Russian Text Quality Review

Overview

Detects Russian-language quality bugs that general linters and LLMs consistently miss in code: broken pluralization in i18n locale files, case-agreement bugs in string concatenation, terminology drift across files, transliterated identifiers, and inconsistent language mixing. Code-aware, not prose-focused.

Core principle: structural Russian mistakes that survive eslint and type-checks but break production UIs. Prose quality is delegated to other tools.

Activation

Run this skill ONLY if BOTH conditions hold:

The project contains ≥10 total Cyrillic characters across code and locale files (excluding node_modules, .git, dist, build, and .gitignored paths).
At least one file contains ≥5 consecutive Cyrillic characters (rules out single stray tokens).

If either condition fails, respond with Not activated — no Russian content detected and stop. Never run on English-only projects.

Full activation pseudo-code: references/activation-logic.md.

What this skill checks

#	Check	Detection	Severity	Confidence
1	Plural structure — i18next/ICU/vue-i18n Russian plurals missing CLDR categories (`one`/`few`/`many`/`other`)	Deterministic parse of locale files + call sites	Error	High
2	Terminology drift — same source key translated to different Russian terms across files, or known-synonym variance	Normalized value comparison	Warning	Medium
3	Concatenation case bug — Russian preposition/verb governing non-nominative case followed by string interpolation	Heuristic regex + closed trigger list	Warning	Medium
4	Language mixing — inconsistent comment language in a single file, or inline Russian literals outside i18n	Per-file line-language ratios	Warning	Medium
5	Transliterated identifiers — `polzovatel`, `tovar`, `kolichestvo` etc.	Dictionary + phonotactic + file-context (3-tier)	Info	Low

Full detection algorithms live in references/plural-rules.md (check #1), references/concatenation-triggers.md (check #3), and references/mixing-and-transliteration.md (checks #4 and #5). Check #2 terminology detection is described in the workflow below and in .claude/russian-glossary.example.yml.

What this skill does NOT check

Delegate these to purpose-built tools:

Out of scope	Recommended tool
Typography (« », em-dash —, non-breaking spaces)	talkstream/ru-text
Spelling	yaspeller
ё/е restoration	eyo / eyo-kernel
UX writing quality, translationese	talkstream/ru-text

Review workflow

Check activation. If conditions not met, stop.
Detect i18n framework from package.json: i18next, react-intl/@formatjs/*, vue-i18n/@intlify/*, or none. If multiple, run all applicable detection strategies.
Run checks against matching file targets. Skip checks whose file types are absent.
Emit the Markdown report in the format below.
Do NOT modify files without explicit user approval. Findings are suggestions except for #1 plural scaffolding, which may be auto-applied only when the user approves.

Output format

# Russian Text Review — <project name>

## Summary
- Files scanned: N
- Russian content detected in: M files
- Findings: X errors, Y warnings, Z info

## Findings

### Errors (must fix)
#### #1 Plural structure
- [HIGH] path/to/ru.json:L — key `items.count` missing `_few`, `_many` siblings
  - **Why:** Russian CLDR requires `one/few/many/other`. Missing `_few` produces wrong form for n=2,3,4.
  - **Fix:** add `items.count_few`, `items.count_many` entries (scaffolded below).

### Warnings
#### #5 Terminology drift
- [MEDIUM] locales/ru.json vs locales/admin/ru.json — key `user` translates to `пользователь` and `юзер`
  - **Why:** inconsistent terminology confuses users and breaks search.
  - **Fix:** pick one; lock via `.claude/russian-glossary.yml`.

### Info
#### #1 Transliterated identifiers
- [LOW] src/order.ts:12 — identifier `kolichestvo_tovarov` looks like transliterated Russian
  - **Suggestion:** `product_quantity` or keep Cyrillic if that is your project convention.

## External tools we recommend
- Typography / UX writing → talkstream/ru-text
- Spelling → yaspeller
- ё/е → eyo

The "External tools we recommend" block appears only when the scan finds Russian content. On English-only projects, the skill does not run at all.

Common mistakes when using this skill

Running on English-only projects. Honor the activation guard. Do not bypass it to "check anyway".
Flagging typography, spelling, or ё/е. These are explicit non-goals — delegate, do not duplicate.
Treating medium-confidence heuristics as errors. Checks #3, #4, #5 emit warnings or info; never promote them to errors.
Auto-fixing without approval. Only #1 plural scaffolding can be offered as an auto-fix, and only after the user approves the suggested diff.
Matching key_plural suffixes. This skill targets i18next v4 format only. Old v3 projects using key_plural are out of scope in v1: when detected, emit a single Info finding noting the v3 file was skipped, then stop plural checks for that file.

References

references/plural-rules.md — CLDR Russian rules + per-framework detection algorithms (i18next, ICU, vue-i18n) with test cases. Includes Step 2.5 for locale values with count-placeholders but no plural siblings.
references/concatenation-triggers.md — closed list of triggers for check #3 with exclusions
references/mixing-and-transliteration.md — detection pseudo-code and thresholds for checks #4 and #5
references/transliteration-dict.yml — seed dictionary used by check #5
references/activation-logic.md — activation pseudo-code + test cases
references/examples.md — 16 before/after cases sourced from real projects
.claude/russian-glossary.example.yml — template for user-provided terminology glossary

russian-text-quality