HERA: Harness-Environment Co-Evolution for Reliable Agentic Abstention

arXiv:2610.06563. (Preprint)

(* co-first authorship)

Han Luo*

University of Leeds / Southwest Jiaotong University

Bingbing Wen*

University of Washington

Guang Yang

Zora Zhiruo Wang

Pan Lu

Lucy Lu Wang

University of Washington / Allen Institute for AI

HERA: Harness-Environment Co-Evolution for Reliable Agentic Abstention

Abstract

HERA improves agentic abstention by jointly evolving agent harnesses and executable environments. It constructs verified pairs of feasible and infeasible tasks through controlled environment mutations, then uses agent failures to guide harness adaptation and generate new tasks. On held-out tasks, the evolved harness improves abstention accuracy from 61.7% to 83.3% and feasible-task completion from 68.3% to 76.7%. The same harness transfers to 19 additional LLMs without model-specific optimization, improving abstention accuracy by 15.3 percentage points on average.

Materials

Project Preprint arXiv Code 
@misc{luo2026hera,
  title={HERA: Harness-Environment Co-Evolution for Reliable Agentic Abstention},
  author={Han Luo and Bingbing Wen and Guang Yang and Zora Zhiruo Wang and Pan Lu and Lucy Lu Wang},
  year={2026},
  eprint={2610.06563},
  archivePrefix={arXiv},
  primaryClass={cs.AI},
  url={https://arxiv.org/abs/2610.06563},
}