What problem does it solve? Deciding whether to delegate work to a subagent, which subagent to pick, and which Claude model tier to assign is error-prone and inconsistent. This Skill gives the main agent explicit routing rules so delegation decisions are deliberate, cost-aware, and bounded, while the main agent stays the orchestrator and prompt-loop owner. ## Core Features & Use Cases - Model-tier routing: Assign cheaper tiers (haiku) to mechanical implementation and stronger tiers (opus) to planning, architecture, and safety-critical work, escalating only when needed. - Subagent selection: Map tasks to generic subagents (Explore, Plan, general-purpose, claude-code-guide) or repo custom agents such as tier/coder agents and domain specialists like test-specialist or docs-maintainer. - Bounded retry and background execution: Enforce follow-up-before-relaunch limits, ban recursive re-spawning loops, and run independent work via background subagents with a structured prompt template. - Use Case: When facing a multi-file refactor, dispatch a mid-tier coder subagent with a full context pack (objective, scope, constraints, validation commands), then review its output and own the final handoff instead of grinding through the change inline. ## Quick Start Ask the agent to decide whether this task should be delegated to a subagent and which model tier fits, then have it dispatch the work with a complete context pack.
