I18n

Skill Syo-M/codex-frontend-skills/.agents/skills/i18n

Codex-native frontend skills, custom agents, profiles, and an evidence-backed evaluation harness for React, Next.js, Vite, and Astro.

Install

npx -y skills add Syo-M/codex-frontend-skills --skill i18n

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

24 days oldThe repository was created 24 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Internationalization for catalogs, ICU messages, Intl formatting, locale routing, language switching, localized validation, and RTL. 日本語の依頼例:「多言語対応」「翻訳」「ロケール切替」「日付や通貨表示」「RTL対応」。

SKILL.md

4.0 KB, as published. Nobody here has run it

Internationalization

No hardcoded user-facing strings

Every user-visible string comes from a message catalog keyed by ID — never a literal in JSX. This is the one rule that, if skipped, makes everything else impossible to retrofit.
Use the framework-idiomatic library: next-intl (Next.js App Router), react-i18next / react-intl (Vite SPA), Astro's i18n routing + a catalog. One per repo (dependency ask-first rule applies).
Keys are semantic and namespaced (checkout.payment.submit), not the English text. Keep catalogs shallow (≤3 levels) and run a missing/orphaned-key check in CI (i18next-parser --fail-on-update, or the library's extractor) — wire it as a gate, don't rely on memory.
Translator-facing context: provide a description for ambiguous keys; never concatenate sentence fragments across keys (word order differs per language).

Plurals, gender, interpolation — ICU MessageFormat

Never build plurals with count === 1 ? 'item' : 'items' — many languages have 3–6 plural categories. Use ICU {count, plural, ...} and let the library + locale data decide.
Interpolate variables through the catalog ({name}), never via template-string concatenation around a translated fragment.
Rich/linked text: use the library's component interpolation (<Trans> / t.rich), not dangerouslySetInnerHTML on a translated string (see frontend-security).

Formatting — always `Intl`, never hand-rolled

Dates/times: Intl.DateTimeFormat with an explicit locale and time zone. Store/transport UTC; render in a deliberately chosen zone; label it when ambiguous (consistent with data-viz).
Numbers/currency/units/lists: Intl.NumberFormat (with style: 'currency' + currency code), Intl.ListFormat, Intl.RelativeTimeFormat. No manual thousands separators, decimal marks, or "3 days ago" strings — these differ per locale.
Currency: format with the data's currency, not the UI locale's; never assume $.

Routing & locale negotiation

Locale in the URL (/ja/... subpath, or domain) so pages are linkable, cacheable, and indexable — not only a cookie/header. Emit hreflang alternates and set <html lang> (and dir) per request.
Negotiate initial locale from the URL → stored preference → Accept-Language, with a sane default; let users switch explicitly and persist it.
SSR/RSC: load only the active locale's catalog on the server and pass it down; don't ship every language's messages to the client (bundle bloat). Code-split catalogs by locale/route.

RTL & layout

Drive direction from data: <html dir="rtl"> for RTL locales; never hardcode LTR assumptions.
Use CSS logical properties (margin-inline, inset-inline-start, text-align: start) through the active styling approach. Physical left/right assumptions break RTL.
Mirror directional icons (arrows, chevrons) in RTL; don't mirror logos or media. Test one RTL locale (e.g. ar) as a first-class case, plus a pseudo-locale for length/encoding.

Locale-aware input & validation

Accept locale-formatted input (decimal comma, native digits, local date order) and parse to a canonical form before validating with zod (see frontend-security) — validate the parsed value, store canonical.
Names/addresses/phone: don't assume Western structure (first/last, state/ZIP). Keep fields permissive; avoid regex that rejects valid non-Latin input.
Sorting/search: locale-aware collation (Intl.Collator), not byte order.

Testing

Pseudo-localization (accented + padded strings) in a story/build catches truncation, clipping, and hardcoded strings early (pairs with visual-regression over both LTR and one RTL locale).
Stories cover at least one non-English and one RTL locale for layout-sensitive components (see storybook).