Harness Handbook: Making Evolving Agent Harnesses Readable,Navigable, and Editable
Quick Answer
The Harness Handbook introduces a behavior-centric representation for evolving AI agent harnesses, enhancing behavior localization and edit-plan quality through static analysis and LLM-assisted structuring.
Quick Take
The Harness Handbook introduces a behavior-centric representation for evolving AI agent harnesses, enhancing behavior localization and edit-plan quality through static analysis and LLM-assisted structuring. This approach addresses the challenges of modifying large, tightly coupled harnesses by linking behaviors to their source code, ultimately improving the efficiency of code modifications across diverse open-source projects.
Key Points
- The Handbook synthesizes behavior representations from harness codebases using static analysis.
- Behavior-Guided Progressive Disclosure (BGPD) aids in navigating from high-level behaviors to implementation details.
- Improvements in behavior localization and edit-plan quality were observed on diverse modification requests.
- The approach reduces planner token usage while enhancing performance on scattered sites and cross-module interactions.
- Evolving complex agentic systems requires precise identification of code locations for effective edits.
Paper Resources
📖 Reader Mode
~2 min readAbstract:The capability of a modern AI agent depends not only on its foundation model but also on its harness, which constructs prompts, manages state, invokes tools, and coordinates execution. As models, APIs, environments, and requirements evolve, the harness must be continually modified. Before such a change can be made, a developer or coding agent must identify all code locations that implement the target behavior. This is difficult because production harnesses are large, tightly coupled, and behaviorally distributed, while modification requests describe what the system should do and repositories are organized by files and modules. Code search, repository indexing, and long-context processing ease inspection, but still leave this behavior-to-code mapping to be recovered by hand. Behavior localization is therefore a central bottleneck in harness evolution. We introduce the Harness Handbook, a behavior-centric representation synthesized automatically from a harness codebase via static analysis and LLM-assisted structuring, linking each behavior to its corresponding source. We also introduce Behavior-Guided Progressive Disclosure (BGPD), which guides agents from high-level behaviors to relevant implementation details and verifies candidate locations against the current source. On diverse modification requests from two open-source harnesses, Handbook-Assisted planning improves behavior localization and edit-plan quality while using fewer planner tokens, with the largest gains on scattered sites, rarely executed paths, and cross-module interactions. Evolving complex agentic systems thus depends not only on generating edits, but also on determining where those edits should be made.
| Comments: | 29 pages, 6 figures. Project page: this https URL |
| Subjects: | Artificial Intelligence (cs.AI); Software Engineering (cs.SE) |
| Cite as: | arXiv:2607.13285 [cs.AI] |
| (or arXiv:2607.13285v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2607.13285 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Ruhan Wang [view email]
[v1]
Tue, 14 Jul 2026 21:39:55 UTC (645 KB)
— Originally published at arxiv.org
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from arXiv cs.AI
See more →Automatic Ordinary Differential Equations Discovery For Biological Systems Using Large Language Model Powered Agentic System
The MEDA system utilizes large language models and symbolic regression to autonomously discover ordinary differential equations for biological systems, achieving strong structural recovery and biologically plausible models. It outperforms existing methods by integrating domain knowledge and mechanistic constraints, demonstrating effective retrieval and extrapolation capabilities.