HN 提问：你们是如何运行代理群的？
1 分•作者: sermakarevich•20 天前
我经历了一些演变，从带有 `/clear` 和“执行下一个任务”提示的待办事项列表，到清理大部分技能/插件/claudemd，使用集中的 beads 数据库，在 SQLite 上开发 ask_human mcp，并创建了一个带有 UI 的轻量级 Python 编排器：https://github.com/sermakarevich/fleet。
我的用例包括：
* 用于并行启动不相关单个任务的集中式接口
* 采用规范驱动开发方法进行编码：https://github.com/sermakarevich/sddw - 我可以使用 beads --deps 链接 sdd 工作流步骤
* 代码库审查并触发大量小型任务并行执行
几篇文章：
* https://simonwillison.net/2025/Oct/5/parallel-coding-agents/
* https://zachwills.net/i-managed-a-swarm-of-20-ai-agents-for-a-week-here-are-the-8-rules-i-learned/
* https://juhache.substack.com/p/on-the-road-to-agent-swarms
您今天如何运行代理集群？有哪些用例？
查看原文
I passed through some evolution, from todo list with /clear + "take next task" prompt to cleaning most of the skills/plugins/claudemd, using centralized beads db, developing ask_human mcp on sqllite and creating lightweight python orchestrator with ui: https://github.com/sermakarevich/fleet.<p>My use cases are:<p>- centralized interface for spawning unrelated single tasks in parallel<p>- coding with spec driven development approach: https://github.com/sermakarevich/sddw - I can chain sdd workflow steps with beads --deps<p>- reviewing codebase and triggering many small tasks to be executed in parallel<p>Few articles:<p>- https://simonwillison.net/2025/Oct/5/parallel-coding-agents/<p>- https://zachwills.net/i-managed-a-swarm-of-20-ai-agents-for-a-week-here-are-the-8-rules-i-learned/<p>- https://juhache.substack.com/p/on-the-road-to-agent-swarms<p>How do you run agent swarm today? What are the use cases?