Conciseness

The ~35-line body is lean: the Quick Reference bullets are concrete directives with zero padding, and everything Claude needs is compressed. It is not 5 because the intro sentence explains screen-reader behavior Claude already knows ("Screen readers read table data cell-by-cell..."), the Check section restates the Quick Reference bullets ("Verify tables use proper th elements with scope, caption or aria-labelledby"), and the Code Review section repeats the title garbled mid-sentence ("interactive states that affect Use semantic table markup for screen readers").

Actionability

Concrete, specific guidance throughout: "Use <th> elements for headers with scope='col' or scope='row'", "Add <caption>", "Use id/headers for complex tables with merged cells", and review-output instructions like "Flag exact elements, roles, labels, focus behavior, or keyboard interactions". Per the rubric's code-vs-instruction note, the absence of code in this instruction-only skill is not penalized; it stays at 4 rather than 5 because no inline example of good vs. bad table markup appears in the body and "id/headers for complex tables" is given without showing how.

Workflow Clarity

The Check → Fix → Explain → Code Review section flow gives a coherent review process, and the Quick Reference adds prioritization heuristics ("Prefer one strong table-semantics finding over multiple weaker enhancements", "missing scope is often a stronger finding than missing <caption>"). It is not 5 because no explicit ordering of the checks is given in the body (the "Check native semantics first" sequence lives only in the frontmatter), and the relationship between the Check and Code Review sections is left implicit.

Progressive Disclosure

The body is a concise overview with well-signaled one-level-deep references: "For full implementation details, code examples, and framework-specific guidance, see `references/rule.md`" — and references/rule.md exists (155 lines) with exactly that content, including the full HTML example. No nested references, no bulk content inlined that belongs elsewhere. This matches the 5 anchor for a clear overview with well-signaled single-level references.

Specificity

The description lists several concrete actions — "Check native semantics first, then inspect keyboard behavior, focus flow, accessible names, and screen-reader output" — matching the anchor for several specific actions with minor gaps. It is not a 5 because the garbled splice "related to Use semantic table markup for screen readers" blurs what the skill actually does with tables (th/scope, caption), and coverage of the concrete table-semantics actions lives only in the body, not the description.

Completeness

Both parts are explicit: what ("Check native semantics first, then inspect keyboard behavior, focus flow, accessible names, and screen-reader output") and when ("Use when reviewing rendered HTML, interactive components, or design-system patterns related to..."). It is not 5 because the when-clause is awkwardly phrased — the mid-sentence "Use semantic table markup for screen readers" read as a templated title rather than a concrete trigger phrase, leaving the trigger condition muddier than the rubric's 5-anchor examples.

Trigger Term Quality

Natural terms a user would say are present: "rendered HTML", "screen readers", "semantic table markup", "interactive components". This fits the good-coverage anchor rather than 3 (it goes beyond a couple of generic keywords) but falls short of 5 because common variations like "data tables", "table headers", "accessibility", or "a11y" are missing.

Distinctiveness Conflict Risk

The table-markup mention gives it a niche, but the trigger context "reviewing rendered HTML, interactive components, or design-system patterns" plus the generic a11y checklist "keyboard behavior, focus flow, accessible names" would apply equally to sibling accessibility review skills (focus order, alt text, ARIA), so overlap risk with similar skills remains. It is not 4 because the overlap is more than minor — most of the when-clause is boilerplate shared across a rule family — and not 2 because "semantic table markup for screen readers" does anchor a specific niche.