Run task evals across multiple Claude models, compare results side-by-side, and optimise. Use when you want to benchmark a skill across models, compare haiku vs sonnet vs opus performance, run multi-model comparison or benchmark reports, identify model-specific gaps versus universal plugin gaps, evaluate whether a skill works for all model tiers, or validate a skill before publishing it to the registry.