multi-agent-cn in Practice: Building Your AI Team with One Skill

What Is multi-agent-cn?

multi-agent-cn is a ClawHub skill implementing a dispatcher-executor pattern: one coordinator agent breaks down tasks and delegates to persistent specialist sub-agents. Instead of one AI doing everything, you get a coordinated team.

Core Architecture

The skill provisions five persistent sub-agents by default. Each has a role, name, and personality defined in the skill config. The dispatcher (main session) receives requests, breaks them down, and routes subtasks to appropriate specialists. The key: sub-agents are persistent across sessions. They accumulate memory and develop specialized knowledge over time.

What Actually Works

Parallel execution: Independent tasks get dispatched simultaneously. Research and code review happen at the same time. On complex projects, elapsed time drops significantly.

Specialization over time: Sub-agents handling certain task types consistently develop better judgment for those tasks. Our QA agent has a long checklist it wouldn't have known about on day one.

Accountability: When something goes wrong, it's traceable. The dispatcher log shows what was assigned to whom and when.

What Needs Careful Handling

Task decomposition: Ambiguous tasks lead to confused delegations. We spent significant time tuning the dispatcher prompts for our workflow.

State sync: When multiple sub-agents work in parallel on related tasks, they can step on each other. We added explicit handoff protocols — sub-agents announce completion and wait for coordinator acknowledgment before the next batch starts.