Lean body that assumes Claude's competence — no 'what is OpenSearch' preamble, just code blocks, tables, and tight prose where each sentence earns its place, matching the 'lean and efficient' anchor over the 'could be tightened' level 2.

Fully executable, copy-paste-ready examples throughout: `_rank_eval` and hybrid search HTTP requests, k-NN index mappings, search-pipeline config, and Python test functions, satisfying the 'fully executable code/commands' anchor.

Seven clearly sequenced steps each carrying explicit assert-based validation (recall≥0.85, delta≥−0.10, ES/OS delta<0.05) with an anti-patterns table mapping failures to fixes, meeting the 'clear sequence with explicit validation steps' anchor.

A single well-sectioned SKILL.md with only one-level-deep external references (the OpenSearch docs link and named sibling skills) and no nested/deep references; no bundle files exist, so the inline code is appropriately organized rather than mis-split.

Lists multiple concrete actions — 'Author OpenSearch relevance tests with Search Relevance Workbench (judgment lists, query sets, experiments)', '`_rank_eval` API', and 'hybrid BM25 + neural ranking eval' — matching the 'lists multiple specific concrete actions' anchor rather than the partial level 2.

Answers both what ('Author OpenSearch relevance tests with…') and when via an explicit 'Use when an OpenSearch index turns on neural or hybrid search, or when a move off Elasticsearch has to prove relevance parity' clause, not the implied-only level 2.

Natural terms a user would say are well covered ('OpenSearch relevance tests', 'neural search', 'hybrid search', 'relevance parity', 'move off Elasticsearch') in third-person voice, above the 'some relevant keywords' level 2.

A clear OpenSearch-specific niche (neural/hybrid eval plus ES→OS migration parity) that distinguishes it from the generic and elasticsearch sister skills, unlikely to trigger for the wrong skill.