pdf

用于处理所有 PDF 相关任务，包括读取、创建、编辑、转换、内容提取、页面处理、表单填写、扫描件解析、PDF转化word文件和PDF转化飞书云文档。用户提供、提及或要求生成 PDF 时使用。

Files21

View on GitHub

SKILL.md

GitHub

PDF Processing Guide

Overview

This guide covers essential PDF processing operations using bundled Python libraries and scripts. For editing an existing PDF, read editing.md. For advanced features, JavaScript libraries, and detailed examples, see reference.md. If you need to fill out a PDF form, read forms.md and follow its instructions.

Rules

  • For editing scenarios (such as document translation or editing), prioritize modifying the original PDF copy.
  • For tasks where layout and visuals matter (such as reading, creating, editing or translating reports, invoices, slides, figures, paperwork etc.) or for filling and validating interactive PDF forms: 1.Prefer visual review: render PDF pages to PNGs and inspect them. 2.After each meaningful update, re-render pages and verify alignment, spacing, and legibility.
  • Prefer PyMuPDF for existing-PDF operations. Use pypdf only for canonical AcroForm handling, encryption/repair edge cases, or operations PyMuPDF cannot perform reliably; use pdfplumber only as a table/edge-case fallback.
  • Use the bundled Python libraries and scripts for all default workflows. Do not invoke a system CLI when PyMuPDF, pypdf, pdfplumber, reportlab, or Pillow can perform the operation.
  • When extracting text, check page count, empty pages, repeated-page hashes, replacement characters, and a visual sample. Do not assume a larger character count is more accurate.
  • Install the locked Python dependencies with python -m pip install -r requirements.txt.
  • If the user has no explicit file format requirements, keep the same file extension as the source file. If the user explicitly specifies an output format, follow it.
  • Do not convert a PDF to Word or a Lark/Feishu document merely because the task involves substantial content editing, reformatting, removing images, changing layout, or producing editable intermediate content. The complexity of the edit does not change the default deliverable.
  • Route to PDF-to-Word only when the user explicitly requests Word, DOCX, or a Word-editable deliverable, or provides an editable Word file as the document to modify. Generic references to a “document,” “report,” “file,” “proposal,” or “editable file” do not by themselves mean Word. When routed, follow pdf-to-word.md. This workflow owns the complete PDF analysis, DOCX reconstruction, structural audit, and visual QA process.
  • Route to PDF-to-Doc only when the user explicitly requests a Lark/Feishu cloud document, Lark/Feishu Doc, Wiki, or provides an existing Lark/Feishu document as the target to modify. Do not interpret the generic words “doc” or “document” as a Lark/Feishu document without clear platform context. When routed, follow pdf-to-doc.md. This workflow owns the complete PDF analysis, Lark XML generation, batched append with fetch readback, and visual QA process.
  • Route to PDF Translation only when the user explicitly requests translating a PDF's text content to another language while preserving the original layout as a PDF. Same-script variants (Traditional ↔ Simplified Chinese, en-US ↔ en-GB, etc.) run inline as a short single-agent path; cross-language translation (English ↔ Chinese, CJK ↔ Latin, etc.) must launch a multi-agent workflow via create_agent. When routed, follow references/workflows/translation.md.

Quick Start

import pymupdf

with pymupdf.open("document.pdf") as document:
    print(f"Pages: {document.page_count}")
    text = "\n\n".join(
        page.get_text("text", sort=True) for page in document
    )

For a reusable command-line path, run python scripts/extract_pdf_text.py input.pdf output.txt. Add --structured to write page text, blocks, words, and coordinates as JSON.

Python Libraries

PyMuPDF - Default for Existing PDFs

Use PyMuPDF (pymupdf) first for reading, text extraction, rendering, page manipulation, coordinates, images, annotations, and inspection. It respects visible page boundaries on PDFs that reuse a larger shared content stream and usually returns cleaner Unicode than pypdf, pdfplumber, or pdfminer.six defaults.

Always use sort=True for human reading order. For multi-column or card layouts, preserve coordinates with "blocks", "words", or "dict" and reconstruct the semantic order instead of flattening blindly.

Extract Text and Coordinates

import pymupdf

with pymupdf.open("input.pdf") as document:
    for page in document:
        text = page.get_text("text", sort=True)
        blocks = page.get_text("blocks", sort=True)
        words = page.get_text("words", sort=True)

Render Pages

import pymupdf

with pymupdf.open("input.pdf") as document:
    matrix = pymupdf.Matrix(2, 2)
    for index, page in enumerate(document, start=1):
        pixmap = page.get_pixmap(matrix=matrix, alpha=False)
        pixmap.save(f"page_{index}.png")

Use a 2x matrix as a practical preview default. Increase it when small text or fine visual details require higher resolution.

Merge PDFs

import pymupdf

output = pymupdf.open()
for pdf_path in ["doc1.pdf", "doc2.pdf", "doc3.pdf"]:
    with pymupdf.open(pdf_path) as source:
        output.insert_pdf(source)
output.save("merged.pdf", garbage=4, deflate=True)
output.close()

Split PDF

import pymupdf

with pymupdf.open("input.pdf") as source:
    for index in range(source.page_count):
        output = pymupdf.open()
        output.insert_pdf(source, from_page=index, to_page=index)
        output.save(f"page_{index + 1}.pdf", garbage=4, deflate=True)
        output.close()

Extract Metadata

import pymupdf

with pymupdf.open("document.pdf") as document:
    metadata = document.metadata
    print(metadata.get("title"))
    print(metadata.get("author"))

Rotate Pages

import pymupdf

with pymupdf.open("input.pdf") as document:
    page = document[0]
    page.set_rotation((page.rotation + 90) % 360)
    document.save("rotated.pdf", garbage=4, deflate=True)

Extract Tables

Try PyMuPDF first. Use pdfplumber only as a fallback when its table strategies are needed.

import pymupdf

with pymupdf.open("document.pdf") as document:
    for page in document:
        finder = page.find_tables()
        for table in finder.tables:
            print(table.extract())

pypdf - AcroForms and Low-Level PDF Operations

Keep pypdf for canonical AcroForm field-tree inspection/filling, encryption cases not covered by the current PyMuPDF workflow, and low-level object repair. Follow forms.md for forms. Do not use pypdf.extract_text() as the default text extractor.

reportlab - Create PDFs

Basic PDF Creation

from reportlab.lib.pagesizes import letter
from reportlab.pdfgen import canvas

c = canvas.Canvas("hello.pdf", pagesize=letter)
width, height = letter

# Add text
c.drawString(100, height - 100, "Hello World!")
c.drawString(100, height - 120, "This is a PDF created with reportlab")

# Add a line
c.line(100, height - 140, 400, height - 140)

# Save
c.save()

Create PDF with Multiple Pages

from reportlab.lib.pagesizes import letter
from reportlab.platypus import SimpleDocTemplate, Paragraph, Spacer, PageBreak
from reportlab.lib.styles import getSampleStyleSheet

doc = SimpleDocTemplate("report.pdf", pagesize=letter)
styles = getSampleStyleSheet()
story = []

# Add content
title = Paragraph("Report Title", styles['Title'])
story.append(title)
story.append(Spacer(1, 12))

body = Paragraph("This is the body of the report. " * 20, styles['Normal'])
story.append(body)
story.append(PageBreak())

# Page 2
story.append(Paragraph("Page 2", styles['Heading1']))
story.append(Paragraph("Content for page 2", styles['Normal']))

# Build PDF
doc.build(story)

Subscripts and Superscripts

IMPORTANT: Never use Unicode subscript/superscript characters (₀₁₂₃₄₅₆₇₈₉, ⁰¹²³⁴⁵⁶⁷⁸⁹) in ReportLab PDFs. The built-in fonts do not include these glyphs, causing them to render as solid black boxes.

Instead, use ReportLab's XML markup tags in Paragraph objects:

from reportlab.platypus import Paragraph
from reportlab.lib.styles import getSampleStyleSheet

styles = getSampleStyleSheet()

# Subscripts: use <sub> tag
chemical = Paragraph("H<sub>2</sub>O", styles['Normal'])

# Superscripts: use <super> tag
squared = Paragraph("x<super>2</super> + y<super>2</super>", styles['Normal'])

For canvas-drawn text (not Paragraph objects), manually adjust font the size and position rather than using Unicode subscripts/superscripts.

Common Tasks

Read Scanned PDFs with Multimodal Vision

Render the pages with python scripts/convert_pdf_to_images.py scanned.pdf pages/, then inspect the PNG files directly with the multimodal model. Preserve page order and page boundaries when transcribing or summarizing visible text. Render at higher resolution when small or faint text is not legible.

This workflow supports reading and layout understanding but does not create a searchable text layer or deterministic word-level coordinates. If the user requires either, explain that an external OCR backend outside this skill is needed.

Add Watermark

import pymupdf

with pymupdf.open("document.pdf") as document, pymupdf.open("watermark.pdf") as watermark:
    for page in document:
        page.show_pdf_page(page.rect, watermark, 0, overlay=True)
    document.save("watermarked.pdf", garbage=4, deflate=True)

Extract Images

from pathlib import Path

import pymupdf

output_dir = Path("images")
output_dir.mkdir(exist_ok=True)

with pymupdf.open("input.pdf") as document:
    seen = set()
    for page_number, page in enumerate(document, start=1):
        for image_number, image in enumerate(page.get_images(full=True), start=1):
            xref = image[0]
            if xref in seen:
                continue
            seen.add(xref)
            data = document.extract_image(xref)
            path = output_dir / f"p{page_number}_img{image_number}.{data['ext']}"
            path.write_bytes(data["image"])

Password Protection

import pymupdf

permissions = int(
    pymupdf.PDF_PERM_ACCESSIBILITY
    | pymupdf.PDF_PERM_PRINT
    | pymupdf.PDF_PERM_COPY
)
with pymupdf.open("input.pdf") as document:
    document.save(
        "encrypted.pdf",
        encryption=pymupdf.PDF_ENCRYPT_AES_256,
        owner_pw="ownerpassword",
        user_pw="userpassword",
        permissions=permissions,
    )

Quick Reference

Task Best Tool Command/Code
Render pages to images PyMuPDF page.get_pixmap(matrix=...)
Merge PDFs PyMuPDF output.insert_pdf(source)
Split PDFs PyMuPDF insert_pdf(..., from_page=i, to_page=i)
Extract text PyMuPDF page.get_text("text", sort=True)
Edit existing PDF content PyMuPDF Follow editing.md
Extract tables PyMuPDF, then pdfplumber fallback page.find_tables()
Extract images PyMuPDF document.extract_image(xref)
Add annotations/watermarks PyMuPDF page.add_*_annot() / show_pdf_page()
Create PDFs reportlab Canvas or Platypus
Read scanned PDFs PyMuPDF + multimodal vision Render pages, then inspect the images
Fill PDF forms pypdf (see forms.md) Keep canonical field-tree handling
PDF to Word pipeline pdf-to-word.md Parse PDF, rebuild editable DOCX, structure audit and page-by-page visual QA
PDF to Feishu Doc pipeline pdf-to-doc.md Parse PDF, generate Lark XML, batched append with fetch readback and page-by-page visual QA
PDF translation (route by script) translation.md Same-script → direct substitute; cross-language → protect highlight fills and image positions

Next Steps

  • For PyMuPDF coordinate extraction and advanced processing, see reference.md
  • For JavaScript libraries (pdf-lib), see reference.md
  • If you need to fill out a PDF form, follow the instructions in forms.md
  • For troubleshooting guides, see reference.md

Related in documents

applying-brand-guidelines-anthropics-claude-cookbooks

This skill applies consistent corporate branding and styling to all generated documents including colors, fonts, layouts...

by anthropics
31,757

docx

Comprehensive document creation, editing, and analysis with support for tracked changes, comments, formatting preservati...

docx
by anthropics
24,530

pdf

Comprehensive PDF manipulation toolkit for extracting text and tables, creating new PDFs, merging/splitting documents, a...

pdf
by anthropics
24,530