This handles the full lifecycle of writing papers for NeurIPS, ICML, ICLR, ACL, AAAI, and COLM. It starts from your research repo, explores your code and results, then produces complete first drafts instead of asking permission for every section. The core philosophy is proactive: Claude should deliver something concrete you can react to, not block on clarifications. The most important rule here is citation handling. It never generates BibTeX from memory and integrates with Semantic Scholar, arXiv, and optionally Exa MCP to verify every reference programmatically. Includes LaTeX templates, conference checklists, and writing philosophy from researchers like Karpathy and Steinhardt. If you're targeting systems venues like OSDI or SOSP, use systems-paper-writing instead.
npx skills add https://github.com/orchestra-research/ai-research-skills --skill ml-paper-writing