running-skill-edd-cycle

Evaluation-driven development structures tests before building skill content. Define 3+ evaluation scenarios that reflect real user needs, run them to establish performance baselines, then iterate on skill logic while measuring against those criteria. This ensures skills solve actual problems before deployment.

Yes. Multi-model validation lets you run the same evaluation scenarios across different Claude models and collect feedback loops. This identifies which models best suit your skill's requirements and guides model-selection decisions for production deployment.

Run your initial evaluation scenarios once to capture baseline metrics, then use those results as reference points for subsequent iterations. Baseline testing reveals whether skill changes improve or degrade performance and tracks progress across development cycles.

Use evaluation-first workflows when you need to validate that a skill addresses real requirements, compare multiple implementations, or ensure consistency across Claude versions. This approach prevents wasted effort on skills that don't meet functional criteria.

Yes. The methodology provides structured evaluation schemas and traceable observation criteria that satisfy compliance prompts and model-compatibility ratings used in formal skill lifecycle reviews and validation documentation.