What problem does it solve?
Reviewing machine-translated content across 24 languages is error-prone: brand names get mistranslated, internal links get localized and break navigation, MDX syntax errors break builds, and semantic inversions (like swapping proof-of-stake and proof-of-work) slip through. This Skill provides a deterministic review framework so every translation PR is evaluated against the same authoritative rules.
Core Features & Use Cases
- ETHGlossary-as-authority enforcement: Verifies every translated term against the ETHGlossary API, classifying deviations as critical issues that can be auto-fixed.
- Severity triage and scoring: Distinguishes critical (build-breaking, navigation-breaking, semantic) issues from warnings, and scores each locale on a 5-category rubric (Brand Name Preservation, Technical Accuracy, Semantic Fidelity, Terminology Consistency, Tone/Register).
- Multi-agent review roles: Splits large PR reviews across structural, terminology, and semantic agents, with per-language rules for 13 non-Latin-script languages.
- Use Case: When reviewing a Japanese translation PR on an
intl/pending-dev branch, the Skill checks that Solidity stays Latin per its always_latin rule, flags a translated /governance href as critical, and produces a per-language quality score plus findings file update.
Quick Start
Review the Korean translations on the intl/pending-dev branch and report critical issues with a quality score.