Skip to content
xray

Roadmap

What’s next and what’s deliberately deferred. How the pieces work lives in Architecture; the why behind each decision lives in the decision records.

Planned

Section titled “Planned”
  • More frameworks & metaframeworks. A second adapter (Solid or Preact) to prove the ADR 0008 chunk contract on a non-React host, then the Vite metaframeworks (Astro, TanStack Start, Solid Start) where SSR/hydration and dev-client bootstrap differ from a plain SPA. The two axes are independent: a UI-framework adapter is a small unit of work against a settled contract; metaframework support is mostly a bootstrap + SSR problem, and largely framework-shared.
  • Animated enter/exit. An opt-in viewTransition for the skeleton→content swap (crossfade, or a shared-name morph), feature-detected and reduced-motion gated. Decided but not built; touches the prod adapter and per-framework swap. Decision: ADR 0012.
  • Capture fidelity. Carry author custom properties conditionally instead of freezing var()/grid tracks; reflow text via the engine with pretext as a capture-side oracle. Decisions: ADR 0019 (carried intent + engine reflow) and ADR 0021 (the ink-driven IR pipeline). The payload track is resolved: the shipped Distil stack (Flatten, Template, dead-decl drop, invisible-element skip, Superset, 2D Crop) takes ~⅕–¼ off raw, more on overlay-heavy views. Atomic CSS factoring was measured and rejected (compressed worse, ADR 0007); quantize and the sub-threshold ink drop are postponed and depth-LOD shelved (ADR 0022).
  • Tailwind class passthrough (opt-in). For Tailwind projects, read the config and keep the layout-relevant utility classes on the bone rather than extracting their computed decls into the plate’s own CSS, so the app’s already-loaded global Tailwind styles the skeleton for zero extra CSS. The biggest CSS-payload win for Tailwind apps, and it avoids the entropy problem that ruled out self-contained atomic factoring (ADR 0007): the utility classes are already on the wire, so no new rule text ships. Opt-in because it trades ADR 0003 self-containment (the skeleton would depend on the app’s Tailwind stylesheet); the self-contained default stays for everyone else. Keep layout classes, drop or override paint (a bone is a grey rect); watch arbitrary values, JIT, and purge. No ADR yet.

Deferred / known limits

Section titled “Deferred / known limits”
  • Extension distribution. The dev bridge (ADR 0010, ADR 0030) speaks a versioned, allowlisted CDP protocol, auto-connects on dev loopback origins, and activates anywhere else via the toolbar action — but it still loads unpacked: a GitHub Releases zip is the next step, the Chrome Web Store after. Tall/lazy (IntersectionObserver) content may need a driven scroll at short emulated heights; Input.* is already within the protocol’s allowlist, so that ships as a package update.
  • Bones in the Blink. The dev capture Blink is a cyan bloom, a negative invert blink, and a scan-line sweep over the element’s box. Overlaying the actual captured bones (a true x-ray of what was captured) drifts vertically: reconstructing the captured flow layout inside an absolutely-positioned overlay misplaces first-child margins, capture-vs-current height differences, and stale-sized stitched children. Revisit once the Sweep gives multi-width data to reconstruct faithfully, or anchor a small overlay per captured root rather than one box for the whole plate.
  • Capture-walker prod-stripping. The captureWalker (ADR 0018) is a dev-only concern but ships to prod as small dead code in v1. A build-time prop strip (the xray plugin removing the captureWalker attribute, letting the walker tree-shake) is deferred until the weight is worth it.
  • Next.js integration. Deliberately a roadmap item, not an ADR yet. Tackle non-React UI adapters and Vite-based metaframeworks first; they should define the shared adapter/bootstrap constraints. A Next-specific decision record comes only when a spike is close enough to decide load-bearing scope (modern Next, App Router, Turbopack, generated modules, dev sidecar, etc.).
  • Non-width axes (dark mode, orientation): out of scope for v1.
  • Pseudo-element rules. Dropped in v1; revisit if a real app leans on ::before/::after for layout.
  • Degenerate replaced boxes. Leaves with a near-zero axis grow to their parent’s content box when they are an only child (overflow-painting charts); broader low-ink heuristics deferred.