Why it is worth attention

It provides a rigorous, principled methodology for LaTeX-to-Word conversion that emphasizes semantic mapping, validation, and acceptance gates, treating each manuscript as a unique project with adaptable steps.

Who it is for

Academics and researchers who need to convert LaTeX manuscripts to Word for journal submission or collaboration
Developers building AI agents for document conversion workflows
Technical writers or editors managing complex LaTeX-to-Word transitions
Users of Claude or Codex skills seeking a structured conversion approach

Use cases

Converting a PhD thesis from LaTeX to a publisher's Word template with proper cross-references and citations
Preparing a journal article in Word format while preserving LaTeX numbering and scientific content
Enabling collaboration with Word-only co-authors on a LaTeX-based document
Auditing and validating a converted DOCX for correct fields, styles, and high-risk objects

Strengths

Default to smallest reliable approach (Pandoc with reference DOCX) before escalating
Uses read-only scripts (`latex_manifest.py`, `docx_audit.py`) for evidence without altering files
Mandates desktop Word validation and explicit user opt-out for static references
Explicitly handles high-risk objects like cross-references, citations, and numbering semantics

Considerations

Requires user to run exact Word/Zotero steps and re-audit, not fully automated
Depends on host-exposed toolchain; compilation evidence may be pending if toolchain unavailable
Skill is designed for Claude/Codex environments and may not work outside those agents

README quick start

LaTeX to Word Workflow

An agent skill for planning, executing, and reviewing project-specific academic LaTeX-to-Word conversions.

Current version: 3.0.0

Purpose

LaTeX-to-Word conversion is highly dependent on the manuscript, target template, citation workflow, numbering semantics, and local Word environment. This skill therefore provides a lean routing contract, detailed on-demand guidance, read-only evidence tools, and acceptance gates instead of a fixed conversion pipeline.

The agent should:

understand the complete LaTeX project and compile it when the host exposes the toolchain, otherwise record compilation evidence as pending;
inspect the target Word requirements and preserve original files;
start with the smallest reliable approach, normally Pandoc with a reference DOCX, and add project-local processing only for observed needs;
prove the LaTeX–Pandoc–Word semantic mapping with representative content; a reference DOCX does not align differently named styles or create target behavior that Pandoc does not emit;
preserve scientific content and numbering semantics ahead of editability;
verify rather than merely claim live citations;
probe Zotero capability only through host-exposed evidence and prefer confirmed live fields unless static citations are explicitly required or probing fails;
generate and validate native Word cross-references by default, allowing static references only after an explicit informed user opt-out;
validate high-risk objects and disclose approved exceptions;
preserve prose chemical formulae and units as ordinary subscript/superscript text unless they are genuine mathematical expressions;
use the user's desktop Word as the authoritative renderer for Word-specific fields, numbering, and template behavior;
hand off exact Word/Zotero steps for the user to run, and independently re-audit the refreshed file when the user returns it.

Installable skill

The only installable and released skill directory is latex-to-word-workflow/.

Repository-level behavior evals are kept in tests/. They are versioned with the skill so trigger and workflow expectations can be reviewed, but are excluded from git archive release artifacts.

git clone https://github.com/hajimi-kun/latex-to-word-workflow.git
Copy-Item -Recurse .\latex-to-word-workfl

Description

Agent Skill for polished LaTeX-to-Word conversion with live Zotero citations

Related repositories

Similar projects matched by category, topics, and programming language.

lopopolo

Featured

harness-engineering

Harness Engineering is a methodology for improving coding agent outputs by carefully crafting the environment around them—providing curated context, tools, and executable constraints that encode an organization’s nonfunctional requirements and cumulative lessons.

AI & Machine LearningAI Agents

2,390

slvDev

Featured

esp32-ai

A 28.9 million parameter language model runs on an $8 ESP32-S3 microcontroller entirely on-device, generating simple stories at about 9.5 tokens per second.

AI & Machine LearningLarge Language Models

1,960

littledivy

Featured

mimic

mimic captures traffic from any iOS or web app and automatically generates a Python client library that lets you call the app's API like a regular library.

AI & Machine Learning

1,482