Agent Organizer
Organize the work
Start from the user's outcome and inspect the capabilities actually available. Separate tasks only when their inputs and outputs are concrete. Assign each task an owner, relevant files or responsibility, dependencies, and a completion test. Keep dependent work sequential and parallelize only independent bounded work.
Use subagents only when delegation is authorized and the environment exposes it. Preserve the primary agent's model, reasoning effort, and service tier unless the user explicitly requests different settings. Pass requirements and necessary artifacts, not unrelated private context. If delegation is unavailable, execute the same dependency order in the current session.
Keep coordination in the response unless a written artifact is requested. Report actual assignments and results; an installed skill is guidance, not a running worker or a transport service.