The Route Command

You've built the multi-agent architecture — the orchestrator and subagents are defined. But how do you actually trigger a multi-agent workflow from a normal conversation? The answer is the route command, and it's beautifully simple.

Add one line to your main agents.md: "When I say 'route content-pipeline', hand off the task to the orchestrator at projects/content-pipeline/orchestrator.md." That's it. You've just connected your conversational interface to your multi-agent workflow.

The route command is a trigger. You're in a normal conversation with your agent, and you say: "route content-pipeline: write a blog post about AI memory systems." Your agent reads the orchestrator file, switches into orchestration mode, and begins delegating to subagents. The research subagent fires up, does its work, returns results. The writer subagent takes over, produces the draft. The orchestrator compiles and returns the finished product to you.

New workflow = new folder + new route. Want to add a lead research pipeline? Create projects/lead-research/, add the orchestrator and subagent files, then add a route line: "When I say 'route lead-research', use projects/lead-research/orchestrator.md." You now have two independent multi-agent workflows, both accessible with a single command.

This pattern scales indefinitely. Each route is independent. Each workflow has its own orchestrator and subagents. They don't interfere with each other. Your main agents.md stays clean — just a list of route commands pointing to project directories. The complexity lives in the project folders, not in your primary agent configuration.

The elegance of this design is that your default agent stays simple and focused for general tasks. Multi-agent workflows are invoked explicitly when you need them. You get the best of both worlds: a quick, responsive agent for simple tasks and powerful multi-agent pipelines for complex work.