📄 中文摘要
大型语言模型（LLMs）受到其上下文限制的制约，即在任何给定时间内能够处理的令牌数量。尽管LLMs在能力上取得了显著提升，但这些限制在过去两年中并没有显著增加，通常最大约为1,000,000，而基准测试通常在200,000以下报告更好的质量结果。因此，合理管理上下文以适应这些限制对于获得良好的模型结果至关重要。子代理提供了一种简单而有效的方法来处理更大任务，而不会过度消耗编码代理的宝贵资源。
子代理
出处: Subagents
发布: 2026年3月17日
📄 English Summary
Subagents
Large Language Models (LLMs) are constrained by their context limit, which defines how many tokens they can manage in their working memory at any given time. Despite significant improvements in LLM capabilities over the past two years, these limits have not increased much, typically maxing out around 1,000,000 tokens, while benchmarks often report better quality results below 200,000. Effectively managing context to fit within these limits is crucial for achieving great results from a model. Subagents offer a straightforward yet effective way to tackle larger tasks without excessively depleting the coding agent's valuable resources.