The body is lean and task-oriented — every section pairs a one-line rationale with executable code and assumes Claude's competence; the only repetition (the worked example) is justified by combining the three gates into a single end-to-end test.

Provides complete, copy-paste-ready JavaScript with real imports (Ajv, vega-lite schema, vega-lite compile) and a fully executable worked example rather than pseudocode.

A sequenced six-step "How to use" list and three explicit validation gates (schema validity → structural assertions → compilation) give a clear order; the anti-patterns table reinforces the sequence, and this validation context is not destructive/batch so no cap applies.

SKILL.md is a concise overview that delegates render/composition/transform/interaction/snapshot coverage to a single one-level-deep, repeatedly-signaled reference (references/advanced-spec-tests.md), which exists on disk.

Lists three concrete actions — "Validate Vega + Vega-Lite specifications against the JSON Schema", "test cross-engine compatibility", and "verify data-binding correctness" — matching the multi-action anchor.

Explicitly states what it does up front and follows with a clear "Use when application code generates Vega/Vega-Lite JSON specs at runtime… before render, or before a Vega major-version upgrade" trigger clause.

Covers natural domain terms a user would say ("Vega", "Vega-Lite", "JSON Schema", "BI builders", "spec templating", "Vega major-version upgrade") with good variation rather than jargon-only language.

Carves a distinct niche (spec validation, not rendering) and explicitly disambiguates from sister skills via "Pair with d3-snapshot-tests when Vega specs render to SVG; pair with chartjs-snapshot-tests when rendered to Canvas."