Claude Code 的 effort level 该怎么选？各档位适用场景与用户实践

Оригинальный заголовок: What effort level do you use for what in Claude Code?

Заголовок и краткое изложение на выбранном языке ожидают перевода.

Краткий обзор ИИ

Claude Code 文档建议按任务选 effort level：low 用于自己复核的快速改动如重命名，medium 用于范围明确的功能（现为 Opus 5.5 和 Sonnet 5.5 默认），high 用于边界情况关键的 bug 修复，max 只留给需要 Claude 独立攻克的难题，因为它容易过度思考。

Полный текст

Полный текст на выбранном языке ожидает перевода. Пока показан оригинал.

The docs suggest:

- low for quick changes you review yourself, like a rename
- medium for features with a clear scope (now the default on Opus 5.5 and Sonnet 5.5)
- high for bug fixes where edge cases matter
- max only for hard problems you leave Claude to work through alone, since it tends to overthink

Skills and subagents can also set their own level with effort: in their frontmatter.

I run xhigh for pretty much everything.

What's your setup? Do you stick to one level, or change it per task? Does anyone run subagents at low to save tokens?