Skip to content

bbylw/pdf

BranchesTags

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

12 Commits
pdf
pdf
 
 
README.md
README.md
 
 

Repository files navigation

PDF Skill

An Amp agent skill for high-quality PDF creation, form filling, and manipulation — using an HTML-first, print-first workflow.

Overview

This skill teaches Amp to produce premium, publication-grade PDFs by first designing pixel-perfect print-ready web pages, then rendering them to PDF via Playwright/Chromium. It also provides a complete toolkit for PDF form filling (both fillable and non-fillable forms), extraction, and manipulation.

When to Use

  • HTML → PDF (default): Premium / brand / presentation / magazine / report style; covers, TOC, chapter pages, cards, charts, complex tables; rapid visual iteration.
  • ReportLab: Minimal plain-text PDF or coordinate-based drawing.
  • Form filling scripts: Fill PDF forms (fillable or non-fillable).
  • pypdf / pdfplumber: Merge, split, extract PDFs.

Core Workflow

1. Content Layering   →  Cover / TOC / Body / Appendix
2. Page Skeleton      →  Unified header/footer, margins, grid
3. Design Tokens      →  Colors, fonts, spacing (4/8pt system)
4. Print-safe CSS     →  @page, break-*, print-color-adjust
5. Data-driven HTML   →  Template / Jinja rendering
6. Playwright Export  →  page.pdf(print_background=True)
7. QA Checklist       →  Pagination, truncation, alignment, color

Scripts

All scripts are located in the scripts/ directory:

Script Description
create_premium_pdf.py Generate multi-page premium HTML and export to PDF
convert_pdf_to_images.py Convert PDF pages to PNG images
check_fillable_fields.py Check if a PDF has fillable form fields
extract_form_field_info.py Extract fillable field metadata to JSON
fill_fillable_fields.py Fill fillable PDF form fields from JSON
extract_form_structure.py Extract text labels, lines, checkboxes from non-fillable PDFs
check_bounding_boxes.py Validate bounding boxes before filling
fill_pdf_form_with_annotations.py Fill non-fillable PDFs via text annotations
create_validation_image.py Create validation overlay images

Quick Start — Premium PDF

python scripts/create_premium_pdf.py \
  --company "Northstar Dynamics" \
  --month "March 2026" \
  --html report.html \
  --output report.pdf

Use --html-only to preview the HTML layout before exporting to PDF.

Quick Start — Form Filling

# 1. Check if the PDF has fillable fields
python scripts/check_fillable_fields.py form.pdf

# 2a. Fillable → extract fields, fill, and export
python scripts/extract_form_field_info.py form.pdf fields.json
python scripts/fill_fillable_fields.py form.pdf field_values.json output.pdf

# 2b. Non-fillable → extract structure, annotate, and export
python scripts/extract_form_structure.py form.pdf structure.json
python scripts/fill_pdf_form_with_annotations.py form.pdf fields.json output.pdf

# 3. Verify output
python scripts/convert_pdf_to_images.py output.pdf verify/

Design Principles

  • Print-first: Design for fixed page sizes (A4/Letter), not infinite scroll.
  • Design tokens: Unified colors, typography, and spacing across all pages.
  • Explicit pagination: Section breaks, avoid element splitting, segmented long tables.
  • Browser validation: Always verify in print preview before exporting.
  • Preserve HTML: Keep the HTML source alongside the PDF for easy iteration.

Default CSS Baseline

:root {
  --bg: #f6f8fb;
  --surface: #ffffff;
  --text: #1f2937;
  --muted: #6b7280;
  --brand: #2563eb;
  --radius: 12px;
}

@page { size: A4; margin: 14mm; }

html, body {
  -webkit-print-color-adjust: exact;
  print-color-adjust: exact;
}

.page { break-after: page; }
.card, .table-block, .chart-block { break-inside: avoid; }

Dependencies

Task Library
HTML → PDF export Playwright (Chromium)
Edit / merge / split / metadata pypdf
Text & table extraction pdfplumber
OCR pytesseract + pdf2image

File Structure

pdf/
├── SKILL.md          # Skill definition & instructions
├── reference.md      # Design reference & Playwright export baseline
├── forms.md          # Form filling workflow (fillable & non-fillable)
├── LICENSE.txt       # License
├── README.md         # This file
└── scripts/
    ├── create_premium_pdf.py
    ├── convert_pdf_to_images.py
    ├── check_fillable_fields.py
    ├── extract_form_field_info.py
    ├── extract_form_structure.py
    ├── fill_fillable_fields.py
    ├── fill_pdf_form_with_annotations.py
    ├── check_bounding_boxes.py
    └── create_validation_image.py

License

© 2025 Anthropic, PBC. All rights reserved. See LICENSE.txt for details.

PDF 技能

一个用于高质量 PDF 创建、表单填写和文档处理的 Amp 智能体技能——采用 HTML 优先、打印优先 的工作流。

概述

本技能让 Amp 通过先设计像素级精准的可打印网页，再经由 Playwright/Chromium 渲染为 PDF，从而生成出版级精美 PDF。同时提供完整的 PDF 表单填写（支持可填写和不可填写表单）、提取和处理工具集。

使用场景

  • HTML → PDF（默认）：精美 / 品牌感 / 演示级 / 杂志风 / 报告风；封面、目录、章节页、卡片、图表、复杂表格；快速迭代视觉风格或多次改版。
  • ReportLab：极简纯文本 PDF 或坐标绘制。
  • 表单填写脚本：填写 PDF 表单（可填写或不可填写）。
  • pypdf / pdfplumber：合并、拆分、提取 PDF。

核心工作流

1. 内容分层    →  封面 / 目录 / 正文章节 / 附录
2. 页面骨架    →  统一页眉页脚、页边距、栅格
3. 设计令牌    →  颜色、字体、间距（4/8pt 体系）
4. 打印安全样式 →  @page、break-*、print-color-adjust
5. 数据驱动渲染 →  模板 / Jinja / 字符串模板
6. Playwright 导出 →  page.pdf(print_background=True)
7. 质量验收    →  分页、截断、对齐、色彩、可读性

脚本工具

所有脚本位于 scripts/ 目录下：

脚本 描述
create_premium_pdf.py 生成多页精美 HTML 并导出 PDF
convert_pdf_to_images.py 将 PDF 页面转换为 PNG 图片
check_fillable_fields.py 检查 PDF 是否含有可填写表单字段
extract_form_field_info.py 提取可填写字段元数据为 JSON
fill_fillable_fields.py 根据 JSON 填写可填写表单字段
extract_form_structure.py 从不可填写 PDF 中提取文本标签、线条、复选框
check_bounding_boxes.py 填写前验证边界框
fill_pdf_form_with_annotations.py 通过文本注释填写不可填写 PDF
create_validation_image.py 创建验证叠加图像

快速上手 — 精美 PDF

python scripts/create_premium_pdf.py \
  --company "Northstar Dynamics" \
  --month "March 2026" \
  --html report.html \
  --output report.pdf

使用 --html-only 可先预览 HTML 排版效果，再导出 PDF。

快速上手 — 表单填写

# 1. 检查 PDF 是否有可填写字段
python scripts/check_fillable_fields.py form.pdf

# 2a. 可填写 → 提取字段、填写、导出
python scripts/extract_form_field_info.py form.pdf fields.json
python scripts/fill_fillable_fields.py form.pdf field_values.json output.pdf

# 2b. 不可填写 → 提取结构、注释填写、导出
python scripts/extract_form_structure.py form.pdf structure.json
python scripts/fill_pdf_form_with_annotations.py form.pdf fields.json output.pdf

# 3. 验证输出
python scripts/convert_pdf_to_images.py output.pdf verify/

设计原则

  • 打印优先：以固定页面尺寸（A4/Letter）为设计基准，而非无限滚动页面。
  • 设计令牌：所有页面统一颜色、字体和间距。
  • 显式分页：章节换页、避免元素拆分、长表格分段。
  • 浏览器验证：导出前始终在打印预览中验证。
  • 保留 HTML：保存 HTML 源文件，便于后续视觉迭代。

默认 CSS 基线

:root {
  --bg: #f6f8fb;
  --surface: #ffffff;
  --text: #1f2937;
  --muted: #6b7280;
  --brand: #2563eb;
  --radius: 12px;
}

@page { size: A4; margin: 14mm; }

html, body {
  -webkit-print-color-adjust: exact;
  print-color-adjust: exact;
}

.page { break-after: page; }
.card, .table-block, .chart-block { break-inside: avoid; }

依赖库

任务 库
HTML → PDF 导出 Playwright (Chromium)
编辑 / 合并 / 拆分 / 元数据 pypdf
文本与表格提取 pdfplumber
OCR 识别 pytesseract + pdf2image

目录结构

pdf/
├── SKILL.md          # 技能定义与指令
├── reference.md      # 设计参考与 Playwright 导出基线
├── forms.md          # 表单填写工作流（可填写与不可填写）
├── LICENSE.txt       # 许可证
├── README.md         # 本文件
└── scripts/
    ├── create_premium_pdf.py
    ├── convert_pdf_to_images.py
    ├── check_fillable_fields.py
    ├── extract_form_field_info.py
    ├── extract_form_structure.py
    ├── fill_fillable_fields.py
    ├── fill_pdf_form_with_annotations.py
    ├── check_bounding_boxes.py
    └── create_validation_image.py

许可证

© 2025 Anthropic, PBC. 保留所有权利。详见 LICENSE.txt。

About

PDF Skill

Resources

Readme
Activity

Stars

0 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases

No releases published

Packages

 
 
 

Contributors

Languages