Reproduce “HerAgent: A Block-Level Checkpointing Harness for Coding-Agent Environment Setup” 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
Xiang Li, Siyu Lu, Federica Sarro, Claire Le Goues, He Ye
HerAgent groups related setup commands into functional stages, validates each, checkpoints successful stages, and repairs failures locally while saving accepted fixes into a replayable artifact. On Installamatic-Bench, it reaches 85.0% with GPT-4o-mini versus Installamatic’s 52.0%; on SetupBench, GPT-4.1 HerAgent reaches 70.4% versus OpenHands’s 46.3%. These results use different benchmark oracles and published attempt protocols, and do not establish universal recovery.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.