自改进智能体运行框架论文提出统一架构并在 SWE-bench Lite 验证
Harnesses That Improve Themselves: A Unifying Framework for Self-Improving Agent Harnesses
AI 导读
arXiv 论文 arXiv:2611.22299 提出自改进智能体运行框架的统一架构,把上下文与状态、智能体控制、动作接口、验证与恢复四个可编辑组件与诊断、补丁、验证、部署的四阶段循环配对。
正文 · 原文
Abstract:Whether a foundation model succeeds as an autonomous agent depends as much on its runtime software as on its weights. This runtime, the agent harness, manages context and memory, steers multi-turn control, executes tool actions in sandboxes, and verifies work before stopping. Because models differ in how they use tools and how they fail, a static harness tuned for one model or task often stalls, wastes tokens, or regresses on another. This bottleneck has driven a shift from prompt tuning to self-improving agent harnesses: systems that inspect their own execution traces, separate runtime bugs from model limits, and rewrite their own code. We present a unifying architectural and design framework for self-improving agent harnesses. The framework pairs four editable runtime components (context and state, agent control, action interfaces, and verification and recovery) with a four-stage loop that diagnoses failures, proposes code patches, validates candidates, and deploys updates. Using this framework, we show how optimizers trace errors across long trajectories and discover reusable runtime mechanisms such as pre-stop verifiers, loop breakers, and persistent memory. We further show why full-harness rewrites overfit or break across modules, and how modular updates with validation gates generalize across models while bounding token cost. To keep self-modification safe, we establish practical defenses against recurring failure modes including prompt bloat, benchmark gaming, and evaluator tampering. Finally, experiments on SWE-bench Lite validate this framework, showing that gated modular evolution outperforms prompt reflection and full-harness rewrites across models and repositories at a fraction of the token cost.
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2611.22299 [cs.AI] |
| (or arXiv:2611.22299v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2611.22299 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Nehal Kathrotia [view email]
[v1]
Tue, 17 Nov 2026 06:55:41 UTC (112 KB)
来源:arXiv cs.AI · arxiv.org