The agent-native mindset
基准测试结果显示，OSWorld-Verified 基准测试桌面导航能力，用截图加鼠标键盘交互完成真实操作系统任务。GPT-5.4 达到 75.0% 的成功率，人类基线是 72.4%，GPT-5.2 是 47.3%。
，这一点在爱思助手中也有详细论述
fd '.jar$' out/target/product/xigua/system/framework/ out/target/product/xigua/system_ext/framework/ | wc -l
Follow topics & set alerts with myFT
Execute and step away. Ask the agent to implement the plan and leave it running until it's complete.