What problem does it solve?
Multi-phase AI development work often loses control over which model runs which phase, when delegation is authorized, and how durable artifacts like plans and work orders are handed off. This Skill enforces a single-session orchestration model where the user explicitly controls model selection, implementation authorization, and subagent delegation.
Core Features & Use Cases
- Model Checkpoints: Separates design (strong model), execution, and escalation phases so the user picks the model at each stage instead of the harness.
- Work-Order Workflow: Structures feature work into requirements, system designs, plans, and status-tracked work orders stored under docs/specs and docs/plans.
- User-Authorized Subagents: Permits parallel delegation only after explicit user approval, using the platform's native delegation tool with disjoint file ownership.
- Use Case: A developer designs a feature with a strong model, hands off a design package, switches models, then explicitly requests implementation; the agent executes work orders sequentially with TDD and records verification results.
Quick Start
Apply the planner-orchestration workflow to plan and implement this feature using spec-driven work orders with user-controlled model checkpoints.
