mad-teams

Parallel agent execution requires standardized team templates and dispatch protocols to prevent file conflicts, duplicated work, and inconsistent synthesis. Automated decision logic evaluates task count, file ownership, and estimated duration to manage team composition and lifecycle safely.

Parallel agent teams are suited for scenarios with three or more independent tasks, disjoint file ownership, and estimated durations exceeding thirty minutes. Automated smart dispatch protocols evaluate these requirements and gracefully fall back to sequential subagent execution when conditions are not met.

Cost optimization in multi-agent workflows uses built-in model selection guidance, assigning Sonnet for teammates and Opus for leads. This hierarchical model selection reduces operational costs by up to 40% while maintaining output quality during parallel execution.

Setting up a parallel research swarm involves selecting from six pre-configured team templates like Research Swarm or Review Board. The lead agent automatically spawns parallel researchers to investigate multiple technical topics and coordinates a review board to validate findings.

Automated task dispatch for complex development workflows is supported through smart dispatch protocols. The system applies team dispatch decision rules, manages agent lifecycles, optimizes costs through model selection, and provides graceful fallback to sequential execution.

Pre-configured team templates include Parallel Validation and PR Review Board patterns for multi-agent scenarios. These templates provide standardized composition patterns to automate validation and review processes across independent tasks.