ls-max-parallel

Task decomposition for parallel execution breaks a request into independent subtasks, maps dependencies, and executes them concurrently in waves. This Skill orchestrates that split, spawns parallel agent calls, and merges outcomes into a single coherent result.

Wave-based execution maps dependencies, spawns parallel agent Task calls, and validates completeness before merging results across recursion depth. This approach runs independent subtasks concurrently in ordered waves to accelerate complex workflows.

An ensemble strategy executes the same task across multiple models to compare outputs. This Skill supports ensemble comparison alongside recursive and sweep strategies, allowing you to prioritize quality or breadth during parallel execution.

Parallel decomposition applies to codebase refactoring and research scenarios by executing independent exploration components in concurrent waves. It produces a unified refactor plan with completeness checks to ensure thorough end-to-end coverage.

Merging parallel outputs requires a mandatory synthesis step that merges findings, flags contradictions, and validates completeness. This Skill always runs final merged results to ensure all independent component outputs form a coherent whole.

Recursive parallelization should be avoided when subtasks have heavy dependencies preventing concurrent execution. If independent components cannot run in wave order or outputs cannot be merged, the orchestration overhead outweighs the benefits.