spawn

To run parallel agent tasks in isolated environments, launch a specialized agent runtime that applies workspace scoping and identity resolution. This isolates concerns by assigning subtasks to purpose-built agents across subprocess, tmux, or docker backends.

Agent orchestration coordinates specialized runtimes to perform focused subtasks, while workspace scoping isolates their access to specific directories and resources. It enforces identity resolution, budget allocation, and communication channels across execution backends.

Yes, you can spawn agents in Docker or tmux for visual execution. The runtime supports multiple backends including subprocess for headless execution, alongside tmux and docker, allowing you to choose the environment matching your task isolation and monitoring requirements.

To monitor agent heartbeat and manage task delivery, the runtime provides built-in inbox and heartbeat monitoring channels alongside task delivery mechanisms. These features ensure active coordination and track specialized agent progress within scoped workspaces.

The best way to allocate budgets to specialized agents is through the runtime's built-in budget control feature, which enforces resource limits during task orchestration. This ensures spawned agents operate within defined constraints while maintaining isolated workspace access.

You should not use dynamic agent spawning for simple sequential tasks lacking workspace isolation or specialized identity needs. It is designed for parallelizing complex subtasks, isolating distinct responsibilities, and building purpose-built specialists requiring dedicated communication channels.