pdf

name	pdf
description	Read, extract, and analyze PDF documents. Triggers: pdf, read pdf, extract from pdf, analyze pdf, .pdf file path.
effort	medium
context	inline

/pdf — PDF Processing

Extract and analyze PDF documents. Always respond in the user's language.

Algorithm

1. Determine Input

• $ARGUMENTS = path to PDF → go to step 2
• $ARGUMENTS = empty or "find" → search for PDFs first:

  Glob "**/*.pdf"
  

  List found PDFs with file sizes. Ask which to process.

2. Read PDF

Read file.pdf                  # small PDFs (<10 pages)
Read file.pdf pages="1-5"      # large PDFs, chunk by 5-20 pages

Important: Maximum 20 pages per Read request.

3. Analyze Structure

Report immediately:

• Total page count
• Document structure (sections, headings)
• Content type (technical doc, requirements, API spec, report, legal)

4. Process Content

Based on content type, extract:

Content Type	Extract
Technical doc	Key concepts, architecture decisions, API endpoints
Requirements	User stories, acceptance criteria, constraints
API spec	Endpoints, request/response schemas, auth
Report	Findings, metrics, recommendations
Legal/Contract	Key terms, dates, obligations

5. Present and Suggest Next Steps

Based on content type, suggest:

• Technical doc → "Create reference file? Save to project memory?"
• Requirements → "Create tasks with /plan?"
• API spec → "Generate types/interfaces?"
• Report → "Summarize findings? Extract data tables?"
• Tabular data → "Export with /excel?"

Operations

Command	Action
/pdf file.pdf	Read and summarize
/pdf file.pdf 1-5	Read specific pages
/pdf analyze file.pdf	Deep analysis with extraction

Rules

• Always report page count and document structure first
• For large PDFs (>10 pages), ask before reading all pages
• Preserve table structures when extracting data
• Offer to convert to markdown for easier reference
• Suggest next: /plan for requirements, /excel for tabular data, /docs for technical content

Gotchas

• Max 20 pages per Read request — always chunk large PDFs. Reading 100 pages at once will fail.
• Scanned PDFs (images, not text) return empty content. The Read tool handles OCR for images but quality varies.
• Tables in PDFs often lose structure — offer to reconstruct as markdown or export via /excel.