knowledge-digest

Files (4)

• SKILL.md
• _meta.json
• prompt_context.md
• skill.toml

SKILL.md (first lines)

# KnowledgeDigest — Unified Learning Content Converter

## Overview

KnowledgeDigest converts textbooks, PDFs, or topic descriptions into personalized, multimodal learning experiences. It analyzes source content, then generates any combination of: handwritten-style notes (PDF), interactive quiz webpages (HTML), slides (PDF+PPTX), mind maps (image+Mermaid), and audio courses (MP3). All output is adapted to the learner's grade level and interests.

## Workflow

### Phase 1: Gather User Input

1. Identify what the user has provided:
   - Uploaded PDF/textbook file (optional)
   - Topic/direction description
   - Grade level (elementary / middle school / high school / university / professional)
   - Expected output format(s)

2. If no PDF/textbook uploaded and no source materials specified (only topic/direction provided):
   - Ask user:
     - **Option A:** "I have materials, uploading now"
     - **Option B:** "No materials, please search and generate courseware about [topic]"
   - If user selects B:
     - Use search tools to collect authoritative materials on the topic
     - Organize into structured content, generate a basic courseware PDF
     - Send PDF to user for confirmation: "This is the basic material I compiled for [topic], please confirm if usable?"
     - Continue after user confirmation

3. **Default output formats** (if user does not specify): mindmap + slides (PDF only) + quiz

### Phase 2: Content Analysis

Parse the PDF or structured content to extract:

**Document Parsing:**
- Identify chapter structure (chapters, sections, subsections)
- Extract heading hierarchy and table of contents
- Identify body text, images, tables, formulas, and other elements

**Core Concept Extraction:**
- Identify core concepts and key terms in each chapter
- Extract definitions, theorems, formulas, and important content
- Mark difficult points and key knowledge

**Learning Objective Analysis:**
- Infer learning objectives for each chapter
- Identify prerequisite knowledge requirements
- Analyze dependencies between knowledge points

**Output structured analysis results in this format:**

```json
{
  "document_info": {
    "title": "Document title",
    "total_pages": 100,
    "language": "zh/en",
    "subject": "Subject area"
  },
  "chapters": [
    {
      "chapter_id": "1",
      "title": "Chapter title",
      "page_range": [1, 20],
      "sections": [
        {
          "section_id": "1.1",
          "title": "Section title",
          "core_concepts": ["Concept 1", "Concept 2"],
          "key_terms": [
            {"term": "Term", "definition": "Definition"}
          ],
          "learning_objectives": ["Objective 1", "Objective 2"],
          "difficulty": "easy/medium/hard",
          "prerequisites": ["Prerequisite knowledge"]
        }
      ]
    }
  ],
  "knowledge_graph": {
    "nodes": ["Concept node list"],
    "edges": [{"from": "Concept A", "to": "Concept B", "relation": "depends/contains/related"}]
  }
}
```

**Parsing Rules:**
1. Chapter Recognition — Identify hierarchy based on font size, bold, numbering, etc. Handle documents without clear chapter markers by logically segmenting.
2. Concept Extraction — Identify bolded, highlighted, boxed important content. Extract proper nouns and term definitions. Identify formulas and theorems.
3. Difficulty Assessment — Assess based on concept abstraction level, prerequisite knowledge, and content complexity.
4. Quality Assurance — Ensure all chapters identified, verify knowledge point coverage completeness, check accuracy of concept definitions.

### Phase 3: Generate Requested Formats

Based on user-selected output formats, generate each in sequence. For each format, follow the corresponding section below.

### Phase 4: Deliver Assets

After all generation is complete:
- Only return file paths, no previews allowed
- No inline display of images/PDFs/audio/video in conversation
- Audio/video files must not auto-play

Present to user using deliver_assets format:
```
<deliver_assets>
<item>
<path>file path</path>
</item>
</deliver_assets>
```

---

## Supported Output Formats

| Format | Output | Description |
|--------|--------|-------------|
| `notes` | `{topic}_notes.pdf` | Handwritten-style notes (annotated on original or generated from scratch) |
| `quiz` | `{topic}_quiz.html` | Minimalist interactive HTML quiz with instant feedback |
| `slides` | `{topic}_slides.pdf` + `{topic}_slides.pptx` | Visual slides |

From badmarsh/librefang, MIT.

More from badmarsh/librefang