Reproduce “Lean4Agent: Formal Modeling and Verification for Agent Workflow and Trajectory” 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
Ruida Wang, Jerry Huang, Pengcheng Wang, Xuanqing Liu, Luyang Kong, Tong Zhang
Lean4Agent checks workflows by assigning machine-checkable expectations to steps and comparing execution traces with them. It combines structural checks, Hoare-style contracts, trajectory diagnosis, and failure-guided workflow revision. Passing workflows were associated with 14.80- and 9.07-percentage-point differences on two selected tasks; LeanEvolve added 7.47 points on a hard SWE-Bench-Verified subset. These checks establish specified-predicate consistency, not full agent correctness.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.