Turns syllabi and lecture slides into exhaustive, source-attributed review notes — every claim traceable, every value verbatim, nothing fabricated. Designed for medical students, validated across surgery, preventive medicine, and anesthesiology.
$ /review ~/Desktop/HMC/大三下/外科
Phase 0 · 扫描 + 分类
大纲 ✅ 2023级-《外科学总论》教学大纲.pdf
日历 ✅ 嵌入大纲各章的学时分配表
PPT 8 files covering 9 theory chapters
分类置信度: 9/9 auto-approved (≥85%)
Phase 1 · 提取 [██████████] 9/9 ✅
PDF: 01绪论 02休克 03感染... → 175K chars
PPTX: 水电解质 创伤 营养... → 81K chars
Phase 2 · 交叉验证
Ch1-Ch9: ✅ confirmed
Ch10 综合实训: — practice-only, excluded
0 fatal gaps · 0 warnings
Phase 3 · 生成
→ 外科学总论-复习笔记.md (exhaustive extraction, 83 h3 sections)
→ 外科学总论-复习笔记.html (sidebar TOC with search, print-ready)
→ 外科学总论-记忆卡.anki.csv
Output preview (Surgery, Chapter 6 — Surgical Infection):
项目 内容 病原 破伤风梭菌 (G⁺ anaerobic bacillus), produces tetanospasmin 潜伏期 Typically 6–10 days (shorter = more severe) 典型表现 Risus sardonicus → Trismus → Opisthotonos → Respiratory failure 预防 Debridement + active immunization (toxoid) + passive: TAT 1500 IU or TIG 250–500 IU 治疗 Metronidazole + TAT/TIG + Diazepam + Tracheostomy if needed 可能的考点
PPT末页思考题 → 破伤风的临床表现和防治（掌握级）
大纲双重点 → 脓毒症概念及qSOFA三项快速筛查
This skill was designed and iterated through real clinical medicine exam preparation. Three courses validated:
- Surgery (考查课, 9 chapters, 8 slide decks) — student reported full exam coverage
- Preventive Medicine (考试课, 6 chapters, 13 slide decks + independent calendar) — matched original notes
- Anesthesiology (考查课, 10 chapters, 5 slide decks + review guide) — first-pass generation
- Syllabus as single source of truth — every output claim must trace back to a syllabus line
- Three-source cross-validation — syllabus, teaching calendar, and lecture slides compared chapter by chapter; discrepancies explicitly annotated rather than silently resolved
- No content generation — organize, extract, and classify only; never fabricate knowledge points
- Explicit gap annotation — missing materials are clearly marked and visually distinct from confirmed content
- Source attribution — every chapter traces back to its originating files; numerical values are transcribed verbatim
git clone https://github.com/Wanix2026/review.git ~/.claude/skills/review
pip3 install python-pptx python-docx openpyxl pdfminer.sixRestart Claude Code. The skill loads automatically.
Organize your course folder:
外科学/
├── 大纲.pdf ← syllabus (required)
├── 课件/ ← lecture slides
│ ├── 01绪论.pdf
│ ├── 02休克.pdf
│ └── ...
└── 复习重点.docx ← exam guide (optional)
In Claude Code:
/review ~/Desktop/外科学
The skill scans, classifies, extracts, cross-validates, and generates. Confirm once at the start, then wait. Output:
外科学/
├── 外科学-复习笔记.md ← Markdown source
├── 外科学-复习笔记.html ← sidebar TOC + search + print
├── 外科学-记忆卡.anki.csv ← Anki import
└── 外科学-思维导图.md ← Mermaid chapter map
From PDF/Word files: drop the file into the course folder. The skill auto-detects, cleans (strips explanations, student answers, separators), and exports .docx for import into 考试宝 or other flashcard apps.
From web platforms (人卫/学习通): open the exam page in your browser, press F12 to open the Console, paste this extraction script:
// Dump all visible text from the exam page
console.log(document.body.innerText);Copy the Console output, paste it into Claude Code with: "Clean this question bank — keep only question type, stem, options, and correct answer. Remove explanations, student answers, and separators. Export as .docx."
| Condition | Behavior |
|---|---|
| Question types provided | Structured sections per question type (A1/A2/B1/名词解释/简答) |
| No types given (default) | Pure knowledge organization — no type labels |
Phase 0 — File Discovery & Classification
Directory scan, automatic file-type classification
Content-based validation (12 Chinese filename aliases)
Guard check: syllabus and teaching calendar must exist
Four-layer chapter matching (number → content sample → keyword → calendar bridge)
Environment dependency pre-check
Phase 1 — Full Extraction (incremental, cached)
SHA-256 file fingerprinting, incremental mode
Format diagnosis: PDF (text/scanned/encrypted), DOCX, DOC, PPTX, XLSX
Post-extraction integrity check with garbled text detection
Structured cache at .course-cache/ with JSON Schema constraints
Resume-from-breakpoint support
Phase 2 — Three-Source Cross-Validation
16-state decision table (syllabus × calendar × slides)
Four-tier gap classification with actionable responses
Conflict resolution: calendar priority, explicit evidence preservation
Maximum 2 feedback rounds before forced output
Phase 3 — Review Note Generation
Fixed three-section chapter template (requirements → content → likely exam topics)
Content depth enforcement: verbatim numerical transcription, step-by-step preservation
Mandatory post-generation completeness audit with coverage report
Source attribution on every chapter
HTML output with sidebar TOC, scroll-spy, mobile responsive, print stylesheet
course-directory/
├── .course-cache/ # Structured extraction cache
├── [Course]-复习笔记.md # Markdown source
├── [Course]-复习笔记.html # HTML with TOC and section navigation
├── [Course]-复习笔记.pdf # Print-optimized PDF (optional)
├── [Course]-思维导图.md # Mermaid chapter mindmap
├── [Course]-记忆卡.anki.csv # Anki flashcard import
└── [Course]-题库导出.docx # Question bank export (optional)
.claude/skills/review/
├── SKILL.md # Full workflow specification
├── install.sh # One-command environment setup
├── README.md
├── scripts/
│ ├── extract.py # Unified extraction (6 formats)
│ ├── verify.py # Post-extraction integrity check
│ ├── to_html.py # Markdown to HTML with sidebar TOC
│ ├── anki.py # Anki CSV flashcard export
│ └── mindmap.py # Mermaid mindmap generator
└── references/
├── template.html # HTML output template
├── schema.md # JSON Schema for cache files
└── checklist.md # Quality assurance checklist
The skill enforces strict accuracy constraints:
- All knowledge points must have verifiable source file origins
- Numerical values transcribed verbatim — no rounding, no format changes, no unit dropping
- Procedural steps listed individually — merging or summarizing is forbidden
- Comparison tables preserved in table form — never flattened to prose
- Syllabus sub-items tracked for full coverage — any item below 90% coverage triggers re-extraction
- Post-generation audit runs automatically, with results appended to output
| Capability | Description |
|---|---|
| Format support | PDF (text layer, scanned, encrypted), PPTX, DOCX, DOC, XLSX |
| Intelligent classification | Filename + content dual validation, 12 Chinese alias patterns |
| Three-source validation | 16-state decision table with explicit conflict annotation |
| Incremental updates | SHA-256 fingerprinting with fast update mode |
| Resume from breakpoint | Extraction state persisted; restart from interruption point |
| Multi-project tracking | Global project registry remembering per-course configuration |
| Anki export | Auto-extract comparison tables, terminology, and numerical facts to CSV |
| Mindmap generation | Mermaid chapter structure visualization |
| Question bank processing | Web platform extraction, cleaning, and .docx export |
| Tool | Purpose |
|---|---|
| pdf2txt (pdfminer.six) | PDF text extraction |
| python-pptx | PPTX slide text extraction |
| python-docx | Word document processing |
| openpyxl | Excel spreadsheet processing |
MIT