swarm-plan

Task decomposition for parallel execution breaks an approved specification into one task per component with explicit file ownership. This generates a dependency-ordered swarm plan that assigns deterministic execution waves for conflict-free parallel dispatch across multiple contributors.

Dependency graph planning prevents write conflicts by grouping tasks into execution waves using disjoint write sets. Each component task receives explicit file ownership, ensuring independent work runs in parallel without collisions across the validated swarm plan.

A swarm workflow plan is a dependency-ordered execution structure that splits complex work safely across multiple contributors. You need one for multi-component implementation efforts requiring one task per component, explicit write sets, and parallelizable execution waves before dispatch.

Generating a parallel task plan requires an approved specification as input to decompose it into a dependency-ordered swarm plan. The specification must be approved first to ensure the resulting task decomposition accurately reflects validated acceptance criteria and component boundaries.

The dependency-ordered swarm plan outputs a validated JSON file saved in the .claude/state/swarm directory. This plan file contains dependency references, explicit file ownership write sets, and deterministic wave assignment structured for conflict-free dispatch.

Execution waves for multi-module refactoring require disjoint write sets, meaning tasks with overlapping file dependencies cannot run in the same wave. This limits parallelism when components share significant file ownership, forcing sequential wave execution for conflict avoidance.