Orchestrator Skill

Task decomposition breaks multi-thousand-line specs into atomic tasks under 100 LOC each, with ~500 lines of context per task. This prevents AI hallucinations by narrowing focus and establishing explicit dependencies, ensuring deterministic code generation across discovery, architecture, and specification phases.

Atomic task decomposition with explicit dependencies and acceptance criteria prevents hallucinations by constraining AI context to ~500 lines and limiting task scope to under 100 LOC. Phase-ordered execution and dependency mapping ensure logical implementation order and correct behavior.

Yes. Task decomposition maps BDD scenarios to Domain-Driven Design components, generating tasks that address business behavior and architectural patterns. This integration ensures tasks align with both behavioral requirements and domain structure.

Decomposition automatically establishes explicit task dependencies and outputs a `tasks.md` file with phase metadata and context-loading guidance. This ordering ensures tasks execute in a logically correct sequence that respects architectural constraints.

Use decomposition for features requiring extensive changes, large specifications (5000+ lines), or high-complexity requirements where AI focus and correctness are critical. It's especially valuable when implementation spans multiple phases or involves multiple architectural components.

Decomposition produces `changes/[change-id]/tasks.md` with phase-ordered atomic tasks, explicit dependencies, acceptance criteria, and context constraints. Each task includes guidance for loading relevant code sections, enabling step-by-step implementation with minimal AI misinterpretation risk.