Claude Duet 以权限立方体链运行演示
视频展示 Claude Duet 合盖运行：ALLOW/ASK/DENY 权限图、CLAUDE.md 规则、晨间简报和子代理索引，模型只准备、由人拍板。
It is becoming f*cking modular A solo builder just showed Claude Duet running as a chain of locked cubes with the laptop shut. Sent without yes: 0. Drafts waiting. Nothing moves alone. What the video shows: - Permission map live: ALLOW (read, edit memory) · ASK (bash, mail.send, calendar.move) · DENY (edit CLAUDE.md, payments, delete) - CLAUDE.md rules on screen: “What you never do without my yes” - send, spend money, delete - Morning brief spinning up: calendar · inbox · promises · drafts ready - Team index of subagents: inbox-triage, meeting-prep, promise-keeper, checker - Cubes chained together, each one a file, a rule, a worker - Model prepares. You decide. One word: YES Most agents either wait for you or run until they break something. This one comes to you, finishes the work, and stops at the ask. Anthropic’s version is still a research preview. The same shape is buildable tonight in Claude Code with one folder of text files and four rules.
这条是引用着另一条写的：「用 Claude Code + Opus 5.5 自建 Claude Duet 教程」