Reproduce “Reconfigurable Action Theories: A Formal Foundation for Dynamically Composable Agent Harnesses” 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
Jianmin Ji
Reconfigurable Action Theories (RAT) models how changes to tools, providers, and policies alter a harness’s available actions and outcomes. Each module packages action laws and an initializer, while configurations record active modules and world state. In scripted tests across three harnesses, causal evaluation and ASP matched all 76 recorded outcomes; these checks cover finite hand-built models, not full harness behavior.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.