Reproduce “Harnesses That Improve Themselves: A Unifying Framework for Self-Improving Agent Harnesses” 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
Deep Shah, Nehal Kathrotia
The authors present a framework for improving fixed-model agents by editing the runtime that manages context, actions, tools, and verification. It diagnoses failures from execution traces, proposes scoped code changes, and validates them before deployment. On held-out tasks, modular evolution uses 46.7k runtime tokens per task and transfers better to a second model, but scores 40.0% versus 42.5% for unrestricted rewrites.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.