/<slug>.md. That would close this in one comment.
Unchanged from last round
The <header>-inside-<main> change is still correct and still worth landing — it genuinely restores a missing h1 on PackDetail.vue, p/supported-models/index.astro and platform/serverless-animation.astro. Splitting it out from the model-twin half would unblock it immediately.
Also still open: src/pages/workshop/models/[slug].astro:48-52 carries a comment this PR falsifies ("markdown-twin.ts drops HEADER along with the site chrome").
…nto maanil/fe-2922-model-markdown-twins # Conflicts: # apps/website/src/lib/markdown-twin.test.ts
…content, and drop only the Workshop spinner
benjcooley
left a comment
There was a problem hiding this comment.
Re-review of d47506c50 — blockers resolved, approving
| Severity | Count |
|---|---|
| BLOCKER | 0 (2 resolved) |
| SHOULD FIX | 2 (both new, non-blocking) |
| TRIVIAL | 2 |
This is a real fix round. All four findings from my last two passes are addressed with code, not acknowledgements, and two of the fixes are better than what I suggested. Threads resolved.
What actually changed
The empty-twin blocker is fixed twice over. writeMarkdownTwins now short-circuits on page.body === '', so the path lands in skipped and stays out of writeSectionIndexes and writeFullText. Then unlinkTwin goes a step further than I asked and strips the rel="alternate" type="text/markdown" link out of the built HTML, so a contentless page does not advertise a twin that was never written. I only asked for the first half; the second half closes the loop properly.
role="status" → data-twin-omit. Replacing the ARIA-role sniff with an explicit opt-out on WorkshopLoading.vue is the right call and cleaner than my "target the loading component" hint. I re-checked the other role="status" users on indexable pages — HubspotFormEmbed.vue:129 and MobileDownloadEmailForm.vue:147 — both are v-if/state-gated and render empty server-side, so nothing leaks from narrowing the filter. The showcase components (ModelsHeroSection, ModelCreationsSection, AIModelsSection, ProductShowcaseSection) have no loading markup at all, so /models.md stays clean.
The fixture blocker is fixed at the root. The test now does renderToString(h(WorkshopLoading, { label: 'Loading' })) and injects the component's actual SSR output, so data-twin-omit is verified against reality instead of a hand-written guess. And the model-page tests split honestly: "writes no twin and drops the markdown link for a spinner-only page" is today's build, while the hero test is labelled "(post-FE-2942 shape)". That is the correct framing and I appreciate it being handled that way rather than defending the old fixture.
isBanner — nav removed, comment restated correctly. Stale comment in pages/workshop/models/[slug].astro deleted.
Also verified: unlinkTwin keys off pathname, not the alternate source, so /models/ correctly unlinks /models/index.html rather than models/showcase/index.html. And the alias/showcase/flag coverage I thought had been dropped from the model pages block is alive in the Models pages block that arrived with the base merge.
SHOULD FIX (non-blocking)
1. The PR description is now wrong and will mislead whoever flips the flag. It still says flipping MODEL_PAGES_INDEXABLE "gives all 148 canonical model pages one" twin. After this change they get none — the SSR body is spinner-only, so the empty-body guard skips every one of them until FE-2942 server-renders the hero. The code is honest about this (the test name says so); the description is not. Please update it so FE-2947 does not get flipped expecting twins.
2. unlinkTwin's regex is string-coupled to BaseLayout.astro with nothing tying them together. TWIN_LINK = /<link rel="alternate" type="text\/markdown"[^>]*>/ matches today's Astro output, but the tests hand-write that markup rather than deriving it from the layout. Reorder rel and type in BaseLayout.astro:184-189, or slip an attribute between them, and the link silently survives while pointing at a .md that was never written — a 404 for agents, and no test fails. Same fixture-vs-reality gap as the original blocker, far lower stakes. Cheapest guard: assert in a test that BaseLayout's rendered <head> matches TWIN_LINK.
TRIVIAL
data-twin-omitis opt-in, so it is now a convention someone has to remember for the next loading/placeholder component. Nothing leaks today; worth a line in the converter's doc comment.- Two adjacent describe blocks named
Models pagesandmodel pagesdiffer only in case. Worth renaming one to say what it covers.
Note on verification
CI was still pending when I looked, so this approval is from reading the code, not a green run. The new test file adds a top-level await renderToString(...) and imports a .vue SFC into src/lib/ — both fine under Vite, but worth a glance at the run before merge.
…nto maanil/fe-2922-model-markdown-twins
benjcooley
left a comment
There was a problem hiding this comment.
Re-check of cfd790ca7 — no code change, approval stands
cfd790ca7 is a merge of maanil/fe-2921-models-sitemap, not a fix commit. All five files this PR owns are byte-identical to d47506c50, the commit I approved:
SAME apps/website/src/components/workshop/WorkshopLoading.vue
SAME apps/website/src/integrations/markdown-twins.ts
SAME apps/website/src/lib/markdown-twin.test.ts
SAME apps/website/src/lib/markdown-twin.ts
SAME apps/website/src/pages/workshop/models/[slug].astro
My approval is already recorded against this head and reviewDecision is APPROVED, so nothing is blocking. Two things worth recording.
The description fix landed, but the Summary now contradicts itself
Thanks for adding this — it is exactly the honest framing I asked for:
If a page's twin body comes out empty, the writer skips it… On this stack a model page is still spinner-only, so flipping the switch today publishes no model twins rather than empty ones.
But the opening paragraph three lines above it still says:
Flipping it for FE-2947 gives all 148 canonical model pages one, and aliases and reserved paths still get none.
Those two sentences disagree inside the same section, and the first one is what a reader skims. Worth deleting or qualifying the "gives all 148… one" clause.
Also a ticket mismatch: the test calls the future SSR work post-FE-2942 (markdown-twin.test.ts:377) while the description calls it post-#18899. One of them is wrong.
New verification after the base merge
The merge pulled in routes/models/app.astro and reworked workshop-release-gate.ts, so I re-checked the one gap I had flagged and dismissed last round: unlinkTwin only runs on the empty-body path, not the isExcludedFromSitemap path, so in principle a sitemap-excluded page could ship while still advertising a twin that was never written. I walked every branch of isExcludedFromSitemap against what actually ships:
app.astro— hardcodednoindex, so no link is emitted. (This is the one the merge touched.)routes/models/[slug].astro—noindex={!isIndexableModelPage(...)}, which covers models with the flag off, aliases and workflow slugs./models/showcaseand/cinematic-studio— inNOINDEX_ROUTES, soshouldNoindexsuppresses the link./workshop/*—workshop-release-gaterm -rfsdist/workshop, so those pages never ship./p/supported-models/<alias>— returnsAstro.redirectbefore BaseLayout renders, and[slug].md.tsfilterscanonicalSlugentries out.
Every sitemap-excluded page that ships is also noindex, so it emits no markdown link in the first place. The empty-body-only placement of unlinkTwin is complete in practice, not just by luck. Confirmed rather than assumed this time.
workshop-release-gate moving from existsSync to the assets map does not interact with the twin writer either — both hook astro:build:done, and either ordering is harmless since the gate only removes dist/workshop, which the twin writer skips anyway.
Still open from last round (non-blocking)
TWIN_LINK in markdown-twins.ts is a string-coupled regex against BaseLayout.astro's markup with no test tying the two together. Reorder rel and type in the layout and the link silently survives, pointing at a .md that was never written. A test asserting BaseLayout's rendered head matches TWIN_LINK would close it.
…wn link unlinkTwin now throws, naming the built file, when it finds no markdown alternate link to remove, and matches the link whatever its attribute order.
|
@benjcooley thanks for both reviews. Here is where each point landed. 1.
2. Your re-review:
Checks: |
benjcooley
left a comment
There was a problem hiding this comment.
Re-review of 20150a17a + 58805c39d — approval stands, one new latent build-breaker
| Severity | Count |
|---|---|
| BLOCKER | 0 |
| SHOULD FIX | 2 (1 new, 1 carried) |
| TRIVIAL | 0 (previous one fixed) |
Both new commits are net improvements. TWIN_LINK is now order-independent (/<link\b[^>]*\btype="text\/markdown"[^>]*>/), with a test that puts the attributes in reverse order and splits the tag across lines — and [^>] matches newlines in JS, so that genuinely covers the multi-line case. The post-FE-2942 (#18899) relabel closes the ticket mismatch. Both new behaviours have tests. Nothing here changes my approval.
SHOULD FIX (new): the throw is one content-light page away from breaking pnpm build
unlinkTwin now throws when it finds no link to remove. Work out when that can fire:
- The link is absent exactly when
shouldNoindexwas true at render time (BaseLayout.astro:184). - The empty-body check is only reached when
isExcludedFromSitemapis false. shouldNoindex = noindexProp || isNoindexPathname(pathname), andisNoindexPathnameimpliesisExcludedFromSitemap, so that half is unreachable.
Which leaves exactly one trigger: a page that passes noindex as a prop, is not covered by NOINDEX_ROUTES, and renders an empty twin body. The invariant the throw assumes — "no content implies there was a link" — is not enforced anywhere; it depends on BaseLayout's noindex prop and indexing.ts's NOINDEX_ROUTES staying in agreement, and they are separate sources of truth.
One page already sits in that state. pages/platform/serverless-animation.astro:11 passes noindex, and /platform/serverless-animation is not in NOINDEX_ROUTES. It escapes the throw only because its <header> happens to contain an h1 and some copy. Make that page island-rendered — the exact pattern /models/<slug> uses — and the production build dies with … has no markdown twin link to remove, a message that names neither noindex nor the sitemap config, so the next person gets to rediscover this chain from scratch.
Same root cause also produces a live oddity today: because isExcludedFromSitemap cannot see the prop, /platform/serverless-animation is noindex yet still gets a .md written and folded into llms-full.txt. The PR description even advertises that twin gaining an h1.
Both problems collapse into one fix. BaseLayout.astro:125 already emits the ground truth into the built HTML:
{shouldNoindex && <meta name="robots" content="noindex, nofollow" />}writeMarkdownTwins has source.html in hand at that point, so checking it for that meta and continue-ing would mean noindex pages get no twin at all, and the throw becomes genuinely unreachable rather than accidentally unreached. That also removes the need for the noindex prop and NOINDEX_ROUTES to stay manually in sync for the twin writer's purposes.
SHOULD FIX (carried, stakes now higher)
Still no test binding TWIN_LINK to BaseLayout's real rendered output — the fixtures hand-write the markup. Loosening the regex reduced the risk of a mismatch, but the throw raised the cost of one: a mismatch used to mean a dangling link, and now means a hard build failure. That becomes load-bearing the moment FE-2947 flips the flag, because unlinkTwin will then run across all 148 model pages in one build. A test that renders BaseLayout's head and asserts TWIN_LINK matches it would make the regex safe to rely on.
To be clear: neither item blocks. Nothing breaks today, and fail-loud is the right instinct here — it is the placement of the guard, not the guard itself, that I would change.
…removal A page that never advertised a twin (rendered noindex) is a valid state, so unlinkTwin now leaves it alone and throws only when a link is still present after TWIN_LINK runs.
|
@benjcooley thanks for the re-review. Both points are in e064209. Where the guard sits now. A page with no twin link isn't an error. It means the page rendered noindex and never offered a twin, so there's nothing to remove. The failure worth catching is a link that's still there after Proof
Pages that pass the
I haven't fixed those two here. Your fix (skip any page whose built HTML carries the robots noindex meta) is the right shape, but it collides with #19362. That PR puts the robots meta on every page of preview and non-production builds, so reading the meta would skip every twin on those builds, and llms.txt link validation there would likely break. The options are: add the two paths to |
benjcooley
left a comment
There was a problem hiding this comment.
Re-review of e064209e6 — the build-breaker is fixed, and fixed better than I suggested
| Severity | Count |
|---|---|
| BLOCKER | 0 |
| SHOULD FIX | 0 |
| TRIVIAL | 1 |
Approval stands. One commit, 6 lines of source, exactly on target.
I proposed reading the robots meta out of the built HTML. Inverting the guard is the better answer:
if (unlinked.includes('type="text/markdown"')) throw ...
if (unlinked !== html) await writeFile(path, unlinked, 'utf8')This checks the postcondition that actually matters — "is a dangling markdown link still here?" — instead of inferring it from "did the replace change anything." It keeps the fail-loud property while dropping the false positive, and it needs no second source of truth about whether a page is noindex. Fewer moving parts than my version.
Walking it back through the cases: a prop-noindex page with an empty body (the pages/platform/serverless-animation.astro shape that would have killed the build) now leaves the file untouched and reports skipped; an indexable empty page has its link removed and written; a link TWIN_LINK fails to match still throws. The <LINK> uppercase-tag fixture is a good choice for the throw test — it makes the regex miss on purpose and proves the residue check catches it, which is the case that matters.
TRIVIAL
The residue check shares the regex's blind spot on quoting and casing: type='text/markdown' or TYPE="..." would slip past both TWIN_LINK and includes(). Astro emits lowercase tags and double quotes so this cannot happen today, and I would not change it — noting it only so the assumption is written down somewhere.
Worth flagging as a follow-up rather than a finding, since it predates this PR: a prop-noindex page like /platform/serverless-animation still gets a .md written and folded into llms-full.txt, so it stays discoverable through that channel even though it carries noindex and no longer advertises a twin link. Your commit message takes the position that a page which never advertised a twin is a valid state, which I agree with — the open question is just whether llms-full.txt should list one. Not this PR's problem.
CI was still pending when I looked, no failures.
## Summary **Deploy gate: comfy-router#46 is merged and live (2026-09-30). Its follow-up comfy-router#48 (serves `/hub/models.md` instead of 301ing it to a /workflows 404) must be deployed before this merges. After the merge, purge Cloudflare for `/hub/models*`, `/hub/workflows*` and the 30 workflow names in `hub-workflow-names.json`. The redirects ship as 307; FE-3083 flips them to 308 later.** This branch combines every Workshop /models SEO PR, so merging it to main replaces merging them one by one. It also launches: Comfy-Org#19405 commits `launchedModelPages = 'all'`, so all 148 model pages become indexable in the same deploy as the URL move. Part of FE-2912, FE-2942 (PR 2 is FE-3086), FE-2974 (/minimax 308 is FE-3087) Fixes FE-2978, FE-2989, FE-2991, FE-2990, FE-2959, FE-2981, FE-2977, FE-2979, FE-3014, FE-2936, FE-2980, FE-2975, FE-3085, FE-2972, FE-2935, FE-2950, FE-2940, FE-2930, FE-2985, FE-2957, FE-2961, FE-2986, FE-2921, FE-2922, FE-2947 ## Changes Router follow-up: comfy-router#48 (`/hub/models.md` routing, 324/324 router tests), deploy before this merges. The included PRs, in merge order. Every head was checked as contained at the final tip. | PR | Ticket | What | Ben | |---|---|---|---| | Comfy-Org#18887 | FE-2978 | Pricing Product markup and full video `uploadDate` | approved | | Comfy-Org#18889 | FE-2989 | A unique meta description on every model page | approved | | Comfy-Org#18890 | FE-2991 | Model page JSON-LD, emitted once noindex lifts | approved | | Comfy-Org#18892 | FE-2990 | Cleaner tag pills, guard against duplicate summaries | approved | | Comfy-Org#18893 | FE-2959, 2981, 2977, 2979 | Social tags, hub title, Organization data. The FE-2977 title was confirmed by the owner | approved | | Comfy-Org#19362 | FE-3014 | Every non-production build is noindex | approved | | Comfy-Org#19366 | FE-2936 | Model titles lead with "API & Playground" | approved | | Comfy-Org#19402 | FE-2980 | Every internal link ends with a slash, plus a CI check | approved | | Comfy-Org#19419 | FE-2975 | Footer links the catalogue on every page and locale | approved | | Comfy-Org#19476 | FE-3085 | `NOINDEX_ROUTES` decides noindex for serverless-animation | approved; re-approval pending on a one-line follow-up (CodeRabbit approved) | | Comfy-Org#19359 | FE-2974 | One typed redirect list that generates `vercel.json` | approved | | Comfy-Org#19369 | FE-2972 A | The old to new model URL map | approved | | Comfy-Org#18891 | FE-2935 | Model content shows whatever the flag says | approved | | Comfy-Org#18899 | FE-2942 | Model pages render in the static HTML | approved | | Comfy-Org#18898 | FE-2950 | Catalogue h1 and A to Z links in the static HTML | approved | | Comfy-Org#19375 | FE-2940 | Related models leave no page unlinked | approved | | Comfy-Org#19385 | FE-2930 | Examples show the prompt that made them | approved | | Comfy-Org#19394 | FE-2985 | Render guard e2e: no loader, no layout shift | approved | | Comfy-Org#19357 | FE-2957 | robots.txt stops blocking our scripts and styles | approved | | Comfy-Org#19360 | FE-2961 | Turning Workshop off keeps every Models page | approved (a CodeRabbit changes-requested review is still open) | | Comfy-Org#19386 | FE-2986 | One registry of every Models address | approved | | Comfy-Org#19421 | FE-2972 B1 | Model pages served at `/hub/models` | approved | | Comfy-Org#19422 | FE-2972 B2 | Every old model URL redirects (307) | approved | | Comfy-Org#19424 | FE-2972 B3 | Workflow pages at `/hub/workflows` with redirects | approved | | Comfy-Org#19399 | FE-2921 | One switch decides model robots and sitemap listing | approved | | Comfy-Org#19404 | FE-2922 | Clean markdown twins for model pages | approved | | Comfy-Org#19405 | FE-2947 | Lifts noindex on every canonical model page | approved | | comfy-router#46 | FE-3075 | Serves `/hub/models*` from the website and routes `/hub/workflows` by manifest. Merged and live | merged | These fixes exist only on this branch, because they only break once two PRs meet: - `scripts/check-models-html.ts` reads each page at `model.href` and skips models without a page (f1df1e4). - `e2e/fixtures/modelsCatalogue.ts` reads `dist/hub/models` (f87bb4c). - `routerIdsByModelPage` accepts an optional `href`, since Comfy-Org#19421 made it optional and Comfy-Org#19405 no longer typechecked (9783819). - `ModelStatus` links a successor only when it has a page, for the same reason on Comfy-Org#18899 (c4761d8). - The model breadcrumb in `modelJsonLd.ts` points at `/hub/models/` instead of the `/models/` redirect (eb985b6). Its label still says "Models" while the nav says "Hub". - Twin, meta-description and entry tests use the hub pages and the new tag set (86ac3b0, d4ffc0f). - The render guard finds the catalogue through `getRoutes().workshop`, and its `test.fail` is gone now that Comfy-Org#18898 renders the h1 (831499f). - A new test renders a real model page and checks its markdown twin has the model's h1 (9a6c5cb). - Main's newer catalogue and workflow e2e cases use the hub addresses (2aeda77). - `resetWorkshopAccountSource` resets `decision`; a clean auto-merge of main's Comfy-Org#19454 into Comfy-Org#18891 left it assigning a variable that no longer exists (c28c7a7). The conflict resolutions, briefly: `ALTERNATE_TWIN_SOURCES` stays deleted as Comfy-Org#18898 wants, so `/hub/models.md` comes from the real catalogue. The hub keeps Comfy-Org#18898's single static h1 instead of the `WorkshopHero` h1 that main's Comfy-Org#19320 added. `ExamplesTab` keeps Comfy-Org#19385's figure and caption with main's video poster. The Comfy-Org#19399/Comfy-Org#19405 indexing tests were rewritten for hub paths. The Comfy-Org#19359 redirect test no longer builds `/enterprise//` from slash-ended routes. The Comfy-Org#19369 URL-map test imports `PROVIDER_NAMES` from its new home. The full table of 47 resolutions is in the integration report. ## Review Focus - Launch rides along with the URL move. If launch should stay a separate step, set `launchedModelPages` in `apps/website/src/config/model-page-launch.ts` to `new Set()` before merging. - Main's Comfy-Org#19320 gives the catalogue island its own h1 through `WorkshopHero`. This branch keeps Comfy-Org#18898's static h1 and renders only the subtitle in `ModelsCatalogue.vue`, so the hub has exactly one h1. - Comfy-Org#19476 overlaps the open community PR Comfy-Org#18591 (same sitemap exclusion for serverless-animation). One of them should be closed once this lands. ## QA - [x] Gates: `pnpm typecheck` and website typecheck clean, `pnpm lint` 0 errors, eslint on `apps/website` clean, `pnpm knip` clean - [x] Website unit: 547 files, 10347 passed, 4 skipped - [x] Build A (`build:e2e`, CI env): 1193 pages; release gate, `check:hreflang`, `validate:llms-txt-links` (225 links), `validate:jsonld` (947 pages), `check:link-slashes`, `check:models-html` (148 pages) all pass - [x] Build B (`WORKSHOP_IN_BUILD=0`): same 182 Models pages as the on build, no `dist/workshop`, `/` is noindex; with the flag stubbed on, no Run, no Hub nav and no Firebase requests - [x] Build C (preview shape): 945 of 945 pages noindex, no canonicals except redirect stubs, no hreflang in the sitemap - [x] E2E on Build A: 644 of 645 passed; the one mobile failure passed on rerun. Earlier load flakes (navigation, demos, fdct, wan-3.0, network isolation, render guard) all passed when rerun alone - [x] Launch: 148 model pages indexable with canonicals, 149 hub entries in the sitemap, 148 non-empty twins with the model h1, 148 in `llms-full.txt`, 0 titles over 60, 30 workflow pages still noindex - [x] Redirects: 853 rules, 782 to /hub all 307, the MiniMax rows 307, every /hub destination built - [x] comfy-router#46 under `wrangler dev` in front of this build: 312 router tests pass; 148 old model URLs (both slash forms) 307 then 200; 30 workflow names served by the website and their old URLs 307; a non-manifest workflow 302 with no-store; `manifest.json` passes through; `/workflows/` and bare `/hub` unchanged - [x] After the router deploys and this merges: purge Cloudflare, then curl `https://comfy.org/hub/models/` for a 200 and three old `/models/<slug>/` URLs for one 307 each - [x] Rich Results Test on `/pricing/` and `/learning/animations/character-sheet/`, then "Validate fix" on Merchant listings and Videos - [x] Rich Results Test on two live model pages: breadcrumbs valid and the breadcrumb URL returns 200 with no redirect - [x] GPTBot and Googlebot curls on the hub and one model page, then a GSC live test on both - [x] Resubmit the sitemap in Search Console and URL-inspect about 20 model pages on day 0 and day 14 - Not covered: Rich Results and GSC need production; the redirect matrix used a local shim for `vercel.json`, since `astro preview` ignores it --------- Co-authored-by: GitHub Action <action@github.com> Co-authored-by: benjcooley <benjcooley@gmail.com> Co-authored-by: Deep Mehta <42841935+deepme987@users.noreply.github.com>
|
Superseded by the integration branch #19494, which combined this PR with every other Workshop SEO slice and has been merged to main. Closing this one. |
Summary
Model pages get a markdown copy for AI readers the same way every other page does: through the existing twin writer, and only when the page is indexable. #19399's switch already decides that for robots and the sitemap, and the twins, the
rel="alternate" type="text/markdown"link and thellms-full.txtentry all follow it. With the switch off (today) no model page gets a twin. Flipping it for FE-2947 makes the canonical model pages eligible for one (aliases and reserved paths never are), but a page only gets its twin once its HTML has real content, which comes with FE-2942 (#18899).This PR makes those copies worth reading, and makes sure an empty one is never published:
data-twin-omitonWorkshopLoading), so the island's "Loading" line never lands in a twin. Otherrole="status"copy is kept..md, nollms-full.txtor section-index entry, and itsrel="alternate" type="text/markdown"link is removed from the built HTML. On this stack a model page is still spinner-only, so flipping the switch today publishes no model twins rather than empty ones.<header>that sits inside<main>. The converter dropped every<header>as site chrome, but the model page hero is a<header>holding the h1, provider, summary and price, so the twin came out with no title. Only the site banner (outside main) is chrome now.Fixes FE-2922 (https://linear.app/comfyorg/issue/FE-2922)
Stacked on #19399
Changes
src/lib/markdown-twin.ts: drop elements markeddata-twin-omit; treat a header inside main as content.src/integrations/markdown-twins.ts: empty-body guard (no twin, no twin link). A page with no twin link to remove is left as is; the build fails, naming the file, only if atype="text/markdown"link is still there after removal.src/components/workshop/WorkshopLoading.vue:data-twin-omiton the spinner.src/pages/workshop/models/[slug].astro: remove a comment this PR makes false.src/lib/markdown-twin.test.ts: a spinner-only model page (realWorkshopLoadingSSR output) gets no twin and loses its link; the hero fixture is labelled as the post-FE-2942 (feat(website): render each model page in the static HTML #18899) shape.Changed from the ticket:
[slug].md.tsendpoint or catalogue endpoint. The twin writer already turns the SSR HTML (feat(website): render each model page in the static HTML #18899) into a clean copy, so one path covers every page./models.mdstays the showcase until the hub has its own content./p/supported-models.mdand/platform/serverless-animation.md). Those twins only gain lines; nothing is removed.Review focus
isBanner: a header outside main is still dropped, so the site header never leaks in.isIndexableModelPagethroughisExcludedFromSitemap..mdfiles can't carry a robots meta, so keeping them out of the index on previews would need anX-Robots-Tagheader rule invercel.json.