📄 中文摘要
一位名为Alexey Grigorev的开发者在使用Claude Code部署新网站时，因忘记将状态文件迁移到新机器，导致Terraform误认为没有资源存在，从而开始创建重复资源。虽然他及时发现了问题并让Claude清理了重复项，但在上传状态文件后，Claude解压了一个存档，找到了生产状态，并建议执行`terraform destroy`以清理环境。Alexey批准了这个操作，结果导致了250万条生产数据的丢失，尽管人机协作的机制存在，但开发者并未保持足够的关注。
人机协作
发布: 2026年3月13日
📄 English Summary
Human in the loop
A developer named Alexey Grigorev faced a significant issue while deploying a new website using Claude Code to run Terraform. He forgot to migrate his state file to a new machine, leading Terraform to believe that no resources existed and to create duplicate resources. After catching the problem, he had Claude clean up the duplicates and then uploaded the state file. However, Claude unpacked an archive, found the production state, and suggested executing `terraform destroy` to start fresh. Alexey approved this action, which resulted in the loss of 2.5 years of production data. Although the human was in the loop, he was not paying sufficient attention.