This repository provides a .codex SKILL designed to automate the creation of ML/AI review papers for arXiv. It scaffolds an IEEEtran LaTeX project, manages content generation through an issue-driven workflow, and verifies BibTeX citations, aiming to significantly accelerate the drafting process for researchers.

How It Works

The project implements an end-to-end, issue-driven workflow inspired by agent-designer methodologies. It leverages AI (tested with GPT-5.2) to generate a paper structure, draft content section by section based on user-approved plans, and refine the output. The core arxiv-paper-writer SKILL orchestrates planning, writing, and compilation, ensuring a scaffolded LaTeX project with verified citations.

Quick Start & Requirements

The workflow is initiated with two prompts. The first prompt defines the paper's topic, leading the agent to perform a literature pass, draft a framework, propose titles, and generate a planning markdown file with clarification questions. The second prompt delegates decisions on title and content inclusion, allowing the agent to proceed with generating a complete, compilable LaTeX project. A working LaTeX environment (e.g., pdflatex + bibtex or latexmk) is required. The system was tested on macOS using GPT-5.2 (Extra High).

Highlighted Details

• Implements an end-to-end, issue-driven paper writing workflow.
• Scaffolds an IEEEtran two-column LaTeX project suitable for arXiv review articles.
• Includes automated BibTeX citation verification.
• Provides experimental Claude Code support via .claude → .codex integration.

Maintenance & Community

The provided README does not contain specific details regarding maintainers, community channels (like Discord/Slack), or project roadmaps.

Licensing & Compatibility

The README does not specify a software license. Therefore, compatibility for commercial use or closed-source linking cannot be determined from the provided information.

Limitations & Caveats

The Claude Code support is explicitly noted as experimental. The workflow's effectiveness may depend on the quality of the underlying AI model (tested with GPT-5.2) and the user's ability to guide the planning and decision-making stages.