Formato puro: `npm run format` (dos pases; siete ficheros no convergían en el
primero) sobre todo lo que .prettierignore no excluye (web/ congelado,
.claude/, artefactos generados). Este commit va en .git-blame-ignore-revs.
Neutralidad medida fichero a fichero, compilando y minificando con esbuild (y
con el compilador de Svelte, cliente y servidor, para .svelte) la versión de
HEAD y la formateada: 1 031 ficheros de código idénticos; 9 CSS que solo
difieren en espacios pegados a un paréntesis (`var( --x )` frente a
`var(--x)`), que CSS no tokeniza como significativos; 510 Markdown sin
compilación posible (docs:check los valida). Los dos generados y el guard que
dependía del formato se resolvieron en el commit anterior.
Verificación sobre este árbol: npm run lint exit 0 · gate entero verde salvo
focus-census, arreglado en el commit anterior y re-ejecutado (9/9): check:gate
OK (src/ y scripts/ a cero) · docs:check 0/0 · arts/blocks/packs/rtl/
translations/agent · eidos:lint 0 inválidos · apps:check (boot 27/27, 0
errores) · suite 465/466 → 466/466 con el guard corregido · build de la raíz
exit 0.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
El flip fue ATOMICO (13/13 con escritor y lector vivos, control negativo en
los siete componentes, ni un canal muerto en silencio) y lo refutado fue
documental:
- STUMBLES.md SS7 entra por fin al arbol: la anotacion existia sin commitear
(fallo de staging del coordinador - la raiz del repo no estaba en las
rutas). Deja escrito lo importante: el renombre no rompio nada por
DISCIPLINA de procedimiento (git ls-files por nombre), no porque un guard
cruce escrituras del provider contra lecturas de receta - S5 sigue abierto.
- PLAN-background: la anotacion decia "las nueve menciones de aqui abajo" y
tres quedaban arriba - ahora dice su alcance exacto.
- next-features SS13 gana las dos deudas del adversarial: la sonda estandar
es CIEGA a 12 de los 13 canales (PROPS sin translate/rotate/
gridTemplateColumns/animationDuration - el control positivo dio 0 diffs
con el canal roto a proposito) y el export muerto
getFloatingContentCSSVars que construye por plantilla el nombre publico
retirado de menubar.
Fe de erratas del commit e100ea520: son 31 ficheros, no 32.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Closes the provider→recipe direction of the CSS-var drift stumble the
architecture-faithful way: the morfo is the source of truth. A component now
declares the functional custom properties its soma provider writes and its eidos
recipe reads:
cssVars: [{ name: 'progress', … }, { name: 'angle', … }] // → --knob-progress, --knob-angle
- types: `MorfoCssVar` + optional `cssVars` on `Morfo` (additive; the 133
existing morfos are unaffected).
- schema: validates `cssVars` (array of `{ name, description? }`).
- compile: exposes `contracts.cssVars` as full names `--{kebab}-{name}`.
- knob morfo declares its two contract vars.
- guard (recipe-css-contract.test.ts): enumerates morfos via `import.meta.glob`
and verifies BOTH sides honour each declaration — the provider writes
`--{kebab}-{name}` AND the recipe reads it — so renaming one side without the
other fails loudly. Proven non-vacuous (a bogus cssVar is flagged on both
sides).
Why A1 and not a grep cross-check (A2): in the CSS a provider-written functional
var and a consumer-override alias are indistinguishable, and providers publish
hook vars the recipe doesn't consume — a grep gives ~17 false positives (proven
earlier). The morfo declaration is what disambiguates.
npm run check 59 (baseline, 0 in touched files); eidos recipe-css-contract 24/24;
morfo suite green (only the pre-existing dialog-role test fails); docs:check 0.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Empirically confirmed the grep approach can't distinguish a provider-written
functional var from a consumer-override alias (identical in CSS), and providers
publish API-hook vars the recipe doesn't consume (17 such, none read by var()).
Only A1 (morfo cssVars declaration) can close the provider→recipe direction.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Closes the recipe→theme direction of the soma→eidos CSS-var drift stumble. New
check in recipe-css-contract.test.ts: every NO-FALLBACK `var(--x)` in a
component recipe must resolve to a token declared somewhere in the eidos CSS
tree (foundation + shared layers + the component's own self-declarations). A
`var(--x, default)` stays exempt — those are runtime-optional by construction
(provider/floating vars whose fallback IS the default), matching the existing
`collectOwnPublicVariables` rule.
This is the exact class that silently broke the Knob (`--color-neutral-content`,
`--state-hover`, `--color-neutral-bg` resolved to nothing, only visible in the
render). The guard immediately caught 3 MORE real phantoms — fixed, not silenced
(silencing would be the anti-pattern of a guard exempting the gap it exists to
catch):
- card-group: `--color-content-default` → `--color-content-primary`
- link-preview: `--leading-body` → `--leading-normal`
- textarea: `--font-family-body` → `--font-family-primary`
Full eidos suite 290/290. The provider→recipe direction (renaming a
provider-written `--knob-angle` breaking the recipe) remains open as option A
(morfo `cssVars` declaration or a grep cross-check) — noted in STUMBLES.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Adds a '### Authoring notes' subsection to soma.md §6 grounded in the real
defs — state<T>() vs $state, role optionality on a Provider, Without<>/
PrimitiveDivAttributes, and gesture pointermove/up ownership. Closes STUMBLES
#9; only #7 (soma->eidos CSS-var contract, design-open) remains.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Knob now built across all 5 layers, validating the exercise. #7 confirmed
first-hand: the Knob build hit phantom theme tokens + uncontracted provider
CSS vars (silent drift), exactly as predicted.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
dev