paper-write

The paper-write skill is a highly specialized automation tool for academic researchers and ML engineers. It orchestrates a structured, section-by-section LaTeX drafting process, taking primary input from a PAPER_PLAN.md and NARRATIVE_REPORT.md. By integrating with the Codex MCP ecosystem, it leverages a dedicated reviewer model (typically GPT-5.4 or other configured LLMs) to provide critical, adversarial feedback on drafted sections, ensuring higher quality than single-model self-review. The skill handles venue-specific formatting for major conferences like ICLR, NeurIPS, ICML, and IEEE publications automatically, including managing anonymous submission requirements and bibliography integration through DBLP/CrossRef.

• Automated generation of LaTeX project structures including main.tex, math_commands.tex, and section-specific files.

• Venue-aware template management to ensure compliance with conference-specific style files and formatting constraints.

• Adversarial review loop via Codex MCP to identify logical gaps, formatting errors, or weak claims before finalizing draft sections.

• Support for real-time figure and table insertion, mapping generated visual content from figure/latex_includes.tex into the paper body.

• Intelligent bibliography management, fetching verified BibTeX entries to eliminate hallucinated citations.

• Backup workflows that ensure existing work is preserved in dated directories before any overwriting occurs.

• Ensure you have executed /paper-plan to generate the necessary PAPER_PLAN.md before starting the writing process.

• Target venues include ICLR, NeurIPS, ICML, CVPR, ACL, AAAI, and various IEEE transactions or conferences.

• Use anonymous mode by default for blind peer review, but toggle ANONYMOUS to false for camera-ready versions.

• Monitor the MAX_PAGES constant, as the skill provides guidance on page limits which differ between ML conferences and IEEE standards.

• Leverage the shared references located in ../shared-references/ to improve prose style, maintain citation discipline, and conduct final submission-readiness checklists.