{
  "text": "OpenAI最近为Codex最新版\n也就是0.145.0版\n推出了全新的Multi-Agent V2\n也就是更加稳定的多智能体系统\n这次升级之后\n我们终于可以在Codex中\n更加灵活的使用和管理Sub-Agent\n目前Multi-Agent V2已经进入相对稳定阶段\n它不仅支持为不同的子Agent\n配置不同的模型和推理层级\n还可以控制并发数量\n当任务恢复之后\n也能快速继续保持原来的角色分工\n它还提供了更加清晰的agent导航和管理能力\n这意味着Codex正从过去的单个超级agent\n逐渐升级成为一个真正的agent调度系统\n它并不是简单的启动几个subagent\n让他们各自完成各自的任务\n而是让一个主agent充当整个任务的调度者\n主agent可以先理解和拆解任务\n再把代码探索\n方案设计\n功能开发\n测试验证代码审查等不同工作\n分别交给不同的subagent\n等这些subagent完成各自的任务之后\n主agent还会统一收集\n判断并且合并他们的执行结果\n最终形成完整的解决方案\n更值得关注的是\nCodex现在这种工作方式\n在设计思路上\n已经越来越接近Cloud Code的\nDynamic Workflow功能\n而且最近Graph Engineering\n这个概念在AI编程领域非常火\n像过去我们更多是在研究如何写好一条prompt\n或者是让一个agent在loop中不断执行\n检查和修复\n而graph engineering关注的是更高一层的问题\n也就是如何把多个agent和多个工具\n多个验证点以及不同的反馈循环\n组织成一张可以实际执行的任务图\n像cloudcode的dynamic workflows\n本身就是graph engineering的一种典型实现\n它可以根据用户的目标来生成一套工作流\n再通过多个subagent条件分制\n并行任务和验证点\n完成更加复杂的任务\n而这一次Codex通过multiagent的v2\n让用户能够更加自由的构建和配置不同的subagent\n虽然目前它还不能完全等同于\nCloudcode的dynamic workflows\n也还不是一套完整的graph workflow运行时\n但从主agent的调度\n角色分工\n并行之行和结果汇总这些能力来看\nCodex的Multi-Agent V2已经具备了非常鲜明的Graph Engineering特征\n本期视频就为大家深度演示Codex中它新增的Multi-Agent V2这个功能\n它在多种场景中的使用方式\n以及如何配置才能让Codex的Multi-Agent V2更加接近Graph Engineering\n好 想使用这个新特性非常简单\n我们直接将Codex升级到最新版\n我们只需要在中轮命令行中\n执行codex update这条命令\n就可以将我们的codex cri升级到最新版\n在codex桌面版的最新版中\n我们同样可以使用multiagent v2这个新特性\n为了让大家能够更直观的感受到\n在codex中subagent它的灵活性有多强\n在这里我先让codex列出\n我已经配置好的subagent\n然后我们再调用subagent来启动一个任务进行测试\n好,Codex这里他列出了我配置好的这4个subagent\n第1个subagent他的任务是code review\n然后我为他设置的模型是kimi k3\n第2个subagent他的使用场景是ui产品设计与评审\n然后我为他设置的模型也是kimi k3\n然后第3个subagent他的任务也是代码审查\n然后我为他设置的模型是minimax模型\n然后第4个subagent他的任务也是代码审查\n然后我为他设置的模型是gpt5.6soul模型\n下方键的话 我们在codex中就可以来调用 我们已经配置好的这些subagent\n也可以直接让codex根据任务的复杂度来自动派生不同的subagent\n在codex中这里的主agent 我使用的模型就是gbt5.6soul模型\n然后我们就可以来测试一下 让主agent来调用 我们已经配置好的subagent来执行任务\n这里我输入的提示词是 让他用这三个subagent对代码进行对抗选择\n然后我们就直接发送 当我们发送这个任务之后 这个任务会被当前的主agent接受 也就是我们设置的gbt5.6送模型\n然后这个主agent就会调用我们刚才配置好的这三个subagent 对我们的项目代码进行对抗性审查\n好 这里codex提示三个审查轨道都已经启动\n在这里我们就可以看到正在运行的这三个subagent\n然后我们可以点击第一个进行查看\n好 第一个subagent 他这里正在运行 已经运行了一分钟\n然后我们再点击查看第二个agent 这个agent调用的是minimax模型\n我们点击 好 这里我们就可以看到他正在执行\n然后我们再点击查看第三个subagent 第三个subagent\n我是让他单独调用的pyagent来执行的代码审查任务\n也就是说 在codex的这些subagent中 我们不仅可以为这些subagent配置非openai官方的模型\n比如说第三方的这些kimi模型或者是mimax等模型\n我们还能让subagent来调用不同的这些工具 比如说这里调用pyagent\n像这样的话 我们就可以将不同任务分配给不同的模型\n这样就可以更加节省codex中gpt5.6 它的token消耗\n好 再等了几分钟之后 这里就提示审查完成 而且还发现了这些严重的bug\n而且我们还可以点击查看这些不同subagent他们的输出结果\n这里是这个使用了minimax模型的subagent 然后这是它的输出结果\n这里是发现了比较严重的代码中的问题 它输出的这些内容都非常详细\n然后这三个subagent他们的执行结果最后都会被主agent进行整理和分析 并给出最终的结论\n像这样的话我们就完成了在codex中调用不同的subagent来完成代码的对抗性审查\n然后我们可以看一下这些subagent他们的配置方式\n我们可以直接在codex中点击查看\n这里是这个subagent它的完整路径\n它会放在codex的agents这个文件夹中\n然后这是这个文件名\n这里就是这个subagent它的名称\n然后这里就是描述\n也就是使用kmmk3模型进行代码审查\n然后这里就是具体的提示词\n在下面的这个模型这里\n我们给它设置的就是KimiK3\n因为我是使用的CCSwitch\n所以在模型提供者这里\n就是设置的CCSwitch\n在模型的推理级别这里\n我设置的是Hi\n相见的话我们就可以在Codex中\n用我们配置的Subagent来调用\n第三方的模型\n然后我们可以打开CCSwitch\n在这里我们可以看到Codex\n我们直接点击这个Codex\n然后在这里我们就看到\n我添加了KimiK3\n然后我们点击编辑可以看一下我是如何配置的\n然后这里就是模型提供者的这个名称\n这里是Kimi的URL\n然后这里就是API key\n在这里就是Kimi他的API的链接\n然后这里这个默认模型就是Kimi key3模型\n在prompt cache这里我设置成了enable\n在模型映射这里就对应Kimi key3\n像这样的话我们就可以通过CC switch这个项目\n将KIMI的API转换为支持Codex的API的格式\n我们就可以在Codex的Subagent中来调用这个模型\n也可以将主Agent模型改成第三方模型\n但我这里还是默认使用的GPT5.6SOM模型\n因为为主模型设置一个更加强大的模型\n它在调度和分配这些Subagent的时候\n会执行的更加精准和高效\n然后我们再看一下这个Subagent\n看一下它的配置\n然后再下\n然后在下面的参数这里这里就是模型名称\n这里就是模型提供商\n这里就是他的思考级别\n因为minimax模型他本身就兼容codex的这种ap格式\n所以在这里就不需要为minimax的这些API通过ccswitch进行转换了\n因为我们刚才调用的kimi的API\n它是不兼容codex的API格式的\n所以我们刚才需要用ccswitch进行转换\n但minimax的API它兼容codex\n所以我们就不需要进行转换就可以直接调用\n然后我们再看一下这个subagent\n它是调用了py agent给它设置的模型\n就是gpt5.6soul模型\n然后给他设置的推理级别是high\n相见的话\n我们在codex中就分别设置了几个不同的subagent\n并且让他们调用了不同的模型\n甚至还可以让他们调用不同的工具\n比如说这个subagent中让他调用了py agent\n而且在这个UI产品设计的subagent中\n我还让他调用了scale\n也就是让他使用了superdesign这个scale\n在codex的插件市场\n我们就可以看到这个superdesign\n像这些subagent我们就可以手动创建\n也可以让codex直接帮我们创建\n然后我们输入任务\n让它创建一个用于漏洞扫描的subagent\n要求这个subagent使用深度安全扫描这个scale\n然后模型使用gpd5.6so\n思考级别设为extra high\n然后我们点击发送\n好这里提示创建完成\n而且它创建的这个subagent现在正在运行\n它正在分析我们的项目代码库\n然后我们还可以让codex在执行任务的时候动态的分配subagent\n我输入的任务是派生5个subagent对代码进行对抗审查\n分别调用不同的gpt系列模型以及kimi模型和mimax模型\n然后我们就发送让codex动态的派生不同的subagent为我们执行任务\n然后我们就看到他动态派生了这几个subagent\n我们可以先点开看一下第一个subagent\n好 第一个sub agent 因为调用的是kimi k3模型 然后调用次数过多 这里被限制了 然后我们再看一下第二个 第二个sub agent 使用的是minimax的模型 然后他正在执行 没有遇到速率限制\n然后我们再看一下 这个使用gpd5.6soul模型的这个sub agent 好 这里他正在执行\n然后我们再继续看 这是另一个sub agent 这里也正在执行\n相见的话我们就在codex中\n让codex动态的派生了不同的subagent来同时执行任务\n而且codex他支持subagent内嵌subagent\n为了节省时间这里就不再为大家去演示了\n通过我们刚才的测试\n可以发现codex的multiagentv2\n这个多subagent的功能已经非常成熟\n而且非常灵活\n我们既可以手动设置这些subagent\n也可以让codex动态派生不同的subagent\n而且我们还可以为这一些不同的subagent\n分配不同的模型\n还能设置他们的思考级别\nCodex它变得也越来越灵活\n在多agent方面的能力也越来越强大\n甚至已经具备graph engineering的初级范式\n在很大程度上\n我们甚至可以使用Codex\n来完全替代cloudcode\n好 本期视频就做到这里\n欢迎大家点赞 关注和转发\n谢谢大家观看",
  "language": null,
  "duration": 688.21625,
  "segments": [
    {
      "start": 0.0,
      "end": 4.92,
      "text": "OpenAI最近为Codex最新版",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.09060935129093219,
      "no_speech_prob": 2.1192469654152113e-11,
      "compression_ratio": 1.1866666666666668
    },
    {
      "start": 4.92,
      "end": 7.5,
      "text": "也就是0.145.0版",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.09060935129093219,
      "no_speech_prob": 2.1192469654152113e-11,
      "compression_ratio": 1.1866666666666668
    },
    {
      "start": 7.5,
      "end": 10.6,
      "text": "推出了全新的Multi-Agent V2",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.09060935129093219,
      "no_speech_prob": 2.1192469654152113e-11,
      "compression_ratio": 1.1866666666666668
    },
    {
      "start": 10.6,
      "end": 13.58,
      "text": "也就是更加稳定的多智能体系统",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.09060935129093219,
      "no_speech_prob": 2.1192469654152113e-11,
      "compression_ratio": 1.1866666666666668
    },
    {
      "start": 13.58,
      "end": 14.74,
      "text": "这次升级之后",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.09060935129093219,
      "no_speech_prob": 2.1192469654152113e-11,
      "compression_ratio": 1.1866666666666668
    },
    {
      "start": 14.74,
      "end": 16.68,
      "text": "我们终于可以在Codex中",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.09060935129093219,
      "no_speech_prob": 2.1192469654152113e-11,
      "compression_ratio": 1.1866666666666668
    },
    {
      "start": 16.68,
      "end": 19.48,
      "text": "更加灵活的使用和管理Sub-Agent",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.09060935129093219,
      "no_speech_prob": 2.1192469654152113e-11,
      "compression_ratio": 1.1866666666666668
    },
    {
      "start": 19.48,
      "end": 23.44,
      "text": "目前Multi-Agent V2已经进入相对稳定阶段",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.09060935129093219,
      "no_speech_prob": 2.1192469654152113e-11,
      "compression_ratio": 1.1866666666666668
    },
    {
      "start": 23.44,
      "end": 25.82,
      "text": "它不仅支持为不同的子Agent",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.09060935129093219,
      "no_speech_prob": 2.1192469654152113e-11,
      "compression_ratio": 1.1866666666666668
    },
    {
      "start": 25.82,
      "end": 28.2,
      "text": "配置不同的模型和推理层级",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.09060935129093219,
      "no_speech_prob": 2.1192469654152113e-11,
      "compression_ratio": 1.1866666666666668
    },
    {
      "start": 28.2,
      "end": 29.98,
      "text": "还可以控制并发数量",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.09060935129093219,
      "no_speech_prob": 2.1192469654152113e-11,
      "compression_ratio": 1.1866666666666668
    },
    {
      "start": 29.98,
      "end": 31.28,
      "text": "当任务恢复之后",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06016272336698426,
      "no_speech_prob": 1.9529305256282825e-11,
      "compression_ratio": 1.3140243902439024
    },
    {
      "start": 31.28,
      "end": 34.26,
      "text": "也能快速继续保持原来的角色分工",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06016272336698426,
      "no_speech_prob": 1.9529305256282825e-11,
      "compression_ratio": 1.3140243902439024
    },
    {
      "start": 34.26,
      "end": 38.06,
      "text": "它还提供了更加清晰的agent导航和管理能力",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06016272336698426,
      "no_speech_prob": 1.9529305256282825e-11,
      "compression_ratio": 1.3140243902439024
    },
    {
      "start": 38.06,
      "end": 41.92,
      "text": "这意味着Codex正从过去的单个超级agent",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06016272336698426,
      "no_speech_prob": 1.9529305256282825e-11,
      "compression_ratio": 1.3140243902439024
    },
    {
      "start": 41.92,
      "end": 45.64,
      "text": "逐渐升级成为一个真正的agent调度系统",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06016272336698426,
      "no_speech_prob": 1.9529305256282825e-11,
      "compression_ratio": 1.3140243902439024
    },
    {
      "start": 45.64,
      "end": 48.3,
      "text": "它并不是简单的启动几个subagent",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06016272336698426,
      "no_speech_prob": 1.9529305256282825e-11,
      "compression_ratio": 1.3140243902439024
    },
    {
      "start": 48.3,
      "end": 50.66,
      "text": "让他们各自完成各自的任务",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06016272336698426,
      "no_speech_prob": 1.9529305256282825e-11,
      "compression_ratio": 1.3140243902439024
    },
    {
      "start": 50.66,
      "end": 53.9,
      "text": "而是让一个主agent充当整个任务的调度者",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06016272336698426,
      "no_speech_prob": 1.9529305256282825e-11,
      "compression_ratio": 1.3140243902439024
    },
    {
      "start": 53.9,
      "end": 56.76,
      "text": "主agent可以先理解和拆解任务",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06016272336698426,
      "no_speech_prob": 1.9529305256282825e-11,
      "compression_ratio": 1.3140243902439024
    },
    {
      "start": 56.76,
      "end": 57.8,
      "text": "再把代码探索",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06016272336698426,
      "no_speech_prob": 1.9529305256282825e-11,
      "compression_ratio": 1.3140243902439024
    },
    {
      "start": 57.8,
      "end": 58.48,
      "text": "方案设计",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06016272336698426,
      "no_speech_prob": 1.9529305256282825e-11,
      "compression_ratio": 1.3140243902439024
    },
    {
      "start": 58.48,
      "end": 59.26,
      "text": "功能开发",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06016272336698426,
      "no_speech_prob": 1.9529305256282825e-11,
      "compression_ratio": 1.3140243902439024
    },
    {
      "start": 59.26,
      "end": 61.98,
      "text": "测试验证代码审查等不同工作",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.09120666481063751,
      "no_speech_prob": 1.84096349276075e-11,
      "compression_ratio": 1.1589595375722543
    },
    {
      "start": 61.98,
      "end": 64.56,
      "text": "分别交给不同的subagent",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.09120666481063751,
      "no_speech_prob": 1.84096349276075e-11,
      "compression_ratio": 1.1589595375722543
    },
    {
      "start": 64.56,
      "end": 67.68,
      "text": "等这些subagent完成各自的任务之后",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.09120666481063751,
      "no_speech_prob": 1.84096349276075e-11,
      "compression_ratio": 1.1589595375722543
    },
    {
      "start": 67.68,
      "end": 69.68,
      "text": "主agent还会统一收集",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.09120666481063751,
      "no_speech_prob": 1.84096349276075e-11,
      "compression_ratio": 1.1589595375722543
    },
    {
      "start": 69.68,
      "end": 72.26,
      "text": "判断并且合并他们的执行结果",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.09120666481063751,
      "no_speech_prob": 1.84096349276075e-11,
      "compression_ratio": 1.1589595375722543
    },
    {
      "start": 72.26,
      "end": 74.88,
      "text": "最终形成完整的解决方案",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.09120666481063751,
      "no_speech_prob": 1.84096349276075e-11,
      "compression_ratio": 1.1589595375722543
    },
    {
      "start": 74.88,
      "end": 75.96,
      "text": "更值得关注的是",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.09120666481063751,
      "no_speech_prob": 1.84096349276075e-11,
      "compression_ratio": 1.1589595375722543
    },
    {
      "start": 75.96,
      "end": 78.26,
      "text": "Codex现在这种工作方式",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.09120666481063751,
      "no_speech_prob": 1.84096349276075e-11,
      "compression_ratio": 1.1589595375722543
    },
    {
      "start": 78.26,
      "end": 79.28,
      "text": "在设计思路上",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.09120666481063751,
      "no_speech_prob": 1.84096349276075e-11,
      "compression_ratio": 1.1589595375722543
    },
    {
      "start": 79.28,
      "end": 81.5,
      "text": "已经越来越接近Cloud Code的",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.09120666481063751,
      "no_speech_prob": 1.84096349276075e-11,
      "compression_ratio": 1.1589595375722543
    },
    {
      "start": 81.5,
      "end": 83.28,
      "text": "Dynamic Workflow功能",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.09120666481063751,
      "no_speech_prob": 1.84096349276075e-11,
      "compression_ratio": 1.1589595375722543
    },
    {
      "start": 83.28,
      "end": 85.44,
      "text": "而且最近Graph Engineering",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.09120666481063751,
      "no_speech_prob": 1.84096349276075e-11,
      "compression_ratio": 1.1589595375722543
    },
    {
      "start": 85.44,
      "end": 88.24,
      "text": "这个概念在AI编程领域非常火",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.09120666481063751,
      "no_speech_prob": 1.84096349276075e-11,
      "compression_ratio": 1.1589595375722543
    },
    {
      "start": 88.24,
      "end": 92.02,
      "text": "像过去我们更多是在研究如何写好一条prompt",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07474837984357562,
      "no_speech_prob": 1.417616371512942e-11,
      "compression_ratio": 1.2666666666666666
    },
    {
      "start": 92.02,
      "end": 95.34,
      "text": "或者是让一个agent在loop中不断执行",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07474837984357562,
      "no_speech_prob": 1.417616371512942e-11,
      "compression_ratio": 1.2666666666666666
    },
    {
      "start": 95.34,
      "end": 96.98,
      "text": "检查和修复",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07474837984357562,
      "no_speech_prob": 1.417616371512942e-11,
      "compression_ratio": 1.2666666666666666
    },
    {
      "start": 96.98,
      "end": 100.28,
      "text": "而graph engineering关注的是更高一层的问题",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07474837984357562,
      "no_speech_prob": 1.417616371512942e-11,
      "compression_ratio": 1.2666666666666666
    },
    {
      "start": 100.28,
      "end": 102.88,
      "text": "也就是如何把多个agent和多个工具",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07474837984357562,
      "no_speech_prob": 1.417616371512942e-11,
      "compression_ratio": 1.2666666666666666
    },
    {
      "start": 102.88,
      "end": 105.78,
      "text": "多个验证点以及不同的反馈循环",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07474837984357562,
      "no_speech_prob": 1.417616371512942e-11,
      "compression_ratio": 1.2666666666666666
    },
    {
      "start": 105.78,
      "end": 109.0,
      "text": "组织成一张可以实际执行的任务图",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07474837984357562,
      "no_speech_prob": 1.417616371512942e-11,
      "compression_ratio": 1.2666666666666666
    },
    {
      "start": 109.0,
      "end": 110.9,
      "text": "像cloudcode的dynamic workflows",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07474837984357562,
      "no_speech_prob": 1.417616371512942e-11,
      "compression_ratio": 1.2666666666666666
    },
    {
      "start": 110.9,
      "end": 114.42,
      "text": "本身就是graph engineering的一种典型实现",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07474837984357562,
      "no_speech_prob": 1.417616371512942e-11,
      "compression_ratio": 1.2666666666666666
    },
    {
      "start": 114.42,
      "end": 117.84,
      "text": "它可以根据用户的目标来生成一套工作流",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07474837984357562,
      "no_speech_prob": 1.417616371512942e-11,
      "compression_ratio": 1.2666666666666666
    },
    {
      "start": 117.84,
      "end": 120.78,
      "text": "再通过多个subagent条件分制",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.11247910261154175,
      "no_speech_prob": 1.5150531870733452e-11,
      "compression_ratio": 1.2229299363057324
    },
    {
      "start": 120.78,
      "end": 122.68,
      "text": "并行任务和验证点",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.11247910261154175,
      "no_speech_prob": 1.5150531870733452e-11,
      "compression_ratio": 1.2229299363057324
    },
    {
      "start": 122.68,
      "end": 124.56,
      "text": "完成更加复杂的任务",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.11247910261154175,
      "no_speech_prob": 1.5150531870733452e-11,
      "compression_ratio": 1.2229299363057324
    },
    {
      "start": 124.56,
      "end": 127.84,
      "text": "而这一次Codex通过multiagent的v2",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.11247910261154175,
      "no_speech_prob": 1.5150531870733452e-11,
      "compression_ratio": 1.2229299363057324
    },
    {
      "start": 127.84,
      "end": 132.4,
      "text": "让用户能够更加自由的构建和配置不同的subagent",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.11247910261154175,
      "no_speech_prob": 1.5150531870733452e-11,
      "compression_ratio": 1.2229299363057324
    },
    {
      "start": 132.4,
      "end": 135.4,
      "text": "虽然目前它还不能完全等同于",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.11247910261154175,
      "no_speech_prob": 1.5150531870733452e-11,
      "compression_ratio": 1.2229299363057324
    },
    {
      "start": 135.4,
      "end": 137.08,
      "text": "Cloudcode的dynamic workflows",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.11247910261154175,
      "no_speech_prob": 1.5150531870733452e-11,
      "compression_ratio": 1.2229299363057324
    },
    {
      "start": 137.08,
      "end": 140.72,
      "text": "也还不是一套完整的graph workflow运行时",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.11247910261154175,
      "no_speech_prob": 1.5150531870733452e-11,
      "compression_ratio": 1.2229299363057324
    },
    {
      "start": 140.72,
      "end": 142.7,
      "text": "但从主agent的调度",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.11247910261154175,
      "no_speech_prob": 1.5150531870733452e-11,
      "compression_ratio": 1.2229299363057324
    },
    {
      "start": 142.7,
      "end": 143.6,
      "text": "角色分工",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.11247910261154175,
      "no_speech_prob": 1.5150531870733452e-11,
      "compression_ratio": 1.2229299363057324
    },
    {
      "start": 143.6,
      "end": 146.76,
      "text": "并行之行和结果汇总这些能力来看",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.11247910261154175,
      "no_speech_prob": 1.5150531870733452e-11,
      "compression_ratio": 1.2229299363057324
    },
    {
      "start": 146.76,
      "end": 152.62,
      "text": "Codex的Multi-Agent V2已经具备了非常鲜明的Graph Engineering特征",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.11395221516705942,
      "no_speech_prob": 2.99932891023591e-11,
      "compression_ratio": 1.2553956834532374
    },
    {
      "start": 152.62,
      "end": 159.36,
      "text": "本期视频就为大家深度演示Codex中它新增的Multi-Agent V2这个功能",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.11395221516705942,
      "no_speech_prob": 2.99932891023591e-11,
      "compression_ratio": 1.2553956834532374
    },
    {
      "start": 159.36,
      "end": 161.76,
      "text": "它在多种场景中的使用方式",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.11395221516705942,
      "no_speech_prob": 2.99932891023591e-11,
      "compression_ratio": 1.2553956834532374
    },
    {
      "start": 161.76,
      "end": 168.32,
      "text": "以及如何配置才能让Codex的Multi-Agent V2更加接近Graph Engineering",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.11395221516705942,
      "no_speech_prob": 2.99932891023591e-11,
      "compression_ratio": 1.2553956834532374
    },
    {
      "start": 168.32,
      "end": 170.8,
      "text": "好 想使用这个新特性非常简单",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.11395221516705942,
      "no_speech_prob": 2.99932891023591e-11,
      "compression_ratio": 1.2553956834532374
    },
    {
      "start": 170.8,
      "end": 173.74,
      "text": "我们直接将Codex升级到最新版",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.11395221516705942,
      "no_speech_prob": 2.99932891023591e-11,
      "compression_ratio": 1.2553956834532374
    },
    {
      "start": 173.74,
      "end": 175.9,
      "text": "我们只需要在中轮命令行中",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.11529455429468399,
      "no_speech_prob": 2.536048454571116e-11,
      "compression_ratio": 1.3084415584415585
    },
    {
      "start": 175.9,
      "end": 178.14,
      "text": "执行codex update这条命令",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.11529455429468399,
      "no_speech_prob": 2.536048454571116e-11,
      "compression_ratio": 1.3084415584415585
    },
    {
      "start": 178.14,
      "end": 181.92,
      "text": "就可以将我们的codex cri升级到最新版",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.11529455429468399,
      "no_speech_prob": 2.536048454571116e-11,
      "compression_ratio": 1.3084415584415585
    },
    {
      "start": 181.92,
      "end": 184.58,
      "text": "在codex桌面版的最新版中",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.11529455429468399,
      "no_speech_prob": 2.536048454571116e-11,
      "compression_ratio": 1.3084415584415585
    },
    {
      "start": 184.58,
      "end": 188.32,
      "text": "我们同样可以使用multiagent v2这个新特性",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.11529455429468399,
      "no_speech_prob": 2.536048454571116e-11,
      "compression_ratio": 1.3084415584415585
    },
    {
      "start": 188.32,
      "end": 190.7,
      "text": "为了让大家能够更直观的感受到",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.11529455429468399,
      "no_speech_prob": 2.536048454571116e-11,
      "compression_ratio": 1.3084415584415585
    },
    {
      "start": 190.7,
      "end": 193.94,
      "text": "在codex中subagent它的灵活性有多强",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.11529455429468399,
      "no_speech_prob": 2.536048454571116e-11,
      "compression_ratio": 1.3084415584415585
    },
    {
      "start": 193.94,
      "end": 196.04,
      "text": "在这里我先让codex列出",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.11529455429468399,
      "no_speech_prob": 2.536048454571116e-11,
      "compression_ratio": 1.3084415584415585
    },
    {
      "start": 196.04,
      "end": 197.84,
      "text": "我已经配置好的subagent",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.11529455429468399,
      "no_speech_prob": 2.536048454571116e-11,
      "compression_ratio": 1.3084415584415585
    },
    {
      "start": 197.84,
      "end": 202.54,
      "text": "然后我们再调用subagent来启动一个任务进行测试",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.11529455429468399,
      "no_speech_prob": 2.536048454571116e-11,
      "compression_ratio": 1.3084415584415585
    },
    {
      "start": 202.54,
      "end": 206.04,
      "text": "好,Codex这里他列出了我配置好的这4个subagent",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.1769591905794091,
      "no_speech_prob": 2.323824292715937e-11,
      "compression_ratio": 1.8438818565400843
    },
    {
      "start": 206.34,
      "end": 208.74,
      "text": "第1个subagent他的任务是code review",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.1769591905794091,
      "no_speech_prob": 2.323824292715937e-11,
      "compression_ratio": 1.8438818565400843
    },
    {
      "start": 208.94,
      "end": 211.14,
      "text": "然后我为他设置的模型是kimi k3",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.1769591905794091,
      "no_speech_prob": 2.323824292715937e-11,
      "compression_ratio": 1.8438818565400843
    },
    {
      "start": 211.44,
      "end": 215.44,
      "text": "第2个subagent他的使用场景是ui产品设计与评审",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.1769591905794091,
      "no_speech_prob": 2.323824292715937e-11,
      "compression_ratio": 1.8438818565400843
    },
    {
      "start": 215.74,
      "end": 217.94,
      "text": "然后我为他设置的模型也是kimi k3",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.1769591905794091,
      "no_speech_prob": 2.323824292715937e-11,
      "compression_ratio": 1.8438818565400843
    },
    {
      "start": 218.24,
      "end": 221.44,
      "text": "然后第3个subagent他的任务也是代码审查",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.1769591905794091,
      "no_speech_prob": 2.323824292715937e-11,
      "compression_ratio": 1.8438818565400843
    },
    {
      "start": 221.54,
      "end": 224.14,
      "text": "然后我为他设置的模型是minimax模型",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.1769591905794091,
      "no_speech_prob": 2.323824292715937e-11,
      "compression_ratio": 1.8438818565400843
    },
    {
      "start": 224.44,
      "end": 227.94,
      "text": "然后第4个subagent他的任务也是代码审查",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.1769591905794091,
      "no_speech_prob": 2.323824292715937e-11,
      "compression_ratio": 1.8438818565400843
    },
    {
      "start": 228.24,
      "end": 231.44,
      "text": "然后我为他设置的模型是gpt5.6soul模型",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.1769591905794091,
      "no_speech_prob": 2.323824292715937e-11,
      "compression_ratio": 1.8438818565400843
    },
    {
      "start": 231.44,
      "end": 237.14,
      "text": "下方键的话 我们在codex中就可以来调用 我们已经配置好的这些subagent",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.1767153263092041,
      "no_speech_prob": 2.5043703222316083e-11,
      "compression_ratio": 1.4573378839590443
    },
    {
      "start": 237.14,
      "end": 242.44,
      "text": "也可以直接让codex根据任务的复杂度来自动派生不同的subagent",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.1767153263092041,
      "no_speech_prob": 2.5043703222316083e-11,
      "compression_ratio": 1.4573378839590443
    },
    {
      "start": 242.44,
      "end": 248.24,
      "text": "在codex中这里的主agent 我使用的模型就是gbt5.6soul模型",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.1767153263092041,
      "no_speech_prob": 2.5043703222316083e-11,
      "compression_ratio": 1.4573378839590443
    },
    {
      "start": 248.24,
      "end": 254.74,
      "text": "然后我们就可以来测试一下 让主agent来调用 我们已经配置好的subagent来执行任务",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.1767153263092041,
      "no_speech_prob": 2.5043703222316083e-11,
      "compression_ratio": 1.4573378839590443
    },
    {
      "start": 254.74,
      "end": 260.54,
      "text": "这里我输入的提示词是 让他用这三个subagent对代码进行对抗选择",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.1767153263092041,
      "no_speech_prob": 2.5043703222316083e-11,
      "compression_ratio": 1.4573378839590443
    },
    {
      "start": 260.54,
      "end": 269.76,
      "text": "然后我们就直接发送 当我们发送这个任务之后 这个任务会被当前的主agent接受 也就是我们设置的gbt5.6送模型",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.1776452099835431,
      "no_speech_prob": 3.464889139492833e-11,
      "compression_ratio": 1.4052044609665428
    },
    {
      "start": 270.02,
      "end": 277.78,
      "text": "然后这个主agent就会调用我们刚才配置好的这三个subagent 对我们的项目代码进行对抗性审查",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.1776452099835431,
      "no_speech_prob": 3.464889139492833e-11,
      "compression_ratio": 1.4052044609665428
    },
    {
      "start": 278.2,
      "end": 281.58,
      "text": "好 这里codex提示三个审查轨道都已经启动",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.1776452099835431,
      "no_speech_prob": 3.464889139492833e-11,
      "compression_ratio": 1.4052044609665428
    },
    {
      "start": 282.04,
      "end": 285.38,
      "text": "在这里我们就可以看到正在运行的这三个subagent",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.1776452099835431,
      "no_speech_prob": 3.464889139492833e-11,
      "compression_ratio": 1.4052044609665428
    },
    {
      "start": 285.38,
      "end": 288.18,
      "text": "然后我们可以点击第一个进行查看",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.11732603002477575,
      "no_speech_prob": 3.069747928075017e-11,
      "compression_ratio": 1.5732217573221758
    },
    {
      "start": 288.18,
      "end": 292.34,
      "text": "好 第一个subagent 他这里正在运行 已经运行了一分钟",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.11732603002477575,
      "no_speech_prob": 3.069747928075017e-11,
      "compression_ratio": 1.5732217573221758
    },
    {
      "start": 292.34,
      "end": 297.18,
      "text": "然后我们再点击查看第二个agent 这个agent调用的是minimax模型",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.11732603002477575,
      "no_speech_prob": 3.069747928075017e-11,
      "compression_ratio": 1.5732217573221758
    },
    {
      "start": 297.18,
      "end": 301.3,
      "text": "我们点击 好 这里我们就可以看到他正在执行",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.11732603002477575,
      "no_speech_prob": 3.069747928075017e-11,
      "compression_ratio": 1.5732217573221758
    },
    {
      "start": 301.3,
      "end": 305.74,
      "text": "然后我们再点击查看第三个subagent 第三个subagent",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.11732603002477575,
      "no_speech_prob": 3.069747928075017e-11,
      "compression_ratio": 1.5732217573221758
    },
    {
      "start": 305.74,
      "end": 310.7,
      "text": "我是让他单独调用的pyagent来执行的代码审查任务",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.11732603002477575,
      "no_speech_prob": 3.069747928075017e-11,
      "compression_ratio": 1.5732217573221758
    },
    {
      "start": 310.7,
      "end": 318.7,
      "text": "也就是说 在codex的这些subagent中 我们不仅可以为这些subagent配置非openai官方的模型",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.1689793005921787,
      "no_speech_prob": 2.1080302434195453e-11,
      "compression_ratio": 1.3846153846153846
    },
    {
      "start": 318.7,
      "end": 322.9,
      "text": "比如说第三方的这些kimi模型或者是mimax等模型",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.1689793005921787,
      "no_speech_prob": 2.1080302434195453e-11,
      "compression_ratio": 1.3846153846153846
    },
    {
      "start": 322.9,
      "end": 328.5,
      "text": "我们还能让subagent来调用不同的这些工具 比如说这里调用pyagent",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.1689793005921787,
      "no_speech_prob": 2.1080302434195453e-11,
      "compression_ratio": 1.3846153846153846
    },
    {
      "start": 328.5,
      "end": 332.9,
      "text": "像这样的话 我们就可以将不同任务分配给不同的模型",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.1689793005921787,
      "no_speech_prob": 2.1080302434195453e-11,
      "compression_ratio": 1.3846153846153846
    },
    {
      "start": 332.9,
      "end": 337.7,
      "text": "这样就可以更加节省codex中gpt5.6 它的token消耗",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.1689793005921787,
      "no_speech_prob": 2.1080302434195453e-11,
      "compression_ratio": 1.3846153846153846
    },
    {
      "start": 337.7,
      "end": 343.4,
      "text": "好 再等了几分钟之后 这里就提示审查完成 而且还发现了这些严重的bug",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.09961740413825669,
      "no_speech_prob": 2.4179616642250323e-11,
      "compression_ratio": 1.4376996805111821
    },
    {
      "start": 343.4,
      "end": 348.1,
      "text": "而且我们还可以点击查看这些不同subagent他们的输出结果",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.09961740413825669,
      "no_speech_prob": 2.4179616642250323e-11,
      "compression_ratio": 1.4376996805111821
    },
    {
      "start": 348.1,
      "end": 353.44,
      "text": "这里是这个使用了minimax模型的subagent 然后这是它的输出结果",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.09961740413825669,
      "no_speech_prob": 2.4179616642250323e-11,
      "compression_ratio": 1.4376996805111821
    },
    {
      "start": 353.44,
      "end": 358.78,
      "text": "这里是发现了比较严重的代码中的问题 它输出的这些内容都非常详细",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.09961740413825669,
      "no_speech_prob": 2.4179616642250323e-11,
      "compression_ratio": 1.4376996805111821
    },
    {
      "start": 358.78,
      "end": 366.82,
      "text": "然后这三个subagent他们的执行结果最后都会被主agent进行整理和分析 并给出最终的结论",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.09961740413825669,
      "no_speech_prob": 2.4179616642250323e-11,
      "compression_ratio": 1.4376996805111821
    },
    {
      "start": 366.82,
      "end": 374.36,
      "text": "像这样的话我们就完成了在codex中调用不同的subagent来完成代码的对抗性审查",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.12114373031927615,
      "no_speech_prob": 2.8953006658838376e-11,
      "compression_ratio": 1.5035714285714286
    },
    {
      "start": 374.36,
      "end": 378.06,
      "text": "然后我们可以看一下这些subagent他们的配置方式",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.12114373031927615,
      "no_speech_prob": 2.8953006658838376e-11,
      "compression_ratio": 1.5035714285714286
    },
    {
      "start": 378.06,
      "end": 380.58,
      "text": "我们可以直接在codex中点击查看",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.12114373031927615,
      "no_speech_prob": 2.8953006658838376e-11,
      "compression_ratio": 1.5035714285714286
    },
    {
      "start": 380.58,
      "end": 382.8,
      "text": "这里是这个subagent它的完整路径",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.12114373031927615,
      "no_speech_prob": 2.8953006658838376e-11,
      "compression_ratio": 1.5035714285714286
    },
    {
      "start": 382.8,
      "end": 386.22,
      "text": "它会放在codex的agents这个文件夹中",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.12114373031927615,
      "no_speech_prob": 2.8953006658838376e-11,
      "compression_ratio": 1.5035714285714286
    },
    {
      "start": 386.22,
      "end": 387.86,
      "text": "然后这是这个文件名",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.12114373031927615,
      "no_speech_prob": 2.8953006658838376e-11,
      "compression_ratio": 1.5035714285714286
    },
    {
      "start": 387.86,
      "end": 390.52,
      "text": "这里就是这个subagent它的名称",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.12114373031927615,
      "no_speech_prob": 2.8953006658838376e-11,
      "compression_ratio": 1.5035714285714286
    },
    {
      "start": 390.52,
      "end": 392.22,
      "text": "然后这里就是描述",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.12114373031927615,
      "no_speech_prob": 2.8953006658838376e-11,
      "compression_ratio": 1.5035714285714286
    },
    {
      "start": 392.22,
      "end": 395.28,
      "text": "也就是使用kmmk3模型进行代码审查",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.12114373031927615,
      "no_speech_prob": 2.8953006658838376e-11,
      "compression_ratio": 1.5035714285714286
    },
    {
      "start": 395.28,
      "end": 397.36,
      "text": "然后这里就是具体的提示词",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.13831754316363418,
      "no_speech_prob": 3.106596230262326e-11,
      "compression_ratio": 1.5593220338983051
    },
    {
      "start": 397.36,
      "end": 399.28,
      "text": "在下面的这个模型这里",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.13831754316363418,
      "no_speech_prob": 3.106596230262326e-11,
      "compression_ratio": 1.5593220338983051
    },
    {
      "start": 399.28,
      "end": 401.36,
      "text": "我们给它设置的就是KimiK3",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.13831754316363418,
      "no_speech_prob": 3.106596230262326e-11,
      "compression_ratio": 1.5593220338983051
    },
    {
      "start": 401.36,
      "end": 403.36,
      "text": "因为我是使用的CCSwitch",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.13831754316363418,
      "no_speech_prob": 3.106596230262326e-11,
      "compression_ratio": 1.5593220338983051
    },
    {
      "start": 403.36,
      "end": 405.18,
      "text": "所以在模型提供者这里",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.13831754316363418,
      "no_speech_prob": 3.106596230262326e-11,
      "compression_ratio": 1.5593220338983051
    },
    {
      "start": 405.18,
      "end": 406.82,
      "text": "就是设置的CCSwitch",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.13831754316363418,
      "no_speech_prob": 3.106596230262326e-11,
      "compression_ratio": 1.5593220338983051
    },
    {
      "start": 406.82,
      "end": 408.7,
      "text": "在模型的推理级别这里",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.13831754316363418,
      "no_speech_prob": 3.106596230262326e-11,
      "compression_ratio": 1.5593220338983051
    },
    {
      "start": 408.7,
      "end": 409.94,
      "text": "我设置的是Hi",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.13831754316363418,
      "no_speech_prob": 3.106596230262326e-11,
      "compression_ratio": 1.5593220338983051
    },
    {
      "start": 409.94,
      "end": 412.34,
      "text": "相见的话我们就可以在Codex中",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.13831754316363418,
      "no_speech_prob": 3.106596230262326e-11,
      "compression_ratio": 1.5593220338983051
    },
    {
      "start": 412.34,
      "end": 415.02,
      "text": "用我们配置的Subagent来调用",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.13831754316363418,
      "no_speech_prob": 3.106596230262326e-11,
      "compression_ratio": 1.5593220338983051
    },
    {
      "start": 415.02,
      "end": 416.26,
      "text": "第三方的模型",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.13831754316363418,
      "no_speech_prob": 3.106596230262326e-11,
      "compression_ratio": 1.5593220338983051
    },
    {
      "start": 416.26,
      "end": 418.52,
      "text": "然后我们可以打开CCSwitch",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.13831754316363418,
      "no_speech_prob": 3.106596230262326e-11,
      "compression_ratio": 1.5593220338983051
    },
    {
      "start": 418.52,
      "end": 420.34,
      "text": "在这里我们可以看到Codex",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.13831754316363418,
      "no_speech_prob": 3.106596230262326e-11,
      "compression_ratio": 1.5593220338983051
    },
    {
      "start": 420.34,
      "end": 422.02,
      "text": "我们直接点击这个Codex",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.13831754316363418,
      "no_speech_prob": 3.106596230262326e-11,
      "compression_ratio": 1.5593220338983051
    },
    {
      "start": 422.02,
      "end": 423.72,
      "text": "然后在这里我们就看到",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.13831754316363418,
      "no_speech_prob": 3.106596230262326e-11,
      "compression_ratio": 1.5593220338983051
    },
    {
      "start": 423.72,
      "end": 425.2,
      "text": "我添加了KimiK3",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.13831754316363418,
      "no_speech_prob": 3.106596230262326e-11,
      "compression_ratio": 1.5593220338983051
    },
    {
      "start": 425.2,
      "end": 428.56,
      "text": "然后我们点击编辑可以看一下我是如何配置的",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.13111705349800282,
      "no_speech_prob": 2.3911213287153288e-11,
      "compression_ratio": 1.3597122302158273
    },
    {
      "start": 428.56,
      "end": 431.16,
      "text": "然后这里就是模型提供者的这个名称",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.13111705349800282,
      "no_speech_prob": 2.3911213287153288e-11,
      "compression_ratio": 1.3597122302158273
    },
    {
      "start": 431.16,
      "end": 432.92,
      "text": "这里是Kimi的URL",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.13111705349800282,
      "no_speech_prob": 2.3911213287153288e-11,
      "compression_ratio": 1.3597122302158273
    },
    {
      "start": 432.92,
      "end": 434.66,
      "text": "然后这里就是API key",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.13111705349800282,
      "no_speech_prob": 2.3911213287153288e-11,
      "compression_ratio": 1.3597122302158273
    },
    {
      "start": 434.66,
      "end": 438.26,
      "text": "在这里就是Kimi他的API的链接",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.13111705349800282,
      "no_speech_prob": 2.3911213287153288e-11,
      "compression_ratio": 1.3597122302158273
    },
    {
      "start": 438.26,
      "end": 441.1,
      "text": "然后这里这个默认模型就是Kimi key3模型",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.13111705349800282,
      "no_speech_prob": 2.3911213287153288e-11,
      "compression_ratio": 1.3597122302158273
    },
    {
      "start": 441.1,
      "end": 443.94,
      "text": "在prompt cache这里我设置成了enable",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.13111705349800282,
      "no_speech_prob": 2.3911213287153288e-11,
      "compression_ratio": 1.3597122302158273
    },
    {
      "start": 443.94,
      "end": 447.08,
      "text": "在模型映射这里就对应Kimi key3",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.13111705349800282,
      "no_speech_prob": 2.3911213287153288e-11,
      "compression_ratio": 1.3597122302158273
    },
    {
      "start": 447.08,
      "end": 451.12,
      "text": "像这样的话我们就可以通过CC switch这个项目",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.13111705349800282,
      "no_speech_prob": 2.3911213287153288e-11,
      "compression_ratio": 1.3597122302158273
    },
    {
      "start": 451.12,
      "end": 455.74,
      "text": "将KIMI的API转换为支持Codex的API的格式",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.13359007492564082,
      "no_speech_prob": 1.892394747848858e-11,
      "compression_ratio": 1.3333333333333333
    },
    {
      "start": 455.74,
      "end": 460.12,
      "text": "我们就可以在Codex的Subagent中来调用这个模型",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.13359007492564082,
      "no_speech_prob": 1.892394747848858e-11,
      "compression_ratio": 1.3333333333333333
    },
    {
      "start": 460.12,
      "end": 463.06,
      "text": "也可以将主Agent模型改成第三方模型",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.13359007492564082,
      "no_speech_prob": 1.892394747848858e-11,
      "compression_ratio": 1.3333333333333333
    },
    {
      "start": 463.06,
      "end": 466.0,
      "text": "但我这里还是默认使用的GPT5.6SOM模型",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.13359007492564082,
      "no_speech_prob": 1.892394747848858e-11,
      "compression_ratio": 1.3333333333333333
    },
    {
      "start": 466.0,
      "end": 469.62,
      "text": "因为为主模型设置一个更加强大的模型",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.13359007492564082,
      "no_speech_prob": 1.892394747848858e-11,
      "compression_ratio": 1.3333333333333333
    },
    {
      "start": 469.62,
      "end": 472.68,
      "text": "它在调度和分配这些Subagent的时候",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.13359007492564082,
      "no_speech_prob": 1.892394747848858e-11,
      "compression_ratio": 1.3333333333333333
    },
    {
      "start": 472.68,
      "end": 475.08,
      "text": "会执行的更加精准和高效",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.13359007492564082,
      "no_speech_prob": 1.892394747848858e-11,
      "compression_ratio": 1.3333333333333333
    },
    {
      "start": 475.08,
      "end": 477.74,
      "text": "然后我们再看一下这个Subagent",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.13359007492564082,
      "no_speech_prob": 1.892394747848858e-11,
      "compression_ratio": 1.3333333333333333
    },
    {
      "start": 477.74,
      "end": 479.5,
      "text": "看一下它的配置",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.13359007492564082,
      "no_speech_prob": 1.892394747848858e-11,
      "compression_ratio": 1.3333333333333333
    },
    {
      "start": 479.5,
      "end": 480.0,
      "text": "然后再下",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.13359007492564082,
      "no_speech_prob": 1.892394747848858e-11,
      "compression_ratio": 1.3333333333333333
    },
    {
      "start": 480.0,
      "end": 482.16,
      "text": "然后在下面的参数这里这里就是模型名称",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.176933707260504,
      "no_speech_prob": 4.789818194850248e-11,
      "compression_ratio": 1.367816091954023
    },
    {
      "start": 482.4,
      "end": 483.94,
      "text": "这里就是模型提供商",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.176933707260504,
      "no_speech_prob": 4.789818194850248e-11,
      "compression_ratio": 1.367816091954023
    },
    {
      "start": 484.2,
      "end": 486.0,
      "text": "这里就是他的思考级别",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.176933707260504,
      "no_speech_prob": 4.789818194850248e-11,
      "compression_ratio": 1.367816091954023
    },
    {
      "start": 486.24,
      "end": 490.86,
      "text": "因为minimax模型他本身就兼容codex的这种ap格式",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.176933707260504,
      "no_speech_prob": 4.789818194850248e-11,
      "compression_ratio": 1.367816091954023
    },
    {
      "start": 490.86,
      "end": 496.72,
      "text": "所以在这里就不需要为minimax的这些API通过ccswitch进行转换了",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.12087770512229518,
      "no_speech_prob": 2.3981599692191047e-11,
      "compression_ratio": 1.416058394160584
    },
    {
      "start": 496.72,
      "end": 499.22,
      "text": "因为我们刚才调用的kimi的API",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.12087770512229518,
      "no_speech_prob": 2.3981599692191047e-11,
      "compression_ratio": 1.416058394160584
    },
    {
      "start": 499.22,
      "end": 502.26,
      "text": "它是不兼容codex的API格式的",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.12087770512229518,
      "no_speech_prob": 2.3981599692191047e-11,
      "compression_ratio": 1.416058394160584
    },
    {
      "start": 502.26,
      "end": 505.3,
      "text": "所以我们刚才需要用ccswitch进行转换",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.12087770512229518,
      "no_speech_prob": 2.3981599692191047e-11,
      "compression_ratio": 1.416058394160584
    },
    {
      "start": 505.3,
      "end": 508.02,
      "text": "但minimax的API它兼容codex",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.12087770512229518,
      "no_speech_prob": 2.3981599692191047e-11,
      "compression_ratio": 1.416058394160584
    },
    {
      "start": 508.02,
      "end": 510.9,
      "text": "所以我们就不需要进行转换就可以直接调用",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.12087770512229518,
      "no_speech_prob": 2.3981599692191047e-11,
      "compression_ratio": 1.416058394160584
    },
    {
      "start": 510.9,
      "end": 513.6,
      "text": "然后我们再看一下这个subagent",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.12087770512229518,
      "no_speech_prob": 2.3981599692191047e-11,
      "compression_ratio": 1.416058394160584
    },
    {
      "start": 513.6,
      "end": 516.72,
      "text": "它是调用了py agent给它设置的模型",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.12087770512229518,
      "no_speech_prob": 2.3981599692191047e-11,
      "compression_ratio": 1.416058394160584
    },
    {
      "start": 516.72,
      "end": 519.24,
      "text": "就是gpt5.6soul模型",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.12087770512229518,
      "no_speech_prob": 2.3981599692191047e-11,
      "compression_ratio": 1.416058394160584
    },
    {
      "start": 519.24,
      "end": 522.02,
      "text": "然后给他设置的推理级别是high",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08870569659817604,
      "no_speech_prob": 2.2626187382024376e-11,
      "compression_ratio": 1.480565371024735
    },
    {
      "start": 522.02,
      "end": 523.02,
      "text": "相见的话",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08870569659817604,
      "no_speech_prob": 2.2626187382024376e-11,
      "compression_ratio": 1.480565371024735
    },
    {
      "start": 523.02,
      "end": 527.02,
      "text": "我们在codex中就分别设置了几个不同的subagent",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08870569659817604,
      "no_speech_prob": 2.2626187382024376e-11,
      "compression_ratio": 1.480565371024735
    },
    {
      "start": 527.02,
      "end": 529.54,
      "text": "并且让他们调用了不同的模型",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08870569659817604,
      "no_speech_prob": 2.2626187382024376e-11,
      "compression_ratio": 1.480565371024735
    },
    {
      "start": 529.54,
      "end": 533.1,
      "text": "甚至还可以让他们调用不同的工具",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08870569659817604,
      "no_speech_prob": 2.2626187382024376e-11,
      "compression_ratio": 1.480565371024735
    },
    {
      "start": 533.1,
      "end": 536.54,
      "text": "比如说这个subagent中让他调用了py agent",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08870569659817604,
      "no_speech_prob": 2.2626187382024376e-11,
      "compression_ratio": 1.480565371024735
    },
    {
      "start": 536.54,
      "end": 539.7,
      "text": "而且在这个UI产品设计的subagent中",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08870569659817604,
      "no_speech_prob": 2.2626187382024376e-11,
      "compression_ratio": 1.480565371024735
    },
    {
      "start": 539.7,
      "end": 541.28,
      "text": "我还让他调用了scale",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08870569659817604,
      "no_speech_prob": 2.2626187382024376e-11,
      "compression_ratio": 1.480565371024735
    },
    {
      "start": 541.28,
      "end": 544.6,
      "text": "也就是让他使用了superdesign这个scale",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08870569659817604,
      "no_speech_prob": 2.2626187382024376e-11,
      "compression_ratio": 1.480565371024735
    },
    {
      "start": 544.6,
      "end": 546.48,
      "text": "在codex的插件市场",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08870569659817604,
      "no_speech_prob": 2.2626187382024376e-11,
      "compression_ratio": 1.480565371024735
    },
    {
      "start": 546.48,
      "end": 548.82,
      "text": "我们就可以看到这个superdesign",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08870569659817604,
      "no_speech_prob": 2.2626187382024376e-11,
      "compression_ratio": 1.480565371024735
    },
    {
      "start": 548.82,
      "end": 551.82,
      "text": "像这些subagent我们就可以手动创建",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.12374022548184073,
      "no_speech_prob": 2.871641986701423e-11,
      "compression_ratio": 1.3581081081081081
    },
    {
      "start": 551.82,
      "end": 554.04,
      "text": "也可以让codex直接帮我们创建",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.12374022548184073,
      "no_speech_prob": 2.871641986701423e-11,
      "compression_ratio": 1.3581081081081081
    },
    {
      "start": 554.04,
      "end": 555.4,
      "text": "然后我们输入任务",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.12374022548184073,
      "no_speech_prob": 2.871641986701423e-11,
      "compression_ratio": 1.3581081081081081
    },
    {
      "start": 555.4,
      "end": 559.06,
      "text": "让它创建一个用于漏洞扫描的subagent",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.12374022548184073,
      "no_speech_prob": 2.871641986701423e-11,
      "compression_ratio": 1.3581081081081081
    },
    {
      "start": 559.06,
      "end": 562.96,
      "text": "要求这个subagent使用深度安全扫描这个scale",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.12374022548184073,
      "no_speech_prob": 2.871641986701423e-11,
      "compression_ratio": 1.3581081081081081
    },
    {
      "start": 562.96,
      "end": 565.36,
      "text": "然后模型使用gpd5.6so",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.12374022548184073,
      "no_speech_prob": 2.871641986701423e-11,
      "compression_ratio": 1.3581081081081081
    },
    {
      "start": 565.36,
      "end": 567.48,
      "text": "思考级别设为extra high",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.12374022548184073,
      "no_speech_prob": 2.871641986701423e-11,
      "compression_ratio": 1.3581081081081081
    },
    {
      "start": 567.48,
      "end": 568.98,
      "text": "然后我们点击发送",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.12374022548184073,
      "no_speech_prob": 2.871641986701423e-11,
      "compression_ratio": 1.3581081081081081
    },
    {
      "start": 568.98,
      "end": 571.28,
      "text": "好这里提示创建完成",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.12374022548184073,
      "no_speech_prob": 2.871641986701423e-11,
      "compression_ratio": 1.3581081081081081
    },
    {
      "start": 571.28,
      "end": 574.82,
      "text": "而且它创建的这个subagent现在正在运行",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.12374022548184073,
      "no_speech_prob": 2.871641986701423e-11,
      "compression_ratio": 1.3581081081081081
    },
    {
      "start": 574.82,
      "end": 577.44,
      "text": "它正在分析我们的项目代码库",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.12374022548184073,
      "no_speech_prob": 2.871641986701423e-11,
      "compression_ratio": 1.3581081081081081
    },
    {
      "start": 577.44,
      "end": 582.36,
      "text": "然后我们还可以让codex在执行任务的时候动态的分配subagent",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.10455973836399565,
      "no_speech_prob": 2.0698996336387943e-11,
      "compression_ratio": 1.4774436090225564
    },
    {
      "start": 582.36,
      "end": 587.18,
      "text": "我输入的任务是派生5个subagent对代码进行对抗审查",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.10455973836399565,
      "no_speech_prob": 2.0698996336387943e-11,
      "compression_ratio": 1.4774436090225564
    },
    {
      "start": 587.18,
      "end": 591.74,
      "text": "分别调用不同的gpt系列模型以及kimi模型和mimax模型",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.10455973836399565,
      "no_speech_prob": 2.0698996336387943e-11,
      "compression_ratio": 1.4774436090225564
    },
    {
      "start": 591.74,
      "end": 597.86,
      "text": "然后我们就发送让codex动态的派生不同的subagent为我们执行任务",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.10455973836399565,
      "no_speech_prob": 2.0698996336387943e-11,
      "compression_ratio": 1.4774436090225564
    },
    {
      "start": 597.86,
      "end": 602.14,
      "text": "然后我们就看到他动态派生了这几个subagent",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.10455973836399565,
      "no_speech_prob": 2.0698996336387943e-11,
      "compression_ratio": 1.4774436090225564
    },
    {
      "start": 602.14,
      "end": 604.78,
      "text": "我们可以先点开看一下第一个subagent",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.10455973836399565,
      "no_speech_prob": 2.0698996336387943e-11,
      "compression_ratio": 1.4774436090225564
    },
    {
      "start": 604.78,
      "end": 618.12,
      "text": "好 第一个sub agent 因为调用的是kimi k3模型 然后调用次数过多 这里被限制了 然后我们再看一下第二个 第二个sub agent 使用的是minimax的模型 然后他正在执行 没有遇到速率限制",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.17265491915824718,
      "no_speech_prob": 3.3438390945050855e-11,
      "compression_ratio": 1.5863453815261044
    },
    {
      "start": 618.66,
      "end": 624.54,
      "text": "然后我们再看一下 这个使用gpd5.6soul模型的这个sub agent 好 这里他正在执行",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.17265491915824718,
      "no_speech_prob": 3.3438390945050855e-11,
      "compression_ratio": 1.5863453815261044
    },
    {
      "start": 625.02,
      "end": 629.1,
      "text": "然后我们再继续看 这是另一个sub agent 这里也正在执行",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.17265491915824718,
      "no_speech_prob": 3.3438390945050855e-11,
      "compression_ratio": 1.5863453815261044
    },
    {
      "start": 629.1,
      "end": 631.32,
      "text": "相见的话我们就在codex中",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.09517624727479972,
      "no_speech_prob": 1.97535564294693e-11,
      "compression_ratio": 1.4035087719298245
    },
    {
      "start": 631.32,
      "end": 636.78,
      "text": "让codex动态的派生了不同的subagent来同时执行任务",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.09517624727479972,
      "no_speech_prob": 1.97535564294693e-11,
      "compression_ratio": 1.4035087719298245
    },
    {
      "start": 636.78,
      "end": 640.56,
      "text": "而且codex他支持subagent内嵌subagent",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.09517624727479972,
      "no_speech_prob": 1.97535564294693e-11,
      "compression_ratio": 1.4035087719298245
    },
    {
      "start": 640.56,
      "end": 643.78,
      "text": "为了节省时间这里就不再为大家去演示了",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.09517624727479972,
      "no_speech_prob": 1.97535564294693e-11,
      "compression_ratio": 1.4035087719298245
    },
    {
      "start": 643.78,
      "end": 645.24,
      "text": "通过我们刚才的测试",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.09517624727479972,
      "no_speech_prob": 1.97535564294693e-11,
      "compression_ratio": 1.4035087719298245
    },
    {
      "start": 645.24,
      "end": 648.02,
      "text": "可以发现codex的multiagentv2",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.09517624727479972,
      "no_speech_prob": 1.97535564294693e-11,
      "compression_ratio": 1.4035087719298245
    },
    {
      "start": 648.02,
      "end": 650.92,
      "text": "这个多subagent的功能已经非常成熟",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.09517624727479972,
      "no_speech_prob": 1.97535564294693e-11,
      "compression_ratio": 1.4035087719298245
    },
    {
      "start": 650.92,
      "end": 652.2,
      "text": "而且非常灵活",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.09517624727479972,
      "no_speech_prob": 1.97535564294693e-11,
      "compression_ratio": 1.4035087719298245
    },
    {
      "start": 652.2,
      "end": 655.08,
      "text": "我们既可以手动设置这些subagent",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.09517624727479972,
      "no_speech_prob": 1.97535564294693e-11,
      "compression_ratio": 1.4035087719298245
    },
    {
      "start": 655.08,
      "end": 658.8,
      "text": "也可以让codex动态派生不同的subagent",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.09517624727479972,
      "no_speech_prob": 1.97535564294693e-11,
      "compression_ratio": 1.4035087719298245
    },
    {
      "start": 658.8,
      "end": 661.6,
      "text": "而且我们还可以为这一些不同的subagent",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.12058196196684966,
      "no_speech_prob": 2.1596124194500632e-11,
      "compression_ratio": 1.2268370607028753
    },
    {
      "start": 661.6,
      "end": 663.58,
      "text": "分配不同的模型",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.12058196196684966,
      "no_speech_prob": 2.1596124194500632e-11,
      "compression_ratio": 1.2268370607028753
    },
    {
      "start": 663.58,
      "end": 665.84,
      "text": "还能设置他们的思考级别",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.12058196196684966,
      "no_speech_prob": 2.1596124194500632e-11,
      "compression_ratio": 1.2268370607028753
    },
    {
      "start": 665.84,
      "end": 668.4,
      "text": "Codex它变得也越来越灵活",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.12058196196684966,
      "no_speech_prob": 2.1596124194500632e-11,
      "compression_ratio": 1.2268370607028753
    },
    {
      "start": 668.4,
      "end": 671.3,
      "text": "在多agent方面的能力也越来越强大",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.12058196196684966,
      "no_speech_prob": 2.1596124194500632e-11,
      "compression_ratio": 1.2268370607028753
    },
    {
      "start": 671.3,
      "end": 675.04,
      "text": "甚至已经具备graph engineering的初级范式",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.12058196196684966,
      "no_speech_prob": 2.1596124194500632e-11,
      "compression_ratio": 1.2268370607028753
    },
    {
      "start": 675.04,
      "end": 676.66,
      "text": "在很大程度上",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.12058196196684966,
      "no_speech_prob": 2.1596124194500632e-11,
      "compression_ratio": 1.2268370607028753
    },
    {
      "start": 676.66,
      "end": 678.0,
      "text": "我们甚至可以使用Codex",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.12058196196684966,
      "no_speech_prob": 2.1596124194500632e-11,
      "compression_ratio": 1.2268370607028753
    },
    {
      "start": 678.0,
      "end": 679.86,
      "text": "来完全替代cloudcode",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.12058196196684966,
      "no_speech_prob": 2.1596124194500632e-11,
      "compression_ratio": 1.2268370607028753
    },
    {
      "start": 679.86,
      "end": 682.08,
      "text": "好 本期视频就做到这里",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.12058196196684966,
      "no_speech_prob": 2.1596124194500632e-11,
      "compression_ratio": 1.2268370607028753
    },
    {
      "start": 682.08,
      "end": 684.6,
      "text": "欢迎大家点赞 关注和转发",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.12058196196684966,
      "no_speech_prob": 2.1596124194500632e-11,
      "compression_ratio": 1.2268370607028753
    },
    {
      "start": 684.6,
      "end": 685.82,
      "text": "谢谢大家观看",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.12058196196684966,
      "no_speech_prob": 2.1596124194500632e-11,
      "compression_ratio": 1.2268370607028753
    }
  ],
  "segmented_transcription": {
    "status": "SEGMENTED_TRANSCRIPT_OK",
    "created_at": "2026-07-23T23:42:40",
    "manifest": "/Users/sagawa/AI_WORK/video_notes/url/20260723_234117__Graph_Engineering范式_Codex_Multi-agent_V2支持Kimi_MiniMax_GPT多模型混用_动态派生subagent_并行执/segmented_work/chunk_manifest.json",
    "audio": "/Users/sagawa/AI_WORK/video_notes/url/20260723_234117__Graph_Engineering范式_Codex_Multi-agent_V2支持Kimi_MiniMax_GPT多模型混用_动态派生subagent_并行执/audio_16k_mono.wav",
    "audio_sha256": "dfda00f72fa280db70c3772d0722463934cf3ce45d797e3ecb74e673a7ddcb6b",
    "audio_duration": 688.21625,
    "model": "mlx-community/whisper-large-v3-turbo",
    "language_hint": null,
    "condition_on_previous_text": false,
    "chunk_count": 2,
    "resumed_chunks": 0,
    "merged_segment_count": 217,
    "final_segment_end": 685.82,
    "tail_repeat_run": 1,
    "tail_repeat_text": "我输入的任务是派生5个subagent对代码进行对抗审查",
    "qc_ok": true,
    "qc_errors": []
  }
}