paper-writing

The paper-writing skill provides a comprehensive, end-to-end orchestration for academic research paper production. Designed for AI-driven research environments, this skill automates the transition from raw research notes or narrative reports into submission-ready LaTeX manuscripts. It chains five core research phases: structural planning, automated figure and plot generation, LaTeX writing, document compilation, and a multi-round improvement loop utilizing advanced review models. By managing the full lifecycle of a manuscript—including citation scaffolding, architecture diagramming, and claim-evidence matrix validation—it allows researchers to focus on core concepts while the agent handles the heavy lifting of composition, formatting, and iterative refinement.

• Orchestrates the full pipeline: /paper-plan, /paper-figure, /paper-write, /paper-compile, and /auto-paper-improvement-loop.

• Supports various academic venues (ICLR, NeurIPS, ICML, CVPR, IEEE, etc.) with automated style and page limit management.

• Integrates multiple figure generation methods: figure-spec for JSON-to-SVG architecture diagrams, Mermaid for flowcharts, and AI-driven image generation for qualitative methods.

• Implements rigorous 'submission' mode with automated audit gating (proof-checker, claim-audit, citation-audit) using internal verification tools.

• Features cross-model reviewer routing (GPT-5.4, Claude, Gemini) to ensure objective, adversarial-style review that avoids single-model blind spots.

• Ideal for converting structured NARRATIVE_REPORT.md files into high-quality scholarly documents.

• Supports both 'draft' mode for rapid iteration and 'submission' mode for verified, final-stage audit compliance.

• Configuration is highly flexible via command-line arguments: set venue, improvement rounds, checkpoint behavior, and illustration backends.

• Operates entirely via a file-based, lightweight architecture, ensuring no vendor lock-in; compatible with Claude Code, Codex CLI, Cursor, Trae, and other agentic environments.

• Use as a primary tool when you need to transform research findings into a polished PDF automatically while maintaining control over the narrative and claim integrity.