生产级智能体三层架构
Production Hermes Agent Stack
要点
- 让 Hermes 只负责判断、目标和时机，不要把过程细节塞进智能体
- Skill 写成精简的流程说明，确定性操作交给智能体可调用的 CLI
- 必须接遥测：是否在故障、花了多少 token、是哪个智能体改了文件
- 故障信号要回灌给智能体，让它根据遥测自己修复而不是等人盯
- 关键任务没有这套闭环，上线后很难知道谁改坏了、花超在哪
原帖开头
This is how I've been building production agents lately. Hermes - owns judgement, objectives, timing, etc. Skills - procedural knowledge, not code. Still very slim CLIs - tools for the agents that are deterministic Then the important part is the telemetry. If you're running mis …
