LaTeX Forge

Convert .docx and .pdf sources into editable LaTeX project folders.

Default Workflow

Use the unified client first:

python scripts/latex_forge.py convert input.docx output-project
python scripts/latex_forge.py convert input.pdf output-project

For non-technical users who want a local graphical client:

python scripts/latex_forge_gui.py

For Windows distribution, build the app EXE and installer EXE:

powershell -ExecutionPolicy Bypass -File packaging\build_windows.ps1

Expected output:

output-project/
  main.tex
  figures/
  references.bib
  conversion-report.md

For Chinese documents, compile with XeLaTeX or LuaLaTeX. The converters detect CJK text and default to:

\documentclass[UTF8]{ctexart}

If ctex is unavailable, rerun with:

python scripts/latex_forge.py convert input.docx output-project --engine xeCJK

Commands

Inspect a file before converting:

python scripts/latex_forge.py inspect input.pdf
python scripts/latex_forge.py inspect input.docx

Convert and attempt a local LaTeX build when a TeX engine exists:

python scripts/latex_forge.py convert input.docx output-project --compile

Run the converters directly only when debugging:

python scripts/docx_to_latex.py input.docx output-project
python scripts/pdf_to_latex.py input.pdf output-project

Run self-tests after editing the skill:

python scripts/smoke_test.py

Conversion Contract

DOCX conversion should preserve or approximate:

• Headings as section commands
• Ordered and unordered lists
• Tables as tabular
• OMML equations as approximate LaTeX math
• Footnotes as \footnote{...}
• Images copied into figures/
• Caption-styled paragraphs attached to adjacent figures or tables

PDF conversion should:

• Extract text from searchable PDFs
• Create a compilable draft main.tex
• Refuse scanned/image-only PDFs with a clear OCR-required message
• Avoid claiming semantic fidelity for PDF tables, equations, and references

Review Gate

Before final delivery to a user:

1. Confirm main.tex, figures/, references.bib, and conversion-report.md exist.
2. Read conversion-report.md for warnings.
3. Inspect and repair equations, tables, captions, citations, and cross-references.
4. Compile with XeLaTeX when available, especially for Chinese documents.
5. Tell the user when OCR, manual equation repair, table cleanup, or citation reconstruction remains.

Detailed Reference

Read references/conversion-notes.md when:

• A conversion has warnings
• A source document contains complex equations, tables, captions, references, or Chinese text
• A PDF appears scanned or low-quality
• The user asks how to install, run, or validate the skill outside Codex