task-orchestration

Model selection optimizes cost and performance by matching task complexity to model capability. Use Haiku for lightweight tasks, Sonnet for balanced work, and Opus for complex reasoning. Task-orchestration automates this choice at implementation start to minimize latency and expense.

Yes, parallelization distributes independent tasks across subagents concurrently. Task-orchestration decomposes work into code, documentation, and research subtasks, spawning parallel agents to execute them simultaneously and synthesize results, reducing total execution time.

Context budget management allocates token limits efficiently across subagents. Task-orchestration tracks and distributes context to each parallel task, preventing overflow while maximizing reasoning depth and preserving information flow between delegated work streams.

Orchestrator-first planning decomposes multi-file efforts upfront, assigning tasks to appropriately-sized models and delegating execution to parallel subagents. Task-orchestration applies this at the start of any non-trivial effort to determine scope, model choice, and delegation strategy.

Context preservation maintains information flow when delegating to subagents. Task-orchestration synthesizes results from parallel work, ensuring each subagent has sufficient context for its task and final outputs integrate cohesively into the main implementation.

Aggressive delegation minimizes main-agent load on complex, multi-dimensional tasks. Task-orchestration triggers automatic delegation when work spans multiple domains or files, creating subagents for specialized tasks and reducing bottlenecks in the primary execution path.