Reproduce “Skip the Compressor: Coding Agents Fix More Issues When They Pick Whole Files” 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

Xiaocong Yang, Rishit Chatterjee, Himanshu Udupi, Vikram Ramavarapu, ChengXiang Zhai

CQ2-CARE selects candidate files for a coding agent and shows their contents whole and verbatim. In a test-graded comparison on 300 Python issues, GPT-5.6 resolved 43.7% with CQ2-CARE versus 37.3% with LongCodeZip at matched per-issue context budgets. The result is limited to one-patch calls on SWE-rebench and does not establish superiority over compression on the same files.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.