Open-source platform for 3D AI agents. Turn text or a photo into a rigged, animated GLB avatar, give it an LLM brain, memory and a wallet, and embed it anywhere with one web component. Remote MCP server, x402 payments, Three.js, browser-native.
JavaScript 3d-generationai-agentsanimationavatardigital-humangenerative-aiglbgltfimage-to-3dllmmcpmcp-servermodel-context-protocolsolanatext-to-3dthreejsweb-componentswebglwebxrx402
fix(play-a11y): land dialog focus immediately, not only on the next frame
nirholas572a326fcdd9
diff --git a/src/game/a11y.js b/src/game/a11y.jsindex cdeef09da..073bddd0a 100644--- a/src/game/a11y.js+++ b/src/game/a11y.js@@ -130,11 +130,20 @@ export function trapFocus(root, { initialFocus } = {}) {// the panel's async content arrives; -1 keeps it out of the Tab order.if (!root.hasAttribute('tabindex')) root.setAttribute('tabindex', '-1');const target = initialFocus || focusableIn(root)[0] || root;- // Defer one frame: panels focus-trap in their constructor, before the card- // has been laid out, and focusing a zero-size element is a silent no-op.- requestAnimationFrame(() => {+ const applyFocus = () => {if (root.isConnected) try { target.focus({ preventScroll: true }); } catch { /* detached */ }- });+ };+ // Twice, on purpose. The next-frame call is what handles the common case:+ // panels focus-trap in their constructor, before the card has been laid out,+ // and focusing a zero-size element is a silent no-op. But a frame is not+ // guaranteed to arrive on demand - /play drives rAF off a WebGL compositor+ // that stalls under load, and a backgrounded tab stops delivering frames+ // altogether - and a deferred-only focus leaves `aria-modal` promising a trap+ // that never closed, with the keyboard still on the page behind the dialog.+ // Focusing now costs nothing when it no-ops and holds the promise when no+ // frame ever comes.+ applyFocus();+ requestAnimationFrame(applyFocus);let released = false;return () => {