Extend loop-engineering with an optional multi-agent execution path in which the long-lived main actor continues to own intake, research, the clarifying interview, plan authoring, human plan approval, judgment, GitHub delivery, scheduled check-ins, independent-review orchestration, and merge readiness, while an implementation-capable subagent may own Code + Verify and mechanical review-finding fixes.
The skill must remain portable across agent harnesses and must not require a harness-specific agent definition. It should resolve an implementation worker from agents the current project, user environment, or harness already exposes. If no compatible worker is available, it must preserve the current single-agent flow without weakening any planning, verification, review, or reporting gate.
The primary reader of this issue is a separate agent session that will plan and implement the change through Loop Engineering without access to the design conversation that produced this issue.
This initial impl