i18n-localization

i18n Localization

Use this skill to plan, review, or harden localization work across product UI, documentation, marketing pages, emails, and web/app routing surfaces.

Canonical Vocabulary

Term	Meaning
locale	Language plus optional regional conventions, such as en, en-US, or ar-EG
message catalog	Source of translated strings and message metadata
ICU message	Message format that models variables, plurals, selects, and ordinals safely
pseudo-locale	Test locale that expands and marks strings to reveal hardcoded text and layout clipping
RTL	Right-to-left layout direction for languages such as Arabic and Hebrew
fallback	Intentional behavior when a locale, key, or translation is missing

Dispatch

$ARGUMENTS	Mode	Output
Empty	Localization audit	Inspect the current surface, identify locale risks, and return a prioritized remediation plan.
App, route, feature, or file path	Implementation plan	Map the affected strings, routing, formatting, QA, and release checks for that scope.
Locale or market list	Locale readiness review	Verify language, region, formatting, routing, fallback, and translation handoff requirements.
Existing plan or diff	Review	Check for missed strings, unsafe concatenation, formatting bugs, catalog drift, RTL gaps, and missing tests.
Natural language outside localization	Redirect	Explain the scope boundary and route to the better skill.

When the user gives no arguments, start with a localization audit of the relevant repository or product surface. Do not ask for locale preferences until the codebase and user request leave a real decision unresolved.

Reference File Index

File	Read When
references/localization-checklist.md	Planning, implementing, reviewing, testing, or releasing localization work

Critical Rules

1. Treat localization as a product contract, not a translation pass. Cover routing, formatting, accessibility, search metadata, analytics labels, emails, and support surfaces when they are in scope.
2. Never concatenate translated fragments to build a sentence. Use full messages with variables, ICU/select/plural forms, or the framework's equivalent message format.
3. Keep user-visible strings out of business logic and shared constants unless the system already has a catalog abstraction for that layer.
4. Use locale-aware formatters for dates, times, numbers, currencies, lists, relative times, units, and names. Do not hand-roll formatting with string templates.
5. Separate language from region. en, en-US, and en-GB can imply different copy, dates, currency, measurement units, and legal content.
6. Define fallback behavior explicitly. Missing translations should fail visibly in development and degrade intentionally in production.
7. Include pseudo-locale checks for expansion, accenting, bracket markers, interpolation safety, and truncation.
8. Include RTL and bidirectional text checks when any Arabic, Hebrew, Persian, Urdu, or mixed-direction user content can appear.
9. Preserve stable message keys. Do not churn catalogs or invalidate translation memory unless the source message meaning changes.
10. Keep translators away from code syntax. Provide notes, screenshots, variable descriptions, character constraints, and product context.

Workflow

1. Scope The Surface

Identify the surfaces affected by the request:

• UI components, server-rendered pages, native screens, emails, PDFs, docs, notifications, error states, empty states, and transactional copy.
• Route structure, locale negotiation, redirects, canonical URLs, hreflang, sitemap entries, and persisted user locale preferences.
• Data-driven content, CMS fields, markdown/docs content, legal text, pricing, units, and market-specific availability.

If the request is implementation work, read the existing i18n library, routing conventions, catalog format, and test patterns before proposing changes.

2. Extract And Model Messages

For each string-bearing surface:

• Move user-visible strings into the existing catalog or message layer.
• Use full-sentence messages with named variables.
• Model plural, select, gender, ordinal, and count-sensitive text with ICU-style variants or the framework's native equivalent.
• Add translator descriptions for ambiguous labels, product terms, placeholders, and variables.
• Keep stable keys tied to semantic meaning, not English wording alone.

Avoid generic keys such as title, subtitle, or buttonText unless they are nested under a stable feature namespace.

3. Localize Formatting

Check all locale-sensitive values:

• Date, time, timezone, relative time, and calendar assumptions.
• Number, decimal, percent, currency, accounting, compact notation, and unit formatting.
• Lists, names, addresses, phone numbers, postal codes, sorting, casing, and search normalization.
• Form validation, parsing, masks, and server/client round-trip behavior.

Prefer platform formatters such as Intl.*, framework helpers, or vetted library APIs.

4. Route And Persist Locale

Verify:

• How the app chooses a locale: URL prefix, domain, account setting, cookie, browser language, or explicit selector.
• Whether redirects are deterministic and cache-safe.
• Whether canonical URLs, alternate links, metadata, Open Graph, sitemap, robots rules, and docs navigation match the locale strategy.
• Whether APIs, background jobs, and emails receive the same locale context as the visible app.

Do not mix locale negotiation and authentication or billing policy unless the product already couples those concepts.

5. Test And QA

At minimum, recommend or add checks for:

• Catalog completeness and unused key detection.
• Missing interpolation variables.
• Pseudo-locale rendering.
• Snapshot or visual coverage for expanded strings and RTL layout where relevant.
• Formatting tests for representative locales, including at least one non-US locale.
• Accessibility checks for language attributes, direction, screen reader labels, and translated aria text.

For review tasks, report what the available tests prove and what still requires visual or translator QA.

Progressive Disclosure

• Start with this SKILL.md for routing, hard rules, and output shape.
• Read references/localization-checklist.md only when planning, reviewing, or releasing concrete localization work.
• Do not preload unrelated frontend, SEO, copywriting, or JavaScript guidance.
• Use the existing project i18n library and catalog format before introducing new tooling.

Output Shape

Return concise, implementation-oriented output:

• Findings for audits and reviews, ordered by severity with file or route references.
• Plan for implementation requests, ordered by dependency: routing/catalog foundation, extraction, formatting, QA, release.
• Acceptance Criteria that include catalog completeness, pseudo-locale, formatting, fallback, and route/SEO checks when relevant.
• Open Questions only for unresolved choices that cannot be grounded in the codebase or user request.

Do not rewrite marketing copy unless the user explicitly asks for copy editing. Do not perform general frontend redesign unless localization exposes a concrete layout or directionality defect.

Validation Contract

Before declaring this skill complete after edits:

uv run wagents validate
uv run wagents eval validate
uv run python audit.py skills/i18n-localization
uv run wagents package i18n-localization --dry-run

Completion criteria:

• Skill and eval validation pass.
• Audit score is A or all remaining findings are explicitly accepted.
• Package dry-run passes.
• A smoke review of the dispatch table covers empty, explicit, implicit, and negative-control prompts.