What problem does it solve? Choosing which model tier (haiku, sonnet, opus) to run a skill on is guesswork without measurement. This Skill automates the search for the cheapest tier that still meets a quality bar, so you avoid overpaying for expensive models or under-serving with weak ones. ## Core Features & Use Cases - Tier-by-tier climbing: Runs the variant sweep at each model tier, cheapest first, and stops as soon as a tier produces a stable winner at or above the success bar. - Strict graduation rule: A tier only counts as sufficient when the winner's score meets the bar and the outcome is "winner" or "tie" — underpowered or saturated results never graduate. - Durable artifact: Writes a validated model-profile.json recording per-tier rungs and the cheapest sufficient tier (lives_at). - Use Case: You have a classification skill and want to know if haiku is good enough. Run the ladder; if haiku's winner scores 0.92 against a 0.9 bar, the ladder stops there and records lives_at as haiku, skipping the costlier tiers. ## Quick Start Run the model ladder for my classify-text skill to find the cheapest model tier where it lives.
