Reproduce “Dittany: Contribution-Guided Repair Framework for MAS Applications” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Hanqi Sun, Yuchen Shao, Hanyi Si, Beibei Xu, Lisheng Wang, Chengcheng Wan

Dittany repairs multi-agent system tasks by tracking required results, acceptance conditions, and dependencies. It corrects actions when existing inputs suffice and adds or revises work when they do not. It reuses completed results only when they pass relevant checks. Across 9,000 instances from 30 applications, it achieves 84.76% task accuracy versus 63.48% for Reflexion, within a finite evaluation set.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.