Reproduce “L3M: A Verifiable Harness and Runtime for Self-Extending and Self-Improving Agents” 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
Carson Eisenach, Robert Joseph George, Vincent Quenneville-Belair, Kate Gwimm, Dean Foster
L3M lets agents turn repeated procedures into reusable tools without expanding their granted authority. Its runtime checks tool source, declared effects, and permission bounds, while durable state supports resumable work. On 37 selected Terminal-Bench tasks, tame L3M achieved 27.0% pass@3, versus 41.7% with shell access and review off; this measures task success only under those conditions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.