Voxyz_ai 推荐 Opus 5.5 自动化代码审查：夜间自动修复 Bug
Opus 5.5 is really good at automation, so before you go to bed, put it on the 𝗻𝗶…
提供了可直接复制粘贴的具体 Prompt 模板，展示了如何将 AI 能力嵌入日常开发工作流，结构清晰且具备实操性。
Opus 5.5 is really good at automation, so before you go to bed, put it on the 𝗻𝗶𝗴𝗵𝘁 𝘀𝗵𝗶𝗳𝘁 too.
Opus 5.5 在自动化方面表现出色，所以在你睡觉前，也把它放到 𝗻𝗶𝗴𝗵𝘁 𝘀𝗵𝗶𝗳𝘁（夜班）上吧。
Set up a Claude Code routine and every night it goes back over the code you pushed to main that day. For each bug it finds, it opens a draft PR with 𝗮 𝘁𝗲𝘀𝘁 𝘁𝗵𝗮𝘁 𝗳𝗮𝗶𝗹𝘀 𝗳𝗶𝗿𝘀𝘁 and only passes after the fix. It runs in the cloud, so it keeps going with your laptop closed.
设置一个 Claude Code 例行任务，每晚它都会回顾你当天推送到 main 分支的代码。对于发现的每个 bug，它会打开一个草稿 PR，其中包含 𝗮 𝘁𝗲𝘀𝘁 𝘁𝗵𝗮𝘁 𝗳𝗮𝗶𝗹𝘀 𝗳𝗶𝗿𝘀𝘁（先失败的测试），只有在修复后才会通过。它在云端运行，因此即使你合上笔记本电脑，它也会继续执行。
In the repo you want checked, type /schedule and paste this 👇
在你想要检查的仓库中，输入 /schedule 并粘贴以下内容 👇
"Run every day at 1:07 AM on Opus 5.5 with no connectors attached. Save the instructions below as the prompt, word for word.
"每天凌晨 1:07 在 Opus 5.5 上运行，不附加任何连接器。将以下说明逐字保存为提示词。
First install dependencies and run the full test suite. Note which tests were already failing. If the tests can't run, or there are none, write down what's missing and stop.
首先安装依赖项并运行完整的测试套件。记录哪些测试已经失败。如果测试无法运行，或者没有测试，写下缺失的内容并停止。
Then look at the commits pushed to the default branch in the last 24 hours. If there are none, end the run. Skip any bug that a claude/ branch is already fixing.
然后查看过去 24 小时内推送到默认分支的提交。如果没有，则结束本次运行。跳过任何已被 claude/ 分支修复的 bug。
Only look for real bugs. Ignore style and naming. To call something a bug, you must be able to point to what a comment, the docs, a type, or an existing test says. If a change in behavior looks intentional, just note it. Don't fix it.
只寻找真正的 bug。忽略样式和命名问题。要称某事为 bug，你必须能够指出注释、文档、类型或现有测试所指出的内容。如果行为变化看起来是故意的，只需记录下来，不要修复它。
For each bug, write a failing test first. Make sure the failure message points to that bug, and rule out a broken test or an environment problem. Then change the code until the test passes, touching only what you must, and run the full suite again. Open one draft PR per bug with your evidence, three at most.
对于每个 bug，先编写一个失败的测试。确保失败消息指向该 bug，并排除测试损坏或环境问题的可能性。然后修改代码直到测试通过，仅触及必要的部分，并再次运行完整测试套件。为每个 bug 打开一个草稿 PR，附上你的证据，最多三个。
Never delete, change, or skip existing tests. Push only to claude/ branches. Don't merge anything.
永远不要删除、更改或跳过现有测试。仅推送到 claude/ 分支。不要合并任何内容。
Finish with a handoff note: the commits you checked, the PRs you opened, and anything that looked like a bug but couldn't be proven."
最后写一份交接说明：你检查的提交、你打开的 PR，以及任何看起来像 bug 但无法证实的问题。"
Once it's saved, do a test run with /schedule run. In the morning, open that run on the Routines page on the web and read the handoff note. Green in the run list only means it finished.
保存后，使用 /schedule run 进行一次测试运行。早上，在网页上的 Routines 页面打开该运行记录并阅读交接说明。运行列表中的绿色仅表示已完成。
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