What problem does it solve? Teams running multiple local AI models waste strong-model capacity on routine work and have no signal when a turn stalls. This Skill defines which model tier handles each turn type and requires the agent to speak up when a turn runs far longer than usual. ## Core Features & Use Cases - Tiered Model Routing: Maps turn types (planning, judging, drafting, wrap-up, formatting) to three tiers — strongest_local, wizard_default, and fastest_local — with defined fallbacks. - Cost Discipline: Sends routine drafting and formatting to the cheapest capable model, reserving the strongest model for planning and judging where errors are costliest. - Slow-Turn Warning: When a turn exceeds twice the median wall-clock time of the last ten turns of the same type on the same model, the agent must say so in one line and offer to hand over partial work. - Use Case: An orchestration server assigns a formatting turn to the smallest local model; when the turn drags past twice its historical median, the agent tells the operator it is taking longer than usual and offers to stop or continue. ## Quick Start Apply the model tier table to route each turn type to the cheapest capable model and announce any turn exceeding twice its median duration.
