Acceptance Criteria
- The context-modifier facet sets
data-in="<parent rune>" on every element it gives a context-modifier class, and keeps the class - The tint facet sets
data-tinted="" wherever it adds --tinted, and keeps the class - The 31 context-modifier selectors in Lumina and the skeleton select
[data-in="…"] instead of the class - The 14 static-modifier selectors select the attributes that already say the same fact
staticModifiers is marked deprecated in packages/transform/src/types.ts, pointing at SPEC-196 D5- Engine tests cover
data-in (including a rune nested in a parent it does not declare, which gets none) and data-tinted (set, and absent for tint="custom" with no tokens) - Both contract copies (
contracts/structures.json, packages/lumina/contracts/structures.json) are regenerated and record the new attributes - The CSS coverage test passes, with its expectations updated for selectors that moved to attributes
npm run seo:baseline:check passes, or the baseline is regenerated and its diff shows only the new attributes
Approach
- The facets are
packages/transform/src/facets/modifiers.ts (context and static modifiers) and facets/tint.ts. Emit the attribute in the same place the class is added, so the two cannot diverge. data-in holds the parent's data-rune name, the same key contextModifiers already uses. The value is the declared parent, not every ancestor: it says what the class said and no more.- The names were checked free in D5:
data-in and data-tinted appear nowhere in packages or plugins (data-parent is taken by realm). - Use
refrakt inspect on actions inside bar and on showcase inside bento-cell to see both outputs before moving their CSS. Those two hold 17 of the 31 selectors. - A selector change keeps its specificity in mind:
.rf-actions--in-bar and .rf-actions[data-in="bar"] differ, so check the rules that relied on the old weight.
References
- SPEC-196 D5, and "What exists, measured".
- SPEC-194 D5: the config fields kept and deprecated.
- ADR-048: implementation order, step 1.
Resolution
Completed: 2026-11-09
Branch: claude/rune-composition-layout-convention-hkqrzx
What was done
packages/transform/src/facets/modifiers.ts: the context-modifiers facet sets data-in="<parent>" beside the --in-<parent> class, only for the declared parent. Its contract description gains attributeSelector; the static-modifiers facet is documented as deprecated.packages/transform/src/facets/tint.ts: data-tinted="" wherever --tinted is added; the axis contract lists it.packages/transform/src/types.ts: staticModifiers marked @deprecated (SPEC-196 D5 / SPEC-194 D5); contextModifiers doc points at data-in. contracts.ts and facets/describe.ts carry attributeSelector.- Lumina and skeleton CSS: 31 context-modifier selectors now select
[data-in], 14 static-modifier selectors select data-rune="featured-tier" / data-media-position="cover". packages/lumina/test/css-coverage.test.ts: context modifiers count as covered by .block[data-in="parent"] (bare or :where()), plus new tests that fail on CSS selecting a context- or static-modifier class (variant deltas included). vocabulary-coverage.test.ts records [data-tinted] as intentionally unstyled.packages/cli: inspect --audit expects [data-in="parent"] and no static-modifier classes; scaffold-css writes the attribute form and no static stubs.packages/transform/test/context-modifiers.test.ts: data-in and data-tinted tests.- Contracts regenerated (both copies). Docs updated:
css.md, config-api.md, output-contract.md, composability.md, theme-authoring/overview.md, runes/actions.md. Changeset: transform minor, lumina/skeleton/cli patch.
Notes
- Every retargeted selector wraps the new attribute in
:where() so specificity is unchanged. Two places depend on it: the frost scrim (lumina/styles/dimensions/cover.css) beats the section band scrim by exactly one attribute, and .rf-tier:hover must keep beating the featured tier's shadow. A plain .rf-x[data-in] would have tied or reversed both. - The contract field is named
attributeSelector, not dataAttribute, because elsewhere in the contract dataAttribute holds an attribute name, not a selector. - The SEO baseline did not change: no fixture nests a context-modified rune or resolves a tint.