Increase gap from 2px to 4px in .subtab-pill to prevent the active
tab's box-shadow from visually overlapping adjacent tabs. The previous
2px gap was insufficient for the shadow to render cleanly between tabs,
causing the Style tab to obscure neighboring Content/Attributes/HTML tabs.
Closes#2904
The "Lint changed blog SEO" and "Guard blog URL changes" steps in
landing-page-ci.yml fell back to the literal "HEAD^" when no base SHA was
available. On merge_group (and first-push) events there is no
pull_request.base.sha / before, so BASE became "HEAD^", and the
blog-indexing scripts' assertSafeGitRef (regex /^[A-Za-z0-9_./:-]+$/,
which forbids "^") threw `Unsafe git ref for base: HEAD^`, failing the
merge queue for every PR.
Resolve the fallback to a concrete commit with `git rev-parse HEAD^`
(checkout uses fetch-depth: 0, so it is always available), mirroring
blog-indexing-on-deploy.yml. The resulting 40-hex SHA passes the ref guard.
* ci: trigger PR exploration via maintainer "/explore" comment (no approval)
Add a low-friction way to run the sandbox exploration: a maintainer
comments "/explore" on a PR.
- on: issue_comment (kept workflow_dispatch). The job `if` allows the
comment path only when it is on a PR and the commenter has write access
(author_association OWNER/MEMBER/COLLABORATOR), so randoms cannot trigger
it; untrusted PR code still runs only inside the Docker sandbox.
- Drop the agent-pr-explore environment approval gate: both triggers are
already write-gated and there is no auto-trigger, so the extra manual
approval is redundant. R2 creds are repo-level secrets (no env-scoped
secrets), so they stay available without the environment.
- Feedback: 👀 reaction on the command + a placeholder comment carrying
the report marker (so the run yields one evolving comment), 🚀 on
success, and 👎 + a failure note (with the run link) on failure.
Does not auto-run on every PR, so unrelated PRs stay clean.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* ci: don't clobber a produced report with the /explore failure note
Review: the failure-feedback step ran after the always() report step, so
on the failure-with-report case (sandbox wrote a report then exited
non-zero) it overwrote the just-posted report with the generic "failed
before producing a report" note — losing the useful output.
Guard it: if the report file exists, leave the posted report in place and
skip the failure note/reaction. Only post the short failure note when no
report was produced.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* Add tracking for Automations, Plugin Detail, and Plugin Loop
- RoutinesSection: track new_automation, create, save, cancel, run_now,
edit, pause, resume, delete, history button clicks
- PluginDetailView: track back navigation and use_plugin action
- PluginLoopHome: track clear_active, submit, card_details, card_use
- Extend AutomationsClickProps with new CRUD elements
- Add PluginDetailClickProps and PluginLoopClickProps contracts
* fix: address review comments on plugin/automation tracking
- Extract onBack handler in PluginDetailView to cover both error-path
and success-path back buttons with tracking
- Move create/save tracking from submit button onClick into the form
submit handler to capture keyboard submissions and avoid false
positives from validation failures
* fix: move submit tracking into submit() handler in PluginLoopHome
Same fix as RoutinesSection: tracking now fires inside submit() so
keyboard Enter submissions are captured and the !trimmed guard
prevents false positives.
---------
Co-authored-by: qiongyu1999 <2694684348@qq.com>
* fix(amr): close ACP stdin on abort so vela tears down OpenCode
When an AMR (vela) run is cancelled, attachAcpSession.abort() sent a
`session/cancel` RPC but left the child's stdin open. The vela ACP bridge
keeps running until it sees EOF (or is signaled), and it only shuts down
its private OpenCode `serve` process on a clean exit — so on abort the
OpenCode server lingered until the caller's SIGTERM fallback, and leaked
entirely if the parent was killed before cleanup ran.
End stdin after sending the cancel (mirroring the clean-completion path)
so the agent receives EOF and shuts down its own runtime promptly,
independent of signal timing.
* fix(amr): end stdin on abort even before session/new resolves
Addresses review on #3097: abort() still returned early when sessionId
was unset, so the stdin EOF only happened after session/new completed.
Cancelling during ACP startup (before the session exists) left the
OpenCode-teardown window open until the caller's SIGTERM fallback — and a
parent hard-kill before that could still strand the private OpenCode
process.
Move stdin.end() out of the sessionId guard so abort always closes stdin
when the pipe is writable; gate only the session/cancel RPC on sessionId.
Add a regression test that aborts during startup and asserts stdin is
ended with no session/cancel emitted.
* fix(docker): fix container startup crash due to missing OD_API_TOKEN
* fix(docker): forward OD_API_TOKEN to fix docker container boot loop
* fix(docker): enforce non-empty OD_API_TOKEN for docker-compose
* fix(deploy): automate OD_API_TOKEN generation in installer and close compose loop
* docs(readme): guide manual deployment users to configure OD_API_TOKEN
* docs(readme): align working directory paths for manual deployment instructions
* docs(readme): align working directory paths for manual deployment instructions
* docs(readme): restore git clone context for first-time users
* fix(web): directly import remixicon.css in layout to fix icon font loading
* fix(web): resolve remixicon font not rendering via public/ directory
RemixIcon icons were blank in Chrome and showed garbled PUA codepoints
in Firefox. The font file was referenced with a relative URL in CSS
loaded through @import, which Turbopack does not resolve.
Fix: copy remixicon.woff2 to public/ and change the @font-face src
to an absolute URL (url("/remixicon.woff2")), so the browser fetches
it directly from the server root without relying on the bundler.
Per product feedback: the Homepage button (rendered when a plugin's
manifest carries a `homepage` field that differs from its
`sourceUrl`) sent visitors off to upstream-author or original-source
URLs before they had a chance to explore the plugin via Open Design
itself. On a marketing detail page that's a leak, not a feature.
Removes the conditional block that rendered
`<a class="btn btn-ghost">Homepage ↗</a>` between the GitHub link
and the Share button. The header-action row is now exactly three
controls everywhere: Use this plugin → · Find on GitHub → · Share ↗.
The `plugin.homepage` data field stays available on the
`BundledPluginRecord` shape since the in-app catalog row, JSON-LD,
and any future author-bio surface can still consume it. The
`pcopy.detailHomepage` i18n key (with full 18-locale coverage) stays
for the same reason — `PluginsCopy` is `Partial<>` everywhere and
removing it would mean a 18-locale block edit for zero functional
gain.
apps/landing-page typecheck stays at 0 errors.
Co-authored-by: Joey-nexu <joeylee12629@gmail.com>
* ci: add idempotent provision script for the agent-pr-explore runner
The self-hosted runner's setup was hand-assembled and easy to lose on a
rebuild — most dangerously the codex-acp pin: expect-cli bundles
codex-acp 0.10, which is incompatible with ChatGPT-account auth (every
model rejected); we run 0.15, but any expect-cli reinstall silently
reverts it and breaks the agent.
Add a self-contained, idempotent provision script that brings the
runner's config layer back to a working state and is safe to re-run:
codex model pin (gpt-5.4), the codex-acp 0.15 pin (npm pack + extract +
chmod), deploy-key generation, base-repo git mirror seed/refresh,
pnpm-store/reports dirs, the weekly image-refresh helper + cron, and the
readiness self-check helper. The header documents the manual/secret
steps it intentionally does not automate (base toolchain + colima, the
interactive `codex login`, registering the deploy key on the repo, and
registering the Actions runner service).
Verified idempotent against the live runner (all checks pass, no config
disturbed).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* ci: provision — update codex model key in place, don't truncate config.toml
Review: step 2 overwrote the whole ~/.codex/config.toml with just the
model line whenever the exact pin wasn't already present, dropping any
other Codex settings on a re-run — destructive, contradicting the
idempotent goal. Now: replace an existing `model =` line in place (sed),
append only when the key is absent, and leave the rest of config.toml
untouched. Verified preservation locally.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* ci: provision — create ~/.ssh before ssh-keygen on fresh host
Review: on the fresh-rebuild path this script targets, ~/.ssh usually
does not exist, so `ssh-keygen -f ~/.ssh/od_agent_deploy` fails with
"No such file or directory" and the deploy key (and downstream mirror
bootstrap) never gets created. mkdir -p the key's parent dir (chmod 700)
before keygen, and only print the pubkey when it actually exists.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Packaged diagnostics bundles never contained the daemon or web
`latest.log` — the very logs that hold the agent/critique run flow — so
support exports could not explain "sent prompt to the agent, then
nothing happened" reports.
Root cause: the sidecar `base` means different things per launch path.
tools-dev passes the pre-namespace source root, so
`resolveNamespaceRoot(base, namespace)` is correct. But the packaged
orchestrator launches every child with `base = <namespaceRoot>/runtime`
(apps/packaged/src/{paths,sidecars}.ts) while logs live a level up at
`<namespaceRoot>/logs`. The diagnostics builders re-appended the
namespace and resolved every log to
`<namespaceRoot>/runtime/<namespace>/logs/...` → ENOENT. renderer.log
only survived by accident: the desktop main process wrote it to the
same wrong path the reader looked in.
Add `resolveRuntimeNamespaceRoot(runtime, contract, runtimeMode)` to
`@open-design/sidecar` which walks up out of the `runtime/` dir in
packaged (runtime-mode) launches and falls back to the dev layout
otherwise. Route the desktop renderer-log path and both diagnostics
exporters (desktop IPC + daemon HTTP) through it so writer and reader
stay in lockstep and renderer.log lands next to the desktop log dir.
Tests: sidecar unit specs for both layouts; a daemon export spec that
writes a real `<namespaceRoot>/logs/daemon/latest.log` and asserts the
bundle captures its contents (red on main → ENOENT placeholder, green
here).
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
SkillsSection kept its own skills list in sync after create/delete, but
App-level skills (used by the chat composer) were only loaded at boot.
Propagate a refresh callback so new skills appear in chat immediately.
Fixes#3017
* fix(connectors): expire stale auth credentials
Mark connector credentials as expired when provider reads report auth-shaped failures so Memory stops presenting stale connected apps as healthy.
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix(connectors): avoid expiring grants on platform 401
Only delete connector credentials for provider tool errors attributable to the current connector so Composio platform auth failures do not wipe valid grants.
Co-authored-by: Cursor <cursoragent@cursor.com>
---------
Co-authored-by: Cursor <cursoragent@cursor.com>
* ci: retry PR-context gh calls so a transient API blip doesn't abort the run
The early PR-context gathering calls `gh pr diff`, `gh pr view`, and
`gh api .../files`. gh hits api.github.com under the hood, and a single
transient timeout/5xx there aborts the whole run before any exploration
(seen on #3083: "could not find pull request diff: Get \"https://api.
github.com/...\": net/http timeout"). These were the only network calls
in the run without a retry (source fetch + npm already retry).
Add a small gh_retry helper (4 attempts, linear backoff) and wrap the
three read-only context calls. gh writes nothing to stdout on a failed
API call, so retrying is safe even for the calls piped into the context
file; the retry warning goes to stderr.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* ci: address review — buffer gh retries to files (no paginated duplication)
Review (Siri-Ray): wrapping `gh api --paginate` retries inline while the
context block is redirected to the file means a mid-pagination failure
leaves partial pages in the context, and the retry appends them again —
duplicating the patches section and burning the context budget the agent
reads.
Replace the in-pipe gh_retry with gh_retry_file: each call buffers to its
own file per attempt (`>` truncates on open, so a failed/partial attempt
is discarded before the next), and the context block just cats the
finished files. Fetch PR body + patches to files up front, then assemble.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Reviewers asked for at-a-glance outcomes. Instruct the agent to begin each
"Cases Tested" bullet with a status emoji (✅ pass / ❌ fail / ⚠️ warning /
⚪ inconclusive) and a bold case name, so the report shows which checks
passed or failed without reading each line.
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* ci: clean agent report (write-to-file) + slim artifacts/uploads
Four related cleanups to the agent PR exploration output:
1. Clean report. The PR comment / report.md was assembled by dumping the
entire verbose expect.log (ACP init logs, "Git failed" warnings, the
~24KB echoed prompt, ANSI codes, progress checklist) under the trace
header -- ~28KB of noise. Instead, instruct the agent to write its
final Markdown report to a file via its file-write tool, and have the
runner read that file directly. Verified: Codex writes a clean report
to the given absolute path. Falls back to an inconclusive note if the
agent did not finish.
2. Drop duplicate trace/video. The script copied
playwright-smoke-trace.zip -> playwright-trace.zip (a ~28MB legacy
duplicate) and the webm likewise, and uploaded both to R2. Keep only
the canonical smoke-named artifacts.
3. Slim the GitHub artifact. The trace zips and videos are already on R2;
exclude *.zip / *.webm from the uploaded artifact so it drops from
~56MB to <1MB (report + logs only).
4. Persist report on the runner. Copy the report / agent-report /
expect.log / trace URL to a stable host dir
($HOME/.cache/agent-pr-explore/reports/pr-<n>) so dry runs
(skip_comment) can be inspected without downloading the artifact.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* ci: address review — keep advisory reports + recursive artifact excludes
Review findings on the report/artifact cleanup:
1. Regression fix: the non-app-surface and deterministic-verifier branches
write their pre-baked advisory report (Inconclusive / Pass / Fail) and
never run the agent, so they don't produce agent-report.md. After
switching write_agent_report_artifact to read only agent-report.md they
fell through to the "agent did not write a final report" fallback,
dropping the real advisory (and mis-reporting on .github-only PRs like
this one). Fix: those branches now write their advisory directly to
$agent_report_file — single source of truth for the report body.
2. Recursive artifact excludes: the source Playwright recording lives at
artifacts/playwright-video/<uuid>.webm; non-recursive !*.webm / !*.zip
didn't match the subdirectory. Use **/*.zip and **/*.webm so the slim
actually holds.
3. Drop the now-dangling summary.legacyTrace field (the legacy trace copy
is no longer produced), matching the legacyVideo removal.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* feat(landing): add Community link to top nav
Adds a 'Community' entry to the landing-page nav between Blog and the
Star CTA, linking to /community/ which Cloudflare Pages 302-redirects to
the contributors honor-cards page (currently a Vercel deploy).
Translations added for all 18 locales. The nav slot was previously empty
after Blog because Contact had been intentionally pulled from the bar
(left as a footer + #contact anchor).
* fix(landing): use literal /community/ href so non-default locales don't 404
PerishCode review caught that href('/community/') routes through
localizedHref and produces /zh/community/, /ja/community/, etc. The
_redirects rule only matches the literal /community/ and the
[locale]/[...path].astro catch-all does not generate community pages,
so 17 of the 18 translated locales would have hit a Cloudflare 404.
The destination is a single non-locale-aware external page, so skip
the locale prefix entirely — same shape as the GitHub Star and
Download CTAs.
* feat(landing): host community + ambassadors page first-party
Lands the contributors / ambassadors page as a static asset at
apps/landing-page/public/community/index.html, served at /community/
on open-design.ai. Drops the temporary 302 to the Vercel preview URL
(d5458c46-…vercel.app) — that hostname was a deploy-time UUID Vercel
could recycle, which the reviewer correctly flagged as a follow-up.
The page now opens with an Ambassadors section: vocation, patronage,
covenant — three columns of the program in Renaissance-atelier voice,
with a single Apply on Discord CTA pointing at the ambassador channel
(discord.gg/2p7Ajbxw3h). Maintainers / leaderboards / good-first-issues
sit below as before. Header.tsx comment updated to point at the new
source of truth instead of the deleted redirect rule.
* fix(community): drop time-bound claims, tighten bot heuristic, drop dead CORE_TEAM entry
PerishCode review on ff1cd44b flagged three correctness issues with
the static community page. Addressing each:
* The 'This week's signal' / 'This week's leader' / 'Last 7 days' /
'PRs · 7d' framing made promises a frozen RANKING_SNAPSHOT can't
keep — three weeks from build, the page would be calling the
2026-05-26 leaderboard 'this week's leader.' Renamed to time-neutral
copy ('Recent signal', 'A recent leader', 'Snapshot', 'Recent PRs')
and dropped the snapshot's 'since' field so we don't pin a window
we can't honour. Real refresh pipeline is a follow-up.
* 'Showing first N · resets every 30 minutes' didn't describe the code
(no caching of any kind exists; each page load re-hits /search/issues
and /users/:login). Replaced with a truthful 'Showing first N open
good-first-issues.'
* The bot exclusion heuristic used substring match on bot/cursor/agent,
which would silently drop real logins like 'agentina', 'cursorsmith',
'robothai'. Tightened to a whole-token regex (/(?:^|[-_])(bot|cursor|
agent)(?:$|[-_])/) and dropped 'leon wang' from CORE_TEAM — it had
an embedded space, which GitHub logins never do, so the entry was
unreachable dead code.
---------
Co-authored-by: koki yanlai xu <koki@kokideMacBook-Air.local>
* Add template social sharing menu
* Update plugin share e2e expectations
* Add additional template social share targets
* Remove Bilibili template share target
* Open social share destinations in new tabs
* Address template share review feedback
* Use canonical public plugin share URLs
* Gate public plugin share links by marketplace provenance
* Update plugin share e2e for local-only badges
* Limit public share URLs to official marketplace
* docs(creative-memory): add integration-shape proposal
Captures the integration-boundary decisions the product/pipeline team needs to make before any creative-memory implementation lands in the live generation loop. Covers the four areas lefarcen requested in the issue thread: signal capture points, retrieval insertion into generation/critique, user controls and escape hatches, and how Section 11's raw-events + content-addressed-derivations contract reshapes the boundary.
Doc-only. No code lands with this PR. Each section enumerates the option space, names a working lean, and flags the explicit decision that needs a product/pipeline call. Written to apply to either implementation (PR #1746 or a team-internal one), per the resolved foundation-vs-reference question.
Refs: #1637
* docs(creative-memory): self-contain the integration-shape doc
Per @mrcfps's review on PR #3021: the doc was referencing files that live on the parked add/creative-memory-system branch (PR #1746), not on main, which left the primary references dead from main's perspective.
Fix: inline the essential engine context (external API, storage shape, raw-events / content-addressed-derivations contract) as a Background section, drop the dead links to packages/creative-memory-system/, and replace inline open-questions.md cross-references with terse parenthetical statements that stand on their own.
The doc is now self-contained at the level reviewers need: the §5 decision summary stays the entry point, and no reference depends on an external file. PR #1746 is still linked as the RFC/prototype thread for anyone who wants the full simulation suite or open-questions ledger.
* docs(creative-memory): align §2 with the live composer, declare relationship to ## Personal memory
Per Looper review on PR #3021: §2's 'Generation prompt today' stack did not match apps/daemon/src/prompts/system.ts:454-483, which was a real correctness issue because §2 is the integration contract for prompt placement. As written, an implementer could ship a second memory layer or move the existing personal-memory block without deciding how the two memories interact.
Patch:
- Replace the simplified five-line stack with the live thirteen-block order, gated as the composer gates each block.
- Add Decision A: replace, append, or sibling block relative to the existing ## Personal memory slot. Lean A3 (sibling block, shared precedence).
- Reframe Decision B (insertion point) against the live order. Lean B1 (after ## Personal memory, before custom instructions).
- Add Decision C: precedence wording for the new block, mirroring the existing 'brand wins / skill wins' clause.
- Update §5 decision summary: items renumbered, three new explicit calls (relationship, insertion point against live order, precedence wording).
- Touch the Purpose paragraph to mention the existing ## Personal memory slot, since it is now the headline §2 question.
* docs(creative-memory): name the §2 prefix scope and route raw-events through OD_DATA_DIR
Per Looper review on PR #3021. Two doc-accuracy fixes:
1. §2 'Live composer order' stopped at ## Active skill (block 13), but composeSystemPrompt keeps appending mode-specific blocks after that (plugin / per-stage atom / metadata / deck framework / media contract / Codex imagegen / critique / visual-direction override / connected-MCP / Claude AskUserQuestion). As written, an implementer could treat block 13 as the tail of the stack and miss precedence layers a creative-memory block has to coexist with. Fix: rename the subsection to 'Memory-relevant prefix of the live composer order', add a 'Tail blocks (mode-specific, after ## Active skill)' note that summarizes the tail without listing every gate, and call out API_MODE_OVERRIDE as block 0. Source line numbers cited as a range with the disclaimer that gating + order are the durable shape.
2. §4 raw-events ownership option B wrote '.od/memory/raw_events.jsonl' literally, which hard-codes a repo-rooted path and bypasses the OD_DATA_DIR precedence documented in AGENTS.md FAQ. Packaged installs and NixOS / Home Manager modules already point OD_DATA_DIR at a writable directory because the install root may be read-only. Fix: rewrite option B as '<OD_DATA_DIR>/memory/<userId>/raw_events.jsonl' or a 'raw_events' table inside '<OD_DATA_DIR>/app.sqlite'; update option C wording for consistency; add a paragraph below the table explaining the OD_MEDIA_CONFIG_DIR > OD_DATA_DIR > <projectRoot>/.od precedence so this section cannot be read as endorsing a hard-coded .od/ path.
* docs(creative-memory): correct the raw-events precedence — OD_DATA_DIR only
Per Looper review on PR #3021. The previous patch's precedence paragraph copy-pasted the full chain (OD_MEDIA_CONFIG_DIR > OD_DATA_DIR > <projectRoot>/.od) from AGENTS.md, but that chain is the resolution rule for media-config.json specifically — OD_MEDIA_CONFIG_DIR is a narrower override that relocates only API credentials, not the daemon data root. Applying it to raw events would route preference event data into the credentials directory, which is the wrong contract.
Fix: rewrite the paragraph so raw events follow only the daemon data-root contract (OD_DATA_DIR if set, else <projectRoot>/.od). Add an explicit follow-up paragraph that OD_MEDIA_CONFIG_DIR is NOT part of this precedence, with a one-line explanation of why mentioning it as a contrast — so anyone reading the AGENTS.md chain alongside this doc knows the difference.
* feat(landing-page): localize plugins library across 18 locales
PR #2926 shipped the new `/plugins/` library hub + four kind sub-routes
+ detail pages, but the chrome was English-only — visitors landing on
`/zh/plugins/` saw the old marketplace registry placeholder rendered
by the catch-all instead, and detail pages rendered identical English
copy regardless of locale prefix. This PR brings the plugin surface
to feature parity with `/zh/skills/`, `/zh/templates/`, `/zh/systems/`,
`/zh/craft/`.
## What changes
- New `app/_lib/plugins-i18n.ts` — single source for all plugin chrome
copy (hub, list pages, chip rails, share dialog, detail-page meta
labels). English baseline + 17 locale overrides keyed on
`LandingLocaleCode` (the same short-code shape `localeFromPath()`
returns). Missing keys per locale fall back to English so a
partially-translated locale still renders sensibly. Translations
cover hub copy, four tile titles + blurbs, seven artifact-kind
labels + descriptions, 23 scene-subcategory labels, 18 detail-page
chrome strings, and a six-key share-dialog table with a
per-locale `shareTemplate({title, url})` function (translated for
every locale where `_lib/i18n.ts` already had one — same voice).
- `app/pages/plugins/{,templates/,templates/[kind]/,skills/,systems/,
craft/,[slug]/}/index.astro` — every hardcoded English string now
reads `getPluginsCopy(locale)` keys. Page logic and routing
unchanged.
- New short-code wrappers under `app/pages/[locale]/plugins/` — six
files (hub + three sub-routes + `[kind]/` and `[slug]/`) following
the same pattern `[locale]/skills/index.astro` already uses: each
re-exports the canonical page component and adds a per-locale
`getStaticPaths()` so the build emits 17 locale prefixes per
plugin route. Total plugin-route prerender count goes from ~390 to
~7 000, matching the existing skill/template scaling.
- Catch-all (`[locale]/[...path].astro`) — old `getPublicPlugins` /
`getRegistryCounts` registry rendering removed (placeholder UI
that was never wired to a real marketplace data source). Plugin
routes now live exclusively under `[locale]/plugins/...` short-code
wrappers, so the catch-all stops claiming `'plugins'` as a route
root. The dead-code path also drops a `pluginCounts.all` reference
the title row was reading.
- `.plugins-tile-grid` styles promoted from a scoped `<style>` in the
default-locale hub to global `app/sub-pages.css` so the
short-code wrapper renders the same hub markup without re-mounting
per-page CSS — `display: contents`-style scoping pitfalls in
Astro's per-component CSS scoping made this the cleanest fix.
## Surface area
- [ ] **UI** — new page / dialog / panel / menu item / setting / empty state in `apps/web` or `apps/desktop`
- [ ] **Keyboard shortcut** — new or changed
- [ ] **CLI / env var** — new `od` subcommand or flag, new `tools-dev` flag, or new `OD_*` env var
- [ ] **API / contract** — new `/api/*` endpoint, new SSE event, or changed shape in `packages/contracts`
- [ ] **Extension point** — new entry under `skills/`, `design-systems/`, `design-templates/`, or `craft/`, or change to the skills protocol
- [ ] **i18n keys** — new translation keys (full plugin chrome added across all 18 locales)
- [ ] **New top-level dependency** — adding any new entry to the **root** `package.json`
- [ ] **Default behavior change** — changes what existing users experience without opting in
- [x] **None** — landing-page-only restoration of i18n parity for the plugin surface
## Validation
- `pnpm --filter @open-design/landing-page typecheck` → 0 errors
- `pnpm --filter @open-design/landing-page build:static` → 16 127 pages
built (+6 584 over current main: ~388 plugin detail pages × 17
locale prefixes plus the hub + four sub-routes × 17 locales).
- `copy-example-html.ts` reports `266 entry files + 65 referenced
files`, identical to before — no regression in the asset-mirroring
pipeline.
- Local Playwright smoke (`/zh/plugins/...`):
- `/zh/plugins/` renders `<title>插件库 · Open Design</title>`,
label `插件库`, h1 `407 个可组合的构件。`, four tiles labelled
`模板 / 技能 / 设计系统 / 工艺`.
- `/zh/plugins/templates/video/` renders h1 `48 视频`, scene chips
`全部 / 动效 / 短视频 / 营销 / 产品 / 数据讲解`.
- `/zh/plugins/example-article-magazine/` share dialog renders
`复制下面的文案、然后跳到你想分享的平台粘贴即可` etc., share
template auto-interpolates plugin title + URL into Chinese voice.
- All 18 locale prefixes (`/zh`, `/zh-tw`, `/ja`, `/ko`, `/de`,
`/fr`, `/ru`, `/es`, `/pt-br`, `/it`, `/vi`, `/pl`, `/id`, `/nl`,
`/ar`, `/tr`, `/uk`) → 200 across hub + four sub-routes + sample
detail page.
- English `/plugins/` unchanged (default-locale path bypasses the
`[locale]/...` wrapper).
* feat(landing-page): finish plugins i18n chrome across 18 locales
The first localization pass shipped a partial fix: hub headings, lead
copy, two-level page chrome, detail-page metadata labels, the share
dialog, and the chip rail were still falling back to English on every
non-English locale because plugins-i18n.ts only filled a chrome slice
for `zh` and the file header even claimed "7 artifact-kind labels and
25 scene-subcategory labels are translated" for every locale that did
not yet have those blocks.
Three changes close the visible gap:
1. plugins-i18n.ts: fills the 27 still-missing chrome fields per locale
for zh-tw / ja / ko / de / fr / ru / es / pt-br / it / vi / pl / id /
nl / ar / tr / uk. Includes the 7-key category map, the 23-key
subcategory map, hubHeading / hubLead, the 4 *Label / *Heading /
*Lead triples for the templates / skills / systems / craft hub
pages, the 4 tile blurbs, the 4 browse buttons, sceneLabel, allChip,
the 12 detail-page metadata labels (mode / scenario / platform /
surface / author / manifest id / tags / preview caption / find on
GitHub / homepage / open in new tab) and bucket label map, the
detail share dialog (title / copy link / jump-to), and the
header-side nav.plugins entry. zh receives the same 11 detail-page
and share-dialog labels it was also missing.
2. header.tsx + site-footer.astro: routes the hardcoded "Plugins /
Templates / Skills / Systems / Craft" labels through `nav.*` from
HeaderCopy, so every locale gets its own dropdown trigger and
footer column. Adds `nav.plugins` to HeaderCopy and fills it in 18
locales with the local form ("插件" / "プラグイン" / "Plugins" /
"Plug-ins" / "Plaginy" / "الإضافات" / etc).
3. plugin-row.astro + content-i18n.ts: chip rail. The bundled-plugin
branch now runs raw `mode` / `scenario` slugs through the shared
localizeTaxonomyValue, and that helper now also consults the
plugins-i18n subcategory map before giving up. localizeTaxonomyValue
now returns undefined on a true miss instead of the unknownTag
placeholder, so chips drop quietly instead of showing "Category" /
"分類" / "Categoría" for taxonomy slugs we have not localized yet.
Callers that genuinely want the placeholder (`localizeContentTag`,
blog `category`, system noun) still keep the explicit fallback.
Out of scope and tracked separately: per-plugin title and description
in plugins/_official/* (author-supplied English metadata, ~401 plugins
without an i18n schema in the manifest yet — needs RFC + tooling
before the manifests can be expanded), and adding the long tail of
mode / scenario / category slugs (`code-migration`, `plugin-sharing`,
`tune-collab`, `live-artifacts`, `engineering`, ...) to TAXONOMY_TERMS
so chips render localized labels for every taxonomy value rather than
dropping silently.
* feat(landing-page): cover plugins chip rail long-tail taxonomy slugs
PR #3010's first round localized the high-frequency mode/scenario
chips (prototype, video, image, marketing, design, ...) but left the
~37 mode/scenario and 14 category slugs that show up in real `od.*`
metadata — code-migration, plugin-sharing, design-system, planning,
scenario, refine, discovery, handoff, token-map, tune-collab, orbit,
live-artifacts, engineering, healthcare, hr, sales, support,
default-router, downstream-export, figma-migration, media-generation,
plugin-authoring, validation, 3d-shaders, animation-motion,
audio-music, creative-direction, design-systems, diagrams, documents,
image-generation, marketing-creative, screenshots, slides,
video-generation, web-artifacts, ... — falling through to undefined
and dropping their chip silently on every non-English locale.
The data layer is the source of truth here, so this expansion lands
in `content-i18n.ts:TAXONOMY_TERMS` / `CATEGORY_LABELS` rather than
the plugins-i18n catalog: a single dictionary entry per slug fans out
to every chip-rail consumer (catalog rows, detail metadata, the
templates/[kind] facets) without each consumer touching its own copy.
Translations cover all 17 non-`en` locales. Brand and product nouns
(Figma, Open Design, BYOK, plugin) stay literal; technical taxonomy
slugs get short equivalents that read as chips rather than full
prose. The result on `/ja/plugins/skills/` matches `/plugins/skills/`
chip-for-chip (30 chips both sides) instead of dropping 27 of them
the way the previous iteration did.
* feat(landing-page): read manifest title_i18n / description_i18n on bundled plugins
PR #3010's prior rounds localized chrome and chip rails but the
catalog's most prominent text — each row's plugin name and blurb —
stayed English on every non-English locale. The plugin manifest
schema (`packages/contracts/src/plugins/manifest.ts`) has supported
`title_i18n` and `description_i18n` (Record<locale, string>) on every
manifest from spec v1; ~24 of the 401 first-party manifests already
carry one for `zh-CN`. The reader was just never wired to use them.
This change does the reader half: bundled-plugins.ts captures the
two i18n maps off each `open-design.json`, plugin-row.astro and the
detail page resolve them at render time via two new helpers
(`resolveBundledTitle`, `resolveBundledDescription`) that mirror the
short→long fallback chain documented in the manifest spec
(`htmlLang` like `zh-CN` → short `LandingLocaleCode` like `zh` →
primary tag → `en` → English baseline). The static-paths pass still
runs once for all locales — it has to, since each manifest produces
one URL — but the title/description shown on the rendered page now
reads the locale off `Astro.url.pathname` and picks the right entry
out of the maps.
Verified locally: `/zh/plugins/example-card-twitter/` now reads
"Twitter 分享卡 / 推特金句 / 数据卡, 适合配推文" from the manifest's
existing `zh-CN` block instead of the English baseline.
Plugin-data half follows in a separate commit. The 17 non-English
locales × 401 manifests need backfilling so the reader has something
to resolve to; that's data, not schema, and lands as a sequence of
manifest patches rather than tangled with this code change.
* feat(plugins): translate scenarios bucket title/description across 17 locales
Closes the first chunk of #3028. Eleven scenarios plugins (the
default-scenario bundle for each taskKind: code-migration,
figma-migration, media-generation, new-generation, tune-collab,
plugin-authoring; the default design router; the React / Vue /
Next.js downstream-export starters; and the Refine baseline) get
title_i18n + description_i18n filled for all 17 non-English locales
the landing page serves (zh-CN, zh-TW, ja, ko, de, fr, ru, es,
pt-BR, it, vi, pl, id, nl, ar, tr, uk).
The reader landed in 7ddfe36; this commit is data-only. taskKind
slugs that other docs reference by name (`code-migration`,
`figma-migration`, `tune-collab`, etc.) stay literal in the
descriptions so cross-references still resolve. Brand nouns —
Open Design, Next.js, React, Vue, Figma — also stay literal.
`/ja/plugins/od-code-migration/` now reads
"コードマイグレーション（デフォルトシナリオ）" instead of the English
baseline; `/zh/plugins/skills/` shows "代码迁移（默认场景）" in the
catalog row.
Remaining buckets (image-templates 45, video-templates 50,
examples 140, design-systems 142 = 377 plugins) follow in
subsequent commits in this PR.
* fix(landing-page): drop CJK template wrap when source name is still English
The Chinese / Japanese / Korean fallback templates for craft, skill,
template, system, plugin, and blog text splice the source `name` /
`title` into a CJK sentence frame: ``${name}工艺规则``,
``Open Design 指南：${topic}``, ``${name} は…のスキルです``. When the
underlying SKILL.md / craft markdown / blog frontmatter still ships
an English name (true for ~95% of the catalog today), that produces
mid-sentence script straddling on `/zh/...`, `/zh-tw/...`, `/ja/...`,
`/ko/...` like:
H1 : "Editorial typography hierarchy工艺规则"
Lead : "这条 Open Design 工艺规则定义 Editorial typography hierarchy
的执行标准…"
Plug : "video 插件 · 3D Animated Boy Building Lego"
That reads worse than the all-English fallback, because the visitor
parses the page in two scripts at once.
Adds a `nameNeedsEnglishFallback` guard that fires for the four CJK
locales whenever the spliced-in name has no CJK characters of its
own, and threads it through every `localizeXxxText` helper:
craft, template, system, plugin, skill, blog. When it fires the
helper returns the raw English content untouched, so the section
renders end-to-end in one language. Chrome (header, footer, breadcrumb,
buttons, share dialog) keeps its CJK rendering — only the
title-and-lead block falls back.
Side benefit: the same guard kicks in on the long tail of plugin
manifests still pending `title_i18n` / `description_i18n` backfill
(tracked in #3028), so `/zh/plugins/<bundled>/` no longer pairs a
"video 插件 · 3D Animated Boy Building Lego" title with a Chinese
breadcrumb. The page reads "3D Animated Boy Building Lego" + the
English manifest description, while header / footer / breadcrumbs
stay localized. Once a manifest ships its i18n maps, the chrome and
body re-converge automatically.
Non-CJK non-Latin scripts (ar, vi, ...) keep the previous behavior —
their templates already read tolerably with English names. If that
turns out to be wrong on a real audit, the same guard generalizes by
adding the matching Unicode range and locale set.
* feat(plugins): translate image-templates bucket title/description across 17 locales
44 of 45 image-templates plugins get title_i18n + description_i18n
filled for all 17 non-English locales (zh-CN, zh-TW, ja, ko, de, fr,
ru, es, pt-BR, it, vi, pl, id, nl, ar, tr, uk). Generated via Claude
Sonnet 4.5 over the OpenRouter gateway, ~$1.38 in API spend, 156s
wall-clock. Brand and cultural references stay literal (Open Design,
Lego, Hanfu, Showa, Pokémon, Black Myth: Wukong). Long AI generation
prompts collapse to a 1-2 sentence summary capturing what the plugin
does — the description doubles as catalog blurb on the landing site,
not as the actual generation prompt (which lives in example.html /
the manifest's preview entry).
Skipped: `profile-avatar-realistically-imperfect-ai-selfie` returned
malformed JSON on three retries; will rerun with a tighter prompt in
a follow-up commit. Catalog rows for that plugin keep falling back to
the raw English fields per #3010's reader change, so nothing breaks.
Tracking: closes the image-templates row in #3028.
* feat(plugins): translate video-templates bucket title/description across 17 locales
49 of 50 video-templates plugins get title_i18n + description_i18n
filled for the 17 non-English landing locales. Generated via Claude
Sonnet 4.5 over OpenRouter, ~$1.47 in API spend, 177s wall-clock.
HyperFrames templates, the Three Kingdoms cinematic series, the
Seedance/short-film prompts, and the K-pop / wuxia / anime variants
all get a 1-2 sentence catalog blurb in each locale; brand and
cultural tokens (Black Myth: Wukong, Hanfu, Showa, Pokémon, Three
Kingdoms / 三国志, Lego, Disney, K-pop, HyperFrames) stay literal.
Skipped: `live-action-anime-adaptation-water-vs-thunder-breathing-duel`
returned malformed JSON on three retries; will rerun in followup.
Falls back to the raw English fields per the reader landed in 7ddfe36.
Tracking: closes the video-templates row in #3028.
* feat(plugins): translate examples bucket (117/140) title/description across 17 locales
117 of 140 examples plugins get title_i18n + description_i18n filled
for the 17 non-English landing locales. Generated via Claude Sonnet
4.5 over OpenRouter, $3.94 in API spend, ~13 min wall-clock at
8-way concurrency. Existing zh-CN translations on 24 manifests are
preserved (the merge keeps author-supplied entries and only adds
missing locales).
23 of 140 returned malformed JSON on three retries — the output
likely hit the 4000 max_tokens ceiling on plugins whose description
balloons across 17 locales. Those manifests fall back to English on
non-`en` rendering per the reader landed in 7ddfe36, and will rerun
in a follow-up commit with a larger token budget and a stricter
output schema.
Tracking: closes 117/140 of the examples row in #3028; the remaining
23 stay open in that issue's failure list.
* feat(plugins): translate design-systems bucket (141/142) title/description across 17 locales
141 of 142 design-systems plugins get title_i18n + description_i18n
filled for the 17 non-English landing locales. Generated via Claude
Sonnet 4.5 over OpenRouter, $2.55 in API spend, 301s wall-clock at
8-way concurrency.
Translator script gained two improvements between examples and this
bucket:
- max_tokens bumped from 4000 to 8000 so 17-locale outputs stop
truncating on the long-tail manifests with verbose descriptions
- a balanced-brace JSON extractor that pulls the outermost `{ ... }`
from the response, tolerating trailing prose Claude occasionally
appends after the JSON object.
Result: only 1 manifest (`totality-festival`) failed parse this
batch, down from ~16% on the examples bucket. The next commit
re-runs the prior buckets' failures with the improved script.
Tracking: closes 141/142 of the design-systems row in #3028.
* fix(plugins): backfill 4 plugins that retried green after JSON extractor improvement
dcf-valuation, social-media-dashboard, wireframe-sketch (examples
bucket) and live-action-anime-adaptation-water-vs-thunder-breathing-duel
(video-templates bucket) parse cleanly under the balanced-brace
extractor introduced for the design-systems batch. The remaining
22 failures from the prior runs hit a different parse mode (Claude
emitting unescaped double quotes inside string values when the source
description contains its own English quotes like 'make it professional');
those will need a tighter prompt and rerun.
* fix(plugins): translate the last 22 plugins with quote-handling prompt fix
The 22 stuck plugins all carried English / Chinese double-quoted
phrases inside their description (\"make it professional\",
\"What's inside\", \"电子杂志 × 电子墨水\") that Claude was emitting
back inside JSON string values without escaping, breaking the parse.
Added one rule to the translator prompt — never use a straight double
quote inside a translated string, prefer single quotes / curly quotes
/ CJK 『 』 / 《 》 — and the previously stuck batch sailed through
clean: 22/22 ok, 0 retries, $0.85.
This closes the long tail of #3028:
- scenarios 11/11 ✓
- image-templates 45/45 ✓
- video-templates 50/50 ✓
- examples 140/140 ✓
- design-systems 142/142 ✓
- atoms N/A (filtered from public catalog)
All 388 catalog-visible plugins now ship title_i18n + description_i18n
for all 17 non-English locales the landing page serves.
* fix(plugins): clean up four review-flagged i18n data issues
- apps/landing-page/app/_lib/plugins-i18n.ts:759 — Polish bucket
label `examples: 'Przyklad'` was missing the diacritic; every
other Polish string in the same block uses proper diacritics.
Restore to 'Przykład'. (Reviewer: looper #4364985878.)
- video-templates/cinematic-route-navigation-guide — German
title_i18n.de was a byte-for-byte copy of en ("Cinematic Route
Navigation Guide") while the German description was already
translated. Replace with "Cinematischer Routen-Navigationsleitfaden"
to match the German voice the description sets.
- video-templates/hollywood-haute-couture-fantasy-video-prompt —
Dutch title_i18n.nl was identical to en for the same reason.
Translate the trailing noun phrase: "Hollywood Haute Couture
Fantasy Videoprompt" (mirrors the Dutch description's compound
word style).
- video-templates/video-seedance-three-kingdoms-guanyu-slaying-yanliang —
Korean Hangul `돌진` had leaked into the Turkish description (a
translation-pipeline artifact where the model copied the verb
from the Korean output without translating it). Replace
"saflarına돌진 eder" with the idiomatic Turkish "saflarına dalar".
All four are data-only fixes against existing manifests; no schema
changes, no reader changes. typecheck stays at 0 errors.
* fix(landing-page): localize aria-labels, alt text and BreadcrumbList JSON-LD on plugin detail page
The PR's prior rounds left six accessibility / structured-data
surfaces on `/{locale}/plugins/<slug>/` either entirely English or
mixing English chrome with the localized plugin title. Reviewer
flagged each one across multiple loops; this commit clears them all:
1. `aria-label` on the open-in-new-tab popout no longer reuses the
visible label `pcopy.detailOpenInNewTab` (which carries the
decorative `↗`). Added `detailOpenInNewTabAria` — same wording,
no glyph — and the `<a aria-label>` consumes that key. The
visible link text still ends in `↗`.
2. `<nav class="breadcrumb" aria-label="Breadcrumb">` now reads
`aria-label={pcopy.breadcrumbLabel}`. Eighteen locales filled
("面包屑导航", "パンくずリスト", "Brotkrumen-Navigation",
"Fil d'Ariane", "مسار التنقل", "İçerik haritası", ...).
3. Share-dialog `<button aria-label="Close">` now reads
`aria-label={pcopy.shareDialogClose}`. Eighteen locales filled
("关闭", "閉じる", "Cerrar", "Закрыть", "إغلاق", ...).
4. Three template-literal a11y strings (`${pluginTitle} preview`,
`Open interactive preview for ${pluginTitle}`, `${pluginTitle}
interactive preview`) become function calls
(`pcopy.previewImageAlt(t)`, `previewSummaryAria(t)`,
`previewIframeTitle(t)`) so the sentence frame around the
plugin title rotates with the page locale. Two `<img alt>` call
sites (the static preview at line 210 and the click-to-expand
thumbnail at line 179) both consume `previewImageAlt`.
5. `BreadcrumbList` JSON-LD position-2 now reads
`name: pcopy.hubLabel` instead of hardcoded English `"Plugins"`.
The visible breadcrumb at line 105 already renders
`pcopy.hubLabel`; this aligns the structured data with the
rendered chrome on every locale.
The new function-typed keys deliberately interpolate `pluginTitle`
(which is itself locale-resolved via `resolveBundledTitle`) so the
mixed-language guard from commit 002d457 is preserved: a manifest
without a per-locale title still flows through to a coherent
single-language a11y string because `pluginTitle` falls back to
English along with the rest of the section.
apps/landing-page typecheck stays at 0 errors.
Closes reviewer threads:
- #pullrequestreview-4364985878 (Open in new tab aria)
- #pullrequestreview-4368926224 (Polish typo + plus mixed-language alt/aria)
- #4373... (BreadcrumbList JSON-LD)
- #4374... (aria-label="Close" + aria-label="Breadcrumb")
* fix(landing-page): redirect legacy fa/hu/th /plugins/ paths to canonical
When the new `/{locale}/plugins/...` short-code wrappers landed, the
legacy catch-all `pages/[locale]/[...path].astro` dropped `'plugins'`
from its `paths` list. That intentionally avoids serving stale
marketplace-registry placeholder routes for the modern landing
locales — but it also takes `/fa/plugins/`, `/hu/plugins/`, and
`/th/plugins/` from 200 to 404, because those three legacy locales
live only in the old `_lib/i18n.ts:LOCALES` set and are not part of
`LANDING_LOCALES` (the modern 18-locale list the new wrappers serve).
Three `301`s in `_redirects` send those legacy URLs to the canonical
English `/plugins/...` so SEO and inbound links keep working until
the legacy locale set is retired entirely.
Reviewer thread (#pullrequestreview-4364052045) flagged this as a
non-blocking regression across multiple loops; this commit closes it.
* ci(landing-page): add merge_group trigger so the queue can clear PRs
`landing-page-ci.yml` only fired on `pull_request` and `push:main`,
which meant the required `Validate landing page` and
`Strict PR visual tests` checks never dispatched against the
`merge_group` ref the merge queue creates. The queue then sat at
"awaiting checks" until it timed out and ejected the PR (the
deadlock observed during the 5/26 release window).
Adding a `merge_group: { types: [checks_requested] }` trigger to
the same workflow lets the queued ref reuse the existing job graph,
matching the pattern in `ci.yml` which already wires `merge_group`.
Also drops `plugins/**` into the same paths filter as `pull_request`
since the new bundled-plugins reader (commit 7ddfe364) consumes
those manifests' `title_i18n` / `description_i18n` maps and the
landing-page CI must rerun when manifest data changes.
---------
Co-authored-by: Joey-nexu <joeylee12629@gmail.com>
Track "Save comment" and "Send to chat" button clicks in the comment
popover with a new `comment_popover` area, so we can measure the
distribution of save vs send-to-chat usage.
Co-authored-by: qiongyu1999 <2694684348@qq.com>
Real Codex runs (#3060) explored correctly — verifying 3-4 UI cases with
DOM evidence — but Codex over-planned (6 steps), executed the high-value
ones, then went silent chasing a remaining planned step and tripped
expect-cli's ~180s no-output watchdog, aborting the turn before it emitted
a final report. The run then fell back to an advisory artifact, so the
real findings never reached report.md.
Tighten the prompt so Codex finishes and submits before going idle:
- cap at 3 cases (was 6) and target 2-3, quality over breadth;
- add a CRITICAL instruction stating the runner aborts with no report
after ~3 min of no output, so Codex must stop after 2-3 cases and emit
the complete report in one final turn rather than leaving planned steps
pending or retrying silently.
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
