跳到主内容
@wquguru
精选88VoxAI 产品与模型

Claude Code 发布云端会话指南,Pro/Max 用户可领一次性云额度

The Claude Code team just published a 𝗳𝘂𝗹𝗹 𝗳𝗶𝗲𝗹𝗱 𝗴𝘂𝗶𝗱𝗲 to cloud sessions. You ca…

原文
发到 X
推荐理由

不仅介绍了产品新功能,还直接给出了可复用的 Prompt 模板,结构清晰且具备实操价值,值得创作者参考其结合工具特性提供具体工作流的结构。

The Claude Code team just published a 𝗳𝘂𝗹𝗹 𝗳𝗶𝗲𝗹𝗱 𝗴𝘂𝗶𝗱𝗲 to cloud sessions. You can ask Opus 5.5 to use it to pick out everything in your project that can run in the cloud.

Claude Code 团队刚刚发布了一份关于云端会话的完整指南。你可以让 Opus 5.5 使用它来找出项目中所有可以在云端运行的内容。

Existing individual Pro and Max subscribers can also claim a one-time cloud credit ($𝟭𝟬𝟬 𝗼𝗻 𝗣𝗿𝗼, $𝟮𝟱𝟬 𝗼𝗻 𝗠𝗮𝘅): run /claim-credit in Claude Code by October 7. It expires November 4.

现有的个人 Pro 和 Max 订阅用户也可以领取一次性云额度(Pro 为 $100,Max 为 $250):在 10 月 7 日之前在 Claude Code 中运行 /claim-credit。该额度将于 11 月 4 日过期。

Cloud sessions give every task a fresh virtual machine, and they keep going when you close your laptop. There's no separate charge for the machine, it draws on your plan's usual limits, and it doesn't tie up your laptop's CPU, so you can run tests without holding back: in the guide, a test that failed now and then was fixed and then run 𝟰𝟬 𝘁𝗶𝗺𝗲𝘀 𝗶𝗻 𝗮 𝗿𝗼𝘄 𝘄𝗶𝘁𝗵 𝘇𝗲𝗿𝗼 𝗳𝗮𝗶𝗹𝘂𝗿𝗲𝘀. Small unrelated fixes, and fixes you can only prove by running tests over and over, go to the cloud. Anything that needs a local database, a VPN, a GPU or a phone simulator stays local.

云端会话为每个任务提供一台全新的虚拟机,并且在你关闭笔记本电脑后仍会继续运行。机器没有额外收费,它使用你计划内的常规限额,并且不会占用你笔记本电脑的 CPU,因此你可以毫无保留地运行测试:在指南中,一个偶尔失败的测试被修复后连续运行了 40 次且零失败。一些不相关的修复,以及只有通过反复运行测试才能证明的修复,可以放到云端进行。任何需要本地数据库、VPN、GPU 或手机模拟器的操作则保留在本地。

Send this prompt to Opus 5.5 👇

将此提示发送给 Opus 5.5 👇

"Read this guide: https://claude.dev/blog/claude-code-in-the-cloud/

"阅读这份指南:https://claude.dev/blog/claude-code-in-the-cloud/

Then look for work in this project's TODOs, failing or flaky tests (check CI history and skip markers; don't rerun them locally over and over), places where the docs and code disagree, and recent commits, and sort it into three groups: 1. Work that suits cloud sessions running in parallel, 5 items at most: unrelated tasks that can each be proven done on their own. Write each as one ready-to-use claude --cloud "…" line that says what's wrong, what done looks like, and how to prove it (make the proof heavy, for example run the full test suite 30 times in a row). Keep the items from editing the same files where you can, and give a merge order. 2. Work that has to stay local: anything that needs a local database, a VPN, a GPU or a phone simulator, or where I want to see each change in my own browser right away. 3. Things a cloud session would be missing: CLAUDE.md, skills and agents this project uses that live only in ~/.claude; changes that aren't committed or pushed; variables the tests need that live only in a local .env; a private repo without the Claude GitHub App installed. Mark anything you can't check as needing my confirmation. For each item, point to the exact location, explain why it goes where it does in words a 10-year-old could follow, and for group 3 give the smallest fix. Show me the list first, and don't start any sessions yet."

然后查看该项目 TODO 中的工作、失败或不稳定的测试(检查 CI 历史记录和跳过标记;不要在本地反复重新运行它们)、文档与代码不一致的地方,以及最近的提交,并将它们分为三类: 1. 适合并行运行的云端会话的工作,最多 5 项:彼此无关的任务,每项都可以独立验证完成。将每项写为一行现成的 claude --cloud "…" 命令,说明问题所在、完成的样子以及如何验证(使验证过程充分,例如连续运行完整的测试套件 30 次)。尽量保持项目不要编辑相同的文件,并给出合并顺序。 2. 必须留在本地的操作:任何需要本地数据库、VPN、GPU 或手机模拟器的操作,或者我希望立即在自己的浏览器中看到每次更改的操作。 3. 云端会话会缺失的内容:CLAUDE.md、仅存在于 ~/.claude 中的技能和工作流;未提交或未推送的更改;测试所需的仅存在于本地 .env 中的变量;未安装 Claude GitHub App 的私有仓库。将你无法检查的任何内容标记为需要我确认。 对于每一项,指出确切位置,用 10 岁孩子也能理解的语言解释为什么归入该类,并为第 3 类提供最小的修复方案。先向我展示列表,暂时不要启动任何会话。"

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

关联信息,但可能不是同一事件