Type: Research Paper / Workshop Submission (ICLR 2026)
Summary: A methodology for improving LLM agents by automatically synthesizing a code harness (action verifier) to provide grounding and evaluation, demonstrating improved agent reliability and performance in complex environments.