pdf-to-html

Este extracto es una copia guardada durante la revisión. La fuente externa contiene la versión completa y actual.

---
name: pdf-to-html
description: Converts a PDF into one self-contained, readable HTML file that preserves images, tables, charts and reading order — optionally translating it into another language while keeping every figure. Uses structured extraction (PyMuPDF), font-size-driven layout, compressed base64-inlined images (a single portable file), and mandatory headless-Chrome visual verification. Use whenever someone wants to READ a PDF as a web page or clean document, turn a PDF into HTML, or translate a PDF into another language while keeping its images/tables/charts intact — e.g. "PDF 转 HTML", "把这个 PDF 转成中文网页版", "make this report readable", "translate this PDF but don't lose the charts", "I just want to read this PDF on my phone". Distinct from doc-to-markdown (plain Markdown text) and pdf-creator (Markdown→PDF) — this one produces a styled, image-faithful HTML reading experience.
---

# PDF to HTML

Turn a PDF into a single, self-contained, readable HTML file — images, tables,
charts and reading order preserved — and optionally translate it, keeping every
figure in place.

The pipeline is **extract → look → (translate) → build → verify**. The middle
"look" and final "verify" steps are where faithfulness actually comes from: a PDF
is a layout, not just a text stream, so you read the rendered pages before
building and the rendered HTML before delivering.

This skill runs **inline** (no `context: fork`): translation orchestrates a
Dynamic Workflow, and a subagent cannot spawn one.

## When to use / not use

- **Use** when the goal is to *read* a PDF as HTML/web page, to convert a PDF to
  a styled HTML document, or to translate a PDF into another language while
  keeping its figures and tables.
- **doc-to-markdown** instead if they want plain Markdown text (no styling, figures optional).
- **pdf-creator** instead for the reverse direction (Markdown → PDF).

## What it does NOT do

- **Scanned/image-only PDFs** (no text layer): OCR first (e.g. `ocrmypdf`), then use this.
- **Complex multi-column tables**: cell *text* is preserved and readable, but column
  alignment can flatten into a text flow — PyMuPDF reads a table as text blocks, not a
  grid, so the grid lines are gone. Tables that are *images* in the PDF survive as
  images. If the table's grid structure is essential, use **doc-to-markdown** (pandoc
  rebuilds real tables) or convert that page separately.
- **Pixel-perfect facsimile**: output is a clean *re-flow* that keeps images and
  reading order, not a 1:1 copy of the original page layout.
- **Rewriting**: it translates and re-lays-out; it does not summarize, add a TL;DR,
  or editorialize. Faithfulness is the point (see Fidelity below).

## Dependencies

`uv` (runs Python with inline deps), Google Chrome or Chromium (visual
verification). Python packages come via `uv run --with`: PyMuPDF, Pillow, numpy.
Nothing to pre-install beyond Chrome and uv.

## Workflow

Copy this checklist and tick as you go:

```
- [ ] 1. Extract structure + render pages   (extract_pdf.py)
- [ ] 2. Read pages/*.png — SEE the layout, find content vs decorative images
- [ ] 3. (only if translating) run the translation workflow
- [ ] 4. Build the single-file HTML          (build_html.py)
- [ ] 5. Verify visually                      (verify_render.py → Read every segment)
- [ ] 6. Deliver the .html
```

### 1. Extract

```bash
uv run --with pymupdf python scripts/extract_pdf.py input.pdf
```

Writes `input-build/` with `structure.json` (text blocks with font sizes + image
blocks flagged `decorative`), `images/`, and `pages/` (one PNG per page).

### 2. Look before you build

Read `input-build/pages/*.png`. This is not optional: you need to see the real
layout, confirm which images are content vs decoration, and spot tables/charts.
For a long PDF, read every page; for a short one it's quick. This is also where
you understand the document well enough to translate it well.

### 3. Translate (optional)

Only if the user asked for another languag