pdf-extraction

name	pdf-extraction
description	Extract text from PDF documents efficiently using poppler's pdftotext. Use when reading academic papers, extracting metadata, or searching PDF content.
allowed-tools	Bash, Read
user-invocable	false

PDF Extraction Skill

Extract text from PDF documents efficiently using poppler's pdftotext utility.

Available Script

Use the extraction script at .claude/skills/pdf-extraction/scripts/pdf_extract.sh

Commands

Extract First N Pages (Default Strategy)

For metadata extraction (title, authors, abstract, venue info):

.claude/skills/pdf-extraction/scripts/pdf_extract.sh first "{pdf_path}" 3

Extract Specific Page

.claude/skills/pdf-extraction/scripts/pdf_extract.sh page "{pdf_path}" {page_num}

Extract Page Range

.claude/skills/pdf-extraction/scripts/pdf_extract.sh range "{pdf_path}" {start} {end}

Search for Text

Returns matching lines with line numbers:

.claude/skills/pdf-extraction/scripts/pdf_extract.sh search "{pdf_path}" "query"

Find Pages Containing Query

Identifies which pages contain specific content:

.claude/skills/pdf-extraction/scripts/pdf_extract.sh search-pages "{pdf_path}" "Method"

Extract Entire PDF (Use Sparingly)

Only for short papers (<8 pages):

.claude/skills/pdf-extraction/scripts/pdf_extract.sh all "{pdf_path}"

Strategic Extraction Patterns

For BibTeX Metadata

1. Extract first 2-3 pages (contains title, authors, abstract, venue)
2. If DOI/venue not found, search for specific terms
3. Only extract additional pages if critical info missing

For Paper Summaries

1. Extract first 3 pages for abstract + introduction
2. Use search-pages to locate key sections (Methods, Results, Limitations, Discussion, Conclusion)
3. Extract only the pages containing those sections
4. Target ~4-6 pages of content maximum

Token Efficiency Guidelines

• Never read entire PDFs for papers >8 pages
• Start with minimal extraction, expand only if needed
• Use search commands to locate specific content before extracting pages
• For batch processing: compact context after every 5 papers