arXiv:2608.28363 · Agent 智能体
EvoUndo: Recoverability-Constrained Self-Evolution for LLM Agent Harnesses
EvoUndo：面向 LLM Agent 运行环境的可恢复性约束自演化
EvoUndo: Recoverability-Constrained Self-Evolution for LLM Agent Harnesses
- 类型：arxiv
- 标识：2608.28363
- 链接：https://arxiv.org/abs/2608.28363
- 主分类：agent
- 形态：method
- TLDR：LLM agents increasingly modify their own prompts, tools, middleware, resources, and execution harnesses at runtime. Such self-evolution can improve capability, but a successful mutation may leave persistent effects that cannot be safely reversed in states different from the one in which it was created. We introduce EvoUndo, a framework for representing, synthesizing, diagnosing, and independently verifying recoverability of model-generated self-modifications across counterfactual states. Across 600 unseen one-shot self-evolution tasks, we identify 197 capability-improving mutations that fail r
- 副分类：evaluation
- 待LLM分类：否
- 标题中文：EvoUndo：面向 LLM Agent 运行环境的可恢复性约束自演化
- TLDR中文：LLM Agent 日益在运行时修改自身的 prompt、工具、中间件、资源与执行环境。此类自演化可提升能力，但一次成功的变更可能留下持久影响，在与创建时不同的状态下无法安全回滚。我们提出 EvoUndo，一个用于表示、合成、诊断并独立验证模型生成的自修改在反事实状态间可恢复性的框架。在 600 个未见过的单轮自演化任务中，我们识别出 197 项提升能力但回滚失败 r
- 来源文件：
- /inbox/tom/_candidates/2026-09-01-rag-retrieval-reranking-candidates.json
- /inbox/tom/_candidates/2026-09-01-agent-rag-longcontext-candidates.json