Harness Handbook: Making Evolving Agent Harnesses Readable, Navigable, and Editable
Published in Preprint, 2026
Modifying an AI agent harness requires developers to pinpoint the exact code locations responsible for a given behavior, which is hard as harnesses evolve. We introduce the Harness Handbook, synthesized automatically from a harness codebase via static analysis and LLM-assisted structuring, that maps behaviors to source code. We pair it with Behavior-Guided Progressive Disclosure, a technique that guides agents from high-level behaviors down to relevant implementation details. On modification requests drawn from open-source harnesses, this approach improves behavior localization and token efficiency, particularly for scattered sites, rarely executed paths, and cross-module interactions.
| arXiv | Code | Blog |
Recommended citation: Ruhan Wang, Yucheng Shi, Zongxia Li, Zhongzhi Li, Yue Yu, Junyao Yang, Kishan Panaganti, Haitao Mi, Dongruo Zhou, Leoweiliang. (2026). "Harness Handbook: Making Evolving Agent Harnesses Readable, Navigable, and Editable." Preprint.
Download Paper