Reproduce “Catching Developers in the Flow: Low-Latency Agentic Program Repair at Google Scale” 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
Celal Ziftci, Spencer Greene, Ray Liu, Livio Dalloro, Lorenzo Dini
FlowAgent automatically attempts to repair test failures found after developers create changes, then offers suggestions in existing review or IDE tools. It uses a tool-using language model that iterates on edits using build and test feedback, with filters before execution and after verification. In 195 sampled failures, experts judged 67.18% of fixes correct; this is not a rate for all failures.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.