Repository navigation
feat(website): make the old /models redirects permanent (308) - #20191
Conversation
comfy-router#46 is live on prod and /hub/models/, /hub/workflows/ and /hub/apps/ answer 200 from the website, so the reason to keep these rows temporary is gone. /minimax stays 307 until its page owner signs off (FE-3087).
|
The latest updates on your projects. Learn more about Vercel for GitHub. 2 Skipped Deployments
|
🌐 Website E2ETip All tests passed.
🌾 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 65d257e48111ba1ba4cbb8e9534668b9d38dc8bb
Full docs: ADR-DEVEX-LINT-0015 🔗 Website PreviewWebsite Preview: https://comfy-website-preview-pr-20191.vercel.app Models is built and gated by PostHog (workshop-enabled), using staging Cloud. This commit: https://website-frontend-ak7u37wv3-comfyui.vercel.app Last updated: 2026-10-06T09:41:32Z 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 |
|
Warning Review paused — included plan limit reachedKeep your review moving with free on-demand reviews.
On-demand reviews are free for the next 4 days. Promotion and pricing detailsOn-demand reviews are free for the next 4 days. After that, they cost $0.25 per reviewed file. Review limit detailsOr wait 6 minutes for your next included review. Limit details: You’ve used all 8 included reviews currently available. Review configuration: ⚙️ Run configuration
📒 Files selected for processing (2)
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info⚙️ Run configuration
📒 Files selected for processing (2)
Included review availability: This review used your included allowance. Your plan provides up to 8 included reviews per hour; 3 remain after this review. 📝 WalkthroughWalkthroughThe ChangesModels redirect status
Priority: ➖ Normal Estimated code review effort: 2 (Simple) | ~10 minutes Change: Feature Suggested reviewers: Merge Risk: 🔵 Low · up to Old /models addresses now return permanent redirects to their /hub pages. The destinations were reported as working on production. Browsers and edge caches may retain these 308s, so run the post-deploy URL check before treating the change as done. Security Architecture ReviewSecurity architecture risk: 🔵 Low · up to The redirect destinations and reachable pages remain unchanged. The main risk is recovery: reverting the server configuration may not undo redirects already cached by visitors. Production routing readiness and recovery arrangements were not independently confirmed. Retained concerns
Security review detailsSecurity Blast Radius
Trust Boundaries and Controls
Hardening Proposals
Caution Pre-merge checks failedPlease resolve all errors before merging. Addressing warnings is optional.
❌ Failed checks (1 error)
✅ Passed checks (6 passed)
Full details: Website End-To-End Regression CoverageExplanation The PR changes website runtime behavior: it changes the old ✨ Finishing Touches📝 Generate docstrings
🧪 Generate unit tests (beta)
Comment |
QA: old /models redirects are permanentChecked on 2026-10-05 against the PR preview
Kept temporary: Not covered here: prod after merge. Run the same 786-URL loop with |
Codecov Report✅ All modified and coverable lines are covered by tests. @@ Coverage Diff @@
## main #20191 +/- ##
==========================================
- Coverage 85.44% 85.31% -0.13%
==========================================
Files 3519 3420 -99
Lines 168831 165714 -3117
Branches 44308 43240 -1068
==========================================
- Hits 144257 141381 -2876
+ Misses 22591 22500 -91
+ Partials 1983 1833 -150
Flags with carried forward coverage won't be shown. Click here to find out more.
... and 99 files with indirect coverage changes 🚀 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 | 3 |
| 🟢 Low | 1 |
Panel: 8/8 reviewers contributed findings.
webreviewer-bot
left a comment
There was a problem hiding this comment.
Gate status — author-verified-preview gate is UNMET
The webreview label on this PR was applied by @benjcooley, not by the author @MaanilVerma. On this repo that label is the author's assertion that they checked the preview deploy. Applied by someone else, it asserts nothing.
This verdict covers code only. An approval here must not be read as merge-ready. Gate one (author verifies the preview) is unmet; gate two (code review) is what follows.
Separately: merges here reach PRODUCTION in minutes, and a 308 is cached by browsers and CDNs, often indefinitely — a client that caches a wrong one cannot be rescued by a later deploy. So I did not spot-check. I parsed both sides of the mapping programmatically and then probed the deployed artifact. Results below.
Redirect audit — numbers, base c9326eb → head 264c596
Parsed vercel.json on both sides and diffed the key sets in both directions.
| Check | Result |
|---|---|
| Rows, base → head | 857 → 857 |
| Paths present before and absent after (dropped) | 0 |
| Paths added | 0 |
| Entries whose destination changed | 0 |
| Entries whose destination and status changed | 0 |
Entries whose status changed, false→true (307→308) |
786 |
Entries true→false (308→307) |
0 |
| Row order identical | yes |
Other vercel.json sections changed (headers, rewrites, build keys) |
none |
| Targets that are themselves a redirect source (chains) | 0 (exact match and slash-variant both checked) |
| Loops (A→B→A) / self-redirects (A→A) | 0 / 0 |
Internal destinations not ending in / |
0 |
| Duplicate sources | 0 |
| Route budget | 861 of 2048 |
The 786 flipped rows decompose exactly: 2 × (1 + 360 + 30 + 2) — /models, 148 model slugs + 212 aliases, 30 workflows, 2 apps, each in both slash forms. No row is anything other than a status flip.
Do the targets exist?
Static: all 212 modelAliasUrls entries resolve to a newSlug that is in modelPageUrls (0 orphans); modelPageUrls has 148 unique newSlug values and 0 duplicates, so getStaticPaths on /hub/models/[slug] builds every one. hub-workflow-names.json and hub-app-names.json are snapshot/equality pinned to workflowModels and appModels by their own tests, so those 32 targets are the built set by construction. 786/786 hub destinations map to a built page.
Dynamic — I probed the real artifacts rather than trusting the config:
- All 857 sources against the PR preview deployment (
comfy-website-preview-pr-20191.vercel.app): every one returned the exact configured status andLocation. 786/786 returned 308; the 11 intentionally-temporary rows returned 307. 0 mismatches (the only two diffs were curl normalisinghttps://trust.comfy.orgto.../, not a defect). This also answers where the 308 is emitted: Vercel'spermanent: true, confirmed on the deployed artifact rather than inferred from a source constant. - All 210 internal destinations, on the preview and on production: 210/210 returned 200 on both. Zero 404s, and a 200 means zero second hops.
- Spot-verified the data paths the literal-rows design exists to protect:
/models/catalogue.json,/models/showcaseand/models/<oldSlug>/page.jsonall still 200 on the preview — the redirects do not swallow them.
I cannot see comfy-router's rollout state from this repository, and I am not claiming it. What I can say is measured: production serves all 210 targets 200 today, including /hub/apps/cinematic-studio/ and /hub/workflows/change-material/, which were the two preconditions the deleted constants named.
308 vs 301
308 is correct and I would not change it. These are GET page URLs; nothing posts to them. 308 preserves method rather than letting a client silently rewrite POST→GET, which is the safer failure mode, and Google treats 308 as a permanent signal identically to 301. It is also what Vercel's permanent: true emits and what the other 70 permanent rows in this file already use — switching to 301 would mean hand-writing statusCode and breaking that consistency. No objection.
Sitemap / canonical / llms.txt / robots
- Pulled the generated sitemap (897 URLs) and intersected it with all 857 redirect sources: 0 overlap. Nothing tells a crawler to index a permanently-redirecting URL.
llms.txt(225 links): 0 point at a redirect source.video-sitemap.xml,robots.txt,/models/catalogue.json,/hub/workflows/manifest.json: 0 references to a redirect source. These are the outputsworkshop-release-gate'sfilesLinkingOldModelsdoes not walk, so I checked them by hand.- Canonicals on production for all three target families are self-referential and point at the
/hub/…URL, not at a redirecting one. robots.txtallows/hub.- No conflict visible from this head with the adjacent machinery in
src/config/indexing.ts; the/p/supported-modelsalias sources stay excluded viaMODEL_REDIRECT_PATHNAMES, unchanged here.
Trailing slashes
Both forms of every source are enumerated as literal rows, and every internal destination ends in / (test-enforced at redirects.test.ts:110). The slash form therefore never produces a second hop — confirmed empirically above, where both forms of all 393 sources resolved in one hop. sources are also regex-restricted to [A-Za-z0-9._-], so no slug (e.g. bfl--flux-pro-1.1--generate-images) can smuggle a path-to-regexp metacharacter.
Generator
sync-vercel-redirects.ts is a pure JSON.stringify of toVercelRedirects(siteRedirects) — deterministic — and redirects.test.ts:70 fails if the committed vercel.json drifts. Output is in sync.
Assessment of the existing review comments
The standing CHANGES_REQUESTED is coderabbit's, and its body is a single failing pre-merge check: "add a Playwright regression test under apps/website/e2e/ … or add a concrete explanation to the PR description of why an end-to-end test is not practical."
Disputed as written; the alternative it offers is the right resolution. The e2e suite runs against astro preview, which serves meta-refresh stubs — and astroRedirects deliberately emits no stub at all for old /models addresses (isOldModelsAddress), so Playwright physically cannot observe these status codes. e2e/canonical-redirects.spec.ts:71-73 already says so in a comment. A description note is the fix; see TRIVIAL-2.
On the five Cursor findings:
- High,
redirects.ts:48— "nothing records that comfy-router#46 is live". Confirmed as a documentation gap, resolved on the facts. Production answers 200 on all 210 destinations including every/hub/…target. The concern was correct to raise and is now empirically settled, but nothing in the repo records it — see TRIVIAL-1. - Medium,
redirects.ts:59— apps rows had a separate precondition. Disputed on the facts./hub/apps/cinematic-studio/and/hub/apps/reshoot/both return 200 from production with correct titles and self-canonicals. Both preconditions hold. - Medium,
redirects.ts:55— workflow targets are noindex, so the 308 "drops a search result". Disputed. The old/models/workflows/*URLs are in no sitemap, build no page, and already 307 to the identical destination today. The destination does not change here — only its permanence. 307-to-noindex and 308-to-noindex index the same (i.e. not at all); if anything 308 is the more correct signal for a URL that is meant to die. See TRIVIAL-3 for what I did measure about those targets. - Medium,
redirects.test.ts:263—oldPathscannot see the workflow rows. Already addressed at this head (hubWorkflowSlugsadded tooldPaths,'workflow'added tolandsOnPage). But the comment's parenthetical — that the assertion overstates what it proves — is the sharper half and is still live. See SHOULD FIX-1. - Low,
redirects.ts:47— stage the groups. Disputed. Staging would leave 307s with no remaining gate, since both preconditions are measurably live. The author's reasoning is sound.
Exposure classification
None of the listed categories. Not a catalogue entry, not a new app, not a new feature surface on an existing page, and no flag check is touched — I diffed every line of all three files and the only semantic change is permanent: false → true on 786 existing rows. No flag default, targeting, or check appears in the diff.
Scope vs title
feat for a status-code change is arguable but the title is accurate and the scope is not larger than it implies: +812/−832 is 786 lines of regenerated vercel.json plus 26 lines of source and test. Nothing is hiding in the bulk edit. No objection.
Findings
| Severity | Count |
|---|---|
| BLOCKER | 0 |
| SHOULD FIX | 1 |
| TRIVIAL | 3 |
SHOULD FIX-1 — redirects.test.ts:265: landsOnPage is tautological and cannot fail. The registry entry and the redirect row are both synthesised from the same hubModelSlugs/hubWorkflowSlugs/hubAppSlugs lists, so the kind of the destination is guaranteed by construction. This test is the named safety net for an unretractable change and its most load-bearing-sounding clause carries no signal. Detail inline.
TRIVIAL-1 — redirects.ts:48: the deleted HUB_ROUTER_PENDING/HUB_APPS_ROUTER_PENDING strings were the only in-repo record of the precondition; nothing replaces them.
TRIVIAL-2 — PR description: add the one-line reason an e2e cannot observe these status codes, which is what clears coderabbit's failing pre-merge check.
TRIVIAL-3 — redirects.ts:55/:59: 64 of the 786 rows land on pages production serves noindex, nofollow; measured detail inline. Not a regression, recorded so it is a decision rather than an accident.
CI (check-runs API, head 264c596)
57 success, 75 skipped, 0 failed, 0 pending; combined status success. website-unit, website-build, website-e2e (4 shards), lint-pr/typecheck and repo-checks all green. Noting the obvious: none of that resolves a redirect target, detects a chain, or notices a dropped entry — which is why the audit above is the actual basis for this verdict, not the green ticks.
APPROVE — code only, and the preview gate above is unmet.
The old check only asked whether a destination was some registered page, so a row that pointed at the wrong existing page still passed. Compare each row with model-urls.ts and the hub name lists instead, and note why the rows are permanent.
🔧 Auto-fixes AppliedThis PR has been automatically updated to fix linting and formatting issues.
Changes made:
|
webreviewer-bot
left a comment
There was a problem hiding this comment.
webreviewer-bot — code review, PR #20191 (cycle 2)
Head 25edffc0c584b63b3930b2a9d3373f4aea03fe1f. Base c9326eb. Cycle 2.
Gate status — author-verified-preview gate is UNMET.
Thewebreviewerlabel was applied by @benjcooley (2026-10-05T17:26:29Z), not by the author @MaanilVerma. The author's only labelling action wascursor-review. On this repo the label is the author's assertion that they checked the preview deploy; applied by someone else it asserts nothing. Unchanged from cycle 1.What a preview check could and could not establish here. Not a formality, but also not sufficient: a Vercel preview does serve
vercel.jsonredirects, so the 307→308 flip is observable there (cycle 1 confirmed all 857 sources return their configured status on the PR preview). What a preview cannot establish is the production-only half — comfy-router's forwarding of/hub/*, and the content gates on the workflow/app targets. Those I measured against production directly, below. This verdict covers code only. Readiness is not my call.
Merge state: this PR is CONFLICTING with main (mergeable: false, mergeable_state: dirty; 4 ahead / 39 behind). It cannot be merged or queued in this state regardless of review outcome. That needs the author's rebase. It is a factual merge blocker, not one of my severity findings.
A standing change request blocks this PR and my approval alone does not clear it. @coderabbitai has CHANGES_REQUESTED at d128352 and again at this head 25edffc (08:42:01Z). I cannot dismiss another reviewer's review. Both bodies are, in full, "Pre-merge checks failed. Please resolve the failing checks before merging." — that is a CI snapshot, not a code finding: zero substantive content, no file, no line, no mechanism. It should not be read as a second opinion on the code.
PRIORITY 1 — the noindex question, settled on measurement
Several reviewers have called the 60 /models/workflows/* rows a BLOCKER on the grounds that they 308 onto noindex targets. I measured it end to end rather than inheriting either side's argument.
All 30 workflow targets + both app targets, fetched from production at this head:
| Result | |
|---|---|
| HTTP status | 200 on all 32, num_redirects=0 |
| Redirect hop at the target | none — 0/32, no permanently-cached chain |
<meta name="robots"> |
noindex, nofollow on all 32 |
X-Robots-Tag header |
index, follow on all 32 |
X-Robots-Tag rule in vercel.json matching these paths |
none — the only noindex header rules are /payment/(.*) and /:locale/payment/(.*) |
So I checked both levels, as the /payment/* precedent requires. The noindex is real — it comes from the meta tag, not a header — and where header and meta conflict the more restrictive directive wins, so these pages are genuinely not indexable today.
But the BLOCKER claim is wrong, and here is the number that settles it. The claim requires that the old URLs had index presence a 307 was preserving and a 308 destroys. They have none:
- Both sitemaps, 9,452 URLs total (
sitemap-0.xml1,258 +sitemap-workflows-0.xml8,194): 0 occurrences of/models/workflows/,/hub/workflows/,/models/apps/or/hub/apps/. (Cycle 1 checked one sitemap; I checked both, including the 8k-URL workflows sitemap it missed.) - The old sources already 307 to the byte-identical destination today — verified on production for 8 workflow forms and both app rows.
- This PR changes 0 destinations (proof below). Only permanence changes.
A 307 onto a noindex page and a 308 onto the same noindex page both leave the old URL out of the index. There is no link equity to terminate at a page that was never in a sitemap, builds no page, and has redirected for its whole life. For a URL that is meant to die, 308 is the more correct signal — and it points at the address these pages will keep when they do launch, so it is forward-correct too.
Verdict: the BLOCKER is not sustained. Recorded as TRIVIAL, so it is a decision rather than an accident. Rollout state of the content gates is unknown from the repository and I am not claiming otherwise.
PRIORITY 2 — cycle-1 SHOULD FIX (the assertion that could not fail): RESOLVED
Cycle 1 flagged landsOnPage as tautological — it asked only whether the destination's kind was model|hub|workflow|app, and both the registry entry and the redirect row were synthesised from the same source lists, so a wrong-but-valid destination could never fail it. d128352 replaces it with a frozen-table comparison of full destination strings.
Is the frozen table literally independent? No — and I want to be precise about that. The import chain re-enters the same source files:
redirects.test.ts --imports--> model-urls.ts { modelPageUrls, modelAliasUrls }
--imports--> hub-workflow-names.json
--imports--> hub-app-names.json
redirects.ts --imports--> hub-models.ts --imports--> model-urls.ts (same file)
hubModelSlugs = new Map(modelPageUrls.map(...))
hubModelAliases = new Map(modelAliasUrls.map(...))
hubWorkflowSlugs = hubWorkflowNames.map(...) (same JSON)
hubAppSlugs = hubAppNames.map(...) (same JSON)
So hubModelSlugs is modelPageUrls, one hop away. The expectation and the generator still read the same four files.
But that is no longer the circularity I flagged, and the fix is genuine. The old assertion collapsed the destination to a 4-value enum; the new one compares the exact string the generator emitted against the string rebuilt straight from the data, bypassing redirects.ts entirely. Any error introduced in the generator — the layer this PR actually edits — is now caught. I proved that by mutation rather than by reading it.
Mutation results — own temp clone at 25edffc, deleted afterwards
Command: npx vitest run src/config/redirects.test.ts in apps/website. Baseline 29/29 pass (matches the author's reported 29/29). Every mutation guarded with git diff --quiet so a no-op edit could not masquerade as a pass; restored and verified by blob hash against git rev-parse HEAD:<file> after each; final re-run green.
| # | Mutation | Layer | Result | Failing assertion |
|---|---|---|---|---|
| M1c | one row → /hub/models/this-does-not-exist/ |
redirects.ts |
FAIL | redirects.test.ts:284 (frozen tables) + :68 (sync) |
| M2c | model A's row → model B's existing page | redirects.ts |
FAIL | redirects.test.ts:284 + :68 |
| M6 | workflow A's row → workflow B's existing page | redirects.ts |
FAIL | :284 + :169 + :68 |
| M3b | one group's permanent back to false |
redirects.ts |
FAIL (6 tests) | :284, :169, :68 |
| M7 | delete a whole row | redirects.ts |
FAIL | :284 + :68 |
| M1b | one row → non-existent page | vercel.json |
FAIL | :68 only |
| M2b | wrong-but-valid destination | vercel.json |
FAIL | :68 only |
| M3a | one row permanent: false |
vercel.json |
FAIL | :68 only |
| M4 | delete a whole row | vercel.json |
FAIL | :68 only |
| M1a | newSlug → non-existent |
model-urls.ts |
FAIL | models-url-registry.ts:143 throws at module load |
| M2a | model A + all its aliases → model B's existing page, vercel.json regenerated |
model-urls.ts |
FAIL | registry duplicate guard: "/hub/models/flux-2-max-text-to-image is registered twice" |
M2c is the decisive one: the wrong-but-valid case cycle 1 said could never be caught now fails at redirects.test.ts:284. The author's claim on my cycle-1 comment is accurate. Credit where due — this is a real fix, not a reworded one.
No mutation left the suite green. Nothing uncovered to name. Three independent layers catch a bad mapping: the new frozen-table test (generator errors), the pre-existing sync test at :68 (committed-artifact drift), and models-url-registry.ts module-load guards (unregistered destination, duplicate registration) — the last of which even catches a coherent repoint of a model and all its aliases with vercel.json regenerated.
A process note on my own method, because it nearly produced a wrong PASS. M1a and M2a first came back green. They were false passes from a stale Vite transform cache in node_modules/.vite: the mutation was on disk and git diff confirmed it, but the test ran the old module. Clearing the cache turned both red. Every result in the table above was re-run with the cache cleared. A PASS from this suite is only trustworthy with node_modules/.vite removed first.
redirects.ts +1-0
The single added line is line 46:
// Permanent since comfy-router#46: prod serves /hub/models, /hub/workflows and /hub/apps from this site.A comment. Test-support only — zero effect on emitted behaviour. Confirmed by the structural diff: the generated output is identical in every field but permanent.
Redirect-set diff, re-measured at this head (not inherited from cycle 1)
Parsed both vercel.json files out of git at c9326eb and 25edffc and diffed the key sets in both directions. The head copy I diffed is byte-identical to the one fetched from the GitHub contents API at 25edffc.
| Check | Result |
|---|---|
| Rows, base → head | 857 → 857 |
| Sources added | 0 |
| Sources removed | 0 |
| Destinations changed | 0 |
| Patterned sources | 0 base / 0 head |
permanent false→true |
786 |
permanent true→false |
0 |
| Any other field changed on any row | 0 |
permanent: true count |
60 → 846 |
| Row order identical | yes |
| Duplicate sources | 0 |
Non-redirects sections of vercel.json changed |
none (headers, build keys untouched) |
| Chains — target is itself a redirect source | 0 (exact and slash-variant) |
Flipped rows outside /models |
0 |
All four cycle-1 numbers reproduce exactly. No source and no destination changed this cycle — nothing to enumerate, which is the single most important negative result in this review. The 786 decompose exactly: 2 × (1 + 148 + 212 + 30 + 2) — /models, 148 model slugs, 212 aliases, 30 workflows, 2 apps, each in both slash forms. Measured per group: /models 2, model+alias 720, workflows 60, apps 4.
Do the flipped targets resolve? Sampled 25 of the 181 distinct destinations against production: 25/25 returned 200 with num_redirects=0. No 404, and no 308-onto-a-redirect chain anywhere. Consistent with cycle 1's full 210/210 sweep, and since 0 destinations changed, that sweep still stands.
Deleted precondition constants — treated as a flag-level change, cleared
HUB_ROUTER_PENDING and HUB_APPS_ROUTER_PENDING gated the rollout, and a deleted precondition arrives looking routine, so I checked it the way I would a touched flag check.
- What gated what:
isPermanent = (row) => row.temporaryBecause === undefined(redirects.ts:136). The constants were only ever values oftemporaryBecause; deleting the field is precisely the mechanism that flips a row to 308. Nothing subtler is happening. - Nothing still reads them: repo-wide grep at this head for
HUB_ROUTER_PENDINGandHUB_APPS_ROUTER_PENDING→ 0 hits. No dangling reference. - The mechanism survives for other rows:
temporaryBecauseis still declared at:21and still used by 7 rows (/trust,/login,/share-news×2,/minimax×2 …). Only the/modelsrows lost it.
Exposure classification
None of the listed categories. Not a catalogue entry, not a new app, not a new feature surface on an existing page, and no flag check, default or targeting is touched — I diffed every field of all 857 rows and the only semantic change is permanent: false → true. Rollout state is unknown from the repository; I am not calling anything shielded.
Merge order vs #20198 / #20201
Settled in cycle 1 and unchanged by d128352: this PR adds 0 new redirect sources (measured above), and those gates read source only, never permanent. Not re-derived.
Scope creep from the lint bot
A lint bot has dragged an unrelated app component into a redirect-config PR. 25edffc is actions-user "[automated] Apply ESLint and Oxfmt fixes" and touches src/components/common/WorkflowActionsList.vue (+1−1) — the main frontend app, entirely outside this PR's subject. Not the author's doing.
I glanced at the line as asked: bg-(--primary-background) → bg-primary-background. It is a formatting no-op. packages/design-system/src/css/style.css:561 declares --color-primary-background: var(--primary-background) inside the @theme inline block (opens at :424), so Tailwind v4 emits background-color: var(--primary-background) for both forms, and because the block is inline the light/dark overrides at :228/:384 still apply. No visual change.
Findings
| Severity | Cycle 1 | Cycle 2 |
|---|---|---|
| BLOCKER | 0 | 0 |
| SHOULD FIX | 1 | 0 |
| TRIVIAL | 3 | 3 |
Resolved since cycle 1
- SHOULD FIX-1 (circular assertion) — fixed in
d128352, verified by mutation M2c failing atredirects.test.ts:284. Genuine fix. - TRIVIAL-1 (no in-repo record of the precondition) — addressed by the new comment at
redirects.ts:46. One residual nit inline.
Open — all TRIVIAL, none blocking
- TRIVIAL-1 —
redirects.ts:46: the new comment records the why but drops the "cannot be retracted" warning, which was the operative half. - TRIVIAL-2 —
redirects.ts:56/:60: the 64noindexrows, recorded as a decision with the measurements above. - TRIVIAL-3 —
redirects.test.ts:273: the frozen tables are not literally independent; a one-line comment should stop a future editor weakening the registry guards that actually close the gap.
CI — check-runs API, paginated
The suite did not run at this head. 25edffc has only 4 check-runs: 3 success (dispatch, cla-assistant, Socket Security: Project Report) + 1 neutral (Socket Security: Pull Request Alerts). 0 failure, 0 pending, 0 cancelled. No unit, build, e2e, lint or typecheck job exists at this SHA. Combined status reports success — it is green because almost nothing ran, which is exactly the kind of PASS not to bank.
The real suite is green one commit back at d128352: 74 success, 34 skipped, 0 failure, 0 cancelled, 0 pending, including website-unit, website-build, website-e2e (4 shards), unit / test-shards (1–8), lint-pr / typecheck and lint-pr / repo-checks. No failure anywhere, so nothing is superseded.
d128352 is the last commit touching apps/website, so the website result does carry to head. What does not carry: 25edffc edits a main-app Vue file, and no main-app unit, build or lint-and-format job has run at head — mildly ironic for a lint-autofix commit. I checked that line by hand instead (no-op, above).
My own run differs from CI only in being scoped to one file with the Vite cache cleared; same result, 29/29.
Standard caveat: none of those ticks resolves a redirect target, detects a chain or notices a dropped row. The production measurements above are the basis for this verdict — not the green ticks, and not the author's claims, though on the points I could check the author's claims held.
APPROVE — code only. Consistent with cycle 1: the one SHOULD FIX is genuinely fixed, 0 destinations and 0 sources changed, all 32 gated targets and a 25-row sample of the rest answer 200 in one hop with no chains, and the noindex BLOCKER does not survive measurement.
This approval does not make the PR mergeable. Three things stand outside my gate: the PR is conflicted and needs a rebase; @coderabbitai's CHANGES_REQUESTED is still open and only its author can clear it; and gate one (author verifies the preview) is unmet. Queueing is webshepherd-bot's call, not mine.
…ent-hub-redirects # Conflicts: # src/components/common/WorkflowActionsList.vue
dante01yoon
left a comment
There was a problem hiding this comment.
Reviewed the literal redirect changes and regenerated artifact. All 786 legacy Models rules are permanent; MiniMax stays temporary. Focused redirect and registry tests: 64 passed on an archived copy of this head with existing installed dependencies. Live GET checks returned 200 without redirects for the hub, both app destinations, a workflow destination, and sampled actual model destinations. Latest CI attempts have no failing or pending checks. No blocking findings; full browser QA and exhaustive live destination checks: NOT_RUN.
…l on redirected llms links (Comfy-Org#20198) ## Summary Root `llms.txt` now sends agents to `/hub/models/` (and its `.md` copy) for hosted models and to the Comfy Router docs for calling them from code. The llms.txt link check now covers every built `llms.txt` plus `llms-full.txt`, and fails on a link that hits a redirect (literal or `:param` / `:path*` pattern) or that the build never produced. ## Changes - **What**: - `public/llms.txt`: a Comfy Router bullet under "How agents should use Comfy" (route, `X-API-Key`, schema URL, quickstart and models docs as `.md`); the "model catalog" line names both catalogues (hosted vs. runs in ComfyUI); two "Agent-readable surfaces" rows for `/hub/models.md` and the Router models doc; the Models Overview row links its `.md` copy. `/p/supported-models` links stay as they are. - `validate:llms-txt-links` walks `dist/llms.txt`, `dist/llms-full.txt` and every section `llms.txt`, and names the file with each failure. Three checks per link: redirect source, not in the build (Workflows app routes exempt), canonical drift. - `findRedirectedLinks` matches pattern sources and treats the `.md` copy of a redirected page as redirected. The test that only allowed literal sources in `vercel.json` is replaced by matcher tests. ## Review Focus - [ ] Changed from the ticket: FE-2925 Part A (a generated `/models/llms.txt`) is not built. `/hub/models.md` already lists every hosted model page, so root `llms.txt` points at it instead of a second list. - [ ] Changed from the ticket: per-model schemas are linked through the Router docs. The ticket said they need a key; the docs also publish them at `docs.comfy.org/router-schemas/...` with no key. - [ ] `redirectSourcePattern` copies the logic of `sourcePattern` in `redirects.test.ts`. That file is left alone to avoid a conflict with Comfy-Org#20191; they can share one helper after it merges. - [ ] Redirects that live only in comfy-router (for example `/models` to `/hub/models/` on prod) are not in `vercel.json`, so this check can't see them. Fixes FE-2925 Fixes FE-2988 https://linear.app/comfyorg/issue/FE-2925 · https://linear.app/comfyorg/issue/FE-2988 --------- Co-authored-by: GitHub Action <action@github.com>
…ng or names another page (Comfy-Org#20199) ## Summary The website build now fails when a page advertises a markdown copy that was not built, or whose front matter names a different page as canonical. Today all 897 advertised copies on comfy.org exist and point back correctly, but nothing keeps it that way. Fixes FE-2984 (https://linear.app/comfyorg/issue/FE-2984) Changed from the ticket: - The `Link: rel="canonical"` header for `.md` files (part A) is not in this PR. It needs a `vercel.json` header rule, and an open PR (Comfy-Org#20191) owns that file right now. The rule is in the Linear ticket. - The supported-models twin canonical is already slash-terminated on main, so there is no front-matter fix to make. ## Changes - `src/utils/markdownAlternateAudit.ts` holds the rule: for each page with `<link rel="alternate" type="text/markdown">`, the file must exist and its `canonical:` must equal the page's `<link rel="canonical">`. - `scripts/validate-markdown-alternates.ts` runs that rule over `dist/` and exits 1 on any problem, or when no page advertises a copy at all (so the check can't pass by finding nothing). - `ci-website-build.yaml` runs `validate:markdown-alternates` right after the llms.txt link check. ## QA - [x] Unit: 7 cases in `markdownAlternateAudit.test.ts` pass; 2 fail when the missing-file and canonical checks are disabled - [x] Build: `validate:markdown-alternates` passes on a `build:e2e` build (855 copies), and fails with exit 1 after deleting `dist/cli.md` and stripping the slash from one model copy's canonical - [ ] Not covered: the HTTP `Link` canonical header; that waits on the `vercel.json` change
…-Org#20291) MaanilVerma ## Summary The supported-models section moves into the Hub. The 334 open-weight model file pages keep their content and move to `/hub/models/local/<slug>/`, and every old address sends a one-hop 308 to its new home, so the ~45k clicks per 90 days these pages earn carry over. ## Changes - **What**: - The page, its markdown twin, the index and the `llms.txt` catalogue move from `src/pages/p/supported-models/` to `src/pages/hub/models/local/`. Only URLs change: the title, description, h1, Hugging Face button, folder, docs link, FAQ and JSON-LD stay the same. - New registry kind `local`. `isWorkshopRoute` leaves `/hub/models/local/*` alone, so these pages are always built, indexable, in the sitemap and twinned, also with `WORKSHOP_IN_BUILD=0`. The pages are file routes, so the flag cannot drop them. - Redirects (all 308, generated into `vercel.json`): - Each file page, its slash form and its `.md` match five `/p/supported-models/:slug(a|b|…)` rules that name only built slugs. The 14 retired slugs and `stability-ai` stay 404. - The 11 quantization aliases go to their file's new page. - The 49 partner pages go to the Hub model for the same model or version (13), to the provider's current model (18), or to `/hub/models/` when no Hub page matches (18). They match one `:slug(a|b|…)` rule per destination, 24 in all. `__fixtures__/published-partner-model-pages.json` pins each destination outside the table. A partner slug that leaves the data keeps its row, so its published address keeps working. - `retiredLocalModelPages` (empty for now) sends a file page that a data refresh removes to the page that replaces it. - The index, `.md` and `llms.txt` go to `/hub/models/local/`, `/hub/models/local.md` and `/hub/models/local/llms.txt`. - Footer, nav, drops card, showcase CTA, root `llms.txt` and `llms-full.txt` link the new addresses. The build has 0 links to `/p/supported-models`. ## Review Focus - [x] One PR, not the A/B split: the change is 248 non-test lines (56 of them the partner table). The pages move instead of being duplicated, so no indexable duplicate ever exists. - [x] Route budget: 946 redirects + 4 headers (was 857 + 4). The slug groups stay under Vercel's 2,048-character source limit (test). - [x] Canonical churn: `__fixtures__/published-local-models.json` freezes the 334 published slugs. If `generate-models.ts` turns one into a `canonicalSlug` stub, CI fails. - [x] Partner pages: `.md` twins redirect to the Hub page's twin. The partner branches in the shared copy modules are now unused in the build; they stay for a follow-up cleanup. - [x] Launch-page breadcrumbs ("Models") point at `/hub/models/` through `routes.workshop`. Only the local file pages keep the crumb to `/hub/models/local/`. - [x] The Drops card keeps its link to `/hub/models/local/`, and its copy now describes open-weight model files that you run locally. - [x] Sibling PRs: Comfy-Org#20191 also edits `redirects.ts` and `vercel.json` (merge main and re-run `sync:redirects`). Comfy-Org#20198 edits `public/llms.txt` line 49, and its redirected-link check must expand `:slug(…)` sources like `llms-txt.test.ts` does here. - [x] How Comfy-Org#19786 ports onto this: - Its page and component changes apply to `src/pages/hub/models/local/` instead of `src/pages/p/supported-models/`. Its `public/llms.txt` lines move to the new URLs. - Of its 32 new slugs, 27 are local files and need no rows: the slug groups pick them up from the data. The other 5 are partner slugs (`fish-audio`, `pixverse-6`, `muse-image`, `pruna-p-video-2`, `hy-image-3-5-preview`) and each needs a row in `partner-model-redirects.ts`. A test fails until they have one. - It drops the partner slugs `veo-2` and `veo-3`. Keep their rows: their old addresses are published. - It drops the published file page `minimax-h3-video-vae-fp16` (81 clicks in 90 days). Add it to `retiredLocalModelPages` with its replacement page, or the published-pages test fails. - `/hub/models/local/all/` needs a `models-url-registry` entry, or `workshop-release-gate` throws at build time. - `/p/supported-models/all/` needs a literal 308 row only if Comfy-Org#19786 ships first. Prod returns 404 for it today. Fixes FE-2964, FE-2933 Linear: https://linear.app/comfyorg/issue/FE-2964 , https://linear.app/comfyorg/issue/FE-2933
Summary
The old
/modelsaddresses now redirect permanently (308) to their/hubpages, so search engines move their ranking to the new URLs. comfy-router#46 is live on prod, which was the only reason they were temporary.Changes
temporaryBecausefrom every old Models row inredirects.ts:/models, each old model slug and alias, each/models/workflows/<name>and each/models/apps/<name>, both slash forms (786 rules).vercel.jsonis regenerated withsync:redirects. The tests that pinned these rows to 307 now pin 308./minimaxand/zh-CN/minimaxstay 307. Their reason is still open: the MiniMax page owner has not signed off that/minimax/won't be reused (FE-3087, asked for in review on feat(website): one typed redirect list, generated vercel.json #19359)./trust,/loginand/share-news*stay 307 for their own reasons, so thetemporaryBecausefield stays./minimaxpart is not in this PR. It is a one-line flip once FE-3087 has the sign-off.Review Focus
/hub/models/,/hub/workflows/and/hub/apps/answer 200 from the website, and each old URL reaches a 200 page in one hop./models/apps/*rows had their own reason (router sends/hub/apps/here). That is also true on prod now, so they flip too.Fixes FE-3083 (https://linear.app/comfyorg/issue/FE-3083)
Part of FE-2974 (https://linear.app/comfyorg/issue/FE-2974)