Summary

This project implements the PaperOrchestra multi-agent framework for automating AI research paper writing. It targets users of coding agents (Claude Code, Cursor, etc.) seeking to transform unstructured research materials into submission-ready LaTeX papers, offering significant performance gains over baseline methods without requiring direct API keys or LLM SDKs.

How It Works

The system orchestrates a five-agent pipeline (Outline, Plotting, Literature Review, Section Writing, Content Refinement) based on the PaperOrchestra paper. Each agent's functionality is exposed as a pluggable "skill" containing instructions, reference materials, and deterministic helper scripts. Crucially, all LLM reasoning, web searches, and LaTeX compilation are delegated to the host coding agent, making the framework highly adaptable. An optional agent-research-aggregator skill synthesizes scattered agent histories into the required input format.

Quick Start & Requirements

Installation involves cloning the repository, installing Python dependencies (pip install -r requirements.txt), and symlinking the skills into your coding agent's skill directory. A compatible coding agent with native web search capabilities is essential. Optional integrations like Semantic Scholar, PaperBanana (for figures), or Exa (for search) require separate API keys and setup. Links to the original paper (arXiv:2604.05018) and host integration guides are provided.

Highlighted Details

• Achieves substantial performance improvements on the PaperWritingBench benchmark, with significant win margins in literature review and overall quality.
• Core design principle avoids direct API keys or LLM SDK dependencies, leveraging the host agent's tools.
• Includes an agent-research-aggregator skill to automatically process and structure research notes from various AI coding agent histories.
• Agent prompts are verbatim reproductions from Appendix F of the referenced PaperOrchestra paper.

Maintenance & Community

No specific details regarding active maintenance, community channels (like Discord/Slack), or notable contributors beyond the paper's authors are provided in the README.

Licensing & Compatibility

The project is released under the permissive MIT License, allowing for commercial use and integration into closed-source projects without significant restrictions.

Limitations & Caveats

Effectiveness is highly dependent on the capabilities of the host coding agent, particularly its web search and LLM reasoning tools. Advanced features like high-throughput citation verification or sophisticated figure generation rely on optional, externally configured API keys and integrations. The agent-research-aggregator may require manual review to ensure accurate synthesis of complex research histories.