Ja Proofreading

Skill: Japanese Proofreading (textlint)

Detect and report issues in Japanese text using textlint with Japanese preset rules.

Rule Presets

• textlint-rule-preset-ja-technical-writing — 25 rules for technical docs (sentence length, commas, ら抜き, style consistency, etc.)
• textlint-rule-preset-japanese — 12 general rules (stable, low false-positive)
• textlint-rule-preset-ja-spacing — spacing rules between Japanese and alphanumerics

Steps

Step 1: Identify input

• Text pasted in conversation → write to a temp file, then process
• File path given → use that file directly
• Uploaded file → read from /mnt/user-data/uploads/

Step 2: Set up environment

Use scripts/setup_textlint.sh. Check for an existing setup at /tmp/textlint-ja/; install if missing:

ls /tmp/textlint-ja/node_modules/.bin/textlint 2>/dev/null || bash <skill-dir>/scripts/setup_textlint.sh

Step 3: Run textlint and format output

cd /tmp/textlint-ja
./node_modules/.bin/textlint --config .textlintrc.json --format json <target-file> \
  | python3 <skill-dir>/scripts/format_results.py -

If python3 is unavailable, save JSON to a temp file and parse manually.

Step 4: Report (in Japanese)

Present the formatted output from format_results.py as-is. If the script was not available, report in the following format:

## 校正結果: <ファイル名>

### 概要
- 指摘件数: X件
- エラー（必須修正）: X件
- 警告（推奨修正）: X件

### 指摘一覧

**1行目: 「〇〇〇」**
- ルール: `preset-ja-technical-writing/no-dropping-the-ra`
- 問題: ら抜き言葉が使われています
- 修正案: 「〇〇られ」→「〇〇れ」

...

### 修正済みテキスト（任意）
<corrected version if requested>

Step 4.5: Colloquial contraction check (supplement)

textlint does not catch colloquial contractions. Run this grep after textlint and include findings in the report:

grep -n "てた\b\|てる\b\|てく\b\|でた\b\|でる\b\|ちゃ\|じゃ\|なきゃ" <target-file> \
  | grep -v "^\`\`\`" \
  | grep -v "^    "

Common contractions and corrections:

Exclude matches inside code blocks, quotes, file names, or URLs.

Step 5: Provide corrected text (optional)

If the user asked for corrections, provide the full corrected text. Note which fixes are mechanical and which require judgment.

textlintrc Configuration

Use the following at /tmp/textlint-ja/.textlintrc.json:

{
  "rules": {
    "preset-ja-technical-writing": {
      "sentence-length": {
        "max": 150,
        "skipPatterns": ["\\[.*?\\]\\(.*?\\)"]
      },
      "max-ten": {
        "max": 4
      },
      "no-mix-dearu-desumasu": {
        "preferInBody": "です・ます",
        "preferInHeader": "である",
        "strict": true
      },
      "no-exclamation-question-mark": false,
      "ja-no-weak-phrase": false
    },
    "preset-japanese": true,
    "preset-ja-spacing": {
      "ja-space-between-half-and-full-width": {
        "open": false,
        "close": false
      }
    }
  },
  "filters": {
    "comments": true
  }
}

Rationale for customizations:

• sentence-length.max: 150 — technical docs contain URLs and code; 100 is too strict
• max-ten.max: 4 — 3-comma limit is too tight; relaxed to 4
• no-mix-dearu-desumasu.strict: true — catch style mixing even mid-paragraph and in lists
• no-exclamation-question-mark: false — allow !/? in blog posts
• ja-no-weak-phrase: false — "〜と思います" is context-dependent; skip

Rule Error Reference (Japanese)

Notes

• Code blocks (```) are not checked — this is expected behavior
• URLs and command examples flagged as "long sentences" are false positives; skip them
• If writing style (です・ます vs だ・である) is unclear, scan the whole document and report the dominant style
• When there are many findings, prioritize errors (severity: error) and report only the count of warnings
• With strict: true on no-mix-dearu-desumasu, list-item である-style will error. If the user says "lists can use である", either set strict: false or suppress that rule only
• Step 4.5 hits inside code, quotes, or URLs are false positives — exclude them