Build the frame-overflow: clip | bleed facet per SPEC-116 / ADR-023 — a universal frame facet (policy) gated by a per-guest runtime data-overflowing signal — and land sandbox as its first consumer by reworking the held bleed="crop" PR (refrakt-md/refrakt#530).
64cdb031650ab545ecacaeaefb7eb7c6829frame-overflow to UNIVERSAL_ATTRIBUTE_NAMES, the frame-facet meta list (FRAME_FACET_META), and FramePresetDefinition. resolveFrameChrome reads it and emits data-frame-overflow on the frame target (the host [data-section="media"] zone). Default clip emits nothing.frame-overflow="bleed" resolves on a host whose config.guestFit is clip (or otherwise has no bleed-capable media zone), emit a warn-once naming the rune and pointing to a bleed host — same shape as warnFrameNoTarget.[data-frame-overflow="bleed"] > [data-overflowing], mobile breakpoint): inline-end → the layout-owned boundary (--rf-bleed-room-end, default page gutter — never the raw viewport, SPEC-116 §5), inline-end corners squared, inline-start anchored. Remove the sandbox-specific bleed CSS. v1 ships the variable seam with the default; per-layout --rf-bleed-room-* values land with the deferred desktop/side-by-side work.scrollWidth measurement on the rf-sandbox-resize bridge and the nextBleedState hysteresis + tests. Set data-overflowing unconditionally when overflowing (policy now lives on the host, not the guest). Remove the bleed attribute from the sandbox rune schema (it never shipped — no deprecation needed).frame-overflow reference (Surfaces / frame docs), the runtime-gate model, the clip-host warning; structure-contract/coverage updates.frame-overflow="bleed" on a bleed host emits data-frame-overflow="bleed" on its [data-section="media"] zone; clip/unset emits nothing; resolved via resolveFrameChrome.[data-frame-overflow="bleed"] > [data-overflowing] on a narrow viewport (inline-end to screen, inline-end corners squared, inline-start anchored); no per-rune bleed CSS remains.data-overflowing from measured content width with hysteresis, independent of host policy; the sandbox bleed attribute is removed.frame-overflow="bleed" on a clip host (guestFit: 'clip') is inert and emits a hard build warning naming the rune and a bleed host; covered by an engine test.clip/unset) output is byte-identical to today for every rune (regression-safe); a guest that fits stays inset and rounded.frame-displace end-to-end: it is the closest existing facet (meta read → data-displace on the frame target → CSS), so frame-overflow follows the same read/emit path in resolveFrameChrome.config.guestFit in scope at frame-resolution time — it already is (the displace-mode default added in the guestFit work reads config.guestFit there).width: calc(100% + gutter) assumes the full-width host's mobile padding equals the gutter (true with contentMeasure anchoring), and rounded-iframe corner clipping is historically finicky in Safari. Verify on mobile Safari/Chrome.data-overflowing signal for codegroup/table/datatable is follow-on work, not in this item.frame-anchor's inline keyword: start→end, end→start, center→both) is SPEC-116 §4 follow-on. Keep v1's emitted data-frame-overflow="bleed" forward-compatible (no rename when direction lands).--rf-bleed-room-* with the default (page gutter); the docs/chrome'd-layout values that cap the bleed at the wide track (SPEC-116 §5) land with the desktop work. v1's ≤640px trigger does not expose the collision (docs hides the sidebar/TOC by then).resolveFrameChrome: SPEC-086.packages/behaviors/src/elements/sandbox.ts nextBleedState + measurement, packages/runes/src/tags/sandbox.ts, packages/lumina/styles/runes/sandbox.css).Completed: 2026-06-23
Branch: claude/work-444-frame-overflow
frame-overflow added as a universal frame attribute (packages/runes/src/lib/index.ts schema def + FRAME_FACET_NAMES; packages/runes/src/attribute-presets.ts UNIVERSAL_ATTRIBUTE_NAMES). FramePresetDefinition.overflow (packages/transform/src/types.ts). resolveFrameChrome reads frame-overflow and emits data-frame-overflow="bleed"; FRAME_FACET_META extended (packages/transform/src/engine.ts).data-frame-overflow="bleed" lands on a host whose guestFit !== 'bleed', the marker is stripped (clean output) and warnFrameOverflowClip warns once.[data-frame-overflow="bleed"] > [data-overflowing] in packages/lumina/styles/dimensions/frame.css (mobile breakpoint): width: calc(100% + var(--rf-bleed-room-end, var(--rf-content-gutter))), inline-end corners squared — a layout-owned boundary, not the raw viewport.packages/behaviors/src/elements/sandbox.ts: the resize bridge also reports scrollWidth; updateOverflow toggles data-overflowing unconditionally (policy lives on the host) via the exported nextBleedState hysteresis. No sandbox bleed attribute was added.frames.test.ts); nextBleedState + the always-on detection (behaviors/test/sandbox.test.ts); reference snapshot updated for the new universal attr.frame-overflow row + explainer in site/content/runes/surfaces.md.frame-anchor), side-by-side outer-edge bleed, the per-layout --rf-bleed-room-* values, and generalising the data-overflowing signal to codegroup/table are deferred (SPEC-116 §4–5).