yuanyuanyuan430/ganhuo-textbook-downloader

Ganhuo Textbook Downloader

Use this skill to turn a natural-language textbook request into ranked candidates and, when safe, a downloaded textbook file from TapXWorld/ChinaTextbook.

Workflow

1. Extract province/city, stage, grade, subject, version, semester/book volume, edition, publisher/school clues, and any exact title clues.
   • K-12 examples: 江苏九上译林英语, 安徽合肥三年级语文上册.
   • University examples: 大一高数同济第七版上册, 本科线代同济第五版, 概率论浙大四版.
   • Cover/page image examples: OCR readable text such as 新版译林 九上教材, Know yourself, Welcome to the unit.
2. Search first. Province and city words are used for ranking, not as required path tokens:

python3 "${CLAUDE_SKILL_DIR:-.}/scripts/china_textbook_downloader.py" \
  search "安徽 合肥 小学 三年级 语文 上册" --explain

3. If the user provides a cover/page screenshot, pass it as an OCR clue:

python3 "${CLAUDE_SKILL_DIR:-.}/scripts/china_textbook_downloader.py" \
  search "新版译林 九上 英语" \
  --cover-image "/absolute/path/to/screenshot.png" \
  --explain

4. If the user already supplied extracted cover text, pass it directly:

python3 "${CLAUDE_SKILL_DIR:-.}/scripts/china_textbook_downloader.py" \
  search "九上英语" \
  --ocr-text "新版译林 九上教材 Know yourself Welcome to the unit" \
  --explain

5. If one candidate clearly matches the user's request, download it:

python3 "${CLAUDE_SKILL_DIR:-.}/scripts/china_textbook_downloader.py" \
  download "安徽 合肥 小学 三年级 语文 上册" \
  --out "$HOME/Downloads/ChinaTextbook" \
  --auto

6. If multiple candidates are plausible, show the numbered list and ask the user which one to download. Province preferences and cover/OCR clues are ranking hints.
7. Return the local file path, upstream GitHub path, source URL, split-file status, and ranking reasons.

Commands

Search with province inference:

python3 "${CLAUDE_SKILL_DIR:-.}/scripts/china_textbook_downloader.py" \
  search "广东 深圳 七年级 数学 上册" --limit 5 --explain

Override province when the user implies location outside the text:

python3 "${CLAUDE_SKILL_DIR:-.}/scripts/china_textbook_downloader.py" \
  search "七年级 数学 上册" --province 江苏 --explain

Download by result number:

python3 "${CLAUDE_SKILL_DIR:-.}/scripts/china_textbook_downloader.py" \
  download "上海 高一 英语 必修 第一册" --pick 1

Search university textbooks:

python3 "${CLAUDE_SKILL_DIR:-.}/scripts/china_textbook_downloader.py" \
  search "大学 高等数学 同济 第七版 上册" --limit 5 --explain

Run the built-in check after editing the script:

python3 "${CLAUDE_SKILL_DIR:-.}/scripts/china_textbook_downloader.py" self-test

Province Matching

The script uses province/city clues to boost likely textbook versions. It does not claim an official province-to-textbook assignment.

• For 语文, 道德与法治, and 历史, prefer 统编版 because these are commonly unified subjects.
• For other subjects, use province/city preference hints such as 江苏 -> 苏教版/译林版, 北京 -> 北京版/外研版/北师大版, 上海 -> 沪教版, 深圳/广东 -> 北师大版/沪教版/人教版, and broader 人教版 fallbacks.
• If the user explicitly names a version, the explicit version wins over province preferences.
• If the request is vague or candidate rankings are close, ask the user to choose.

Read references/province-preferences.md only when you need to explain the limitation to the user or update the province preference table.

University Matching

The upstream repository currently stores university resources under 大学/, mainly math-related subjects such as 高等数学, 线性代数, 离散数学, and 概率论.

• Map natural language such as 本科, 高校, 大一, 高数, 线代, 离散, 数统, and 概率统计 into repository terms.
• Use edition and source clues such as 同济, 浙大, 第五版, 第六版, 第七版, 四版, and 原书 as strong matching terms.
• Demote non-main materials such as 答案, 习题, 全解, 解析, 学习辅导, 附册, and 指南 unless the user explicitly asks for them.
• If the requested university edition is absent, report the closest candidates instead of pretending a near miss is correct.

Cover And OCR Matching

The script can use readable text from a cover or page screenshot. It runs local tesseract OCR when available and accepts --ocr-text when text has already been extracted.

• Useful clues: title, version/publisher, grade/volume, unit title, edition, author/editor, school/publisher names.
• Noisy clues from phone/PDF viewer UI are filtered where possible.
• OCR clues help ranking; they are not proof that the exact edition exists in the upstream repository.

Boundaries

• Only download publicly accessible upstream GitHub files. Do not bypass login, payment, copyright, or access restrictions.
• Do not promise that the province ranking is authoritative or current for every school.
• Do not auto-download when grade, subject, or book volume is unclear and multiple candidates exist.
• Do not auto-download a near-miss university edition, such as a fifth edition when the user asked for the sixth edition.
• Split files such as .pdf.1 and .pdf.2 are byte chunks; the script downloads them in numeric order and concatenates them into one .pdf.