请问HN：你是如何管理你的代理群体的？
我经历了一些演变，从使用待办事项列表和“获取下一个任务”的提示，到清理大部分技能/插件/云数据库，使用集中式的数据库，开发基于SQLite的ask_human多通道处理器，并创建了一个带有用户界面的轻量级Python协调器： https://github.com/sermakarevich/fleet。
我的使用案例包括：
- 集中式界面，用于并行生成不相关的单一任务
- 采用规范驱动开发的方法进行编码： https://github.com/sermakarevich/sddw - 我可以使用beads将sdd工作流步骤链接起来 --deps
- 审查代码库并触发多个小任务并行执行
几篇相关文章：
- 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
你今天是如何运行代理群体的？有哪些使用案例？
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?