Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use when adding a new supported application UI language or locale family to all-api-hub, including i18next resources, language selection and normalization, browser extension manifest locales, date localization or natural-language date parsing, documentation-link fallback, locale completeness tests, and validation.
.claude/skills/qixing-jk-add-app-language/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-16 | ✗→✓ | ▲ Improved | 69% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -1% | 0% |
| case-15 | ✗→✓ | ▲ Improved | 34% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 6% | 0% |
| case-14 | ✗→✓ | ▲ Improved | 58% | 0% |
Treat a new language as a cross-runtime compatibility feature. Define one locale contract, then make application resources, browser manifests, auxiliary locale libraries, and tests agree with it.
git status --porcelain and preserve unrelated work and index state.src/constants/i18n.ts, src/utils/i18n/, src/components/LanguageSwitcher.tsx, i18next.config.ts, src/locales/, src/public/_locales/, tests/utils/i18nLocaleValidation.test.ts, and nearby language-specific tests.src/locales/, including public JSON. Use the owning project workflow for independently managed surfaces instead of changing their translation or fallback contract incidentally.tests/utils/i18nLanguage.test.ts.tests/components/LanguageSwitcher.test.tsx.tests/utils/i18nResources.test.ts.tests/utils/docsLocale.test.ts.tests/utils/i18nLocaleValidation.test.ts only for a stable invariant not already covered by its generic supported-language loops, such as required manifest aliases or locale-specific plural constraints.SUPPORTED_UI_LANGUAGES.src/locales/<app-tag>/ with every current namespace and normalized key family. Preserve interpolation tokens and markup, and implement every cardinal plural category reported by Intl.PluralRules(<app-tag>).src/utils/i18n/resources.ts glob-driven unless the resource-loading contract itself must change.pnpm; add custom parsing only for a reproduced gap the library cannot cover.src/public/_locales/<chrome_locale>/messages.json resources. Keep application tags hyphenated and Chrome locale identifiers underscored; add regional alias directories only when the browser locale contract requires them.pnpm run i18n:status and require the new locale to have the complete current key set.pnpm run i18n:extract:ci. Inspect the task-scoped locale diff and account for every unexpected deletion, plural rewrite, empty value, source-language copy, or key-shaped translation value.pnpm build when browser manifest resources or runtime locale imports changed, and verify the built extension contains the expected _locales directories.validate:staged during commit and validate:push during a requested push. Do not manually duplicate a matching hook on an unchanged index or file tree.SUPPORTED_UI_LANGUAGES already feeds its privacy allow-list; add no event solely for the presence of another locale.feat(i18n): add <language> locale for a complete new supported language unless the final diff is a correction to existing support.resolvedLanguage fallback as proof that a requested regional alias was normalized correctly.i18n:extract:ci as a translation-quality check rather than an extraction-integrity check.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-16 | fail→pass | 10,815 | 11,503 | +6% | 1 | 1 | 0% | 2,041 | 3,440 | +69% | 0 | 0 | — |
case-17 | pass→pass | 17,081 | 5,115 | -70% | 1 | 1 | 0% | 2,667 | 2,465 | -8% | 0 | 0 | — |
case-01 | fail→fail | 4,807 | 6,181 | +29% | 1 | 1 | 0% | 152 | 1,862 | +1125% | 0 | 0 | — |
case-02 | fail→fail | 4,955 | 3,962 | -20% | 1 | 1 | 0% | 288 | 1,752 | +508% | 0 | 0 | — |
case-03 | fail→fail | 17,967 | 4,405 | -75% | 1 | 1 | 0% | 3,841 | 1,832 | -52% | 0 | 0 | — |
case-04 | fail→fail | 16,120 | 11,799 | -27% | 1 | 1 | 0% | 2,791 | 3,780 | +35% | 0 | 0 | — |
case-05 | pass→pass | 12,492 | 9,993 | -20% | 1 | 1 | 0% | 2,048 | 2,992 | +46% | 0 | 0 | — |
case-06 | pass→pass | 10,741 | 9,986 | -7% | 1 | 1 | 0% | 1,916 | 3,531 | +84% | 0 | 0 | — |
case-07 | pass→pass | 11,899 | 5,586 | -53% | 1 | 1 | 0% | 1,671 | 2,387 | +43% | 0 | 0 | — |
case-08 | fail→pass | 14,764 | 3,398 | -77% | 1 | 1 | 0% | 2,147 | 2,134 | -1% | 0 | 0 | — |
case-09 | pass→pass | 4,728 | 4,401 | -7% | 1 | 1 | 0% | 872 | 2,188 | +151% | 0 | 0 | — |
case-15 | fail→pass | 12,552 | 8,643 | -31% | 1 | 1 | 0% | 2,293 | 3,084 | +34% | 0 | 0 | — |
case-10 | fail→pass | 17,316 | 23,080 | +33% | 1 | 1 | 0% | 3,088 | 3,267 | +6% | 0 | 0 | — |
case-11 | pass→pass | 12,903 | 6,419 | -50% | 1 | 1 | 0% | 2,226 | 2,683 | +21% | 0 | 0 | — |
case-12 | pass→pass | 10,215 | 4,051 | -60% | 1 | 1 | 0% | 1,675 | 2,230 | +33% | 0 | 0 | — |
case-13 | fail→fail | 14,562 | 8,155 | -44% | 1 | 1 | 0% | 2,168 | 2,908 | +34% | 0 | 0 | — |
case-14 | fail→pass | 9,085 | 5,944 | -35% | 1 | 1 | 0% | 1,665 | 2,636 | +58% | 0 | 0 | — |
case-18 | fail→pass | 14,886 | 6,892 | -54% | 1 | 1 | 0% | 2,587 | 2,877 | +11% | 0 | 0 | — |
case-19 | fail→pass | 14,035 | 6,782 | -52% | 1 | 1 | 0% | 2,532 | 2,723 | +8% | 0 | 0 | — |
case-20 | fail→fail | 2,172 | 4,525 | +108% | 1 | 1 | 0% | 322 | 1,875 | +482% | 0 | 0 | — |
case-21 | pass→fail | 7,466 | 4,870 | -35% | 1 | 1 | 0% | 1,393 | 1,769 | +27% | 0 | 0 | — |
case-22 | pass→fail | 13,387 | 4,651 | -65% | 1 | 1 | 0% | 2,608 | 1,818 | -30% | 0 | 0 | — |
case-23 | fail→fail | 5,534 | 6,118 | +11% | 1 | 1 | 0% | 284 | 1,876 | +561% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 23 cases were attempted, and 16 counted toward the lift figure. The other 7 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +22 percentage points is the difference between those two pass rates over the 16 comparable cases. 3 cases got worse with the skill loaded, and they are included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.