PaperSpine is an academic writing system that guides AI agents through structured stages to build and revise research papers with evidence-aware blueprints and LaTeX-safe output.
The tool addresses the challenge of translating academic contributions into well-structured manuscripts. It works by requiring an AI agent to first learn the target format and study exemplary papers, then confirm the paper's core contributions with the user, design the article unit by unit, and only then draft or rebuild the manuscript. The system outputs to LaTeX, PDF, or Word formats. Rather than offering a flat collection of writing skills, the current version organizes twelve stages of paper development through a single orchestrated skill that reads playbooks from a references directory and uses role information from an agents directory.
The tool suits writers working on journal papers, conference submissions, course reports, technical reports, surveys, and competition papers where output format and structure matter significantly. It is designed for workflows where an AI agent can iteratively refine contributions and manuscript architecture before final writing. The system supports four hosts: Claude Code, Codex, OpenClaw, and Hermes CLI, with the same skill installed to each without mixing between them.
Development activity shows a focus on installation robustness and backward compatibility. The project maintains a clear separation between source code in src/ and generated distribution files in dist/, with a synchronization script handling the build process. Installation scripts for Windows PowerShell and macOS/Linux include legacy cleanup options to remove old PaperSpine directories and prevent duplicate discovery. The tool preserves user configuration across updates while never modifying host settings files, addressing a previous issue where the installer would overwrite settings. A release candidate version introduces a five-stage local research workspace while maintaining compatibility with the previous version's update mechanism.
