swarm

Parallel task execution with Claude agents is achieved by spawning isolated workers with fresh context for each task, enabling scalable completion of multiple issues simultaneously. The skill orchestrates these agents using native coordination primitives and optional validation scripts.

Wave-based task orchestration spawns a new team per execution wave to ensure fresh context. Pre-assigned tasks are distributed by a mayor before spawning, which prevents race conditions and enables deterministic work across parallel agents.

Distributed parallel agent orchestration requires tmux and Agent Mail to support persistent, isolated workers. Local mode does not need these dependencies and instead relies on Claude-native task tooling with fast background workers.

To set up local mode for spawning parallel autonomous workers, run the /swarm command. Local mode uses Claude-native task tooling with fast background workers, requiring no additional environment setup beyond the base Claude integration.

Task validation and automated retries are handled by allowing workers to report progress and issues during parallel execution. Validation checks gate the completion of tasks, ensuring deterministic work before a wave is finalized.

You should avoid using isolated agent spawning if your workflow requires shared context between tasks, as each agent operates with a fresh context. Additionally, distributed mode is unsuitable if your environment lacks tmux and Agent Mail support.