A Qodana workflow with one job per Qodana Cloud project:
- typescript: Qodana for JS in native mode, on the runner's Node 26 and
node_modules (qodana.yaml, project yt-ts-js, QODANA_TOKEN_TS_JS);
- rust: Qodana for Rust, Early Access, in Docker on src-tauri/
(src-tauri/qodana.yaml, project yt-rust, QODANA_TOKEN_RUST). Its
bootstrap installs what rust.yml installs; a stand-in dist/index.html
lets generate_context!() expand.
Both use qodana.recommended with a quality gate of zero problems and
analyse the whole project on every run. A tokens job skips the analysis
for Dependabot and fork PRs, which get no secrets; the token reaches only
the scan step, never `npm ci`. Annotations and artifacts, no PR comment.
Failures reach Discord through notify-ci-failure.yml.
Docs: TECH-SPEC (workflow table, Qodana notes), DEVELOPMENT (+ vi),
CONTRIBUTING, CLAUDE.md.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- ACCESSIBILITY.md (+ docs/i18n/vi): goal (WCAG 2.2 AA, not a conformance
claim), what works with a keyboard, screen readers, zoom, motion and touch,
how it is checked, supported environments, known limitations, how to
report a barrier (severity, response), and a checklist for contributors.
- In the Legal Center (consent: false, English + Vietnamese) and as
/legal/accessibility on the website; the Turnstile page's footer links it
in all eight languages.
- .github: Accessibility barrier issue form (label `accessibility`), an
Accessibility option in the feature request scope, an accessibility
checklist in the PR template.
- README, CONTRIBUTING, DEVELOPMENT (+vi), FEATURES and CLAUDE.md point to
it; SECURITY.md's supported versions move from 1.3.x to 1.5.x.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The coverage scope is spelled out in TECH-SPEC, vitest.config.ts and sonar-project.properties; the table row only says what the command does.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
SonarCloud's Automatic Analysis is off. The new `sonarcloud` job in
ci.yml runs SonarSource/sonarqube-scan-action v8.3.0 after `check`, with
the LCOV report that `check` now uploads; it never installs npm packages,
so SONAR_TOKEN never shares a job with them, and it fails with the
quality gate. Fork and Dependabot PRs skip it (no repository secrets).
sonar-project.properties analyses the tests as tests, imports the
coverage and excludes from it everything Vitest does not measure. Vitest
now also measures src/config, src/store and src/utils, next to the
engine, the Worker and the build plugins. The job passes package.json's
version as sonar.projectVersion, so the "previous version" new-code
period starts again at every release.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- CHANGELOG: v1.1.0 (copyright email and purpose words, Hide scrollbars,
installer copyright string, copyright-year check, the rust-toolchain pin)
- FEATURES: the four split contact lines, purpose words, the Settings row,
the profile's emails
- TECH-SPEC / ARCHITECTURE: editor store v20, hideScrollbars without a
settings bump, theme-init and App applying the scrollbar class, the new
CI step and scripts
- DEVELOPMENT (en + vi), CLAUDE.md: check:copyright / update:copyright
- I18N: 1,253 UI + 574 template keys per locale
- six manifests 1.0.0 -> 1.1.0
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Every claim checked against the code.
- TECH-SPEC: current configs (React 19 / Vite 8 / Tailwind 4, the web vs
Tauri build), Cloudflare hosting and the Worker, store versions and
migrations, the backup format, engine rules added in 1.0, CI as it is.
- ARCHITECTURE: layers, web vs Tauri, routing and start-up, the engine,
state, backup & restore, web delivery, the hardened Tauri shell.
- PACKAGING: tauri.conf and capabilities as shipped, the eleven
name-only commands and what they guard, where desktop data lives, the
first-start recovery of pre-1.0 files, the release workflows; the
auto-update flow that never existed is gone.
- FEATURES, CONTENT-INVENTORY (+ vi): recounted from config (42 genres,
16 store entries, 22 social/donate fields, 7 cross-post platforms,
37 Vietnamese banks); shortcuts that never existed removed.
- DEVELOPMENT (+ vi): main-only branching, every npm script, Rust 1.90,
Cloudflare local dev with test keys, the Rust checks.
- I18N: both namespaces and all seven registration places, including
the check page's own language list; ROADMAP: v1.0.0 shipped + backlog.
- DISCLAIMER (vi): eight languages. The locale template script now
prints the full registration checklist; a stale comment in main.tsx
about the removed settings file is corrected.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Two new locales, 1,562 strings each, complete: titles, every description
block, all 248 content warnings, tech notes, playthrough notes, pinned
comments, playlist copy, and the whole interface.
Brazil is a top-three market for gameplay viewing and Indonesia the largest in
Southeast Asia. Both use Latin script, so neither needed font or layout work.
**docs/I18N.md was wrong about what registering a locale takes**, and this
release is where that bit. It claimed `SUPPORTED_LANGUAGES` was "the only
registration step". It is one of four, and they fail differently:
SupportedLanguage union compile error
CORE_TAGS_BY_LANG compile error (there are TWO such records,
MULTILINGUAL_TAGS compile error not one — the doc mentioned one)
detectBrowserLanguage SILENT
The silent one matters most: the language works perfectly and simply never
auto-selects for a visitor whose browser is set to it. `SUPPORTED_LANGUAGES`
also wasn't type-bound to the union, and every consumer casts, so a mismatch
between the first two compiled cleanly — it now carries a `satisfies` clause.
`tests/i18n/locale-registration.test.ts` exists to close those gaps: it checks
the picker list against the locale directories on disk in BOTH directions, plus
namespace files, flags, native names and tag pools.
Browser-language detection now matches the full BCP-47 tag before the bare
language, or `pt-BR` would be reduced to `pt` and never found. Any other
Portuguese variant resolves to pt-BR rather than falling through to English.
115 UI keys were not hand-translated at all: their English text is
byte-identical to a string already translated in templates.json (graphics
presets, ray-tracing modes, difficulty names, content-warning labels), so they
are copied across. Translating those twice risks a checkbox and the rendered
description disagreeing about what the same option is called.
Deliberately left in English, matching what the Spanish locale already does:
vendor product names (NVIDIA DLSS, AMD FSR, Intel XeSS, Fluid Motion Frames),
platform brands, hardware acronyms, genre terms that are loanwords in both
languages, and example URLs/emails in placeholders. So pt-BR reads 1,344/1,562
and id 1,308/1,562 "differing from English" — the remainder are those
pass-throughs, not gaps.
Both are AI-translated and NOT native-reviewed. Recorded in DISCLAIMER.md
alongside JA/ES/KO/ZH.
Also replaces `__dirname` in the Vite and Vitest configs with
`import.meta.dirname` — Vite 8's native config loader doesn't provide the CJS
global and warned on every run.
v0.36.0 is skipped: the dependency release shipped first and took 0.37.0.
Tests 692 → 698. Validator now 1,562 keys x 8 locales.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
**docs/I18N.md was wrong in a way that would waste a contributor's afternoon.**
It said adding a locale to `SUPPORTED_LANGUAGES` was "the only registration
step". It is one of four, and the missing ones fail differently: the
`SupportedLanguage` union is a compile error, `MULTILINGUAL_TAGS` is a compile
error, but `detectBrowserLanguage`'s own hardcoded copy of the list fails
*silently* — the language works and simply never auto-selects for a visitor
whose browser is set to it. Also noted: `SUPPORTED_LANGUAGES` is not type-bound
to the union and every consumer casts, so a mismatch compiles cleanly.
Also documents the `--copy-english` / `--force` flags on the locale generator,
which only started existing in v0.35.0 despite having been advertised in
package.json since v0.1.
Elsewhere:
- **DEVELOPMENT.md** — the five new npm scripts, including why `format:check`
excludes markdown and locale JSON, and why `typecheck:all` exists.
- **TECH-SPEC.md** — the real CI job, a new Field Limits section, and how
over-limit enforcement derives from the engine's long-ignored `warnings`.
- **ARCHITECTURE.md** — the settings→engine mapping and its two guards, the
two-source validation state, and an honest note that the first Strict Mode
implementation typechecked, passed tests, and did nothing.
- **FEATURES.md** — the settings table still listed `Default Genre` and
`Auto-save Draft`, removed in v0.22.0 and v0.3. Replaced, plus a guardrails
section.
- **README.md** — guardrails in the feature list, and a link to the wiki.
The wiki itself (12 pages: quick start, glossary, field reference, guardrails,
batch/social, import-export, languages, privacy, troubleshooting, FAQ, release
process) is published separately — it lives in its own repository.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Resolves the 4 high-severity esbuild advisories (GHSA-gv7w-rqvm-qjhr,
GHSA-g7r4-m6w7-qqqr) deferred from v0.27.0. esbuild 0.17.0-0.28.0 is
vulnerable; the fix (0.28.1) is only reachable via vite 8 -- vite 7.3.3
pins esbuild ^0.27.0 and cannot reach it. Dev-server-only; the shipped
static bundle was never affected. npm audit: 4 high -> 0.
- vite ^7.3.3 -> ^8.0.16, @vitejs/plugin-react ^4.3.0 -> ^5.2.0,
tsx ^4.16.0 -> ^4.22.4 (vitest 4.1.8 already supports vite 8)
- vite 8 bundles with Rolldown: manualChunks object form -> function
form (vite.config.ts) is the only source change required
- Full gate chain green (typecheck, lint, knip, validate:locales,
500/500 tests, build); dev/preview + Tauri-context build verified
- Bump 5 manifests 0.27.0 -> 0.27.1; CHANGELOG + stack-version docs
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>