Claude Code 从多个 API/MCP 工具拉取数据时，如何处理部分失败？

Оригинальный заголовок: How do you handle partial failures when Claude Code is pulling data from multiple APIs or tools?

Заголовок и краткое изложение на выбранном языке ожидают перевода.

Краткий обзор ИИ

有开发者提出 Claude Code 工作流中的部分失败问题：当 Claude 调用 5 个数据源、其中 1 个超时并返回不完整数据时，是让 Claude 继续并显式标记缺失来源、重试，还是直接终止任务。他更担心 Claude 拿到看似足够有效的数据后自信完成任务，却没意识到关键信息已缺失，并询问实际工作流中是否有人处理得较好。

Полный текст

I've been working with Claude Code workflows that pull live data through APIs/MCP tools and I'm curious about how are you handling partial failures...

Let's say Claude calls 5 sources.. Four return normally, but one time out and returns incomplete data.

Do you let Claude continue and explicitly flag the missing source? Do you retry it, or do you stop the task entirely?

I'm not really worried about obvious errors. I am talking about Claude getting enough valid-looking data to confidently finish the task without realizing something important is missing...

Anyone handling this very well in real Claude Code workflows?