GPT Orchestration

Use the installed GPT Engineer core when available; load its dynamic-workflow reference only for nontrivial scheduling. This is an orchestration entry point, not another routing policy or a reason to load every engineer wrapper. A Skills CLI installation does not resolve dependencies or register agents and hooks.

Preserve the selected parent (gpt-6-astra, gpt-5.6-sol, gpt-5.6-terra, or gpt-5.6-luna). Same-model delegation is explicit by default; mixed-model lanes require a deliberate policy and exact child selection. Never infer execution from profiles or model self-description. If the core or native delegation is absent, explain the limitation and perform suitable work directly; do not switch models or install a framework automatically. Claude native routing stays provider-specific.

For the requested fleet, inspect live capacity and active descendants. State lane and resource budgets. Partition independent decisions or subsystems; assign one accountable integrator. Give each lane objective, owned paths, prerequisites, authority, checks, compact output and stop condition. Preserve dirty work; keep shared files under one writer and isolate parallel candidates. Workers are not alone and must not revert others' edits.