pdf-reader

name: pdf-reader metadata: version: "0.0.1" author: "ShuangShu" description: Read and analyze PDF files, extract text by page range, and produce structured summaries with citations. Use when a user asks to read, summarize, compare, or QA against PDF documents, including long reports, papers, contracts, manuals, and scanned PDFs.

PDF Reader

Overview

Use this skill to read PDF content safely and reproducibly, then return concise answers with page-level evidence.

Quick Workflow

1. Confirm target files and scope.
2. Extract text from the PDF (full file or page range).
3. Clean obvious noise (headers/footers, broken line wraps) without changing meaning.
4. Produce structured output with page citations.
5. If extraction quality is poor, switch to OCR-capable tooling and state limitations.

Extraction Strategy

Use the fastest available method first, then fallback:

1. pdftotext (if installed) for quick extraction:

pdftotext -layout input.pdf -

2. Python pypdf for controlled page-by-page extraction:

from pypdf import PdfReader
reader = PdfReader("input.pdf")
for i, page in enumerate(reader.pages, start=1):
    text = page.extract_text() or ""
    print(f"\\n--- Page {i} ---\\n{text}")

3. If text is mostly empty, treat as scanned PDF and use OCR workflow (for example ocrmypdf + extraction), then clearly mark OCR uncertainty.

Output Contract

Always return:

• 结论: 直接回答用户问题。
• 依据: 关键证据点，附页码（如 p.12）。
• 不确定性: 缺失页、识别错误、扫描噪声等风险。
• 后续建议: 仅在确有必要时给 1-2 条下一步。

Long PDF Handling

• Read in chunks (for example 20-50 pages each pass).
• Keep a running note of entities, terms, and contradictions.
• When asked to compare sections, cite both page positions explicitly.
• Do not claim facts that cannot be located in extracted text.

Safety and Quality

• Preserve meaning; do not "fix" numbers, dates, or legal terms without evidence.
• Keep quotations short and exact when required.
• If page numbering in document differs from PDF index, explain mapping explicitly.