Reproduce “Can Coding Agents Write and Transpile Unit Tests” 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
Gil Pasternak, AJ Langley, Tom Yotam, Ellie Kitanidis
The authors evaluate coding agents’ test suites across 240 open-source Python, C++, and Rust projects, measuring injected-fault detection and behavior coverage. Claude Code achieves a mean mutation score of 0.818, versus 0.625 for human suites, often by writing more tests; scores weaken as projects grow. In C++-to-Rust translation, suites often lose checks, so these results do not establish reliable preservation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.