/document-extract

Extract and analyse content from scanned documents, PDFs, and document images using Sonnet sub-agents.

Usage

code

/document-extract <file-path>
/document-extract +Attachments/meeting-whiteboard.jpg
/document-extract +Attachments/spec-document.pdf
/document-extract +Attachments/handwritten-notes.png --type "meeting notes"

Instructions

This skill uses Sonnet model sub-agents for comprehensive document analysis and text extraction.

Phase 1: Document Loading

•Verify the file exists at the specified path
•Identify document type (PDF, image, photo of document)
•Note any context about document purpose

Phase 2: Comprehensive Analysis (Sonnet Sub-Agents)

Launch these sub-agents using model: "sonnet":

Agent 1: Text Extraction (Sonnet)

code

Task: Extract all text content from the document
- Read the file using the Read tool
- Perform comprehensive OCR on all visible text
- Preserve structure (headings, paragraphs, lists)
- Handle multiple columns if present
- Extract text from tables maintaining structure
- Note any text that is unclear or uncertain
Return: Complete text extraction with structure preserved

Agent 2: Structure Analysis (Sonnet)

code

Task: Analyse document structure and formatting
- Read the file
- Identify document type (letter, form, spec, notes, etc.)
- Map section hierarchy
- Identify tables, lists, and special formatting
- Note headers, footers, page numbers
- Identify logos, stamps, signatures
Return: Document structure map

Agent 3: Content Classification (Sonnet)

code

Task: Classify and categorise the content
- Read the file
- Determine document purpose
- Identify key entities (people, projects, dates, systems)
- Extract action items or tasks
- Find decisions or commitments
- Note any deadlines or dates mentioned
- Identify references to YourOrg projects or systems
Return: Classified content with entity extraction

Agent 4: Quality Assessment (Sonnet)

code

Task: Assess extraction quality and completeness
- Read the file
- Evaluate image/scan quality
- Identify areas with low confidence extraction
- Note any missing or obscured content
- Assess if re-scan might be needed
- Check for multiple pages
Return: Quality report with confidence scores

Phase 3: Compile Report

markdown

# Document Extraction

**Source**: {{filename}}
**Extracted**: {{DATE}}
**Document Type**: {{type}}
**Quality Score**: {{High/Medium/Low}}

## Summary

{{Brief description of document content and purpose}}

## Extracted Content

### Full Text

{{complete extracted text with formatting preserved}}

---

## Structured Data

### Key Information
| Field | Value |
|-------|-------|
| Date | {{if found}} |
| Author | {{if found}} |
| Subject | {{if found}} |
| Reference | {{if found}} |

### People Mentioned
{{list of names with context}}

### Dates & Deadlines
| Date | Context |
|------|---------|
{{dates found}}

### Projects/Systems Referenced
{{links to matching vault notes}}

## Tables Extracted

{{any tables found, formatted as markdown}}

## Action Items Found

- [ ] {{action items extracted from document}}

## Decisions/Commitments

{{any decisions or commitments mentioned}}

## Extraction Notes

### Confidence Assessment
- Overall quality: {{assessment}}
- Unclear sections: {{list any problematic areas}}
- Recommendations: {{re-scan suggestions if needed}}

### Areas of Uncertainty
{{text that couldn't be confidently extracted}}

## Suggested Actions

1. **Create Note**: {{suggest note type and title}}
2. **Link to**: {{suggest related notes}}
3. **Follow up**: {{any actions needed}}

Integration

After extraction, offer to:

•Create a new vault note from the content
•Append to an existing meeting or project note
•Create tasks from action items found

Notes

•Works with photos of whiteboards, handwritten notes, printed documents
•PDF support for text extraction
•Can recognise organization-specific terminology and systems