Okhp3 I18N Page Release

okhp3-i18n-page-release

OverKill Hill P³ · overkillhill.com · github.com/OKHP3

This is the release stage after an exact-pair translation skill and before a localized page is treated as public. It keeps language transformation separate from page delivery and keeps deterministic source drift separate from technical SEO and rendered-content evidence.

Scope

Required inputs

• one reviewed or explicitly draft-marked output from the matching okhp3-translation-<source>-<target> skill;
• an owner-approved locale registry based on assets/locale-registry.example.json;
• the static HTML files named in that registry; and
• the site's stated release policy for each target: indexable or draft-noindex.

If the source locale, target BCP-47 tag, route mapping, site origin, or release state is unknown, return BLOCKED. Do not guess a country from a language-only folder name, and do not change a draft to indexable to make validation pass.

Procedure

1. Confirm the exact-pair translation record: direction, source and target hashes, dictionary, review state, unresolved terms, and output locale. A translation without a review record remains a draft, even if HTML exists.
2. Read the locale registry and its explicit release policy. It is the one source for locale tag, route root, direction, indexability, optional x-default fallback, declaration policy, and exact-pair skill. A filesystem slug such as fr is not a substitute for fr-FR.
3. Run scripts/validate-i18n-page-release.py --help, then use --check on the registry. It verifies each declared HTML file, BCP-47 html lang, self-canonical URL, configured alternate graph, noindex staging policy, and optional og:url equality. It never writes files or fetches a site.
4. Apply the owner-set public-cluster policy. By default, an indexable locale set is a complete, reciprocal cluster: each indexable page names itself and every other indexable equivalent with the registry's BCP-47 tags, plus the declared x-default. Draft-noindex pages remain outside that public cluster until their review gate is complete. This is a project release policy, not a claim that every item is universally required by a search engine.
5. If a browser language-detection adapter is available, inspect the rendered page and retain its report. Reconcile detected body language with html lang, nested lang declarations, and hreflang. A short page, unreliable detector result, or unavailable adapter is UNKNOWN, not a pass or a language-quality failure.
6. Check the language switcher manually or with a browser: it names locales in their own language and maps to the equivalent available route. Do not force a geo-IP or Accept-Language redirect.
7. Return structural results, rendered-language evidence, translation-review evidence, and release state separately. ready-for-release requires all structural checks to pass, a reviewed indexable translation, and an explicit owner decision to publish. This skill never makes that publication.

Boundaries with adjacent skills

Output contract

Return Locale registry, Translation evidence, Structural validation, Rendered-language evidence, Release state, Warnings and unknowns, and Next action. Distinguish pass, fail, warning, unknown, and not-run. A passing structural validator is not native-language certification, search-ranking evidence, or publication confirmation.

Resource routing

• Read references/locale-registry.md before creating or changing a registry.
• Use assets/locale-registry.example.json as a starting shape, never as live site data.
• Run scripts/validate-i18n-page-release.py --help before direct use.
• Read evals/evals.json when changing the skill's validation policy.
• Browser adapters may use an Adobe page-langs-style declared-versus-detected report, but that optional adapter must not become a dependency of the portable core.

Evaluation and release

evals/evals.json covers an indexable reciprocal cluster, a staged noindex draft, head-only duplicate-aware metadata, configured language-declaration and x-default policy, and unavailable rendered-body language evidence. Tests establish structural behavior only. No fresh live browser benchmark, native-review study, or search-performance result has been run for version 1.1.0.

About

Built by Jamie Hill · OverKill Hill P³ Published at github.com/OKHP3 Part of the OKHP3/skillz Agent Skill library. MIT License -- free to use, fork, and adapt. A nod to the source is appreciated.