exam-notes-generator-skill

Npx-installable AI skill for generating exam-ready LaTeX study notes from course materials (PDF, DOCX, PPTX). Includes diagram extraction, OCR, worked numericals, and complete compilation-ready .tex output.

What This Is

This package scaffolds a Windsurf / Copilot-compatible agent skill into any project. The skill turns lecture slides, PDFs, and Word documents into:

• Complete, compilable LaTeX (.tex) documents with proper preamble, TOC, sectioning, equations, and diagrams
• Exhaustive study notes that preserve every concept, formula, and worked example from your sources
• Diagram extraction & embedding with semantic filenames and proper figure placement
• Augmented worked numericals — every taught method gets at least 3 solved examples (source + agent-constructed)
• Flashcards & cheatsheets as alternative outputs

Quick Start

One-shot (no install)

npx exam-notes-generator init

This scaffolds the skill into your current directory under .github/docs/skills/exam-notes-generator/.

Install locally (for repeated use)

npm install -g exam-notes-generator-skill
exam-notes-generator init ./my-course-project

What Gets Installed

.github/
  docs/
    copilot-instructions.md          # AI workspace routing rules
    prompts/
      exam-notes.prompt.md           # Slash-prompt template
    skills/
      exam-notes-generator/
        SKILL.md                      # Full skill specification (~1000 lines)
        references/
          flashcard-format.md         # Q&A card output format
          ocr-strategy.md             # Scanned PDF handling
          soffice-convert.md          # Legacy .doc/.ppt conversion
  reference-notes/
    (placeholder for reference .tex examples)

Skill Workflow (After Install)

1. Place your course materials in the project (e.g., subjects/ML 3/Module 1/, UNIT 1/)
2. Ask your AI assistant:
   • "Make complete study notes from subjects/ML 3/"
   • "Generate exam notes for UNIT 1 and UNIT 2"
   • "Create flashcards from these slides"
3. The skill automatically:
   • Inventories all source files
   • Extracts text, tables, and embedded images
   • Rasterizes every page and runs OCR
   • Classifies diagrams (architecture, flowcharts, plots, etc.)
   • Builds a numerical inventory and augments with extra examples
   • Produces a production-ready .tex file in /outputs/latex/

Key Design Principles

| Principle | Enforcement | |---|---| | Grounded only | Every fact must trace to a specific source file | | LaTeX-first | Output is compilable .tex, not markdown-in-disguise | | Diagrams are content | Every significant figure is extracted, OCR'd, renamed semantically, and embedded | | No compression | Source bullet hierarchy, sub-points, and caveats are preserved | | Numericals mandatory | Every method gets ≥3 worked examples (source + constructed) | | Semantic placement | Diagrams live in the same section as the theory they illustrate |

Example Invocation

After installing the skill into a project, trigger it via:

/exam-notes sourcePath=subjects/QPM/ outputType=notes depth=detailed

Or simply ask your AI:

"Generate complete study-source LaTeX for all units in this workspace."

Requirements

• Node.js >= 18 (for the CLI scaffold tool)
• The AI skill itself runs inside your agent environment (Windsurf, Copilot, etc.)
• Optional but recommended for full functionality: pymupdf, pdfplumber, python-pptx, python-docx, tesseract

License

MIT