pdf

name: pdf description: "Use this skill whenever the user wants to do anything with PDF files — reading, extracting, creating professional documents, filling forms." metadata: { "yiyi": { "emoji": "📄", "requires": {} }

}

PDF Processing

Reading & Extracting Text

Use Python's pypdf via run_python_script. If pypdf isn't installed, call pip_install(["pypdf"]) first.

# extract_pdf.py
import sys
from pypdf import PdfReader

reader = PdfReader(sys.argv[1])
for i, page in enumerate(reader.pages, 1):
    text = page.extract_text() or ""
    print(f"=== Page {i} ===")
    print(text)

Run with run_python_script(script_path="extract_pdf.py", args=["/path/to/document.pdf"]). Then summarize / analyze / answer questions about the captured stdout.

For Tables in PDFs

1. Extract raw text with the script above.
2. Parse the tabular structure from the text output (look for repeating column-aligned rows).
3. If the user wants it as a spreadsheet, write a Python script that uses openpyxl to emit .xlsx (see the xlsx skill).

Creating Professional PDFs

Use the create_pdf.py script to generate beautiful, content-rich PDF documents.

Step 1: Build a JSON structure file

Create a JSON file describing the document content. Be thorough and detailed — include rich content, not just bullet points.

{
  "title": "Document Title",
  "subtitle": "A detailed subtitle explaining the document purpose",
  "author": "Author Name",
  "date": "2024-01-15",
  "header": "Document Title — Confidential",
  "footer": "© 2024 Company Name",
  "theme": "professional",
  "page_size": "A4",
  "body": [
    { "type": "toc" },
    { "type": "heading", "level": 1, "text": "Introduction" },
    { "type": "paragraph", "text": "Write full, detailed paragraphs here. Avoid short one-liners. Each paragraph should be 3-5 sentences with substantive content that provides real value to the reader." },
    ...
  ]
}

Step 2: Run the script

python3 scripts/create_pdf.py structure.json output.pdf

Available Themes

Content Block Types

Headings (3 levels)

{ "type": "heading", "level": 1, "text": "Major Section" }
{ "type": "heading", "level": 2, "text": "Subsection" }
{ "type": "heading", "level": 3, "text": "Sub-subsection" }

Paragraph

{ "type": "paragraph", "text": "Full paragraph text. Write detailed, multi-sentence content.", "indent": false }

Blockquote

{ "type": "quote", "text": "Important insight or notable quote from a source." }

Bullet / Numbered List

{ "type": "list", "style": "bullet", "items": ["First point with explanation", "Second point with detail"] }
{ "type": "list", "style": "number", "items": ["Step one", "Step two"], "start": 1 }

Table

{ "type": "table", "headers": ["Name", "Role", "Department"], "rows": [["Alice", "Engineer", "R&D"], ["Bob", "Designer", "Product"]] }

Code Block

{ "type": "code", "language": "python", "text": "def hello():\n    print('Hello, World!')" }

Key-Value Pairs (clean metadata display)

{ "type": "key_value", "items": [{"key": "Project", "value": "Phoenix"}, {"key": "Status", "value": "Active"}] }

Image

{ "type": "image", "path": "/absolute/path/to/image.png", "width": 150, "caption": "Figure 1: Architecture diagram" }

Layout Controls

{ "type": "divider" }
{ "type": "spacer", "height": 10 }
{ "type": "page_break" }
{ "type": "toc" }

Content Writing Guidelines

When generating PDF content, follow these rules to produce professional documents:

1. Be thorough: Each section should have multiple paragraphs, not just one sentence. Expand on ideas, provide context, and include supporting details.

2. Structure deeply: Use all three heading levels. A good document has:

   • Level 1: Major sections (3-6 per document)
   • Level 2: Subsections (2-4 per major section)
   • Level 3: Specific topics (as needed)

3. Mix content types: Don't just use paragraphs. Include:

   • Tables for comparative data
   • Lists for actionable items or enumerations
   • Quotes for key insights or citations
   • Code blocks for technical content
   • Key-value pairs for metadata/specs

4. Write full paragraphs: Each paragraph should be 3-5 sentences. Avoid one-liners.

5. Always include:

   • Table of Contents ({"type": "toc"} as first body item)
   • Cover page info (title, subtitle, author, date)
   • Header and footer text
   • Proper section numbering in headings

6. Chinese content: The script auto-detects and uses CJK fonts (PingFang on macOS, Noto Sans CJK on Linux, Microsoft YaHei on Windows).

Example: Complete Report Structure

{
  "title": "2024 年度技术报告",
  "subtitle": "人工智能与机器学习应用进展分析",
  "author": "技术部",
  "date": "2024-12-01",
  "header": "2024 年度技术报告",
  "footer": "© 2024 公司名称 · 机密文件",
  "theme": "professional",
  "body": [
    { "type": "toc" },

    { "type": "heading", "level": 1, "text": "1. 执行摘要" },
    { "type": "paragraph", "text": "本报告全面回顾了 2024 年度公司在人工智能和机器学习领域的技术应用进展。报告涵盖了关键项目里程碑、技术架构演进、团队能力建设以及未来发展规划四个核心维度。" },
    { "type": "paragraph", "text": "在过去一年中，我们成功部署了 12 个 AI 驱动的生产系统，处理效率平均提升 43%，客户满意度提高至 94.2%。这些成果的取得离不开团队的不懈努力和技术路线的正确选择。" },
    { "type": "key_value", "items": [
      {"key": "报告周期", "value": "2024 年 1 月 — 12 月"},
      {"key": "项目总数", "value": "12 个生产系统"},
      {"key": "效率提升", "value": "平均 43%"},
      {"key": "客户满意度", "value": "94.2%"}
    ]},

    { "type": "heading", "level": 1, "text": "2. 核心项目进展" },
    { "type": "heading", "level": 2, "text": "2.1 智能客服系统" },
    { "type": "paragraph", "text": "智能客服系统于 2024 年 3 月正式上线，基于大语言模型构建的对话引擎能够处理 85% 的常见客户咨询。系统采用 RAG（检索增强生成）架构，结合企业知识库实现精准问答。" },
    { "type": "paragraph", "text": "截至年末，系统累计处理对话 230 万轮次，平均响应时间 1.2 秒，首次解决率达 78%。与人工客服相比，处理成本降低 62%，同时客户满意度保持在 91% 以上。" },

    { "type": "heading", "level": 2, "text": "2.2 数据分析平台" },
    { "type": "paragraph", "text": "数据分析平台完成了从传统 BI 工具到 AI 驱动分析的升级转型。新平台支持自然语言查询，用户只需描述分析需求，系统即可自动生成可视化报表。" },
    { "type": "table", "headers": ["指标", "升级前", "升级后", "提升幅度"], "rows": [
      ["查询响应时间", "45 秒", "3 秒", "93%"],
      ["日活跃用户", "120", "580", "383%"],
      ["报表生成耗时", "2 小时", "10 分钟", "92%"]
    ]},

    { "type": "heading", "level": 1, "text": "3. 技术架构" },
    { "type": "paragraph", "text": "我们的技术架构遵循微服务设计原则，核心组件包括模型服务层、数据管道、特征工程平台和监控告警系统。各组件之间通过消息队列实现松耦合通信。" },
    { "type": "quote", "text": "好的架构不是一次设计出来的，而是在持续迭代中逐步演进的。我们的架构每季度进行一次评审和优化。" },
    { "type": "list", "style": "bullet", "items": [
      "模型服务层：支持 TensorRT 加速推理，P99 延迟 < 50ms",
      "数据管道：基于 Apache Kafka 的实时流处理，日吞吐量 10 亿条记录",
      "特征工程平台：自动化特征提取和存储，支持在线和离线两种模式",
      "监控告警系统：全链路追踪，异常检测准确率 96%"
    ]},

    { "type": "heading", "level": 1, "text": "4. 未来展望" },
    { "type": "paragraph", "text": "展望 2025 年，我们将重点推进多模态 AI 应用、边缘计算部署和自动化 MLOps 三大方向。预计投入预算 1200 万元，新增技术岗位 15 个。" },
    { "type": "list", "style": "number", "items": [
      "Q1：完成多模态模型评估和选型",
      "Q2：边缘设备适配和性能优化",
      "Q3：MLOps 平台 2.0 上线",
      "Q4：全面评估与规划下一年度"
    ]}
  ]
}

Advanced Operations

Merge / Split / Rotate PDFs (Python pypdf)

python3 -c "
from pypdf import PdfReader, PdfWriter
# Merge
writer = PdfWriter()
for f in ['a.pdf', 'b.pdf']:
    writer.append(f)
writer.write('merged.pdf')
"

OCR Scanned PDFs

Requires: pytesseract, pdf2image

python3 -c "
from pdf2image import convert_from_path
import pytesseract
images = convert_from_path('scanned.pdf')
text = '\n'.join(pytesseract.image_to_string(img, lang='chi_sim+eng') for img in images)
print(text)
"

Fill PDF Forms

See forms.md for detailed form-filling workflow (both fillable and non-fillable forms).

Quick Reference