AgentBug-Smith: Automated Discovery of Real Harness Bugs in Agent Frameworks

A joint paper from UChicago, Fudan, Tsinghua and UIUC introduces AgentBug-Smith, a method to automatically discover and reproduce real harness bugs in open-source agent systems. Top coding agents can fix only about 9% of such bugs.

2026-10-03 ~ 2026-10-03 · 2 related posts

• AgentBug-Smith paper: top coding agents fix just 9% of agent harness bugs — rohanpaul_ai · 2026-10-03

1 near-duplicate retellings: rohanpaul_ai