model-orchestrationlisted

# Model Orchestration Apply intelligence where it most improves a verified, usable outcome. Optimize whole-task cost and delay, including context duplication, coordination, verification, and likely rework, while preserving required quality and authority. These are allocation principles, not a mandatory multi-agent workflow. Explicit project requirements remain binding. ## 1. Keep ownership with useful context The agent holding the user's intent and relevant project context owns the integrated result. It may plan, draft, implement, run checks, and refine directly. Roles describe responsibilities; they do not require separate workers. Keep tightly coupled work together when a handoff would mostly make another agent reconstruct the same understanding. Completing the task in one capable context is a valid orchestration choice, not a failure to delegate. Treat model capability, reasoning effort, and context count as separate choices. Where supported, changing model or effort within an existing context may be more useful than spawning another agent. Reuse applicable routing decisions rather than reopening them for every tool call or phase. ## 2. Put capability at the bottleneck Use stronger reasoning for unresolved design choices, difficult diagnosis, cross-boundary behavior, and consequential mistakes. Choose sufficient capability early when uncertainty or recovery cost warrants it; there is no need to exhaust weaker models first. A stronger model doing the work directly ma