jlpt-n5-reading-passage-creator

name: jlpt-n5-reading-passage-creator description: "Generate JLPT N5 reading passages (short or mid type) in JSON format. Use this skill when the user asks to create, generate, or produce N5 reading passages, comprehension texts, or reading questions. Handles the full pipeline: Japanese core content generation, 5-language translation, file saving, and optional Firebase upload. Trigger on requests like 'N5 okuma metni üret', 'create N5 reading passage', 'short passage oluştur', 'mid passage yaz', 'generate reading comprehension', 'JLPT N5 okuma sorusu oluştur'."

JLPT N5 Reading Passage Creator

Generate N5-level reading passages (core Japanese + 5-language translations).

References

• JSON schemas and field rules: See references/json-schema.md
• N5 scope and quality rules: See references/n5-reading-guidelines.md

Workflow

1. Determine passage type from user request: short (80–150 words, 1–2 questions) or mid (200–350 words, 2 questions). Default: short.
2. Read references/json-schema.md for exact field specifications.
3. Read references/n5-reading-guidelines.md for N5 vocabulary, grammar scope, and quality rules.
4. Read backend/reading/data/n5_passages_{type}_core.json to find the last ID number. New passage ID = n5_reading_{type}_{NNN} (next sequential number, zero-padded).
5. Generate the passage:
   • Write natural N5-level Japanese prose.
   • Decide visual_type (notice, memo, letter, email, none).
   • If visual_type ≠ none: put context text in sentences, boxed content in framed_sentences. Otherwise: use only sentences.
   • Add furigana: <ruby>漢字<rt>reading</rt></ruby> for every kanji word only.
   • Add romaji in Modified Hepburn.
   • Write questions (1–2 for short, 2 for mid) with 4 options each; exactly 1 is_correct: true per question.
   • Sentence IDs: s1, s2… continuous across sentences + framed_sentences.
6. Append new passage to backend/reading/data/n5_passages_{type}_core.json.
7. Validate: Run jlpt-reading-passage-tester skill on the file. Fix any failures, re-run until all pass.
8. Generate translations for all 5 languages (en, tr, de, fr, es):
   • Translate every sentence (both sentences and framed_sentences) naturally and idiomatically.
   • Add mining_text per sentence: localized grammar/chunk explanation + vocabulary list (see mining text headers in json-schema.md).
   • Translate question text and all 4 option texts.
9. Distribute translations: Append each language's translation entry to backend/reading/data/n5_passages_{type}_translations_{lang}.json.
10. Upload prompt: Ask the user whether to upload to Firebase. If yes:

    python3 backend/reading/scripts/upload_passages_v2.py n5
    

    Note: This uploads all N5 passages (short + mid). Existing passages are safely overwritten (Firestore uses set()); the new passage is added alongside them.
11. Output a summary: passage ID, type, question count, languages translated, upload status.

Quality Rules

• All vocabulary and kanji within N5 scope.
• Questions must be answerable ONLY from the passage text.
• Exactly 1 correct option per question; 3 plausible distractors using passage vocabulary.
• Default: 1 passage per run unless user specifies more.