让大模型自进化时能可靠撤销修改，避免永久错误。
EvoUndo: Recoverability-Constrained Self-Evolution for LLM Agent Harnesses
- 用可验证的恢复语言追踪自修改行为，确保每步可逆。
- 在197个有效改进中，新方法成功恢复191个，旧方法全失败。
- 适合研究可信自进化智能体的开发者和安全研究人员。
大型语言模型代理在运行时越来越多地自我修改其提示、工具、中间件、资源和执行框架。这种自进化虽能提升能力，但成功的突变可能留下无法在不同状态中安全回滚的持久影响。本文提出EvoUndo框架，用于表示、合成、诊断并独立验证跨反事实状态下的模型生成自修改的可恢复性。在600个未见过的一次性自进化任务中，识别出197个提升能力但无法通过恢复验证的突变。在原始恢复表示下，传统修复策略仅恢复0/197例自然失败；确定性预言分析在原恢复语言L0下恢复48/197，而扩展恢复演算使实证预言恢复率达191/197。协议锁定的2×2显式表达力干预分离出两个瓶颈：精确状态定位将恢复成功率从0/48提升至38/48（79.2%），当原语言足够时；扩展恢复语言则使预言定义的S1层级中142/143（99.3%）失败得以恢复。在主干gpt-oss-120b上，加入精确地址诊断后，恢复率降至133/143（93.0%）；而Qwen3.8-27B复现保持了定位与表达力效应，但未出现此负面交互，表明该现象具有模型依赖性。结果表明，可靠的代理自进化需协同设计验证机制、状态定位、见证语义与恢复语言表达力，而非仅依赖迭代提示。
原文摘要 · Abstract (English)
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 recoverability verification. Under the original recovery representation, conventional repair strategies recover 0/197 of these natural failures. Deterministic oracle analysis recovers 48/197 under the original recovery language L0, while the extended recovery calculus increases empirical oracle recovery to 191/197. A protocol-locked 2x2 grounding-by-expressivity intervention then separates two bottlenecks: exact state-address grounding increases successful recovery from 0/48 to 38/48 (79.2%) when the original language is sufficient, while extending the recovery language enables recovery on 142/143 (99.3%) failures in the oracle-defined S1 stratum. On the primary gpt-oss-120b backbone, adding exact-address diagnostics to the richer language reduces recovery to 133/143 (93.0%); a Qwen3.8-27B replication preserves the grounding and expressivity effects but not this negative interaction, indicating that the latter is model-dependent. These results indicate that reliable agent self-evolution requires co-designing verification, state grounding, witness semantics, and recovery-language expressivity rather than relying on iterative prompting alone.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。