pdfx — Kit Completo de PDF
docaohieu2808/pdfx
Gera e processa PDFs: Markdown→PDF com IA + OCR, extração, segurança
Ver no GitHubSkill Claude Code para manipulação bidirecional de PDFs: gera PDFs de publicação a partir de Markdown com capa por IA, TOC e syntax highlighting, e também processa PDFs existentes com OCR, extração, merge/split, preenchimento de formulários, marca d'água e criptografia. Inclui 35 subcomandos especializados com suporte nativo a vietnamita.
Instalação
npx convez-ai docaohieu2808/pdfx1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Gerar ebooks e manuais em PDF a partir de Markdown com capa por IA e TOC automático
- Extrair texto, tabelas e imagens de PDFs, incluindo OCR com Gemini
- Mesclar, dividir, criptografar, descriptografar e aplicar marca d'água em PDFs
- Preencher formulários PDF (editáveis ou digitalizados) com suporte a vietnamita
Veredito

Skill robuusta e bem arquitetada que cobre o ciclo completo de trabalho com PDFs — geração de ebooks publication-grade e processamento bidirecional. O suporte nativo a vietnamita e a modularidade (zero dependências externas) são diferenciais legítimos; ferramenta pronta para produção.
README
A tradução para o português deste README ainda está na fila.
pdfx — complete PDF toolkit (Claude Code skill)
A self-contained Claude Code / agent skill for anything PDF.
Unlike typical pdf skills that only read or transform PDFs, pdfx does both directions —
it also produces publication-quality PDFs from Markdown.
One skill, two pillars. Copy the
pdfx/folder into any skills directory and it just works — bundled fonts, modular CLI scripts, reference guides. Depends on no other skill. Optional heavy engines auto-install on first use (or runinstall.sh).
A. Generate beautiful PDFs / ebooks (Markdown → PDF)
Markdown → HTML (python-markdown + Pygments) → WeasyPrint → PDF with published-grade print CSS:
- Designed cover — optional AI art via Gemini (free-tier key), a pre-made image, or a gradient.
- Auto-numbered TOC with dot leaders + real page numbers; optional depth-2 sub-entries.
- Part dividers (
--parts), a back-of-book A-Z index (--index), clickable PDF bookmarks. - Diagrams that actually render: inline
<svg>(shielded frommd_in_html) and```mermaidfences auto-rendered to SVG (Vietnamese-safe labels). - Syntax-highlighted code,
[B]/[I]/[A]/[!]badges, bundled Playfair Display + EB Garamond, A4/6×9, light/dark,--previewto auto-render page PNGs.
PDFX=/path/to/pdfx ; PY=python
$PY "$PDFX/scripts/build_ebook_from_markdown.py" \
--title "DevOps Handbook" --subtitle "Complete Reference" \
--gen-cover "a server scaling into a distributed cloud, glowing data-flow lines" \
--parts "1=Foundations,7=Build & Ship" --toc-depth 2 --index --preview \
--page a4 --theme light --accent "#2563eb" --output handbook.pdf ch01.md ch02.md ...
B. Process existing PDFs — scripts/pdf_process.py (35 subcommands)
| Group | Commands |
|---|---|
| Transform | merge · split · rotate · pages (remove/keep/reorder) · nup · crop |
| Extract | extract-text · extract-tables (+--engine camelot for borderless) · extract-images · to-images · to-html · to-markdown · fonts · links · attachments · bookmarks |
| Security | encrypt · decrypt · unlock · redact (true removal, --ignore-case/--regex) · permissions |
| Optimize | compress · linearize · pdfa (veraPDF-valid PDF/A-2b, --validate) · repair |
| Annotate | watermark · stamp (headers/footers/page numbers) · metadata |
| Convert-in | from-images · from-html · from-url |
| Analyze | info · compare (visual or text diff) · validate (PDF/A via veraPDF) |
| OCR | ocr (tesseract or Gemini, auto for Vietnamese) · ocr-searchable (ocrmypdf text layer) |
| Forms | check_fillable_fields.py → fill_fillable_fields.py (fillable) or fill_pdf_form_with_annotations.py (flat scans — stamps a Unicode font, Vietnamese-safe) |
$PY "$PDFX/scripts/pdf_process.py" compress big.pdf small.pdf --quality ebook
$PY "$PDFX/scripts/pdf_process.py" ocr scan.pdf --lang vie --out scan.txt # Gemini auto for vi
$PY "$PDFX/scripts/pdf_process.py" pdfa report.pdf archive.pdf --validate # PDF/A + veraPDF check
$PY "$PDFX/scripts/pdf_process.py" extract-tables in.pdf --engine camelot --flavor stream
🇻🇳 Vietnamese-aware: Gemini OCR (auto for --lang vie) and Unicode-font form stamping preserve
diacritics where plain tesseract / base-font annotations drop them.
Install
bash install.sh # full toolkit (incl. Mermaid+Chromium ~1.7GB, camelot ~250MB)
bash install.sh --lean # skip the heavy optional engines (auto-install on first use)
install.sh installs Python packages into the chosen venv (set PDFX_PY=…), system tools via
apt/brew/dnf (poppler, qpdf, ghostscript, mupdf-tools, tesseract + Vietnamese, img2pdf, pango/cairo),
the Mermaid CLI, and veraPDF (PDF/A validator), then prints a capability report.
Or piecemeal:
pip install weasyprint markdown pygments pypdf pikepdf pymupdf pymupdf4llm pdfplumber \
reportlab img2pdf pytesseract pdf2image ocrmypdf camelot-py google-genai
# system: qpdf, ghostscript, poppler-utils, mupdf-tools, tesseract-ocr (+ -vie), img2pdf, Pango/cairo
Commands that need an optional package (pymupdf4llm, ocrmypdf, camelot-py) pip-install it on
first use (PDFX_NO_AUTOINSTALL=1 to disable). Set GEMINI_API_KEY (env or ~/.claude/.env) for
AI covers + Gemini OCR; without it, --gen-cover uses a gradient and OCR uses tesseract.
Disk: skill code ~3 MB · veraPDF ~33 MB · Python deps ~250 MB · Mermaid Chromium ~1.7 GB (full only). Lean install ≈ 290 MB; the skill itself is tiny — the weight is optional engines.
Layout
pdfx/
├── SKILL.md # full agent instructions (entry point)
├── install.sh # one-shot dependency installer (--lean for a minimal set)
├── scripts/
│ ├── build_ebook_from_markdown.py # generator (orchestrates pdfx_lib/)
│ ├── pdf_process.py # thin dispatcher over pdfx_ops/
│ ├── pdfx_lib/ # cover · mermaid · markdown_render · toc_index · styles
│ ├── pdfx_ops/ # transform · extract · security · optimize · annotate · convert · analyze · ocr
│ └── *form*.py # form-filling toolkit
├── references/ # weasyprint-recipes · pdf-manipulation · forms · ocr-patterns · gemini-ocr-model · reference
├── assets/fonts/ # bundled display fonts (Playfair Display, EB Garamond) + OFL.txt
└── examples/ # sample chapters (inline SVG + mermaid + glossary) for a smoke test
Smoke test
$PY "$PDFX/scripts/build_ebook_from_markdown.py" --title "pdfx smoke" --input-dir "$PDFX/examples" --output /tmp/pdfx.pdf --preview
$PY "$PDFX/scripts/pdf_process.py" info /tmp/pdfx.pdf
$PY "$PDFX/scripts/pdf_process.py" pdfa /tmp/pdfx.pdf /tmp/pdfx-a.pdf --validate
Author: hieudc · License: MIT (skill code). Bundled fonts are under SIL OFL-1.1 — see
assets/fonts/OFL.txt.