Replace broken hub API thumbnail fetch with local webp lookup.
Each template JSON has a companion -1.webp; we use the first
template's webp per model, served via raw.githubusercontent.com.
178/207 models now have thumbnails.
- model-metadata.ts: add hubSlug field (only where comfy.org/workflows/model/{slug} returns 200)
- models.ts: add hubSlug to Model interface
- ModelHeroSection: use hubSlug for CTA, fall back to /workflows if absent
- [slug].astro: pass hubSlug, expand FAQ to 4 questions, improve pageDescription
- SKILL.md: document hubSlug field with verification note
Resolved conflict in translations.ts: keep models UI keys from this branch
and payment status page keys added in main.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Nodes in the 'essentials' category do not have type 'core'. The check
has been updated to instead use the dedicated `isCoreNode` prop.
No tests currently. The existing tests for this code section all mock
out the relevant code path and properly writing a test for this would
take far more time than I can allocate right now.
┆Issue is synchronized with this [Notion
page](https://app.notion.com/p/PR-11809-Fix-core-node-detection-for-missing-nodes-3536d73d3650815aabb2deb54c4ecec4)
by [Unito](https://www.unito.io)
generate-models.ts:
- Add buildTutorialUrlMap() — reads all locale index*.json files, maps each
template's tutorialUrl to the raw model filenames in that template file
- Adds docsUrl to OutputModel; populated for 106 of 207 models at generate time
model-metadata.ts:
- Remove Wave-N section comments
- Add docsUrl for openai-dall-e, ltxv-api, wan-api, flux-1-kontext-dev
models.ts:
- Forward docsUrl from generated JSON into Model (metadata overrides still apply)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
## Summary
Extends the new unsafe Error assertion lint rule from #11845 to also
reject angle-bracket assertions.
## Changes
- **What**: Adds a `TSTypeAssertion` selector alongside the existing
`TSAsExpression` selector and broadens the lint message to cover Error
type assertions generally.
## Review Focus
This is stacked on #11845 and only addresses the lint-rule bypass for
`<Error>value` syntax.
┆Issue is synchronized with this [Notion
page](https://www.notion.so/PR-11909-fix-catch-angle-bracket-Error-assertions-3566d73d365081f58ecfecfa2e948c33)
by [Unito](https://www.unito.io)
---------
Co-authored-by: bymyself <cbyrne@comfy.org>
*PR Created by the Glary-Bot Agent*
---
## Summary
Updates the "Install via GitHub" CTA buttons on the `/download` page to
deep-link to the install instructions section of the ComfyUI README
(`#installing`) instead of the repo root, so users land directly on
setup steps.
## Changes
- `apps/website/src/config/routes.ts`: add `externalLinks.githubInstall
= 'https://github.com/Comfy-Org/ComfyUI#installing'` (separate from
`externalLinks.github`, which is still used by the navbar/footer/star
badge for the generic repo link).
- `apps/website/src/components/product/local/HeroSection.vue`: switch
the secondary CTA next to "Download Local" from `externalLinks.github`
to `externalLinks.githubInstall`.
- `apps/website/src/components/product/local/EcoSystemSection.vue`: same
swap on the ecosystem-section CTA.
The platform-aware `Download Local` button (Windows/macOS installers via
`useDownloadUrl`) and the generic GitHub social/repo links elsewhere on
the site are unchanged.
## Verification
- `pnpm --filter @comfyorg/website typecheck` — 0 errors
- `pnpm --filter @comfyorg/website test:unit` — 23/23 passing
- `pnpm exec eslint` on changed files — clean
- `pnpm exec oxfmt --check` — clean
- Manual via `pnpm dev` + Playwright DOM assertion on `/download`:
- Hero "INSTALL FROM GITHUB" →
`https://github.com/Comfy-Org/ComfyUI#installing` ✓
- EcoSystem "INSTALL FROM GITHUB" →
`https://github.com/Comfy-Org/ComfyUI#installing` ✓
- Other "GitHub" links (nav, footer, star badge) → unchanged at
`https://github.com/Comfy-Org/ComfyUI` ✓
Per request from #website-and-docs: the local download surfaces should
at least link to the ComfyUI install instructions on GitHub. Companion
change to comfy-org/website#227.
## Screenshots
┆Issue is synchronized with this [Notion
page](https://www.notion.so/PR-11852-fix-website-point-Install-via-GitHub-buttons-to-install-docs-anchor-3546d73d365081fe8370cd675ae8f896)
by [Unito](https://www.unito.io)
---------
Co-authored-by: Glary-Bot <glary-bot@users.noreply.github.com>
*PR Created by the Glary-Bot Agent*
---
Synthesizes lessons from the recent `v1.43.16` backport session (PRs
#11856–#11862) into the `.claude/skills/backport-management/` skill.
**Documentation only — no code, no workflow, no automation changes.**
## What's new
### `SKILL.md`
- **Quick Start** expanded from 7 to 12 steps, surfacing the pre-filter
and target-file-existence check that should run BEFORE per-PR triage
- **New gotcha**: *Cherry-Picked Tests Can Reference Files Added By
Earlier Unbackported PRs* — drop the test, document why
- **New gotcha**: *Backport-Only Compatibility Shims* — when a
refactor-style fix's mechanism relies on changes that aren't on the
older branch, a literal cherry-pick can recreate the bug for extensions
still using the old contract. Real example: #11541 + `LGraphNode.vue`
`handleDrop`
- **New section**: *Path Pre-Filter* under Auto-Skip Categories —
auto-skip PRs touching only `apps/website/`, `browser_tests/`,
`.github/`, `packages/design-system/`, generated types, `.claude/`,
`docs/`, or `*.stories.ts`. Removes 30–50% of candidates without reading
their bodies
### `reference/analysis.md`
- **New subsection**: *Verify Target File Existence* — `git cat-file -e`
before cherry-pick to skip PRs targeting features the branch doesn't
ship, faster than letting cherry-pick fail with modify/delete
- **New section**: *Tiered Triage* — **Tier 1** (core editor
must-haves), **Tier 2** (cloud-distribution only), **Tier 3** (skip)
before per-PR Y/N. Surfaces release-engineering decisions a flat
MUST/SHOULD list obscures
### `reference/discovery.md`
- Reconciliation workflow combining Slack bot list + git gap,
subtracting already-backported PRs (extracted via `Backport of #` grep
on the branch)
- Clarifies that no single source is authoritative
### `reference/execution.md`
- **New Step 0**: *Test-Then-Resolve Pre-Pass* — moved the dry-run loop
to the top of the workflow so you classify clean vs conflict before
triggering automation
- **New inline guard**: *Public-API conflict review* in the manual
cherry-pick loop — consult oracle BEFORE pushing if resolution touches
LiteGraph callbacks, `node.*` methods, or extension-API surfaces
- **Per-PR validation block** (typecheck + targeted unit tests + ESLint
+ oxfmt) before push, in addition to wave verification
- **Three new lessons learned** (19–21) covering the above
- **New PR Body Template** for manual cherry-picks — non-negotiable
conflict-resolution section so reviewers don't have to re-derive
resolution logic from the diff six months later
## Why
1. **Path pre-filtering** caught 35 of 61 candidates as `skip` before
any per-PR analysis was needed during the `v1.43.16` session
(`apps/website/`, CI, tests, design-system). The previous flat
MUST/SHOULD/SKIP rubric meant reading every PR.
2. **Oracle review on PR #11856** (#11541 backport) caught a regression
in `LGraphNode.vue:823` where the upstream PR's removal of the legacy
`handled === true` sync-return path would silently break custom nodes
still using the old `onDragDrop` contract — recreating the very
duplicate-node bug the PR was fixing. The skill had no guidance for this
class of issue.
3. **Two separate backports** (#11180, #11541) hit the same
modify/delete conflict pattern: a test file added on `main` by an
earlier unbackported PR. The skill's *Conflict Triage* table covered
modify/delete generally but didn't surface this specific anti-pattern of
smuggling in test scaffolding without its prerequisites.
4. **Discovery sources** — the skill assumed Slack bot + git gap as
inputs but didn't show how to reconcile them with already-backported
PRs, leading to potential double-cherry-picking.
## Verification
- No code, no workflow, no automation changes — skill documentation only
- `git diff --check` clean
- All four edited files render as valid markdown
- Cross-references between SKILL.md and `reference/*.md` files validated
by line-count check
## Out of scope (intentionally not changed)
- The `pr-backport.yaml` GitHub Action — the skill describes this but
doesn't own it
- The Slack bot — described as *Source 1*, not modified
- The PR title convention `[backport TARGET] ...` — kept as-is
(CodeRabbit's auto-skip filter relies on it)
┆Issue is synchronized with this [Notion
page](https://www.notion.so/PR-11868-docs-skill-improve-backport-management-with-tiered-triage-path-pre-filter-and-public-3556d73d3650814faec3df795567bee3)
by [Unito](https://www.unito.io)
---------
Co-authored-by: Glary-Bot <glary-bot@users.noreply.github.com>
Co-authored-by: GitHub Action <action@github.com>
*PR Created by the Glary-Bot Agent*
---
## Step 2 of 2 — GitHub squash-commit incident recovery for #11460
This is the companion to #11629. After #11629 (the revert) merges, this
PR re-applies the original PR #11460 contents on top of the post-revert
`main`, restoring the intended state of the codebase.
### ⚠️ Sequencing — must merge after #11629
Until #11629 is merged, this PR's diff against `main` is empty (because
`main` currently still contains the squashed bug-dump-ingest commit that
#11629 will revert). After #11629 is merged into `main`, GitHub will
recompute the diff and this PR will show a clean re-add of the same 5
`.claude/skills/bug-dump-ingest/` files.
### Verification
The branch was constructed by:
1. Branching from `glary/revert-pr-11460` (the revert branch in #11629).
2. `git checkout FETCH_HEAD -- .claude/skills/bug-dump-ingest/` from
`refs/pull/11460/head` to restore the exact files.
3. Committing them as a single squash-style commit.
I verified that all 5 file blobs are byte-for-byte identical to the
original squash commit `559922eaa5c129767c22275c206c6877931ac15c` by
comparing git object SHAs:
| File | Object SHA |
|---|---|
| `.claude/skills/bug-dump-ingest/SKILL.md` |
`413737835fa1c996291019483effdd39e6da33e5` |
| `.claude/skills/bug-dump-ingest/reference/examples.md` |
`4fc54a4f14b1359de63f558acca0de48c1b65c57` |
| `.claude/skills/bug-dump-ingest/reference/linear-api.md` |
`57986740df2ee02c19b81059f0f1e00e54c2a042` |
| `.claude/skills/bug-dump-ingest/reference/schema.md` |
`84db1a5818c04ee53a94167092ec76dd814992d4` |
| `.claude/skills/bug-dump-ingest/reference/verify-commands.md` |
`a2c99a43a030ccc3769692d3c09be74132645bb4` |
### Notes
- One commit on `refs/pull/11460/head` (an automated lint commit
`76ca1598e`) modified
`src/renderer/extensions/vueNodes/widgets/components/WidgetChart.test.ts`
to remove an `eslint-disable` directive. That change was **not** in the
original squash commit on `main` (verified via `git show --stat`), and
the directive has separately been removed from `main` by an unrelated
commit (#11550), so re-applying that change would now be a no-op. This
repair PR therefore intentionally restores **only the 5 bug-dump-ingest
files**, matching the original squash commit exactly.
- Branch name is `glary/repair-pr-11460` (the `glary/` prefix is
required by the tooling; otherwise equivalent to GitHub's suggested
`repair-pr-11460`).
Refs: #11460, #11629
┆Issue is synchronized with this [Notion
page](https://www.notion.so/PR-11630-Repair-re-add-bug-dump-ingest-skill-11460-GitHub-squash-commit-incident-recovery--34d6d73d365081acbd54c19316561fa9)
by [Unito](https://www.unito.io)
---------
Co-authored-by: Glary-Bot <glary-bot@users.noreply.github.com>
*PR Created by the Glary-Bot Agent*
---
## Step 1 of 2 — GitHub squash-commit incident recovery for #11460
GitHub flagged the squash commit for #11460
(`559922eaa5c129767c22275c206c6877931ac15c`) as affected by the
squash-commit incident that produced non-deterministic merges. This PR
is **step 1 of 2** in the recovery procedure GitHub asked us to follow.
### What this PR does
Reverts `559922eaa5c129767c22275c206c6877931ac15c` (the affected squash
commit for #11460, "feat: add bug-dump-ingest skill") on top of current
`main`.
### Diff
5 files removed, all under `.claude/skills/bug-dump-ingest/`:
- `.claude/skills/bug-dump-ingest/SKILL.md`
- `.claude/skills/bug-dump-ingest/reference/examples.md`
- `.claude/skills/bug-dump-ingest/reference/linear-api.md`
- `.claude/skills/bug-dump-ingest/reference/schema.md`
- `.claude/skills/bug-dump-ingest/reference/verify-commands.md`
`git revert` applied cleanly with no conflicts.
### Recovery procedure
Per GitHub's instructions:
1. **This PR (step 1)** — Revert the affected squash commit. Removes
both the original changes and any unintended changes that the incident
may have introduced.
2. **Next PR (step 2)** — Re-apply the original PR's changes from
`refs/pull/11460/head`, rebased onto post-revert `main`. Will be opened
as a separate "repair" PR.
### Review notes
- This is a pure revert; please confirm the diff is exactly the inverse
of #11460.
- After this PR is merged, the companion repair PR will land the
original changes back on `main` from a clean source ref, restoring the
intended state of the codebase.
- Branch name is `glary/revert-pr-11460` (the `glary/` prefix is
required by the tool that opened this PR — it's otherwise equivalent to
GitHub's suggested `revert-pr-11460`).
- Code review: Oracle reviewed and found 0 issues
(critical/warning/suggestion all 0). Ready to merge.
Refs: #11460
┆Issue is synchronized with this [Notion
page](https://www.notion.so/PR-11629-Revert-feat-add-bug-dump-ingest-skill-11460-GitHub-squash-commit-incident-recove-34d6d73d3650810f9ed3f6068e4f1511)
by [Unito](https://www.unito.io)
Co-authored-by: Glary-Bot <glary-bot@users.noreply.github.com>
## Summary
Per product feedback, scroll-jacking the page to step through
'Industries that create with ComfyUI' categories feels bad on desktop.
This PR removes the desktop scroll override and switches to hover-driven
imagery on lg+, while preserving the existing pin/scrub interaction on
mobile/touch breakpoints.
## Changes
- **What**:
- Gate `usePinScrub` setup to `(max-width: 1023px)` so the pin/scrub
only runs on touch breakpoints; desktop never engages it.
- Wire `@mouseenter` and `@focus` on each category button to update the
active category on desktop. Click still works on both modes via the
existing `scrollToIndex` (which falls through to a direct ref set when
no ScrollTrigger instance is present).
- Pass the same `(max-width: 1023px)` to `useParallax` via its existing
`mediaQuery` option so parallax doesn't run against the no-longer-pinned
section on desktop.
- Apply the section's `lg:h-[calc(100vh+60px)]` unconditionally on lg+
since pin no longer drives it; mobile height is still managed
dynamically by `usePinScrub`'s `cacheLayout`.
- **Breaking**: None.
- **Dependencies**: None.
## Review Focus
- Mobile path inside `usePinScrub.onMounted` is byte-identical to main —
only the early-return condition gained one extra clause
(`!window.matchMedia('(max-width: 1023px)').matches`), which mobile
evaluates to `false` so it falls through to the existing setup.
- `onCategoryHover` early-returns when `isEnabled` is true, making it a
no-op on mobile (where pin is engaged), so a tap doesn't accidentally
fight the scrub.
- `@focus` is wired alongside `@mouseenter` so keyboard tab navigation
also previews the imagery.
- The previous Lenis-on-macOS workaround from this branch is reverted —
it was only needed because the scroll override existed.
## Screenshots (if applicable)
N/A — interaction change. Test on desktop (≥1024px) by hovering category
labels — imagery should swap with no scroll-jacking. Test on mobile
(<1024px) by scrolling the section; pin/scrub should engage as before.
---------
Co-authored-by: Marwan Ahmed <marwan@Marwans-MacBook-Pro.local>
Co-authored-by: GitHub Action <action@github.com>
Co-authored-by: Alexander Brown <drjkl@comfy.org>
## Summary
Refactor E2E tests added in #11210 that repeated full prior-test bodies
as setup, combining duplicate pairs into single tests with named
`test.step()` blocks.
## Changes
- **What**: In
[`browser_tests/tests/keyboardShortcutActions.spec.ts`](../blob/batch-dispatch/cr-11556/browser_tests/tests/keyboardShortcutActions.spec.ts):
- Merge `Ctrl+Z undoes` + `Ctrl+Shift+Z redoes` → single test with two
`test.step()` blocks.
- Merge `Ctrl+, opens settings dialog` + `Escape closes settings dialog`
→ single test with two `test.step()` blocks.
- **What**: In
[`browser_tests/tests/topbarMenuCommands.spec.ts`](../blob/batch-dispatch/cr-11556/browser_tests/tests/topbarMenuCommands.spec.ts):
- Merge `Edit > Undo` + `Edit > Redo` → single test with two
`test.step()` blocks.
The redo step now reuses the post-undo state from its preceding step
instead of re-creating and re-undoing the node, removing the duplicated
setup the reviewer flagged.
## Review Focus
- Naming of combined tests and `test.step()` labels.
- Note: per @AustinMroz's [comment
thread](https://github.com/Comfy-Org/ComfyUI_frontend/pull/11210#discussion_r3113526265),
location 2 in the issue refers to the `Escape closes settings dialog`
test (which duplicated the `Ctrl+,` test body), not the `Delete` test
(which has unique logic). Treated accordingly.
Fixes#11556
┆Issue is synchronized with this [Notion
page](https://www.notion.so/PR-11835-test-combine-duplicated-undo-redo-and-settings-dialog-E2E-tests-with-test-step-3546d73d365081689df3c56bfbb6f4e4)
by [Unito](https://www.unito.io)
## Summary
Replaces all 7 production `as Error` type assertions with proper
`instanceof Error` narrowing or a new `toError()` helper, and adds an
ESLint rule to prevent new ones. First slice of #11429 (the `as Error`
category — 9 total occurrences, 7 production + 2 in a test file left
untouched).
## Changes
- **What**:
- New `src/utils/errorUtil.ts` exporting `toError(value: unknown):
Error` and `getErrorMessage(value: unknown): string | undefined`.
`toError` returns the value unchanged if already an `Error`, otherwise
wraps it (handles strings, `undefined`, JSON-serializable objects, and
circular refs via `String()` fallback).
- Refactored 7 production call sites:
- `src/services/gateway/registrySearchGateway.ts` — `toError(error)` for
`lastError` assignment in fallback loop
- `src/platform/cloud/onboarding/auth.ts` (×2) — `toError(error)` for
`captureApiError` Sentry calls
-
`src/renderer/extensions/vueNodes/widgets/composables/audio/useAudioRecorder.ts`
— `toError(err)` before forwarding to `options.onError`
- `src/extensions/core/load3d/LoaderManager.ts` — replaced `error as
Error & { response?: ... }` cast inside `isNotFoundError` with
`'response' in error` + nested narrowing
- `apps/desktop-ui/src/stores/maintenanceTaskStore.ts` — inline `error
instanceof Error ? error.message : String(error)`
- `apps/desktop-ui/src/components/maintenance/TaskListPanel.vue` —
inline `error instanceof Error ? error.message : undefined`
- New ESLint rule (`no-restricted-syntax` block named
`comfy/no-unsafe-error-assertion`) banning `TSAsExpression
TSTypeReference[typeName.name='Error']` in `src/**` and `apps/*/src/**`,
with test files (`*.test.ts`, `*.spec.ts`) excluded.
- 12 unit tests for the new helpers in `src/utils/errorUtil.test.ts`.
- **Breaking**: none
- **Dependencies**: none
## Review Focus
- The lint rule is scoped to non-test source files. Test files retain
freedom to use `as Error` for fixture construction; only 2 occurrences
exist (in `teamWorkspaceStore.test.ts` and `errorDialog.spec.ts`) and
they're intentional.
- `toError` is duplicated as inline `instanceof` narrowing in
`apps/desktop-ui/` rather than imported, since the desktop-ui workspace
doesn't share `@/utils/` with the main app and adding a path mapping for
one helper felt heavier than two inline guards.
- Remaining `as`-on-DOM categories (HTMLElement ×133, HTMLInputElement
×55, HTMLCanvasElement ×36, KeyboardEvent ×7, Element ×3, MouseEvent ×2,
Event ×2) are intentionally left for follow-up PRs to keep this one
reviewable.
Refs #11429
┆Issue is synchronized with this [Notion
page](https://www.notion.so/PR-11845-refactor-replace-unsafe-as-Error-assertions-with-type-guards-3546d73d36508137a015c4f9e8708f23)
by [Unito](https://www.unito.io)
generate-models.ts:
- Add thumbnailUrl field to OutputModel
- Fetch representative thumbnail from /api/hub/workflows?tag={slug} at generate time
- Set SKIP_THUMBNAILS=1 to skip network calls for offline/CI use
- Fix error handling: throw on parse failure instead of warn-and-continue
- make run() async; top-level error handler exits with code 1
models.ts:
- Add thumbnailUrl to Model interface
- Remove hardcoded publishedDate/modifiedDate (unreliable for 207 models)
[slug].astro:
- Pass model.thumbnailUrl as ogImage to BaseLayout (falls back to default)
- Remove article:published_time/modified_time meta tags (dates were hardcoded)
- Remove datePublished/dateModified from SoftwareApplication JSON-LD
model-page-discovery.yaml:
- Replace workflow_templates clone + pnpm generator with hub API call
(curl /api/hub/labels?type=model — no repo clone, no Node setup needed)
- Use sparse-checkout for generated-models.json only (no full clone)
- Add duplicate-issue guard: skip creation if open discovery issue exists
- Fix path inconsistency: single checkout at repo root, no working-directory hops
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
*PR Created by the Glary-Bot Agent*
---
Removes the "coming soon" badge from the Parallel Job Execution feature
card on the cloud pricing page (`comfy.org/cloud/pricing`).
## Changes
- `apps/website/src/components/pricing/WhatsIncludedSection.vue`: drop
`isComingSoon: true` from feature11 so it renders with the standard
check icon and no badge.
The `isComingSoon` mechanism (clock icon + yellow badge) is preserved in
the component for future use on other features.
## Note
The FAQ copy elsewhere on the site (`cloud.faq.9.a`) still references
"one active job at a time" and "parallel runs soon". That copy will be
updated separately.
## Verification
- `pnpm typecheck` (website): 0 errors
- `pnpm lint`: clean (1 pre-existing warning unrelated to this change)
- `pnpm format:check`: clean
- `pnpm test:unit` (website): 20 passed
- Visual check via Playwright on local dev server (see screenshot)
## Screenshots
┆Issue is synchronized with this [Notion
page](https://app.notion.com/p/PR-11819-fix-remove-coming-soon-badge-from-parallel-job-execution-3546d73d365081d19060f976095d03ac)
by [Unito](https://www.unito.io)
---------
Co-authored-by: Glary-Bot <glary-bot@users.noreply.github.com>
*PR Created by the Glary-Bot Agent*
---
## Summary
Registers a new nightly feature survey for the Queue Progress Overlay
using the existing feature-survey registry (same pattern as the merged
node-search survey, PRs #8175/#8355/#9934).
- New registry entry `queue-progress-overlay` → Typeform `HZ5saxry`,
threshold **16**, 5s display delay.
- `trackFeatureUsed()` wired at the major user-initiated handlers inside
the overlay so the survey triggers regardless of panel location
(floating-right v1 or docked-left v2).
- Run button and other ActionBar items that the overlay pops over from
are deliberately **not** tracked — tracking is scoped to interactions
that originate inside the job panel / queue progress overlay itself.
## Tracked interactions
Both variants share most sub-components, so tracking is instrumented
once at each logical surface:
- **`QueueProgressOverlay.vue`** (v1 container): `viewAllJobs`,
`interruptAll`, `cancelQueuedWorkflows`, `onClearHistoryFromMenu`,
`toggleAssetsSidebar`, `onCancelItem`, `onDeleteItem`, `inspectJobAsset`
- **`QueueOverlayExpanded.vue`**: job tab switches
- **`JobHistorySidebarTab.vue`** (v2 docked): job tab switches,
`clearQueuedWorkflows`, `onClearHistory`, `onCancelItem`,
`onDeleteItem`, `onViewItem`
- **`JobFilterActions.vue`** (shared): workflow filter + sort mode
selections
- **`JobHistoryActionsMenu.vue`** (shared): docked-history toggle +
run-progress-bar toggle
Deliberately **not tracked** to keep the signal clean:
- Hover handlers (ambient preview behaviour)
- Search-box keystrokes (debounced typing)
- Context menu open and menu-item dispatch — menu actions either bubble
through already-tracked terminal handlers (e.g. inspect-asset →
`onViewItem`) or are secondary operations (copy-id, open-workflow,
download). Avoids double-counting per code review feedback.
## How it works (inherits from existing infrastructure)
1. `surveyRegistry.ts` drives `NightlySurveyController` →
`NightlySurveyPopover`, which handles the Typeform embed.
2. Eligibility already gated on `isNightly && !isCloud && !isDesktop`,
once-per-user, 4-day global cooldown across all surveys, and opt-out.
3. Typeform response routing to #C0ALLT6Q3SQ is handled on the Typeform
side.
## Verification
- `pnpm typecheck` ✅
- `pnpm lint` ✅ (no new warnings)
- `pnpm knip` ✅
- `pnpm test:unit` on `src/components/queue`,
`src/components/sidebar/tabs/JobHistorySidebarTab`,
`src/platform/surveys` → **123/123 passing**
- Pre-commit hooks (stylelint, oxfmt, oxlint, eslint, typecheck) all
pass
- Manual: dev server + backend boot cleanly, app loads without new
runtime errors, `localStorage['Comfy.FeatureUsage']` layout verified to
match what `useFeatureUsageTracker` writes
## Notes
- Survey key `queue-progress-overlay` covers both v1 (floating-right)
and v2 (docked-sidebar) per product guidance: _"This should trigger
regardless of the location of the panel (docked from left or floating on
right)."_ Both surfaces are the same product feature — the survey is
intentionally scoped to the whole job-panel experience.
## Screenshots
┆Issue is synchronized with this [Notion
page](https://www.notion.so/PR-11560-feat-add-queue-progress-overlay-feature-survey-34b6d73d3650819a9a50fd67fd9b5941)
by [Unito](https://www.unito.io)
---------
Co-authored-by: Glary-Bot <glary-bot@users.noreply.github.com>
*PR Created by the Glary-Bot Agent*
---
## Summary
Adds the building blocks for a responsive media system on
`apps/website`, motivated by the gallery video blurriness raised in
Slack. Three independent pieces:
1. **`<SiteVideo>` Vue component + URL helper** — emits a `<video>` with
multiple `<source>` tags, designed to pair with assets named
`${name}-${width}.${format}` on `media.comfy.org`.
2. **`scripts/process-videos.sh`** — local-developer `ffmpeg` helper
that produces VP9/WebM + H.264/MP4 variants and a poster JPG. Not wired
into CI; the team uploads to `media.comfy.org` out-of-band.
3. **Marketing image conventions** — shared `MARKETING_FORMATS` /
`MARKETING_WIDTHS` constants and a README documenting how to render
local marketing images via Astro's built-in `<Picture>` from
`astro:assets`.
This PR is **infrastructure only** — no existing pages are modified.
Adoption (e.g. converting `HeroSection`, gallery videos) is a follow-up.
The new files are added to knip's ignore list with the existing "pending
stacked PR" pattern.
## Why this shape
- **No custom `<Picture>` wrapper.** Astro 5 already ships a
`ResponsiveImage` component (name conflict), and Astro's
`LocalImageProps | RemoteImageProps` discriminated union does not
survive a thin wrapper without unsafe `as` casts. Shared constants give
the consistency benefit at lower cost.
- **No CI media-upload step.** The `Release: Website` workflow currently
only refreshes the Ashby snapshot; wiring GCS uploads into it would
require new secrets and team coordination beyond this PR's scope. The
script runs locally and outputs are uploaded to `media.comfy.org` the
same way as today.
- **Single resolution per `<video>`.** `<source media="...">` inside
`<video>` is unreliable across browsers (Safari ignores it). The script
generates multiple widths so callers can pick one per page; JS-based
selection can be layered on later if metrics demand it.
## What's verified
- `pnpm --filter @comfyorg/website test:unit` — 30 pass (7 new for
`buildVideoSources` / `videoKey`)
- `pnpm --filter @comfyorg/website typecheck` — clean
- `pnpm --filter @comfyorg/website build` — 41 pages built clean
- `pnpm knip` — exit 0
- `oxfmt --check` and `oxlint` clean on all changed files
- `bash -n` on `process-videos.sh` clean; usage and missing-deps paths
exercised manually
- Manual: home page and `/gallery` rendered via `astro dev` — both
unchanged with zero console errors (screenshots attached)
## Review feedback addressed
After Oracle review, three follow-up commits land:
- **`SiteVideo` reactivity** — `sources` is now `computed`; the
`<video>` is keyed on the joined source URLs so it remounts when the
source set changes (browsers don't reload on `<source>` mutation).
- **`SiteVideo` accessibility** — `aria-hidden="true"` only when truly
decorative (no `alt` and no `controls`).
- **Shell script robustness** — probes duration with `ffprobe` and falls
back to `t=0` for clips shorter than 1s; enables `nocaseglob` so
`CLIP.MP4` is picked up.
- **Docs** — clarifies when to use `<SiteVideo>` (lightweight
multi-source) vs `<VideoPlayer>` (captions, controls, scrubber).
## Out of scope (follow-ups)
- Converting existing pages (`HeroSection`, customer detail heros,
gallery) to use the new components. Most current images are CDN-hosted
and migrating them is a separate decision.
- Re-encoding the gallery videos at a higher source width to actually
fix the blurriness — that requires the team to run `process-videos.sh`
against the source clips and re-upload.
- Combining `<SiteVideo>`'s multi-source support with `<VideoPlayer>`'s
rich chrome.
## Screenshots
┆Issue is synchronized with this [Notion
page](https://www.notion.so/PR-11869-feat-website-add-responsive-media-tooling-for-marketing-assets-3556d73d3650818899c7f9ed3204c9a5)
by [Unito](https://www.unito.io)
---------
Co-authored-by: Glary-Bot <glary-bot@users.noreply.github.com>
models.ts is now a 65-line merger file with no slug strings. The regex match
against models.ts would always return empty [], making every model appear new.
Read from generated-models.json instead.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Adding Astro pages as entry points surfaced pre-existing unused dep warnings
(@comfyorg/design-system, tailwindcss) that are unrelated to this PR.
The type export fix (non-export ModelDirectory/Model) is sufficient to resolve
the original knip failures.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
ModelDirectory and Model were exported but not imported by any consumer.
Make them module-private to fix knip unused-export check.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Architecture:
- Replace 1585-line hand-coded models.ts with generated-models.json (207 models,
auto-generated by running pnpm generate:models) + 40-line model-metadata.ts for
editorial overrides (docsUrl, blogUrl, featured) + slim 65-line models.ts that
merges both. No more manually maintained data file.
- Remove 830 per-model translation keys from translations.ts — displayName is now
a plain string from the generator, not a TranslationKey
- models.ts no longer imports TranslationKey; displayName/description are strings
- Add generate:models script to website package.json
Design token fixes (pattern-compliance violations):
- Replace text-[var(--color-primary-comfy-yellow)] with text-primary-comfy-yellow
- Replace text-white/70 with text-primary-comfy-canvas/70 (matches existing components)
- Replace invented --color-accent/--color-text-secondary with real tokens
- Fix h1 text from text-white to text-primary-comfy-canvas (matches HeroSection pattern)
i18n fixes:
- Add models.hero.tutorialCta, models.hero.blogLink, models.whatIs.heading,
models.whatIs.tutorialLink keys
- Use t() for all button labels and UI strings in ModelHeroSection.vue
- partner_nodes added to dirLabel in index.astro
Other:
- H1 now reads "{displayName} in ComfyUI" (FE-421 spec)
- Remove description prop from hero (redundant with What-is section)
- Fix GH Actions discovery: m.slug field now matches generator output
- Update add-model-page skill to document new 3-file architecture
Architecture:
- Replace 1585-line hand-coded models.ts with generated-models.json (207 models,
auto-generated by running pnpm generate:models) + 40-line model-metadata.ts for
editorial overrides (docsUrl, blogUrl, featured) + slim 65-line models.ts that
merges both. No more manually maintained data file.
- Remove 830 per-model translation keys from translations.ts — displayName is now
a plain string from the generator, not a TranslationKey
- models.ts no longer imports TranslationKey; displayName/description are strings
- Add generate:models script to website package.json
Design token fixes (pattern-compliance violations):
- Replace text-[var(--color-primary-comfy-yellow)] with text-primary-comfy-yellow
- Replace text-white/70 with text-primary-comfy-canvas/70 (matches existing components)
- Replace invented --color-accent/--color-text-secondary with real tokens
- Fix h1 text from text-white to text-primary-comfy-canvas (matches HeroSection pattern)
i18n fixes:
- Add models.hero.tutorialCta, models.hero.blogLink, models.whatIs.heading,
models.whatIs.tutorialLink keys
- Use t() for all button labels and UI strings in ModelHeroSection.vue
- partner_nodes added to dirLabel in index.astro
Other:
- H1 now reads "{displayName} in ComfyUI" (FE-421 spec)
- Remove description prop from hero (redundant with What-is section)
- Fix GH Actions discovery: m.slug field now matches generator output
- Update add-model-page skill to document new 3-file architecture
- 103-model registry (100 local + 3 partner nodes) in models.ts with docsUrl/blogUrl fields
- Dynamic [slug].astro route with SoftwareApplication + BreadcrumbList + FAQPage JSON-LD
- ModelHeroSection.vue with partner node support and docs/tutorial CTAs
- "What is X?" section on every model page targeting AI Overviews / PAA
- generate-models.ts parser with API_PROVIDER_MAP for 30+ partner node providers
- Auto-discovery GH Actions workflow (weekly, opens issue on new models)
- add-model-page Glary-Bot skill for non-dev Slack-driven PRs
- Index page listing all 103 models
Closes FE-421
## Summary
Use exact BLAKE3 hash lookups first for missing model/media detection,
and add a separate public-inclusive input asset cache so public input
assets are considered missing-detection candidates without changing the
user-only input assets shown in the UI.
## Changes
- **What**:
- Added `assetService.checkAssetHash()` for `HEAD
/api/assets/hash/{hash}` status-only existence checks.
- Added strict BLAKE3 hash helpers so only `blake3:<64 hex>` media
values and raw 64-hex BLAKE3 model metadata are sent to the hash
endpoint.
- Updated missing media detection to group BLAKE3 candidates by hash,
resolve them through the hash endpoint, and fall back to the legacy
asset list path for invalid/unverifiable/non-hash values.
- Updated missing model detection to use hash lookup for BLAKE3-backed
asset-supported candidates before falling back to the existing node-type
asset matching path.
- Added `assetService.getInputAssetsIncludingPublic()` backed by a
dedicated cache that fetches input assets with `include_public=true` for
missing media fallback checks.
- Kept `assetsStore.inputAssets` user-only for widget/UI display, while
invalidating the public-inclusive missing-detection cache when input
assets may change.
- Added abort handling for paginated asset fetches and shared
public-input cache callers so one aborted caller does not cancel the
shared fetch for other callers.
- Added regression coverage for hash lookup, fallback behavior, abort
paths, public input fallback detection, and cache invalidation.
- **Dependencies**: None.
- **Change size**:
- Production code: 4 files, 400 insertions, 24 deletions, net +376.
- Test code: 4 files, 806 insertions, 59 deletions, net +747.
- Total: 8 files, 1206 insertions, 83 deletions, net +1123.
## Review Focus
- The public-inclusive input asset cache is intentionally separate from
`assetsStore.inputAssets`. The existing store data is user-only and
drives the asset widgets/sidebar, so using it for missing input
detection misses public assets. Making that store public-inclusive would
change UI data semantics; this PR instead keeps the UI dataset unchanged
and adds a missing-detection-specific cache in `assetService`.
- Hash lookup is only used when the workflow exposes a valid BLAKE3
hash. Filename-like values and invalid hash values still use the legacy
fallback path.
- Missing model detection keeps the existing fallback behavior for
non-hash candidates and for hash checks that are invalid or fail
transiently.
- Async model download cache refresh behavior is left unchanged; this PR
avoids coupling model download completion to input asset cache
invalidation.
- No browser/e2e test was added because this changes the missing asset
detection data path, not UI interaction or rendering. The behavioral
coverage is in unit tests for the asset service and the missing
media/model scanners.
## Follow-up Items
- Fix `assetsStore.updateAssetTags()` partial-failure recovery. If
`removeAssetTags()` succeeds and `addAssetTags()` fails, the local model
asset cache can roll back to tags that the backend has already removed;
this should be handled in a focused model asset cache PR.
- Consider extracting shared hash-verification flow used by missing
media and missing model scans after this behavior stabilizes.
- Consider adding a concurrency cap or short-lived request cache for
large workflows with many unique hash lookups.
- Consider splitting `assetService.ts` further, e.g. hash helpers, abort
utilities, and the public-inclusive input asset cache.
- Consider tightening the asset hash service API shape so callers do not
directly depend on HTTP-oriented statuses such as `invalid`.
- Consider adding broader mutation-path coverage for public-inclusive
input cache invalidation once the cache has more consumers.
Linear: FE-534
## Screenshots (if applicable)
Before <false positive / missing image / public asset>
https://github.com/user-attachments/assets/db7ce2a9-b169-4fae-bf9f-98bb93d3ee6d
After
https://github.com/user-attachments/assets/29af9f9e-b536-4fcd-a426-3add40bcb165
┆Issue is synchronized with this [Notion
page](https://www.notion.so/PR-11873-Use-hash-lookup-for-missing-asset-detection-3556d73d36508165babafb16614be0d8)
by [Unito](https://www.unito.io)
Automated refresh of `apps/website/src/data/ashby-roles.snapshot.json`
from the Ashby job board API.
**Flow:**
1. `Release: Website` workflow ran (manual trigger).
2. This PR opens with the regenerated snapshot.
3. `CI: Vercel Website Preview` deploys a preview for review.
4. Merging to `main` triggers the production Vercel deploy.
The snapshot fallback in `apps/website/src/utils/ashby.ts` remains
intact: builds without `WEBSITE_ASHBY_API_KEY` continue to use the
committed snapshot.
Triggered by workflow run `25260868155`.
┆Issue is synchronized with this [Notion
page](https://www.notion.so/PR-11851-chore-website-refresh-Ashby-roles-snapshot-3546d73d365081579f98f13f7b58e611)
by [Unito](https://www.unito.io)
Co-authored-by: christian-byrne <72887196+christian-byrne@users.noreply.github.com>
## Summary
Adds tests for metadata parsers
## Changes
- **What**:
- add test file generation script
- identified & fixed bug in webp exif parsing over-reading
- identified & fix bug in mp3/ogg parser where it would read from a
fixed position instead of relative, causing incorrect reads throwing
RangeError
- added catch in latent + json parsing to resolve errors
┆Issue is synchronized with this [Notion
page](https://www.notion.so/PR-11307-test-add-metadata-parser-coverage-3446d73d36508108ac36dddcec0a54d4)
by [Unito](https://www.unito.io)
---------
Co-authored-by: GitHub Action <action@github.com>
## Summary
Add a new customer story for Groove Jones — Crocs x NFL FOOH holiday
campaign for Dick's Sporting Goods, delivered with Comfy.
## Changes
- **What**:
- New entry in
[`apps/website/src/config/customerStories.ts`](https://github.com/Comfy-Org/ComfyUI_frontend/blob/feat/groove-jones-customer-story/apps/website/src/config/customerStories.ts)
registering slug `groove-jones` with cover image hosted on
`media.comfy.org/website/customers/groove-jones/`.
- Added `customers.story.groove-jones.{category,title,body}` and
`customers.detail.groove-jones.topic-1` … `topic-10` translations (en +
zh-CN) in
[`apps/website/src/i18n/translations.ts`](https://github.com/Comfy-Org/ComfyUI_frontend/blob/feat/groove-jones-customer-story/apps/website/src/i18n/translations.ts).
10 sections matching design sidebar: INTRO, THE OUTPUT, THE PROBLEM, HOW
COMFY SOLVED THE PROBLEM, BRAND-TRAINED LORAS, MULTI-MODEL
ORCHESTRATION, THE PIPELINE, VERSION CONTROL, FINISHING IN NUKE, THE
TAKEAWAY.
- Includes 2 pull quotes (Doug Hogan, Dale Carman), 1 final blockquote +
author card, and 3 inline images.
- Routes `/customers/groove-jones` and `/zh-CN/customers/groove-jones`
are auto-generated by `[slug].astro`.
## Review Focus
- Contributors author card uses `TBD` placeholder names/roles — to be
filled in.
- No `readMoreHref` set yet (no public blog post URL).
- All 4 images uploaded to
`gs://comfy-org-videos/website/customers/groove-jones/` and served via
`media.comfy.org`.
<img width="1000" height="535" alt="Kapture 2026-05-02 at 23 17 04"
src="https://github.com/user-attachments/assets/28654d24-0d49-4303-82ac-b6923cd6bc93"
/>
┆Issue is synchronized with this [Notion
page](https://www.notion.so/PR-11849-feat-add-Groove-Jones-customer-story-3546d73d36508128a64bd6809ad77447)
by [Unito](https://www.unito.io)
Co-authored-by: Amp <amp@ampcode.com>
Fixes#11345
## Summary
`clearModel()` in `SceneModelManager` only traversed and disposed
`THREE.Mesh` instances, leaving `THREE.Points` objects (created by
`handlePLYModeSwitch()` for point-cloud mode) leaking GPU geometry and
material memory on repeated point-cloud loads/clears.
## Changes
- `SceneModelManager.ts`: extend the dispose traversal in `clearModel()`
to also handle `THREE.Points`, mirroring the pattern already used by
`removeAllMainModelsFromScene()`.
- `SceneModelManager.test.ts`: add regression test verifying
`geometry.dispose()` and `material.dispose()` are called for
`THREE.Points` children on `clearModel()`.
┆Issue is synchronized with this [Notion
page](https://www.notion.so/PR-11836-fix-load3d-dispose-THREE-Points-GPU-resources-in-clearModel-3546d73d365081718338e824bc3e737d)
by [Unito](https://www.unito.io)
## Summary
Adds `/llms.txt` at the apex following the [llms.txt
standard](https://llmstxt.org) — a curated, link-based markdown file
that tells LLM-based search agents (ChatGPT search, Perplexity, Claude
search, Google AI Overviews, etc.) what's most important on the site. It
complements `robots.txt` (crawler permissions) and `sitemap-index.xml`
(URL inventory) by giving AI agents a short, prose-friendly index they
can ingest into a context window.
## What's in the file
28 links across 6 sections:
- **Product** (6) — homepage, Local download, Cloud, Cloud pricing, API,
Enterprise
- **Workflows and Gallery** (2) — gallery + community workflows site
- **Customers and Case Studies** (5) — customers index + 4 named studios
(Series Entertainment, Moment Factory, Ubisoft Chord, Open Story
Movement)
- **Developers and Documentation** (4) — docs.comfy.org, ComfyUI repo,
Comfy-Org GitHub org, registry.comfy.org
- **Company** (6) — about, careers, contact, blog, privacy, terms
- **Optional** (5) — `zh-CN` locale variant, long-form enterprise case
studies, blog posts (de-prioritized per spec — agents can skip if
context-limited)
The intro paragraph names the four product surfaces (Local, Cloud, API,
Enterprise), the named customers, and the use-case industries (VFX &
animation, advertising, gaming, eCommerce/fashion) — so an agent that
ingests only the prose still gets the elevator pitch.
## Verification
- All 28 URLs verified live (`HTTP 200`) before commit.
- File is plain markdown — no build step. Astro/Vercel will serve it
from `apps/website/public/llms.txt` exactly as it serves `robots.txt`
(which lives in the same directory and ships at
`https://comfy.org/robots.txt`).
- Will verify on the Vercel preview deploy after this PR opens that
`curl -sI https://<preview>/llms.txt` returns `200` with a sensible
`content-type`. (`robots.txt` currently serves as `text/plain;
charset=utf-8` — `.txt` will likely do the same; that's fine for AI
agents.)
## Decisions
- **No `llms-full.txt` yet.** That variant inlines full prose of key
pages and requires curating substantive content. Deferred to a follow-up
— the marketing-site pages are mostly Vue-rendered hero/feature blocks
rather than long-form prose, so a meaningful `llms-full.txt` would need
either dedicated copy or a build step that flattens i18n strings +
section text. Tracking separately.
- **No comment line in `robots.txt`.** I considered adding a `# AI
agents: see /llms.txt` comment above the `Sitemap:` directive, but
decided against it: (a) the convention is to probe the well-known path
`/llms.txt` directly, not to discover it via robots.txt; (b)
`robots.txt` was just polished in #11823 with a deliberate compact
design and adding a non-standard comment would muddy that; (c) zero
implementations I checked actually parse robots.txt for llms.txt hints.
Easy to add later if needed.
## Context
Third of three follow-ups from the SEO/GEO sweep on 2026-05-02:
1. ~~Comfy-Router: add `X-Content-Type-Options: nosniff` to apex
security headers~~ (separate PR on `Comfy-Org/comfy-router`)
2. ~~Cloudflare: enable "Always Use HTTPS"~~ (dashboard toggle, no PR)
3. **This PR** — add `llms.txt` for GEO discovery
## Testing
- [x] All linked URLs return 200
- [x] File parses as valid markdown
- [ ] Preview deploy serves `/llms.txt` (will verify once preview is up)
┆Issue is synchronized with this [Notion
page](https://app.notion.com/p/PR-11830-feat-website-add-llms-txt-for-GEO-discovery-by-AI-search-agents-3546d73d365081a98c6bfc5301699f64)
by [Unito](https://www.unito.io)
