Describe the request
This makes Agent-driven local full-stack workflows fragile: frontends lose watch/recompile sessions; backends drop mid-debug. Users then must restart services one-by-one again.
Forum staff have described this as terminal rebind failure + intentional GC, not always an intentional product preference — but from a user perspective we need an opt-in keep-alive.
Why a setting matters
- Some users want aggressive cleanup (current behavior).
- Many local-dev users want Agent terminals to behave like normal Editor terminals: survive chat switches.
- Without a setting, Agents cannot reliably “start services and leave them running” across sessions.
Proposed behavior
When keep-alive is enabled:
- Switching Agent chats must not terminate the underlying process.
- Terminals should remain listed and re-bindable (Terminal list / other agents group is fine).
- Optional: only GC terminals after explicit user Stop, or when the workspace folder is closed.
Environment / context
- OS: Windows 11
- Use case: sequential start of multiple FE/BE services via Agent shells; frequent chat switches while coding
- Current workaround: run services in classic Editor user terminals / Tasks (less convenient when asking Agent to start them)
Acceptance criteria
- A documented setting exists in Cursor Settings.
- With the setting ON, Agent-started long-running processes survive chat/workspace switches for at least N hours (or until user stops them).
- With the setting OFF, current GC behavior remains.
Thanks — this would materially improve Agent-assisted local development.
中文摘要（方便你对照）
希望 Cursor 增加设置项：切换 Agent 会话/工作区时，可选择不关闭 Agent 拉起的终端及其中的长驻进程（如 ng serve / spring-boot:run）。
现状约 1 分钟重连宽限期后会杀进程，且没有开关；导致本地起服工作流不稳定。