SkillBridge's whole pitch is AI-terminology fidelity — and the curated dictionaries are good, but they have never had a native-speaker pass. A 2026-06-10 full LLM audit caught and fixed real errors in 7 locales (e.g. Italian had `Slack → "Lento"`, German rendered Discernment as `Diskriminierungsfähigkeiten`), which is exactly the class of thing a native reviewer spots in minutes.
What reviewing means (one locale, ~1–2 hours):
- Open `src/data/.json` — every key is the English source, every value is the translation.
- Skim for: wrong meaning, unnatural register (the file's existing register is documented in docs/TRANSLATION_QA.md), brand terms that should stay English (Claude, Anthropic, MCP, Claude Code, Managed Agents, …), and inconsistent renderings of the same English term.
- Open a PR with corrections (or just an issue listing them — we'll apply).
- We set your locale's `_meta.nativeReview` to `"reviewed"` and credit you in the README QA table.
Locales needing a first native pass (one checkbox each — claim one by commenting):
No coding required — JSON value edits only. CI guards the structure (`check:i18n`, `check:dict-coverage`, `check:locales`), so you can't break anything.
SkillBridge's whole pitch is AI-terminology fidelity — and the curated dictionaries are good, but they have never had a native-speaker pass. A 2026-06-10 full LLM audit caught and fixed real errors in 7 locales (e.g. Italian had `Slack → "Lento"`, German rendered Discernment as `Diskriminierungsfähigkeiten`), which is exactly the class of thing a native reviewer spots in minutes.
What reviewing means (one locale, ~1–2 hours):
Locales needing a first native pass (one checkbox each — claim one by commenting):
No coding required — JSON value edits only. CI guards the structure (`check:i18n`, `check:dict-coverage`, `check:locales`), so you can't break anything.