Infra

OpenAI Agents SDK 新增原生沙箱与模型原生执行框架 The next evolution of the Agents SDK: native sandboxing and harness

OpenAI 为 Agents SDK 引入两大能力：一个更贴近模型原生模式的执行框架，让 agent 在多工具、长任务中更可靠；以及原生沙箱执行，提供安全的计算机环境供 agent 读写文件、运行代码。开发者可接入多家沙箱提供商，并通过 Manifest 抽象保持环境可移植。 OpenAI adds native sandbox execution and a more capable model-native harness to the Agents SDK. Agents can now run in controlled compute environments with file I/O, code execution, and tool use. The update includes a Manifest abstraction for portability across sandbox providers and intends to improve reliability on complex tasks.