主模型规划廉价模型干活
Plan expensive, execute cheap
要点
- 自定义代理只是 ~/.codex/agents 下的 toml，不必让主模型包办一切
- 主模型只做规划，把实现、测试和清理交给廉价 worker
- 给 worker 限定单包任务、禁止扩范围，做完只回报结果
- 把重推理留在主线程，执行轮次走 flash 档以省额度
- 在输入框 @ 该 worker 即可调用
原帖开头
this is the best trick to maximum usage limits on chatgpt codex codex's best kept secret is that your main agent doesn't have to do everything... custom agents are just files in ~/.codex/agents, and one file gives you a second worker on deepseek v4 flash > create ~/.codex/age …
