跳到主内容
精选85elvis技巧与观点

CLAUDE.md为何不断膨胀:指令累积的实证研究

Why does CLAUDE.MD keep growing?

原文
推荐理由

维护CLAUDE.md或AGENTS.md的开发者必看,这份研究用数据揭示了指令膨胀的根源,并给出了加注释这一简单有效的解法,建议直接应用到你的项目配置里。

Why does CLAUDE.MD keep growing?

为什么 CLAUDE.MD 会不断增长?

If you maintain a CLAUDE.md or an AGENTS.md, this one is worth your time.

如果你维护 CLAUDE.md 或 AGENTS.md,这篇值得一读。

(bookmark it)

(收藏它)

This work traces why these files grow without bound. Appending an instruction is free. Deleting one after its rationale is gone costs exponential verification, so nobody removes anything.

这项工作追溯了这些文件无限增长的原因。追加一条指令是免费的。在其理由消失后删除它需要指数级的验证成本,所以没有人会删除任何内容。

Across 247,694 instruction lifetimes in 1,867 repositories, prompts more than tripled over their lifetime and gained 4.9 net instructions per commit. The older an instruction gets, the less likely anyone is to delete it.

在 1,867 个仓库的 247,694 条指令生命周期中,提示词在其生命周期内增长了三倍以上,并且每次提交净增加 4.9 条指令。指令越旧,被删除的可能性就越小。

The proposed fix is comments. Writing down the reasoning behind an instruction removed 99.3% of excess instructions in verifiable settings and improved real agentic instruction-following by up to 23.1%.

提出的解决方案是注释。在可验证的环境中,写下指令背后的推理消除了 99.3% 的多余指令,并将真实代理指令遵循率提高了高达 23.1%。

Paper: https://arxiv.org/abs/2608.11095

论文:https://arxiv.org/abs/2608.11095

Track more trending AI papers in our academy: https://academy.dair.ai/

在我们的学院中追踪更多热门 AI 论文:https://academy.dair.ai/

更进一步:量化金融体系

看懂新闻只是起点——沿量化金融路径,把它变成能交付的工程能力

进入量化体系 →

相似阅读

另一事件,读法相近