Shared PTY, viewer, lock, async tracking, and policy rules are not equally important in every environment. This page breaks them apart so users can map product behavior to embedded, ops, deployment, and remote debugging work.共享 PTY、viewer、输入锁、异步跟踪和策略规则，在不同环境里的重要性并不一样。这个页面把它们拆开，让用户能把产品行为映射到嵌入式、运维、部署和远程调试场景。
This structure is better than a flat feature list because it meets users where their real pain already exists: stateful terminals, manual intervention, long jobs, and safety boundaries.这个结构比平铺功能列表更好，因为它直接对准用户已经存在的痛点：状态型终端、人工接管、长任务和安全边界。
When commands change prompt state, spawn background jobs, or leave partial shell context behind, a shared PTY keeps both operator and AI grounded in the same runtime instead of re-simulating context each round.当命令会改变提示符状态、拉起后台任务或留下半截 shell 上下文时，共享 PTY 能让操作者和 AI 站在同一个运行时上，而不是每轮都重新模拟上下文。
A browser viewer matters when the human needs to inspect prompts, logs, or interactive confirmations in real time while the AI is still helping in the same session.当人需要实时看提示符、日志或交互确认，而 AI 仍在同一会话里帮忙时，浏览器 viewer 的价值会非常直接。
Input lock stops the AI from typing over the operator when the shell is in a sensitive moment. That matters more than usual on machines where a single stray input can alter remote state.输入锁能在 shell 进入敏感时刻时阻止 AI 抢输入。对那些一次误输入就可能改变远端状态的机器，这个能力尤其重要。
Async command tracking lets the agent launch long work and poll it explicitly instead of pretending every remote task behaves like a short shell command.异步命令跟踪让 Agent 能拉起长任务并显式轮询，而不是假装所有远程任务都像短命令一样马上结束。
Policy rules let operators stay in `safe` mode by default and loosen behavior only when the environment, workflow, and trust model actually justify it.策略规则让操作者默认停在 `safe` 模式，只在环境、流程和信任模型真的允许时再放宽行为。
Local demo mode is useful when a team wants to validate the human/AI collaboration model before wiring credentials, hosts, and policy into a real environment.当团队想先验证人机协作模型，再去接入真实凭证、主机和策略时，本地演示模式会很有帮助。
From here, the next step is usually install steps, direct answers, or official links.从这里继续，下一步通常就是去看安装、答案或官方链接。