🛰️ Daily AI Frontier
‹ back to 2026-08-13

Spark-to-Paper: End-to-End Research Paper Generation as a Composable Skill

arXiv cs.CL LLM Agents Zhuoyang Qian, Biao Wu, Yiran Wang, Chris D Yan, Desan Dai, Liangwei Zheng, Jin Jiang, Junsheng Zhang, Wenhao Wang 2026-08-12
Representative image for Spark-to-Paper: End-to-End Research Paper Generation as a Composable Skill

TL;DR - Spark-to-Paper is a 13-skill workflow that turns research ideas into complete papers inside an existing coding assistant. It emphasizes evidence-driven claims, deterministic checks, and editable figures without requiring a separate agent platform.

  • Separates model judgment from executable, verifiable operations and plans experiments before observing results.
  • Achieves 99.5% citation validity and 96.4% figure editability across eight controlled topics.
  • Its full review stack raises fabrication detection from 14% to 92%; adversarial review reaches 74% precision.
  • A manuscript averages 11.9M tokens, $8.10, and 3.2 hours.

view merged work →