What problem does it solve? Translating documentation and localization files by hand or with ad-hoc LLM prompts risks corrupted placeholders, broken Markdown structure, inconsistent terminology, and overwritten originals. This Skill orchestrates a controlled translation pipeline with an approval gate, parallel translator agents, and three tiers of verification so file translations stay structurally intact and terminologically consistent. ## Core Features & Use Cases - Plan-Approve-Translate Flow: Discovery produces a reviewable plan (files, languages, output mode, glossary, wave plan) that requires explicit user approval before any translator dispatches. - Three-Tier Verification: Tier-1 structural parity checks (fence counts, headings, placeholders), Tier-2 LLM-as-Judge semantic review on every chunk, and Tier-3 uncertainty-driven lookups via Wiktionary, MyMemory, and back-translation. - Broad Format Support: Handles md, mdx, json, yaml, po, properties, strings, xliff, and txt files with sidecar, inplace, or i18n output modes and automatic source-language detection. - Use Case: Run "translate README.md to Russian" to get a chunked, glossary-seeded translation written to README.ru.md, with every placeholder like {{user.name}} preserved byte-identical and a summary of verification verdicts. ## Quick Start Ask the assistant to translate README.md to Japanese using the translate skill and approve the generated plan when prompted.
