Local PDF Reader

Local PDF Reader

Overview

This skill handles local PDF-reading tasks from an absolute file path. It supports three common cases:

• Read the full text of a PDF
• Read one page or a page range
• Extract highlighted text or titles from PDF highlight annotations

Use the bundled script first when the environment lacks pdftotext, Python PDF packages, or OCR tools.

Workflow

1. Confirm the file exists and note the page count:

file "/abs/path/to/file.pdf"
bash scripts/read_local_pdf.sh pages "/abs/path/to/file.pdf"

2. Choose the extraction mode:

• Full document text:

bash scripts/read_local_pdf.sh text "/abs/path/to/file.pdf"

• Specific page:

bash scripts/read_local_pdf.sh text "/abs/path/to/file.pdf" 12

• Page range:

bash scripts/read_local_pdf.sh text "/abs/path/to/file.pdf" 12 15

• Highlighted text:

bash scripts/read_local_pdf.sh highlights "/abs/path/to/file.pdf"

3. Post-process for the user's intent:

• If the user asked what is highlighted, report the extracted highlight text grouped by page.
• If the user asked for titles, keep only the title-like lines instead of dumping all text.
• If the user asked for a summary, read the relevant page text first, then summarize.

Heuristics

• Prefer highlights when the user mentions "标黄", "highlight", "annotations", or asks which titles were highlighted.
• Prefer text when the user asks to read page content, extract a section, or summarize the PDF.
• If highlight extraction returns nothing, say that the PDF may not contain saved highlight annotations. Then fall back to page text extraction.
• The bundled script reads text-based PDFs. For scanned PDFs, you may need OCR; this skill does not bundle OCR.

Output Discipline

• Keep page numbers in the result when multiple pages are involved.
• Do not dump the whole document unless the user asked for it.
• When titles are clearly identifiable, return just the titles first and mention that they were extracted from PDF highlights or page text.

Bundled Script

The script compiles a temporary Objective-C helper against macOS PDFKit and supports:

• pages
• text
• highlights

Run it from the skill directory or reference it by absolute path.