The `/html-anything/` landing page was loading 11 screenshots directly
from `raw.githubusercontent.com/nexu-io/html-anything/main/docs/…` —
that's 6.5 MB of raw PNGs per first visit, no AVIF/WebP transcoding,
and a separate cross-origin TLS handshake the rest of the site doesn't
pay. The TODO comment at line 70 of the page has been calling this out
since the page landed.
This PR routes those images through the same pipeline as `hero.png`:
- Source PNGs upload to `static.open-design.ai/landing/assets/html-anything/`
on the existing `open-design-static` R2 bucket. Upload package lives
at `~/Downloads/r2-html-anything-orig/` (README + upload.sh ready);
merge AFTER running that script.
- Page uses `imageAsset('html-anything/<name>.png', { width, quality })`
so CF Image Resizing emits `format=auto` AVIF/WebP variants — same
contract every other landing image already uses.
- Banner is the LCP element; gets `srcset` across 768/1280/1920/2400
(mirroring `heroImageSrcset`) plus its existing `fetchpriority="high"`.
- 10 below-the-fold screenshots become `<LazyImg>` so they pick up the
precise IntersectionObserver (rootMargin 300px) instead of Chrome's
native lazy load that over-prefetches up to 3000px.
Expected on first visit (mobile, AVIF-capable browser):
banner: ~150 KB AVIF (was 2.05 MB PNG)
10 screenshots ~80-120 KB each (was 500-900 KB PNG each)
total ~1.2-1.5 MB (was 6.5 MB)
LCP estimated -800ms to -1.5s
* feat(daemon): add CTA hierarchy static QA pass
Introduce apps/daemon/src/qa/cta-hierarchy.ts exporting a pure
analyseCtaHierarchy(html) that parses generated prototypes with cheerio
and flags three precision-biased findings: multiple-primary CTAs in the
same section, ambiguous-weight (all CTAs share identical class + inline
style), and misleading-prominence (secondary-coded copy like "Learn
more" / "了解更多" styled with primary weight).
CTA candidates come from <button>, <a>, role="button" with btn/button/cta
class markers plus CTA copy keywords covering both English (Get started,
Sign up, Buy, Subscribe, Learn more, ...) and Chinese (立即购买,
立即下单, 了解更多, ...). Weight is inferred from class tokens
(primary/solid/filled/accent/cta) and from non-transparent inline
background-color, matching the inverse of the issue #2251 sample where
the header CTA was rendered with the neutral .btn style.
This PR only ships the pure function plus its tests. HTTP route, CLI
subcommand, and any auto-repair feedback loop are deliberate follow-ups
so the first cut can land without touching the daemon HTTP surface.
Refs #2251
* fix(qa): respect container boundaries in CTA hierarchy heuristics
Two precision fixes from review of #2427:
- computeContainerKey()'s parent fallback keyed by tag name alone, so
flat layouts like <div><a class=btn-primary>...</a></div> repeated
for sibling cards all landed in 'parent:div' and
detectMultiplePrimary() reported a fake shared-section conflict on
what is in fact one CTA per card. Switch to parent-node identity
(positional index of the matched parent within its tag group, same
trick the landmark branch already uses), so each sibling wrapper
gets its own bucket.
- detectAmbiguousWeight() compared signatures across the entire
document, so two unrelated sections each containing one '.btn' CTA
with matching style would trigger 'ambiguous-weight' despite neither
container having 2+ CTAs. The PR body's rule is narrower — 'every
CTA in a container shares the same class + inline style' — so bucket
by containerKey first and only emit the finding for containers with
2+ CTAs whose signatures are identical.
Tests lock both behaviors down:
- sibling <div> card-grid without a landmark ancestor stays under the
multiple-primary threshold;
- one-CTA-per-section pairs stay under the ambiguous-weight threshold.
Fix electron-builder failure in `pnpm exec tools-pack linux build --to appimage`
by adding required description, author, and repository fields to generated
package.json in tools/pack/src/linux.ts.
* fix(daemon): remove 10-item cap from discovery TodoWrite plan prompt
The RULE 3 sentence in DISCOVERY_AND_PHILOSOPHY told the model to write
'a plan of 5–10 short imperative items'. That upper bound caused the agent
to cap every plan at exactly ten steps even when the task genuinely needed
more. The TodoWrite JSON schema imposes no maxItems constraint, so the cap
was entirely prompt-driven.
Replace '5–10 short imperative items' with 'short imperative items covering
the work'. TodoWrite intent, RULE 3 label, and planning-before-building
requirement all survive unchanged.
Red spec: apps/daemon/tests/prompts/discovery-todo-cap.test.ts
* fix(prompts): remove 10-item cap from contracts discovery copy and harden tests
[pass-6,7 BLOCKER] packages/contracts/src/prompts/discovery.ts still had
the old '5-10 short imperative items' wording. apps/web imports
composeSystemPrompt from @open-design/contracts (ProjectView.tsx:43),
so web-originated chat runs were still subject to the cap.
[pass-8 WARNING] discovery-todo-cap.test.ts did not cover the contracts
copy, leaving that path unguarded. Also no guard against semantically
equivalent re-introduction via 'at most / maximum / no more than'.
Changes:
- packages/contracts/src/prompts/discovery.ts: apply same wording fix as
apps/daemon; add inline rationale comment
- apps/daemon/src/prompts/discovery.ts: add inline rationale comment
- apps/daemon/tests/prompts/discovery-todo-cap.test.ts: add 4th assertion
blocking 'at most|maximum|no more than N item' re-introduction
- packages/contracts/tests/system-prompt.test.ts: add 5-assertion suite
guarding the contracts copy and composed prompt output
* refactor(daemon): introduce HTTP Request Adapter + typed Deps (proof on active-context-routes)
Adds a typed HTTP boundary Adapter under apps/daemon/src/http/ that replaces
the untyped ServerContext service-locator pattern (30+ fields, mostly any)
for route handlers. Routes become pure (input, deps) -> Result<output>
functions, unit-testable without Express or supertest.
Six new modules under apps/daemon/src/http/:
- types.ts Result<T,E>, ok(), err(), JsonRouteSpec, Handler,
RouteInputContext, HttpMethod, InputParser
- parse.ts rawInput(req), validationError(message, issues?)
- response.ts sendJson(), sendApiError(), statusForError() +
ERROR_STATUS_BY_CODE map
- origin-guard.ts guardSameOrigin(req, origin) wrapping isLocalSameOrigin
as a Result
- adapter.ts defineJsonRoute(), mountJsonRoute() (only place that
knows about req/res)
- index.ts barrel
active-context-routes.ts migrated as proof of pattern. parsePostActive(),
handlePostActive(), handleGetActive() are now pure functions; postActiveRoute
and getActiveRoute are exported route specs. The wire signature
registerActiveContextRoutes(app, ctx) is preserved so server.ts is untouched.
Spec at specs/current/daemon-http-adapter.md captures the strangler migration
order for the remaining route files (mcp-routes, chat-routes, artifact
routes, etc.) and a StreamRoute follow-up where the Run Orchestrator lands.
Wire-format note: cross-origin response moves from the legacy
{ error: 'cross-origin request rejected' } shape to the structured
{ error: { code: 'FORBIDDEN', message: ... } } shape. Backwards-compatible
via the existing CompatibleErrorResponse = ApiErrorResponse | LegacyErrorResponse
union in @open-design/contracts.
Validation:
- pnpm install (post-rebase, exit 0)
- pnpm --filter @open-design/daemon typecheck (both tsconfig.json and
tsconfig.tests.json silent => pass)
- pnpm --filter @open-design/daemon test: 15 new tests pass
(tests/http/adapter.test.ts + tests/active-context-routes.test.ts).
84 pre-existing failures across 23 files are unchanged and unrelated
to this PR (Windows symlink / short-name / colon-in-filename, upstream
behavior drift, missing plugin marketplace fixtures, and a freshly-
added tools-connectors-cli suite of 38 failures that landed during
the rebase).
Sharpens W4/W5 of specs/current/maintainability-roadmap.md and unlocks
W6 (Run Orchestrator).
* chore: add core-js, electron-winstaller, protobufjs, sharp to pnpm.onlyBuiltDependencies
Two PSI-targeted wins (split from #2599 follow-up).
1. New `resource-hints.astro` mounted in every page's <head> declares
`<link rel="preconnect" href="https://api.github.com" crossorigin>`.
The inline enhancer script on /` issues 3 fetch() calls to
api.github.com right after DOMContentLoaded (stars, latest release,
contributors). Without preconnect each pays a full DNS + TCP + TLS
handshake (~150-300ms) inline with the fetch. With preconnect those
handshakes happen in parallel with HTML parse and all three share one
warmed HTTP/2 connection.
2. Wrap the scroll listener's read + classList write in
requestAnimationFrame. Trackpads and high-rate wheels fire scroll
faster than display refresh, and every callback that hits classList
triggers layout recalc. PSI was attributing ~700ms of "forced reflow"
to the un-throttled version. The rAF gate collapses each burst to one
DOM mutation per frame; `{ passive: true }` is preserved so the
listener still doesn't block the scroll thread.
Same throttling pattern mirrored to `header-enhancer.astro` (used by
every sub-page) and `home-enhancer.astro` (kept in lockstep even
though /` currently uses its own inline copy).
Expected PSI delta:
- "Preconnect to required origins" hint: cleared
- "Forced reflow" diagnostic 700ms → near zero
- LCP: small bonus from earlier GH fetch warm-up (~100-300ms)
The Settings → Automations/Routines section (RoutinesSection.tsx) used
hardcoded English for every label, action, status pill, schedule
description, time suffix, and dialog. With the app language set to
Chinese the whole section stayed in English, breaking the localized
experience.
Introduce a dedicated `routines.*` i18n namespace, wire RoutinesSection
and the SettingsDialog routines header through `useT()`, and add
Simplified + Traditional Chinese translations. Schedule summaries are
now built from interpolated keys instead of concatenated English
literals, and the 12-hour AM/PM suffix is localized too. The remaining
locales fall back to English via the existing `...en` spread, matching
the repo's incremental per-locale translation pattern.
A regression guard in locales.test.ts asserts the representative
`routines.*` keys are actually translated (not English fallback) in
zh-CN and zh-TW.
* feat(deploy): add one-click Docker/Podman Compose installer for Linux and macOS
- Add install.sh with interactive wizard, Podman/Docker runtime detection,
port conflict check, health verification, and systemd user unit creation
- Add update.sh for image pull and restart with health check
- Add uninstall.sh with interactive user data backup before removal
- Unify CLI output styling with step/ok/warn/error/info helpers
- Add install-guide.md documentation
- Add install.test.ts integration test suite
* feat(deploy): add one-click Docker/Podman Compose installer
- interactive setup wizard with port, image, CORS, memory prompts
- automatic Docker/Podman detection with install guidance
- systemd user unit for Linux, health check polling
- update.sh (pull + restart + prune) and uninstall.sh (backup + cleanup)
- node:test integration suite and install-guide.md
* style(deploy): improve POSIX sh compatibility and systemd unit handling
- unify shell shebangs to #!/usr/bin/env bash
- add pipefail option for better error handling
- fix systemd unit for Podman: remove After/Requires when no service
- correct documentation to match actual uninstall behavior
* fix(deploy): address review feedback for installer scripts
- remove curl | sh path, document clone-first only
- isolate tests via docker-compose.override.yml with unique names
- support both --image <ref> and --image=<ref> in update.sh
- add running container detection before install
* docs(install): remove demo scripts and add MCP note
* perf(landing): self-host fonts + inline critical CSS
PageSpeed Insights flagged ~2.3s of render-blocking on /:
globals.css 12.9 KB external link, 160ms
fonts CSS 2.2 KB fonts.googleapis.com, 750ms
+ 4 woff2 ~1200ms each from fonts.gstatic.com
Two changes drop that whole chain:
1. Self-host fonts via @fontsource-variable/{inter,inter-tight,
playfair-display,jetbrains-mono}. Each family ships a single variable
woff2 (covers all weights we use) that Astro bundles into /_astro/*
alongside the rest of the build, served same-origin through CF Pages —
no separate TLS handshake, no Google Fonts CSS round-trip. The CSS
variable names get an extra alias in front (`'Inter Tight Variable',
'Inter Tight', ...`) so a system fallback still works if the package
ever ships under a different family name.
2. `astro.config.ts: build.inlineStylesheets: 'always'` inlines every
emitted <style> into the HTML <head> instead of emitting a separate
/_astro/*.css link. The HTML grows from ~13KB to ~28KB (gzip) but
loses one stylesheet round-trip + the entire @font-face chain that
used to gate text rendering.
Component cleanup: the `<FontStylesheet>` component (preconnect + link to
fonts.googleapis.com) is no longer needed and is deleted, removed from
all 7 places that mounted it. og.astro keeps its own font setup since
it renders to a screenshot.
Expected effect (from PageSpeed Insights "Render-blocking requests"
diagnostic on the previous build):
FCP 1.9s → ~1.2s
LCP 2.2s → ~1.5s
Verified: pnpm typecheck 0 errors, pnpm build 1853 pages 78s, preview
serves /_astro/*.woff2 as font/woff2 same-origin, 0 fonts.googleapis or
fonts.gstatic references in the built HTML.
* perf(landing): include Playfair italic + bump nix pnpm-deps hash
Two follow-ups on the self-host fonts PR:
1. globals.css imported only `@fontsource-variable/playfair-display`,
which ships @font-face for font-style: normal only. The previous
Google Fonts URL included the italic axis (`ital,wght@0,500;1,400;
...`) and several rules (.roman, .work-rule .roman, .sec-rule .roman,
plus 8 other italics across globals.css + sub-pages.css) render
Playfair italics via `font-family: var(--serif); font-style: italic`.
Without the italic face self-hosted, those would fall through to
Times New Roman italic or browser synthesis. Adding
`wght-italic.css` keeps the typography visually equivalent.
2. nix/pnpm-deps.nix uses a fixed-output derivation hash that has to
match the pnpm vendored store; adding the four fontsource packages
changed pnpm-lock.yaml so the hash has to be bumped to the value Nix
reported in CI.
Codex (Looper reviewer) flagged #1 as non-blocking.
* perf(landing): pin fontsource versions exactly per repo guard
`pnpm add` defaulted to caret ranges (`^5.2.8`) but repo guard rejects
non-exact specs ("dependency specs must be exact versions like 1.2.3 or
workspace:*"). That was the actual cause of the Preflight + Validate
workspace failures — pinning to the locked versions Codex reviewer
called out:
@fontsource-variable/inter 5.2.8
@fontsource-variable/inter-tight 5.2.7
@fontsource-variable/jetbrains-mono 5.2.8
@fontsource-variable/playfair-display 5.2.8
`pnpm guard` now passes locally (6/6 tests).
Replaces the HA page `<title>` from the previous descriptive form
("HTML Anything — your local AI agent writes the HTML, you ship
it | Open Design") with a noun-stack that captures three high-value
brand-defense queries:
- "open source html anything" — researching alternatives
- "html anything official" — verifying the canonical source
- "open source html anything official" — precise long tail
Why
- The descriptive title before this change told the story but didn't
match what users actually type when they land on this page from
search. The brand has been growing fast (4.1K+ stars on the repo);
visitors increasingly arrive with intent to verify ("is this the
real one?"), to compare alternatives, or to find the canonical
install path. The new title format leads with the tokens those
queries hit.
- Joey called the change explicitly. The brand has reached a stage
where SEO defense beats descriptive copy on a one-line surface.
What
- New `OPEN_SOURCE_OFFICIAL_TITLE` constant maps every landing-page
locale to its translated title. Pattern: `{open-source} HTML Anything
{official}` with no separator. Word order follows what's natural in
each locale (German: Offiziell at end; French/Romance: adjectives
after the noun; CJK: adjectives before).
- The three explicit copy blocks (`HTML_ANYTHING_COPY_EN`,
`_ZH`, `_ZH_TW`) all reference the constant rather than holding
their own literal title strings, so future title changes update
in one place.
- `derivedHtmlAnythingCopy()` (which builds copy for the 15
non-explicit locales by spreading EN over locale UI strings) now
pulls the title from the same constant via `[locale]` lookup with
EN as a defensive fallback.
- Adds three new `SoftwareApplication.alternateName` entries for the
same query stack. Schema-level signal complements the title-level
signal: the structured-data entity now matches whichever shape the
user typed regardless of UI locale, since these are proper-noun
query variants that don't translate.
Surface area
- Marketing site only. Single file: `app/pages/html-anything/index.astro`.
- No `apps/web`, no `apps/daemon`, no contracts, no CLI.
- No new dependencies.
Validation
- `pnpm --filter @open-design/landing-page typecheck` — 0 errors.
- Local dev: every locale's HA page renders the new title.
- en → `<title>Open Source HTML Anything Official</title>`
- zh → `<title>开源 HTML Anything 官方</title>`
- ja → `<title>オープンソース HTML Anything 公式</title>`
- de → `<title>Open Source HTML Anything Offiziell</title>`
- ar → `<title>HTML Anything مفتوح المصدر الرسمي</title>`
- …same shape across all 18 landing locales.
Co-authored-by: Joey-nexu <joeylee12629@gmail.com>
* fix(landing-page): translate Library/Tutorials, drop dropdown numbers, dedupe Product menu
Three small UX issues surfaced after #2605 landed Tom's full i18n
bundle. None of them block deploy; all of them affect every visitor
who lands on the site, especially the non-English ones.
Issues
1. The new Library / Tutorials nav labels in #2605's header
restructure were hardcoded English. On `/zh/`, `/ja/`,
`/de/`, etc. the rest of the nav was localized but those two
labels stayed "Library" / "Tutorials" inside otherwise-translated
chrome.
2. The four catalog facets inside the Library dropdown carried the
live count badges from the previous top-row treatment ("Skills132",
"设计系统150", etc.). Inside a dropdown they read as a glued-on
suffix rather than a tag, and Joey called them out as "weird —
drop them entirely".
3. The Product dropdown still listed Tutorials as a child item
alongside Open Design and HTML Anything. After #2605 made
Tutorials a standalone top-level link, the duplicate was
confusing — same URL appearing in two places in the same nav row.
4. Chinese (zh and zh-tw) had `skills: 'Skill'` — the singular
English word, untranslated in an otherwise fully Chinese nav.
Fixes
- Extends `HeaderCopy.nav` interface with `library` and `tutorials`
keys; populates both across all 18 landing locales (en, zh, zh-tw,
ja, ko, de, fr, ru, es, pt-br, it, vi, pl, id, nl, ar, tr, uk).
- Updates `header.tsx` to read `headerCopy.nav.library` and
`headerCopy.nav.tutorials` instead of literal strings.
- Removes the four `<span class="dropdown-num">{count}</span>` badges
from inside the Library dropdown items. The numbers still sit on
the catalog index pages themselves (header counts, hero) and on
the homepage; the dropdown is for navigation, not status.
- Removes the Tutorials `<li>` from the Product dropdown — Tutorials
lives at the top-row level only. Updates the Product trigger's
`is-active` set so it no longer highlights when on `/tutorials/`.
- zh/zh-tw `nav.skills` switches to `技能` (the actual translation
every other Chinese localized site uses for "skills" in this
context).
Bonus: catalog-row horizontal padding unified at 24px to match the
fix already applied across `featured-card` / `template-card` /
`system-card` / `source-card` (see PR #2600). The skill list grid
on `/skills/` was the last family still on `padding: 22px 0`.
Surface area
- Marketing site only. `apps/landing-page/app/_components/header.tsx`,
`app/i18n.ts`, `app/sub-pages.css`, `app/globals.css`.
- No `apps/web`, no `apps/daemon`, no contracts, no CLI surfaces.
- No new dependencies.
Validation
- `pnpm --filter @open-design/landing-page typecheck` — 0 errors
- Local dev: `/zh/` shows `资源库 / 教程 / 博客`; `/zh/` Library
dropdown shows `技能 / 设计系统 / 模板 / 工艺` without count badges.
- Local dev: `/` Product dropdown shows only Open Design + HTML
Anything (no Tutorials child); top-row Tutorials link is the
single canonical entry to that page.
* fix(landing-page): keep catalog-row horizontal padding at narrow widths
#2610's catalog-row 24 px fix only addressed the desktop rule.
The `@media (max-width: 720px)` override block still set
`.catalog-row a { padding: 18px 0 }` — zero horizontal — which is
what Joey's screenshot was actually showing on narrow viewports.
Switches the narrow padding to `18px 20px`. Matches the same
narrow inset `.source-card` already uses (line 1058 of this file)
so the catalog-row family stays in lockstep with the rest of the
catalog tile families at narrow widths.
Also drops the now-stale hover override (`padding-left: 8px;
padding-right: 8px`) — that rule existed only to compensate for
the zero-horizontal default and now produces a visible jolt
against the new 20 px base.
* fix(landing-page): featured-card description honors the 24px gutter
Same shorthand fragility pattern that #2600 fixed for template-card,
just one card family further down the file.
`.featured-card p` (the description line on each featured strip
tile, e.g. "Nº 0.001 deck-swiss-international" with the 16-列网格
blurb under it) had `margin: 0 0 12px` — three-arg shorthand,
which expands to TOP RIGHT BOTTOM = LEFT, so margin-left and
margin-right both got reset to 0.
The group rule `.featured-card a > * + *` had set the horizontal
margin to 24px so every text child sits inside a clean 24px gutter
from the card border. The shorthand on `<p>` overrode the group
rule and the description text bled right up against the card's
right edge.
Switches to `margin-bottom: 12px` only. Horizontal margin stays
owned by the group rule, matching the fix template-summary already
got in #2600.
Surfaced by Joey on `/skills/` featured strip — the exact symptom
he's been calling out for the last few rounds. Same shape, different
file.
---------
Co-authored-by: Joey-nexu <joeylee12629@gmail.com>
* chore(landing-page): bring PR #2469 content wholesale onto post-revert main
Step 1 of replicating @pftom's #2469 work without the deploy-blocking
issues that forced #2603. This commit copies the full \`apps/landing-page/\`
diff from #2469's HEAD (`9d2a4f1`) onto current main verbatim — every
i18n bundle, every page rewrite, every \`[locale]/\` wrapper. Subsequent
commits on this branch then surgically restore the SEO fixes that
#2469 silently regressed and configure the sitemap to survive the
Cloudflare Pages 25 MiB limit, so deploy is healthy when this lands.
What's in this commit
- Tom's i18n bundle: \`i18n.ts\` (5377 lines), \`home-page-i18n.ts\`,
\`info-page-i18n.ts\`, \`landing-ui-i18n.ts\`, \`content-i18n.ts\`
(~10K lines total of locale data)
- 18 landing-page locales: en, zh, zh-tw, ja, ko, de, fr, ru, es,
pt-br, it, vi, pl, id, nl, ar, tr, uk
- All existing pages rewritten to consume the new i18n bundle
- Full \`[locale]/<route>/\` wrapper tree for every catalog page
- \`plugin-registry.ts\` rewrite, \`catalog.ts\` adjustments
- \`astro.config.ts\` route + sitemap reconfiguration
- \`public/_headers\`, \`public/_redirects\`, \`public/favicon.svg\` adds
- \`_components/locale-switcher-script.astro\` add
What's intentionally NOT done in this commit (handled in follow-ups
on this same branch):
- Restore brand mark 44px + rounded corners (was lost from #2588)
- Restore HA SoftwareApplication \`alternateName\` array (was lost from #2566)
- Restore HA \`url\` canonical pointing at the landing page (was lost from #2586)
- Restore Product/Library/Tutorials/Blog nav grouping (was lost from #2588)
- Restore catalog-card padding 24px (was lost from #2600)
- Configure sitemap to filter \`[locale]/\` routes so the generated XML
stays under 25 MiB and Cloudflare Pages accepts the deploy
- Add \`/zh-CN/* → /zh/*\` redirects for backwards-compatibility with
any externally-linked OD-canonical locale URLs
Validation so far
- \`pnpm --filter @open-design/landing-page typecheck\` — 0 errors
* fix(landing-page): unblock deploy + restore SEO regressions on top of #2469
Step 2 of replicating @pftom's #2469. The previous commit on this
branch brings #2469's content wholesale; this commit applies the
surgical fixes that make the result actually deploy and preserves
the SEO improvements that #2469 silently regressed.
Fix 1 — sitemap stays under Cloudflare Pages 25 MiB upload limit
- `astro.config.ts` `filter` now drops every `/{locale}/...` route
so the sitemap only emits canonical English URLs.
- Locale variants are still discoverable via the
`<xhtml:link rel="alternate" hreflang="...">` annotations the
`namespaces.xhtml: true` option emits inside each canonical entry.
This is Google's recommended pattern for a multi-language site.
- Verified: post-fix `out/sitemap-0.xml` = 179 KB (was 38.4 MiB
on the prior attempt that forced #2603's revert).
Fix 2 — header brand block restored to the polished version
- Logo `width/height` 36 → 44 (matches PR #2588's brand-mark refresh
for visual weight against the new black speech-bubble glyph)
- `.brand-meta` block ("Studio Nº 01 · Berlin / Open / Earth") removed
from the header bar; the same editorial flourish still lives on the
rotated `.side-rail .rail-text` pseudo-elements at page edges.
Fix 3 — header nav grouped into Library + standalone Tutorials/Blog
- Skills / Systems / Templates / Craft are now children of a Library
dropdown (matches PR #2588's grouping). Each row keeps its count
badge inline; the trigger highlights when any of the four facet
pages is active.
- Tutorials and Blog stay as standalone top-row items (PR #2588's
original decision after Joey's review on the Learn dropdown).
- Contact removed from the header — it was a same-page anchor that
the footer already surfaces.
- Hardcoded "Library" / "Tutorials" labels match the brand-name
pattern: unlocalized across all 18 landing-page locales.
Fix 4 — HA SoftwareApplication entity canonicalized on the LP again
- `alternateName` is back to an explicit array of real query
variants `["html anything", "html-anything", "htmlanything",
"HTML Anything Editor", "The agentic HTML editor"]`. #2469
re-routed it through `copy.schemaAlternateName` which dropped
the literal alias declarations Google needs for spaced-vs-
hyphenated-vs-joined matching. (Restores PR #2566.)
- `url` flips back from `HA_URL` (the GitHub repo) to the LP URL
itself, matching the `BreadcrumbList` block on the same page.
GitHub repo lives in `sameAs` as a peer surface. (Restores PR
#2586. Without this, Google credits the GitHub repo as canonical
for the entity, which is the opposite of what this surface
exists for.)
Fix 5 — catalog-card horizontal padding unified at 24 px
- featured-card 22 → 24, template-card 20 → 24,
system-card 18 → 24, source-card 28 → 24.
- For template-card, also moved horizontal padding into the group
rule exclusively so future siblings join without re-asserting
margin shorthands. (Restores PR #2600.)
Fix 6 — `_redirects` for the locale-code rename
- This bundle uses `zh` / `zh-tw` / `pt-br` / `es` (the codes Tom's
i18n.ts ships). The previous OD landing-page used `zh-CN` /
`zh-TW` / `pt-BR` / `es-ES`. Externally-indexed and inbound-linked
URLs against the old prefixes now 301 to the new canonical.
Validation
- `pnpm --filter @open-design/landing-page typecheck` — 0 errors
- `pnpm --filter @open-design/landing-page build` — completed
successfully; 18,204 pages built; sitemap-0.xml is 179 KB
(well under the 25 MiB Cloudflare Pages limit).
* docs: promote 'open-source alternative to Claude Design' to README H1
Brings the missing README and .gitignore changes from #2469 that the
first wholesale-checkout in this branch missed (the auto-pulled diff
scope was filtered to apps/landing-page/ initially).
What
- Every README.*.md (13 locale variants) now leads with the
"open-source alternative to Claude Design" tagline as a subtitle to
the project name in the H1 / first paragraph. This was @pftom's
brand-positioning commit (`ee851dc`) on the original #2469 branch.
- `.gitignore` adds `growth/**` to keep growth-research scratch out of
the repo.
Why
- The README is one of the highest-PageRank surfaces a GitHub project
exposes to Google. Promoting the "alternative to Claude Design"
framing into the H1/subtitle position makes the project surface for
exactly the query the SEO work in this PR is trying to capture.
- Without this commit, the replicated #2469 in this branch would still
rank against the previous H1 ("Open Design") on GitHub crawls,
letting the SEO win at the LP fall short on the GitHub surface.
This is a strict subset of #2469's content — pure docs, no code,
no behavior change beyond what GitHub renders on the repo overview.
---------
Co-authored-by: Joey-nexu <joeylee12629@gmail.com>
* Enhance landing page with SEO-focused content and FAQ section
- Updated `.gitignore` to include growth directory.
- Modified `astro.config.ts` to prioritize high-intent landing pages for SEO.
- Added new FAQ styles and layout in `globals.css` for better user experience.
- Implemented FAQ section in `page.tsx`, ensuring it aligns with structured data requirements.
- Created dedicated pages for agents and alternatives to Claude Design, enhancing SEO and user navigation.
- Introduced comparison page for evaluating Open Design against competitors.
- Added favicon links component for consistent branding across all pages.
* Add SVG favicon and update favicon links for improved branding
* Enhance landing page with official source pillars for improved branding and navigation
- Added five canonical "official source" pillars to the homepage, reinforcing key links: official site, GitHub repository, releases, documentation, and Discord community.
- Updated URLs for releases, issues, documentation, and license to streamline access and improve user experience.
* Add locale support and enhance landing page with language switcher
- Introduced locale management with a new i18n module, defining multiple languages for the landing page.
- Implemented a locale switcher in the topbar and header, allowing users to select their preferred language.
- Updated global styles for the locale selector and adjusted layout for better responsiveness.
- Enhanced SEO by ensuring localized content is served based on user selection.
- Added a script for automatic locale detection and persistence in local storage.
* Implement localized routing and enhance navigation with href utility
- Added `stripLocaleFromPath` and `localizedHref` functions to manage locale-based URL paths.
- Updated `localePath` to normalize paths based on the detected locale.
- Refactored links in the header, footer, and main page components to utilize the new `localizedHref` function for improved navigation.
- Introduced locale-aware routing for new pages, ensuring consistent user experience across different languages.
- Enhanced SEO with alternate links for localized content in the sub-page layout.
* Update header component to use new logo format
- Replaced favicon image with a new logo in WebP format for improved performance and quality.
- Ensured consistent branding across the landing page with the updated logo.
* Enhance landing page with localization and new UI components
- Introduced a comprehensive localization schema for content management, allowing for multilingual support across various sections.
- Updated the blog and collection schemas to include internationalization (i18n) fields for better content localization.
- Implemented a new official source strip for improved navigation and branding, linking to key resources like the official site and documentation.
- Enhanced the locale switcher functionality, allowing users to select their preferred language with improved UX.
- Updated styles for the locale switcher and added new components for better responsiveness and accessibility.
- Refactored existing components to utilize localized URLs, ensuring a consistent user experience across different languages.
* Implement comprehensive localization and enhance landing page UI
- Introduced new localization features, including `EXTRA_LOCALIZED_HOME_BODY_COPY` and `EXTRA_LOCALIZED_LANDING_UI_COPY`, to support multilingual content across the landing page.
- Updated `astro.config.ts` to integrate internationalization (i18n) settings for the sitemap, improving SEO for localized content.
- Created new files for home page and info page internationalization, defining structured content for various languages.
- Enhanced the locale switcher functionality with improved UX, allowing users to easily select their preferred language.
- Refactored existing components to utilize localized content and URLs, ensuring a consistent experience across different languages.
- Made CSS adjustments for better responsiveness and accessibility in the UI components.
* Enhance landing page with new design templates and localization improvements
- Added support for design templates in the content management system, allowing for better organization and access to design resources.
- Implemented comprehensive localization for blog topics, enhancing multilingual support across various sections of the landing page.
- Updated the header component to include new product menu items, improving navigation and user experience.
- Refactored CSS for improved responsiveness and accessibility, including a new sticky chrome bar for better navigation.
- Enhanced the locale switcher functionality, ensuring a seamless experience for users selecting their preferred language.
* docs(readme): promote 'open-source alternative to Claude Design' tagline to subtitle across locales
Pulls the high-value, low-conflict slices of @pftom's #2469 onto
current main. That PR's branch was based on `af63af3` (May 20) and
diverged from a fast-moving main, so its Tier 1 SEO content is shipped
here as a fresh PR rather than rebased.
What's included
- **5 new SEO landing pages**, English-only for v1:
- `/official/` — brand-authority hub naming every alias the project
is searched as (Open Design, OpenDesign, open-design, opendesign,
Open Design AI, OD); Organization JSON-LD with the same alternateName.
- `/quickstart/` — three-command install path with HowTo JSON-LD,
requirements, expected output, troubleshooting, next steps.
- `/agents/` — 17 BYOK adapters across three tiers, ItemList JSON-LD,
BYOK explainer.
- `/compare/` — evaluation-stage comparison hub vs Claude Design,
Figma Make, v0, Lovable/Bolt, Open CoDesign; mandatory honest-limits
block as FAQPage JSON-LD.
- `/alternatives/claude-design/` — primary commercial-intent page;
TL;DR, why-people-search, BYOK explainer, feature-comparison table,
who-should-pick-which, migration steps, FAQPage JSON-LD.
- **`[locale]/` wrappers** for each of the 5 new pages, generated via
`PREFIXED_LOCALES` from OD's existing `_lib/i18n.ts`. Each wrapper is
a thin re-export of the canonical English page; per-locale translations
are a follow-up.
- **`seo-head.astro` schema upgrades** — `inLanguage` on Article /
WebSite / Blog JSON-LD, `availableLanguage` on the WebSite block,
`og:locale:alternate` for non-current locales, hreflang `x-default`
link.
- **`favicon-links.astro` extracted as a shared component** so every
page emits the same icon set; new sizes (favicon-16x16, favicon-32x32,
android-chrome-192x192, android-chrome-512x512) generated from the
current brand mark; `site.webmanifest` published for PWA / Android
install affordances.
- **`llms.txt`** restructured to lead with brand-alias declaration and
list every new official entry point. The Sister Projects section
introduced in #2566 stays.
What's intentionally NOT included
- @pftom's parallel i18n bundle (`i18n.ts` / `home-page-i18n.ts` /
`info-page-i18n.ts` / `landing-ui-i18n.ts` / `content-i18n.ts`,
~10K lines of locale data). It duplicates and conflicts with OD's
existing `_lib/i18n.ts` + `_lib/home-copy.ts`. Adapting the new
pages to OD's existing system instead keeps the codebase on a
single source of truth; per-locale translations of the new pages
can land as a separate PR.
- Mass-rewrites of existing pages (homepage, HA, Skills/Systems
catalogs, blog detail). #2469's branch base predates a number of
in-flight PRs that touched those files heavily; cherry-picking the
rewrites would re-litigate already-merged work.
- The `[locale]/<existing-route>/` routing tree for Skills, Systems,
Templates, etc. Without translations, mirroring 26 routes per locale
produces duplicate-content signals; that lift belongs with a
per-locale copy push.
Surface area
- Marketing site only. No `apps/web`, no `apps/daemon`, no contracts,
no CLI surfaces. No new dependencies. No env vars.
- New favicon assets are static PNGs generated from the canonical
brand-mark source via `magick`; committed as files, no pipeline.
Validation
- `pnpm --filter @open-design/landing-page typecheck` — 0 errors
- Local dev server: every new English route returns 200; every
`[locale]/<route>/` variant (e.g. `/zh-CN/agents/`) returns 200;
`/favicon.ico` and `/site.webmanifest` resolve.
- Local dev server: `seo-head.astro` emits the new `inLanguage`,
`availableLanguage`, `og:locale:alternate`, and `hreflang=x-default`
signals; `<FaviconLinks />` renders the consolidated icon set
including the new manifest link.
Followups
- Translate the 5 new pages and replace the English-content wrappers
under `[locale]/`.
- Swap `sub-page-layout.astro` to use `<SeoHead />` so the new schema
upgrades reach all sub-pages, not just the homepage and any future
pages that opt into SeoHead directly.
- Mass-update the homepage and HA page along the lines of #2469's
rewrite, but rebased onto current main rather than off `af63af3`.
Co-authored-by: Joey-nexu <joeylee12629@gmail.com>
* feat(landing-page): group header nav into Product / Library / Learn
Reduces the top-level nav from 7 link slots (Product, Skills, Systems,
Templates, Craft, Tutorials, Blog, Contact — plus the locale switcher
and two CTAs) to 3 dropdown groups, eliminating the row-overflow
that forced the brand sub-meta to truncate at narrow desktop widths
and cramming the hamburger panel on tablet.
Why
- The four catalog pages — Skills, Systems, Templates, Craft — share
the same shape: a list page that links into per-record details.
They are facets of one library, not competing peer destinations.
Letting each occupy its own top-row slot was a policy of last
resort when there were only four; with Product newly added and
Tutorials newly added since #2452 + #2266, the row no longer fits.
- Tutorials and Blog are both editorial reading surfaces (videos
vs articles), so they pair naturally under one Learn group.
- Contact in the top nav was always misleading — it links to
`#contact`, a same-page anchor on the homepage CTA section. The
footer already lists it. Promoting it to top-row real estate while
the row was overflowing made no sense.
What
- New `Library` dropdown holds Skills (132) / Systems (150) /
Templates (111) / Craft (11). Each row inside the panel keeps its
count badge inline beside the label, plus a one-line blurb so the
panel reads as a directory rather than an undifferentiated list.
The trigger highlights `is-active` whenever any of the four facet
pages is the active route, so users still see "you are inside
Library" feedback.
- New `Learn` dropdown holds Tutorials and Blog. Same pattern: inline
blurbs, parent-highlight when either child page is active.
- `Contact` removed from the header. The footer surface still lists
it; the top-row capacity it occupied is reclaimed.
- `.dropdown-num` CSS class added for the inline count badge inside
dropdown rows, distinct from the existing `.num` (which floats
above top-row links as a superscript).
Active-state semantics are preserved — every sub-page that already
passes `active="skills"`, `active="systems"`, etc. keeps working
without changes; the new Library trigger reads those same values
to decide whether to highlight itself.
Surface area
- Marketing site only. No `apps/web`, no `apps/daemon`, no contracts,
no CLI surfaces. No new dependencies. No i18n keys (existing
`copy.navSkills` / `copy.navSystems` / `copy.navTemplates` /
`copy.navCraft` / `copy.navBlog` keep working; "Library" and
"Learn" are universal English labels that don't need translation
tables for this iteration).
Validation
- `pnpm --filter @open-design/landing-page typecheck` — 0 errors
- Local dev server at narrow breakpoints (≤1080px) — hamburger panel
flat-expands the new Library / Learn dropdowns; existing handler
closes the panel on link click.
- Local dev server at desktop widths (≥1180px) — top row now reads
Product · Library · Learn, matching the design intent.
* feat(landing-page): polish brand mark — bigger glyph, single-line wordmark
Three small refinements to the header brand block, requested after
the new logo + grouped-nav landed:
1. The "Open Design" wordmark was wrapping to two lines once the new
nav groups (Product · Library · Learn) sat alongside it on narrow
desktop widths. The hamburger fallback already hides the entire
nav at ≤1080px so the wordmark has plenty of horizontal room
there — wrapping was a layout glitch, not a width constraint.
Applies `white-space: nowrap` on `.brand` and adds an explicit
`.brand-name` span so the rule is intent-bearing rather than
relying on the parent inline-flex behaviour.
2. The brand mark felt undersized against the wordmark's optical
weight, especially with the new black mark where the negative
space inside the speech bubble visually shrinks the glyph. Bumps
the displayed size from 36×36 to 44×44 (+22%) in both the header
and the footer, so the proportion matches what users see in the
logo.webp source.
3. Removes the `<span class="brand-meta">` block ("STUDIO Nº 01 ·
BERLIN / OPEN / EARTH"). It was decorative metadata that earned a
third of the brand block's horizontal real estate without
carrying any user-facing function. Drops the JSX, drops the
styles, drops the now-stale `display:none` overrides at the 1180
and 880 breakpoints. The same rotated brand-string still lives on
the side-rail pseudo-elements at the page edges (`.side-rail
.rail-text`) — that surface is the canonical home for that
editorial flourish, not the nav.
Surface area
- Marketing site only. `_components/header.tsx`, `page.tsx` (footer
brand block), and `globals.css`. No `apps/web`, no `apps/daemon`,
no contracts, no CLI. No new dependencies, no i18n changes.
Validation
- `pnpm --filter @open-design/landing-page typecheck` — 0 errors
- Local dev server: header HTML no longer contains `brand-meta`;
the brand block on `/` and any sub-page renders the bigger 44px
glyph + single-line "Open Design" wordmark.
* refactor(landing-page): unbundle Learn dropdown and round the brand mark
Two small follow-ups after looking at the rendered nav:
1. Reverts the Learn dropdown back to two standalone top-row links
(Tutorials, Blog). Rolling exactly two items into a dropdown adds
a click without reclaiming any horizontal space — the Library
grouping is what genuinely cleared the row, and Tutorials + Blog
can live side by side. Active-state semantics are preserved
automatically since both pages already pass `active="tutorials"`
and `active="blog"`.
2. Adds `border-radius: 10px` to the brand-mark image. The source
PNG is a solid-fill square; clipping the corners gives the mark
the modern app-icon silhouette (~22% of side length, matching
the iOS / macOS convention) so it reads as a brand glyph rather
than a raw screenshot next to the wordmark. Applies to both the
header and footer brand blocks via the shared `.brand-mark img`
selector.
Validation
- `pnpm --filter @open-design/landing-page typecheck` — 0 errors
- Local dev server: top row reads `Product ▾ · Library ▾ · Tutorials · Blog`
- Local dev server: brand mark renders with rounded corners on `/`
and any sub-page (header + footer reuse the same `.brand-mark`
styling)
---------
Co-authored-by: Joey-nexu <joeylee12629@gmail.com>
Follow-up to #2566. That PR introduced `url: HA_URL` plus a
two-element `sameAs` on the SoftwareApplication block, which declared
the GitHub repo as the entity's canonical surface — Google uses `url`
to decide which page to attribute structured-data entities to. The
BreadcrumbList block right below uses the landing page URL, so the
two blocks contradicted each other on what the canonical page was.
Since the goal of #2566 was making "html anything" surface the landing
page in search, canonicalizing on GitHub was an inversion of intent.
This commit switches `url` to the landing page URL and keeps
`HA_URL` only in `sameAs` so the GitHub repo is registered as a peer
reference surface, not the canonical one. The two structured-data
blocks now agree.
Surfaced by Looper (PerishCode) on the #2566 review thread; the fix
landed too late to ride the original PR, hence this small follow-up.
Co-authored-by: Joey-nexu <joeylee12629@gmail.com>
Three small additions that close the gap between users searching
"html anything" (with a space) and finding the existing
`/html-anything/` page:
1. JSON-LD `SoftwareApplication.alternateName` is now an array of
real query variants — `html anything`, `html-anything`,
`htmlanything`, `HTML Anything Editor`, plus the existing tagline.
Each entry maps to a query users actually type. Feeding them as
declared alternates is a stronger signal than relying on Google's
tokenizer to figure out hyphen ↔ space equivalence on its own.
Also adds a `sameAs` array linking the canonical landing page and
the GitHub repo so structured-data crawlers see the entity span
both surfaces.
2. `llms.txt` now lists the HA landing page in its canonical entry
points and adds a dedicated "Sister Projects" paragraph that
spells out the name variants and the export targets. LLM crawlers
(Claude, Perplexity, ChatGPT browsing) read this file to decide
what to surface; the previous version made HA invisible to them.
3. The home page footer's Library column now lists "HTML Anything"
as a peer to Skills / Systems / Templates / Craft, which puts a
real body-anchor link to `/html-anything/` on `/`. Nav-dropdown
anchors (the Product menu) carry less SEO weight than an anchor
in a discoverable content area, so this makes the home actually
pass link equity to the sub-page. The label is hardcoded English
intentionally — "HTML Anything" stays unlocalized as a brand name
on every locale, so threading a new key through all 18 home-copy
translations would just churn translators.
## Why now
The HA landing page has been live since #2452 but a "html anything"
query (the variant most users type) doesn't surface us, while the
hyphenated form does. Tokenizer equivalence isn't enough on a young
domain — we need declared signals, an inbound link from a
content-area on the highest-PR page, and machine-readable hints for
LLM crawlers. This PR ships all three.
## Test plan
- [x] `pnpm --filter @open-design/landing-page typecheck` — 0 errors
- [ ] After deploy: `curl https://open-design.ai/html-anything/`
shows `"alternateName":[...]` array in JSON-LD
- [ ] After deploy: `curl https://open-design.ai/llms.txt` includes
a `## Sister Projects` section
- [ ] After deploy: home page footer Library column lists HTML Anything
## Surface area
- [x] `apps/landing-page` only
- [ ] no `apps/web`, no `apps/daemon`, no contracts, no CLI
- [ ] no new dependencies
- [ ] no skill / design-system / template / craft additions
## Followups (not in this PR)
- Mirror the HA hero banner to R2 + Image Resizing (Core Web Vitals)
- Generate a HA-specific Open Graph card so social previews stay on
brand (currently reuses the GitHub-hosted banner)
- Add a `SoftwareSourceCode` JSON-LD block to associate the landing
page with the GitHub repository entity
- Land a Chinese landing page at `/zh-CN/html-anything/` to capture
CN-locale queries
Co-authored-by: Joey-nexu <joeylee12629@gmail.com>
Browsers omit the Origin header on same-origin GET requests per the Fetch
spec, but the no-Origin host-only branch in isLocalSameOrigin only honors
IP-literal entries in OD_ALLOWED_ORIGINS (via ipOnlyExtraOrigins). When a
remote-access proxy (Tailscale Serve, Caddy, etc.) terminates TLS at a
non-loopback hostname listed in OD_ALLOWED_ORIGINS, those legitimate
same-origin GETs to /api/app-config, /api/mcp/servers, etc. get rejected
with "cross-origin request rejected" 403s while the matching POST/PUTs
(which always carry Origin) succeed.
Trust Sec-Fetch-Site: same-origin as a fallback when the Origin header is
absent. The Sec-* prefix is browser-set and cannot be modified by
JavaScript, so a value of "same-origin" attests that the request really
originated from the same origin as the target — a cross-site
<img>/<script> exploit would carry "cross-site" instead, leaving the
existing CSRF defenses intact.
Tests:
- pnpm --filter @open-design/daemon test -- tests/origin-validation.test.ts → 43 passed (5 new)
