Lists multiple specific concrete actions: 'running them on identical inputs', 'diffing outputs, side effects, and errors'. Clearly describes comparing runtime behavior of code versions.

Clearly answers both what ('Compares runtime behavior...by running them on identical inputs and diffing outputs') AND when ('Use when validating a refactor, port, or optimization; when the user asks if two implementations behave the same; or when investigating a suspected regression').

Includes natural terms users would say: 'refactor', 'port', 'optimization', 'two implementations behave the same', 'regression', 'versions'. These are terms developers naturally use when needing this functionality.

Highly distinctive niche focusing on behavioral comparison of code versions. The specific triggers (refactor validation, implementation equivalence, regression investigation) are unlikely to conflict with general code review or testing skills.

The content is lean and efficient, assuming Claude's competence with concepts like differential testing, property-based testing, and normalization. Every section earns its place with actionable tables and concrete examples rather than explanatory padding.

Provides fully executable Python code with Hypothesis integration, concrete normalization strategies in table format, and a complete worked example showing real divergence findings. The harness code is copy-paste ready.

The workflow is clear: decide observables → generate inputs → normalize → compare → cluster divergences. The worked example demonstrates the full sequence with explicit validation (majority voting) and the output format provides a structured checklist for results.

Well-organized with clear sections (observables table, input generation hierarchy, normalization table, worked example, do-not list, output format). References the related skill appropriately. Content is appropriately scoped for a single file without needing external references.