CLAUDE.md 为何不断膨胀？Agent 编码中的灾难性记忆
New term “catastrophic remembering”
做 Agent 工程的同学必看，这个“灾难性记忆”现象和注释修复法很实用，赶紧去读论文，优化你的指令文件管理。
New term “catastrophic remembering”
新术语“灾难性记忆”
CLAUDE.md has a ratchet problem: instructions are easy to add and increasingly hard to delete.
CLAUDE.md 存在棘轮问题：指令容易添加，却越来越难以删除。
Agent prompts may remember rules much longer than maintainers remember why those rules exist.
代理提示可能记住规则的时间远长于维护者记住这些规则存在原因的时间。
Across 1,867 GitHub repositories, agentic instruction files more than tripled over their lifetime, with instruction count rising 226% on average.
在 1,867 个 GitHub 仓库中，代理指令文件在其生命周期内增加了两倍多，指令数量平均增长 226%。
Deletion became less likely as instructions aged, and that decay was steeper in files touched by multiple human authors.
随着指令老化，删除的可能性降低，而这种衰减在由多位人类作者修改的文件中更为明显。
The paper calls this “catastrophic remembering”: the instruction survives, but the reasoning that once justified it disappears.
论文将这种现象称为“灾难性记忆”：指令幸存下来，但曾经证明其合理性的推理却消失了。
The proposed fix is a decades-old software practice: attach a comment recording the failure, hypothesis, and outcome behind each instruction, then hide that comment from the executing model.
提出的解决方案是一种已有数十年历史的软件实践：为每条指令附加一条注释，记录其背后的失败、假设和结果，然后对执行模型隐藏该注释。
In the 51-step controlled IFEval setting, informative comments cut excess prompt size from +211.3% to +1.4%, with last-three-round constraint satisfaction at 44.0% in both arms.
在 51 步受控 IFEval 设置中，信息性注释将多余提示大小从 +211.3% 降至 +1.4%，两组中最后三轮约束满足率均为 44.0%。
Comment-shaped noise did not reproduce the effect, pointing to preserved rationale rather than extra text.
注释形式的噪声并未重现该效果，表明起作用的是保留的推理而非额外文本。
– arxiv. org/abs/2608.11095
– arxiv.org/abs/2608.11095
Title: "Why Does CLAUDE.md Keep Growing? Catastrophic Remembering in Agentic Coding"
标题：“为什么 CLAUDE.md 不断增长？代理编码中的灾难性记忆”
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