kaimingwan/oh-my-kiro/omk-pdf-gen

PDF Generation Skill

Trigger Examples

• "帮我生成一个报价 PDF"
• "把这个内容导出成 PDF"
• "PDF 中文乱码怎么办"
• "generate a comparison PDF for this customer"
• "create a budget proposal PDF"

Approach: HTML + weasyprint

Use HTML + CSS → weasyprint pipeline. Do NOT use reportlab for documents containing CJK text.

Why:

• reportlab CID fonts (STSong-Light) have incomplete glyph coverage — symbols like • render as garbage (e.g. "煉")
• reportlab cannot read macOS SIP-protected TTC fonts (PingFang etc.)
• reportlab canvas API requires manual coordinate positioning — fragile and ugly
• weasyprint uses system fontconfig — correct font resolution out of the box

Workflow

Step 1: Write HTML string in Python

HTML = """\
<!DOCTYPE html>
<html lang="zh">
<head><meta charset="utf-8">
<style>
@page { size: A4; margin: 2cm 2.2cm; }
body { font-family: "Hiragino Sans GB", "Heiti SC", "Noto Sans CJK SC", sans-serif; ... }
</style></head>
<body>...</body></html>
"""

Step 2: Write to temp HTML, convert, clean up

import subprocess, os
html_tmp = output_path.replace('.pdf', '.html')
with open(html_tmp, 'w', encoding='utf-8') as f:
    f.write(HTML)
subprocess.run(['weasyprint', html_tmp, output_path], check=True)
os.remove(html_tmp)

Font Rules (Critical)

Never use these in weasyprint CSS:

• -apple-system — weasyprint doesn't understand Apple system font aliases
• "PingFang SC" — fontconfig often can't resolve it even though macOS has it
• "STSong-Light" — CID font name, not a real font family for CSS

Before generating, verify CJK fonts are available:

fc-list :lang=zh family | head -10

Styling Best Practices

Use standard HTML elements — weasyprint handles them well:

• <table> with CSS border-collapse: collapse for data tables
• <ul>/<li> for bullet lists (no manual bullet symbols needed)
• <h1>-<h3> for headings
• <hr> for dividers
• CSS @page for margins and page size

Recommended CSS skeleton

@page { size: A4; margin: 2cm 2.2cm; }
body { font-family: "Hiragino Sans GB", "Heiti SC", "Noto Sans CJK SC", sans-serif;
       font-size: 11pt; color: #1a1a2e; line-height: 1.7; }
h1 { font-size: 22pt; text-align: center; }
h3 { font-size: 13pt; color: #1a73e8; }
table { width: 100%; border-collapse: collapse; font-size: 10pt; }
th { background: #1a73e8; color: #fff; padding: 7pt 10pt; text-align: left; }
td { padding: 6pt 10pt; border-bottom: 1px solid #e5e7eb; }
tr:nth-child(even) td { background: #f8f9fa; }

When reportlab IS acceptable

English-only documents where you need precise programmatic layout (charts, diagrams, pixel-perfect positioning). Use platypus (SimpleDocTemplate + Paragraph + Table), never raw canvas API.

Reference: tools/gen-73strings-pdfs.py — good example of reportlab platypus for English PDFs.

Anti-Patterns (from real incidents)

Output

Place generated script in tools/gen-<name>-pdf.py. The script should:

1. Define HTML as a string constant
2. Write temp HTML → call weasyprint → remove temp HTML
3. Print the output path on success