委派危险带：中等能力子智能体为何过度信任继承的陈旧状态
原标题：The Delegation Danger Band: Why Mid-Capability Sub-Agents Over-Trust Inherited Stale State
AI 摘要
该研究提出并测量了智能体框架中子智能体继承父智能体完整上下文时的「委派危险带」现象。作者在Qwen3 0.6B至8B同族模型阶梯上对比Reset、Selective、Full三种继承策略，发现对已被推翻的陈旧状态的盲从随能力提升而急剧下降，但净继承危害呈非单调模式：中等能力模型（Qwen3-1.7B）是统计显著的局部最差点，形成危险带。精选式交接（Selective）在所有数据集上优于完整继承，而固定阈值的能力路由器无法跨数据集迁移，因此建议采用人工精选交接。
正文节选
The Delegation Danger Band: Why Mid-Capability Sub-Agents Over-Trust Inherited Stale State Abstract Agent frameworks increasingly delegate work by forking sub-agents; a common default makes the child inherit the parent’s full working context. We measure how the effect of inherited state changes with capability, where denotes clean fork-fresh accuracy. We compare 3 inheritance policies: Reset (fork fresh: base evidence only), Selective (curated handoff: the useful prior conclusion), and Full (imp