@burdenoff/fe-libs 2026.1004.4 → 2026.1004.5

package/dist/shared/assistant/ui/voice/AssistantVoiceSession.js

@@ -3,44 +3,44 @@ import { useAssistantTr as t } from "../../i18n.js";
 import { micIsInterrupt as n, showsAssistantCaption as r, voiceStageState as i } from "../../voice/voiceStageState.js";
 import { VOICE_THINKING_LABEL as a, composeVoiceStatus as o, voiceStatusSuffix as ee } from "../../voice/voiceStatus.js";
 import { spokenSoFar as s } from "../../voice/voiceSession.js";
-import { ASSISTANT_PANEL_ATTR as c, ASSISTANT_VOICE_SESSION_ATTR as l } from "../panelOverlays.js";
-import { AssistantAttachmentButton as u } from "../composer/AssistantAttachmentButton.js";
-import { AssistantAttachmentChips as te } from "../composer/AssistantAttachmentChips.js";
-import { AssistantVoiceOrb as ne } from "./AssistantVoiceOrb.js";
-import { VoiceMiniController as re } from "./VoiceMiniController.js";
-import { VoiceStage as ie } from "./VoiceStage.js";
-import { VoiceTranscriptDrawer as ae } from "./VoiceTranscriptDrawer.js";
-import { escapeBelongsToLayerAbove as oe, voiceSessionEscapeAction as se } from "./voiceSessionEscape.js";
-import { useCallback as d, useEffect as f, useRef as p, useState as m } from "react";
-import { Maximize2 as ce, Minimize2 as le, ScrollText as ue, X as h } from "lucide-react";
-import { jsx as g, jsxs as _ } from "react/jsx-runtime";
+import { ASSISTANT_PANEL_ATTR as te, ASSISTANT_VOICE_SESSION_ATTR as ne } from "../panelOverlays.js";
+import { AssistantAttachmentButton as re } from "../composer/AssistantAttachmentButton.js";
+import { AssistantAttachmentChips as ie } from "../composer/AssistantAttachmentChips.js";
+import { AssistantVoiceOrb as ae } from "./AssistantVoiceOrb.js";
+import { VoiceMiniController as oe } from "./VoiceMiniController.js";
+import { VoiceStage as se } from "./VoiceStage.js";
+import { VoiceTranscriptDrawer as ce } from "./VoiceTranscriptDrawer.js";
+import { escapeBelongsToLayerAbove as le, voiceSessionEscapeAction as ue } from "./voiceSessionEscape.js";
+import { useCallback as c, useEffect as l, useRef as u, useState as d } from "react";
+import { Maximize2 as de, Minimize2 as fe, ScrollText as pe, X as f } from "lucide-react";
+import { jsx as p, jsxs as m } from "react/jsx-runtime";
 //#region src/shared/assistant/ui/voice/AssistantVoiceSession.tsx
-var v = ({ session: v, turns: pe = [], onSaveTranscript: me, onContinueInChat: he, notice: y = null, spoken: b, onInterrupt: ge, onToggleMute: x, attachments: S = [], onAttachFiles: C, onRemoveAttachment: _e, onRetryAttachment: ve, canAttachMore: ye = !0, preparingAttachments: w = 0, spokenChunks: T = null, onEnd: E, capturing: be = !0, interrupting: xe = !1, awaitingEndpoint: Se = !1, proposalChip: D = null, proposalTray: O = null }) => {
-	let k = t(), { muted: A, error: j } = v, M = p(null), N = p(null), [P, F] = m("mini"), I = P === "full", L = P === "mini", R = P === "mini" ? "panel" : P === "panel" ? "full" : "mini", [z, B] = m(!1), V = O !== null && !L, H = p(!1), U = p(x);
-	U.current = x;
-	let W = d(() => {
-		v.phase === "listening" && (A || H.current || s(v) || (H.current = !0, U.current()));
-	}, [A, v]), G = d(() => {
+var h = ({ session: h, turns: ge = [], onSaveTranscript: _e, onContinueInChat: ve, notice: g = null, spoken: _, onInterrupt: ye, onToggleMute: v, attachments: y = [], onAttachFiles: b, onRemoveAttachment: be, onRetryAttachment: xe, canAttachMore: Se = !0, preparingAttachments: x = 0, spokenChunks: S = null, onEnd: C, capturing: Ce = !0, interrupting: we = !1, awaitingEndpoint: Te = !1, holding: w = !1, correcting: Ee = !1, onLetMeFinish: T, onImDone: E, proposalChip: D = null, proposalTray: O = null }) => {
+	let k = t(), { muted: A, error: j } = h, M = u(null), N = u(null), [P, F] = d("mini"), I = P === "full", L = P === "mini", R = P === "mini" ? "panel" : P === "panel" ? "full" : "mini", [z, B] = d(!1), V = O !== null && !L, H = u(!1), U = u(v);
+	U.current = v;
+	let W = c(() => {
+		h.phase !== "listening" && h.phase !== "thinking" || A || H.current || s(h) || (H.current = !0, U.current());
+	}, [A, h]), G = c(() => {
 		H.current && (H.current = !1, U.current());
-	}, []), K = W;
-	f(() => {
+	}, []), De = W;
+	l(() => {
 		let e = () => {
 			document.visibilityState === "hidden" ? W() : G();
 		};
 		return document.addEventListener("visibilitychange", e), () => document.removeEventListener("visibilitychange", e);
-	}, [W, G]), f(() => {
+	}, [W, G]), l(() => {
 		let e = G;
 		return window.addEventListener("focus", e), () => {
 			window.removeEventListener("focus", e), H.current = !1;
 		};
-	}, []), f(() => {
+	}, []), l(() => {
 		M.current?.focus();
-	}, []), e(N, { fallbackSelector: `[${c}]` }), f(() => {
+	}, []), e(N, { fallbackSelector: `[${te}]` }), l(() => {
 		let e = (e) => {
 			if (e.key === "Escape") {
-				let t = se({
+				let t = ue({
 					defaultPrevented: e.defaultPrevented,
-					layerAbove: oe({
+					layerAbove: le({
 						target: e.target,
 						session: N.current,
 						doc: document
@@ -48,7 +48,7 @@ var v = ({ session: v, turns: pe = [], onSaveTranscript: me, onContinueInChat: h
 					transcriptOpen: z
 				});
 				if (t === "ignore") return;
-				e.preventDefault(), t === "close-transcript" ? B(!1) : V ? O?.onClose() : E();
+				e.preventDefault(), t === "close-transcript" ? B(!1) : V ? O?.onClose() : C();
 				return;
 			}
 			if (e.key !== "Tab" || !I) return;
@@ -59,19 +59,28 @@ var v = ({ session: v, turns: pe = [], onSaveTranscript: me, onContinueInChat: h
 		};
 		return window.addEventListener("keydown", e), () => window.removeEventListener("keydown", e);
 	}, [
-		E,
+		C,
 		I,
 		z,
 		V,
 		O
 	]);
-	let q = y !== null && j === null, J = s(v), Y = i({
-		snapshot: v,
-		capturing: be,
-		interrupting: xe,
-		awaitingEndpoint: Se
-	}), X = (() => {
-		switch (Y) {
+	let K = g !== null && j === null, q = s(h), J = T && E && h.phase === "listening" && !A && j === null ? {
+		letMeFinish: T,
+		imDone: E
+	} : null, Y = J !== null;
+	l(() => {
+		if (Y) return;
+		let e = document.activeElement;
+		(!e || e === document.body) && M.current?.focus();
+	}, [Y]);
+	let X = i({
+		snapshot: h,
+		capturing: Ce,
+		interrupting: we,
+		awaitingEndpoint: Te
+	}), Oe = (() => {
+		switch (X) {
 			case "connecting": return k("voiceSession.connecting", "Starting the microphone…");
 			case "endpoint-grace": return k("voiceSession.endpointGrace", "Still listening — take your time");
 			case "interrupting": return k("voiceSession.interrupting", "Stopped — go ahead");
@@ -80,79 +89,90 @@ var v = ({ session: v, turns: pe = [], onSaveTranscript: me, onContinueInChat: h
 			case "mic-error":
 			case "ended": return "";
 			case "listening": return A ? k("voiceSession.muted", "Microphone off") : k("voiceSession.listening", "Listening…");
-			default: return Y;
+			default: return X;
 		}
-	})(), Z = ee({
+	})(), Z = !A && (X === "listening" || X === "endpoint-grace"), Q = Z && w ? k("voiceSession.holding", "Take your time — tap “I’m done” when you finish") : Z && Ee ? k("voiceSession.correcting", "Stopped that answer — listening to your correction") : Oe, $ = ee({
 		tr: k,
-		stage: Y,
+		stage: X,
 		muted: A,
-		label: X
-	}), Q = o(X, Z), $ = Y === "thinking";
-	return /* @__PURE__ */ _("div", {
+		label: Q
+	}), ke = o(Q, $), Ae = X === "thinking";
+	return /* @__PURE__ */ m("div", {
 		ref: N,
 		className: `flex flex-col items-center justify-between bg-bg-sunken ${P === "full" ? "fixed inset-x-0 bottom-0 top-14 z-[60] px-6 py-8" : P === "panel" ? "absolute inset-0 z-20 px-4 py-6" : "absolute inset-x-0 bottom-0 z-20 h-[11.25rem] border-t border-border-subtle px-3 py-3 shadow-lg"}`,
 		role: "dialog",
 		"aria-modal": I || void 0,
 		"aria-label": k("voiceSession.title", "Voice conversation"),
 		"data-testid": "assistant-voice-session",
-		[l]: "",
+		[ne]: "",
 		children: [
-			/* @__PURE__ */ _("button", {
+			/* @__PURE__ */ m("button", {
 				ref: M,
 				type: "button",
-				onClick: E,
+				onClick: C,
 				className: `absolute ${I ? "right-6 top-6 h-14 gap-2 px-5 text-base" : L ? "right-12 top-2 h-8 gap-1.5 px-2.5 text-xs" : "right-3 top-3 h-11 gap-2 px-4 text-sm"} flex items-center justify-center rounded-full border border-border-default bg-bg-elevated text-text-secondary hover:bg-bg-sunken transition-colors focus:outline-none focus-visible:ring-2 focus-visible:ring-border-strong`,
 				title: k("voiceSession.end", "End the voice conversation"),
 				"data-testid": "assistant-voice-end",
-				children: [/* @__PURE__ */ g(h, {
+				children: [/* @__PURE__ */ p(f, {
 					className: I ? "size-6" : L ? "size-4" : "size-5",
 					"aria-hidden": "true"
-				}), /* @__PURE__ */ g("span", {
+				}), /* @__PURE__ */ p("span", {
 					className: `truncate whitespace-nowrap font-medium ${L ? "max-w-[3.5rem] text-xs" : "max-w-[7rem]"}`,
 					children: k("voiceSession.endShort", "End")
 				})]
 			}),
-			/* @__PURE__ */ g("div", {
-				className: `flex flex-col items-center ${L ? "px-3" : `min-h-[1.5rem] ${q ? I ? "pt-14" : "pt-10" : ""}`}`,
-				children: q ? /* @__PURE__ */ g("p", {
+			/* @__PURE__ */ p("div", {
+				className: `flex flex-col items-center ${L ? "px-3" : `min-h-[1.5rem] ${K ? I ? "pt-14" : "pt-10" : ""}`}`,
+				children: K ? /* @__PURE__ */ p("p", {
 					className: "max-w-xs text-center text-xs text-status-warning",
 					"data-testid": "assistant-voice-notice",
-					children: de(y, k)
+					children: me(g, k)
 				}) : null
 			}),
 			(() => {
-				let e = /* @__PURE__ */ g(ne, {
-					session: v,
+				let e = /* @__PURE__ */ p(ae, {
+					session: h,
 					compact: L,
-					connecting: Y === "connecting",
-					onPress: n(Y) ? ge : x,
-					label: n(Y) ? k("voiceSession.interrupt", "Stop the assistant and speak") : A ? k("voiceSession.unmute", "Turn the microphone on") : k("voiceSession.mute", "Turn the microphone off")
-				}), t = r(Y) ? b : J || null, i = j === null ? null : fe(j, k);
-				return L ? /* @__PURE__ */ g(re, {
-					spokenChunks: T,
-					state: Y,
-					caption: t,
-					statusText: Q,
-					statusSuffix: Z,
-					rotateStatus: $,
-					errorText: i,
-					control: e
-				}) : /* @__PURE__ */ g(ie, {
-					state: Y,
-					userCaption: J || null,
-					spokenChunks: T,
-					assistantCaption: b,
+					connecting: X === "connecting",
+					onPress: n(X) ? ye : v,
+					label: n(X) ? k("voiceSession.interrupt", "Stop the assistant and speak") : A ? k("voiceSession.unmute", "Turn the microphone on") : k("voiceSession.mute", "Turn the microphone off")
+				}), t = J ? /* @__PURE__ */ p("button", {
+					type: "button",
+					onClick: w ? J.imDone : J.letMeFinish,
+					className: `flex items-center justify-center rounded-full border font-medium transition-colors focus:outline-none focus-visible:ring-2 focus-visible:ring-border-strong ${L ? "h-7 px-3 text-xs" : I ? "h-11 px-6 text-base" : "h-9 px-4 text-sm"} ${w ? "border-transparent bg-action-primary-bg text-action-primary-text hover:opacity-90" : "border-border-default bg-bg-elevated text-text-secondary hover:bg-bg-sunken"}`,
+					title: w ? k("voiceSession.imDoneHint", "Send what you said now") : k("voiceSession.letMeFinishHint", "Pause sending until you tap “I’m done”"),
+					"data-testid": "assistant-voice-hold",
+					"data-holding": w ? "true" : "false",
+					children: w ? k("voiceSession.imDone", "I’m done") : k("voiceSession.letMeFinish", "Let me finish")
+				}) : null, i = t ? /* @__PURE__ */ m("div", {
+					className: "flex flex-col items-center gap-2",
+					children: [e, t]
+				}) : e, a = r(X) ? _ : q || null, o = j === null ? null : he(j, k);
+				return L ? /* @__PURE__ */ p(oe, {
+					spokenChunks: S,
+					state: X,
+					caption: a,
+					statusText: ke,
+					statusSuffix: $,
+					rotateStatus: Ae,
+					errorText: o,
+					control: i
+				}) : /* @__PURE__ */ p(se, {
+					state: X,
+					userCaption: q || null,
+					spokenChunks: S,
+					assistantCaption: _,
 					expanded: I,
-					statusText: Q,
-					statusSuffix: Z,
-					rotateStatus: $,
-					errorText: i,
-					control: e
+					statusText: ke,
+					statusSuffix: $,
+					rotateStatus: Ae,
+					errorText: o,
+					control: i
 				});
 			})(),
-			D && !V ? /* @__PURE__ */ g("div", {
+			D && !V ? /* @__PURE__ */ p("div", {
 				className: L ? "absolute left-20 right-36 top-2 flex justify-center" : `flex w-full justify-center ${I ? "max-w-[calc(100%-12rem)]" : "max-w-[calc(100%-8rem)]"}`,
-				children: /* @__PURE__ */ g("button", {
+				children: /* @__PURE__ */ p("button", {
 					type: "button",
 					onClick: () => {
 						F("panel"), D.onOpen();
@@ -162,72 +182,72 @@ var v = ({ session: v, turns: pe = [], onSaveTranscript: me, onContinueInChat: h
 					children: D.label
 				})
 			}) : null,
-			O && V ? /* @__PURE__ */ _("div", {
+			O && V ? /* @__PURE__ */ m("div", {
 				className: `absolute z-10 flex flex-col gap-2 overflow-y-auto rounded-lg border border-border-subtle bg-bg-elevated shadow-lg ${I ? "inset-x-6 bottom-24 top-24 mx-auto max-w-xl p-4" : "inset-x-3 bottom-16 top-16 p-3"}`,
 				"data-testid": "assistant-voice-proposal-tray",
-				children: [/* @__PURE__ */ g("div", {
+				children: [/* @__PURE__ */ p("div", {
 					className: "flex justify-end",
-					children: /* @__PURE__ */ g("button", {
+					children: /* @__PURE__ */ p("button", {
 						type: "button",
 						onClick: O.onClose,
 						className: "flex size-8 items-center justify-center rounded-full text-text-secondary transition-colors hover:bg-bg-raised hover:text-text-primary focus:outline-none focus-visible:ring-2 focus-visible:ring-border-strong",
 						"aria-label": O.closeLabel,
 						"data-testid": "assistant-voice-proposal-tray-close",
-						children: /* @__PURE__ */ g(h, {
+						children: /* @__PURE__ */ p(f, {
 							className: "size-4",
 							"aria-hidden": "true"
 						})
 					})
 				}), O.content]
 			}) : null,
-			z ? /* @__PURE__ */ g(ae, {
-				turns: pe,
+			z ? /* @__PURE__ */ p(ce, {
+				turns: ge,
 				onClose: () => B(!1),
-				onSave: me,
-				onContinueInChat: he
+				onSave: _e,
+				onContinueInChat: ve
 			}) : null,
-			C ? /* @__PURE__ */ g("div", {
+			b ? /* @__PURE__ */ p("div", {
 				className: `absolute ${I ? "left-6 top-6" : L ? "left-11 top-2" : "left-3 top-3"}`,
-				children: /* @__PURE__ */ g(u, {
+				children: /* @__PURE__ */ p(re, {
 					onFiles: (e) => {
-						C(e), F((e) => e === "mini" ? "panel" : e);
+						b(e), F((e) => e === "mini" ? "panel" : e);
 					},
-					onPickerOpen: K,
+					onPickerOpen: De,
 					direction: "down",
-					canAddMore: ye
+					canAddMore: Se
 				})
 			}) : null,
-			(S.length > 0 || w > 0) && !L ? /* @__PURE__ */ g("div", {
+			(y.length > 0 || x > 0) && !L ? /* @__PURE__ */ p("div", {
 				className: `absolute ${I ? "bottom-24 left-24 right-24" : "bottom-16 left-16 right-16"}`,
 				"data-testid": "assistant-voice-attachments",
-				children: /* @__PURE__ */ g(te, {
-					attachments: S,
-					onRemove: _e ?? (() => void 0),
-					onRetry: ve ?? (() => void 0),
-					preparingCount: w
+				children: /* @__PURE__ */ p(ie, {
+					attachments: y,
+					onRemove: be ?? (() => void 0),
+					onRetry: xe ?? (() => void 0),
+					preparingCount: x
 				})
 			}) : null,
-			/* @__PURE__ */ g("button", {
+			/* @__PURE__ */ p("button", {
 				type: "button",
 				onClick: () => B((e) => !e),
 				className: `absolute ${I ? "bottom-6 right-6 size-14" : L ? "right-2 top-2 size-8" : "bottom-3 right-3 size-11"} flex items-center justify-center rounded-full bg-bg-elevated text-text-secondary transition-colors hover:bg-bg-raised hover:text-text-primary focus:outline-none focus-visible:ring-2 focus-visible:ring-border-strong`,
 				"aria-pressed": z,
 				"aria-label": k("voiceSession.transcriptTitle", "Voice transcript"),
 				"data-testid": "assistant-voice-transcript-toggle",
-				children: /* @__PURE__ */ g(ue, { className: I ? "size-6" : L ? "size-4" : "size-5" })
+				children: /* @__PURE__ */ p(pe, { className: I ? "size-6" : L ? "size-4" : "size-5" })
 			}),
-			/* @__PURE__ */ g("button", {
+			/* @__PURE__ */ p("button", {
 				type: "button",
 				onClick: () => F(R),
 				className: `absolute ${I ? "bottom-6 left-6 size-14" : L ? "left-2 top-2 size-8" : "bottom-3 left-3 size-11"} flex items-center justify-center rounded-full bg-bg-elevated text-text-secondary transition-colors hover:bg-bg-raised hover:text-text-primary focus:outline-none focus-visible:ring-2 focus-visible:ring-border-strong`,
 				"aria-label": R === "panel" ? k("voiceSession.expandPanel", "Show the conversation in the panel") : R === "full" ? k("voiceSession.expand", "Expand the conversation to full screen") : k("voiceSession.collapse", "Shrink to the compact controller"),
 				"data-testid": "assistant-voice-expand",
-				children: I ? /* @__PURE__ */ g(le, { className: "size-5" }) : /* @__PURE__ */ g(ce, { className: L ? "size-4" : "size-5" })
+				children: I ? /* @__PURE__ */ p(fe, { className: "size-5" }) : /* @__PURE__ */ p(de, { className: L ? "size-4" : "size-5" })
 			})
 		]
 	});
 };
-function de(e, t) {
+function me(e, t) {
 	switch (e) {
 		case "network": return t("voiceSession.noticeNetwork", "Speech recognition lost its connection. Check your network — it will retry.");
 		case "not-capturing": return t("voiceSession.noticeNotCapturing", "The microphone did not start. Check that this site is allowed to use it, then try again.");
@@ -236,7 +256,7 @@ function de(e, t) {
 		default: return e;
 	}
 }
-function fe(e, t) {
+function he(e, t) {
 	switch (e) {
 		case "mic-denied": return t("voiceSession.micDenied", "Microphone access is blocked. Allow it in your browser settings to talk.");
 		case "no-speech-support": return t("voiceSession.noSpeechSupport", "This browser cannot listen. Try a different browser, or type instead.");
@@ -245,4 +265,4 @@ function fe(e, t) {
 	}
 }
 //#endregion
-export { v as AssistantVoiceSession };
+export { h as AssistantVoiceSession };

package/dist/shared/assistant/voice/handsFree.d.ts

@@ -0,0 +1,160 @@
+/**
+ * Hands-free voice, steps H1 and H2 — the pure rules (BOFF hands-free slice).
+ *
+ * Everything here is data in, decision out: no DOM, no timers, no React, for
+ * the reason `voiceSession.ts` gives (this repo's unit runner has no browser).
+ * The hook (`useVoiceSession`) owns the clock and the recogniser; this file
+ * owns what the words and the preference MEAN.
+ *
+ * ★ H1 — the end-of-speech grace is a USER PREFERENCE (2 s default, 3 s, 5 s),
+ * and there is exactly ONE timing owner: `planSubmitTimer`. The old
+ * `SILENCE_SUBMIT_MS` constant is gone; the value the timer waits is the
+ * preference, resolved by `resolveVoiceGraceMs`. Nothing else in the package
+ * waits before submitting a spoken turn (`ENDPOINT_GRACE_MS` only decides when
+ * the stage SHOWS the pause, and is clamped inside the grace).
+ *
+ * ★ H2 — corrections while the assistant is still PREPARING an answer. The
+ * microphone stays open in `thinking`, so a recogniser can deliver three kinds
+ * of result there, and telling them apart is the whole difficulty:
+ *   1. a trailing result of the turn JUST submitted (Chrome finalises late;
+ *      Chrome Android re-sends the whole utterance, sometimes revised) — must
+ *      be ignored, or every spoken question is aborted and sent twice;
+ *   2. noise and fillers ("uh", "hmm") — must not abort a model turn;
+ *   3. a person speaking — a correction: the turn is superseded.
+ * `planThinkingSpeech` is that classification.
+ *
+ * A spoken "yes" never commits a business write. In `thinking` a bare
+ * affirmation is therefore NOT a correction at all: it neither aborts the turn
+ * nor withdraws anything (`classifyWriteReply` is the same classifier the 5a
+ * binder uses). The binder, and the card's button, stay the only way a change
+ * is confirmed.
+ */
+/** The choices offered in the voice settings, shortest first. */
+export declare const VOICE_GRACE_CHOICES_MS: readonly [2000, 3000, 5000];
+export type VoiceGraceMs = (typeof VOICE_GRACE_CHOICES_MS)[number];
+/**
+ * How long the user must stop talking before the turn is submitted, until they
+ * choose otherwise.
+ *
+ * ★ 2000, not 1200: 1.2 s does not survive an ordinary pause. Once a turn is
+ * submitted the reducer discards the rest of the sentence, so cutting someone
+ * off costs the whole round trip AND makes them repeat themselves — waiting
+ * 800 ms longer costs only a moment of silence. The asymmetry is what settled
+ * the number, and it is why the longer choices exist for people who pause more.
+ */
+export declare const DEFAULT_VOICE_GRACE_MS: VoiceGraceMs;
+/**
+ * Read a stored or passed value as one of the offered choices.
+ *
+ * Anything else — a stale number from an older build, a string, `null`, a value
+ * hand-edited into storage — is the default, never a nearby choice: a grace of
+ * 0 or 600000 ms read back from storage would either submit mid-word or never
+ * submit at all.
+ */
+export declare function resolveVoiceGraceMs(value: unknown): VoiceGraceMs;
+/**
+ * The ONE decision about the submit timer.
+ *
+ * `arm` is true only for a listening, unmuted session with something worth
+ * sending, and NOT while the person asked to finish in their own time
+ * ("Let me finish", `holding`). `delayMs` is the preference. The hook re-asks
+ * on every change of the session, so new speech cancels a pending submit and
+ * the turn continues — there is no second timer to forget.
+ */
+export declare function planSubmitTimer(input: {
+    phase: string;
+    muted: boolean;
+    /** Everything heard this turn (`spokenSoFar`). */
+    spoken: string;
+    /** "Let me finish" is on: nothing is submitted until the person says done. */
+    holding: boolean;
+    /** The preference, in ms. */
+    graceMs: number;
+}): {
+    arm: boolean;
+    delayMs: number;
+};
+/**
+ * The comparable words of a transcript: letters, marks and digits only (marks
+ * kept — Tamil and Hindi vowel signs are marks), lower-cased. Punctuation and
+ * capitals a final adds are not speech.
+ */
+export declare function speechWords(text: string): string[];
+/**
+ * Add a newly finalised phrase to the turn heard so far.
+ *
+ * ★ Chrome Android duplicates finals. With `continuous` on, a restarted
+ * recogniser can re-send the phrase it just finalised, and it often re-sends
+ * the WHOLE utterance so far with a few words added. Plain concatenation turned
+ * "when should I sow" into "when should I sow when should I sow ragi" and sent
+ * that. So:
+ *  - a phrase that STARTS WITH everything heard is the same utterance, longer
+ *    or equal: it replaces what was heard;
+ *  - a phrase of two or more words that the turn already ENDS with is a
+ *    repeat: dropped;
+ *  - anything else is new speech: appended.
+ * A single repeated word is kept unless it is the whole turn — "no no" is
+ * emphasis, but a lone "hello" arriving twice is the recogniser.
+ */
+export declare function mergeFinalPhrase(pending: string, phrase: string): string;
+/**
+ * A transcript without the words of the turn it follows, when it starts with
+ * them (a cumulative result: "when should I sow ragi" after "when should I
+ * sow"). Anything else is returned unchanged.
+ */
+export declare function stripSubmittedPrefix(transcript: string, submitted: string): string;
+/**
+ * How long after a turn is submitted its OWN trailing results may still arrive.
+ *
+ * The person was silent for the whole grace (2 s at least), so anything
+ * arriving this soon after the submit and resembling what was sent is the
+ * recogniser finishing the sentence, not a correction. Past it, only an exact
+ * repeat is ignored — a person correcting "plant ragi in June" with "no plant
+ * ragi in July" is speaking, however alike the words are.
+ */
+export declare const TRAILING_RESULT_WINDOW_MS = 1500;
+/**
+ * Is this transcript the recogniser finishing, repeating or revising the turn
+ * that was just submitted?
+ *
+ * `sinceSubmitMs` is null when unknown (no clock): then only an exact repeat
+ * counts, which can only ever err towards treating speech as speech.
+ */
+export declare function isTrailingResultOfSubmitted(transcript: string, submitted: string, sinceSubmitMs: number | null): boolean;
+/**
+ * Is this a person speaking, rather than noise?
+ *
+ * Aborting a model turn costs the turn, so the bar is two letters or digits
+ * that are not all filler. "no" and "ok" pass; "a", "uh" and "hmm" do not. A
+ * single Han, kana or Hangul character passes — "不" is a whole word.
+ */
+export declare function isGenuineUserSpeech(text: string): boolean;
+/** What a result heard while the answer is being prepared amounts to. */
+export type ThinkingSpeechPlan = {
+    kind: "ignore";
+    reason: "empty" | "trailing" | "noise" | "affirmation";
+} | {
+    kind: "supersede";
+    /** What the person said, without the words of the turn it corrects. */
+    text: string;
+};
+/**
+ * Classify a result that arrives while the assistant is PREPARING an answer.
+ *
+ * Ignored: nothing; the recogniser finishing, repeating or revising the turn
+ * just submitted; noise; and a bare affirmation (a spoken "yes" is never a
+ * correction, and never commits a write — see the header). Everything else is
+ * a correction.
+ *
+ * Echo is not checked here: nothing plays during `thinking`. The caller still
+ * passes the previous reply through `isEcho` first, because its tail can sound
+ * for a moment after playback ends.
+ */
+export declare function planThinkingSpeech(input: {
+    transcript: string;
+    /** The turn in flight, as it was sent. */
+    submitted: string;
+    /** Milliseconds since it was sent, or null when no clock was given. */
+    sinceSubmitMs: number | null;
+}): ThinkingSpeechPlan;
+//# sourceMappingURL=handsFree.d.ts.map

package/dist/shared/assistant/voice/handsFree.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"handsFree.d.ts","sourceRoot":"","sources":["../../../../src/shared/assistant/voice/handsFree.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA8BG;AAQH,iEAAiE;AACjE,eAAO,MAAM,sBAAsB,6BAA8B,CAAC;AAElE,MAAM,MAAM,YAAY,GAAG,CAAC,OAAO,sBAAsB,CAAC,CAAC,MAAM,CAAC,CAAC;AAEnE;;;;;;;;;GASG;AACH,eAAO,MAAM,sBAAsB,EAAE,YAAmB,CAAC;AAEzD;;;;;;;GAOG;AACH,wBAAgB,mBAAmB,CAAC,KAAK,EAAE,OAAO,GAAG,YAAY,CAEhE;AAED;;;;;;;;GAQG;AACH,wBAAgB,eAAe,CAAC,KAAK,EAAE;IACrC,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,OAAO,CAAC;IACf,kDAAkD;IAClD,MAAM,EAAE,MAAM,CAAC;IACf,8EAA8E;IAC9E,OAAO,EAAE,OAAO,CAAC;IACjB,6BAA6B;IAC7B,OAAO,EAAE,MAAM,CAAC;CACjB,GAAG;IAAE,GAAG,EAAE,OAAO,CAAC;IAAC,OAAO,EAAE,MAAM,CAAA;CAAE,CAIpC;AAMD;;;;GAIG;AACH,wBAAgB,WAAW,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,EAAE,CAMlD;AA6BD;;;;;;;;;;;;;;;GAeG;AACH,wBAAgB,gBAAgB,CAAC,OAAO,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,GAAG,MAAM,CAQxE;AAED;;;;GAIG;AACH,wBAAgB,oBAAoB,CAAC,UAAU,EAAE,MAAM,EAAE,SAAS,EAAE,MAAM,GAAG,MAAM,CAelF;AAMD;;;;;;;;GAQG;AACH,eAAO,MAAM,yBAAyB,OAAO,CAAC;AAK9C;;;;;;GAMG;AACH,wBAAgB,2BAA2B,CACzC,UAAU,EAAE,MAAM,EAClB,SAAS,EAAE,MAAM,EACjB,aAAa,EAAE,MAAM,GAAG,IAAI,GAC3B,OAAO,CAYT;AAyBD;;;;;;GAMG;AACH,wBAAgB,mBAAmB,CAAC,IAAI,EAAE,MAAM,GAAG,OAAO,CAOzD;AAED,yEAAyE;AACzE,MAAM,MAAM,kBAAkB,GAC1B;IAAE,IAAI,EAAE,QAAQ,CAAC;IAAC,MAAM,EAAE,OAAO,GAAG,UAAU,GAAG,OAAO,GAAG,aAAa,CAAA;CAAE,GAC1E;IACE,IAAI,EAAE,WAAW,CAAC;IAClB,uEAAuE;IACvE,IAAI,EAAE,MAAM,CAAC;CACd,CAAC;AAEN;;;;;;;;;;;GAWG;AACH,wBAAgB,kBAAkB,CAAC,KAAK,EAAE;IACxC,UAAU,EAAE,MAAM,CAAC;IACnB,0CAA0C;IAC1C,SAAS,EAAE,MAAM,CAAC;IAClB,uEAAuE;IACvE,aAAa,EAAE,MAAM,GAAG,IAAI,CAAC;CAC9B,GAAG,kBAAkB,CASrB"}

package/dist/shared/assistant/voice/handsFree.js

@@ -0,0 +1,104 @@
+import { classifyWriteReply as e } from "../writeReplyBinder.js";
+//#region src/shared/assistant/voice/handsFree.ts
+var t = [
+	2e3,
+	3e3,
+	5e3
+], n = 2e3;
+function r(e) {
+	return t.find((t) => t === e) ?? 2e3;
+}
+function i(e) {
+	let t = r(e.graceMs);
+	return {
+		arm: e.phase === "listening" && !e.muted && !e.holding && e.spoken.trim().length > 0,
+		delayMs: t
+	};
+}
+function a(e) {
+	let t = e.toLowerCase().replace(/[^\p{L}\p{M}\p{N}]+/gu, " ").trim();
+	return t ? t.split(" ") : [];
+}
+var o = (e, t) => t.length <= e.length && t.every((t, n) => e[n] === t), s = (e, t) => t.length <= e.length && t.every((n, r) => e[e.length - t.length + r] === n), c = (e, t) => {
+	if (t.length === 0) return !0;
+	for (let n = 0; n + t.length <= e.length; n += 1) if (t.every((t, r) => e[n + r] === t)) return !0;
+	return !1;
+};
+function l(e, t) {
+	let n = Array(t.length + 1).fill(0);
+	for (let r of e) {
+		let e = Array(t.length + 1).fill(0);
+		for (let i = 1; i <= t.length; i += 1) e[i] = r === t[i - 1] ? (n[i - 1] ?? 0) + 1 : Math.max(n[i] ?? 0, e[i - 1] ?? 0);
+		n = e;
+	}
+	return n[t.length] ?? 0;
+}
+function u(e, t) {
+	let n = a(e), r = a(t);
+	return r.length === 0 ? e : n.length === 0 || o(r, n) ? t.replace(/\s+/g, " ").trim() : r.length >= 2 && s(n, r) ? e : `${e} ${t}`.replace(/\s+/g, " ").trim();
+}
+function d(e, t) {
+	let n = a(t);
+	if (n.length === 0) return e.replace(/\s+/g, " ").trim();
+	let r = e.split(/\s+/).filter(Boolean), i = 0;
+	for (let t = 0; t < r.length; t += 1) {
+		let o = a(r[t] ?? "");
+		if (o.length !== 0) for (let a of o) {
+			if (a !== n[i]) return e.replace(/\s+/g, " ").trim();
+			if (i += 1, i === n.length) return r.slice(t + 1).join(" ").trim();
+		}
+	}
+	return e.replace(/\s+/g, " ").trim();
+}
+var f = 1500, p = .6;
+function m(e, t, n) {
+	let r = a(e), i = a(t);
+	return r.length === 0 ? !0 : i.length === 0 ? !1 : r.length === i.length && r.every((e, t) => e === i[t]) ? !0 : n === null || n >= 1500 ? !1 : c(i, r) ? !0 : o(r, i) && r.length - i.length >= 2 ? !1 : l(r, i) / Math.max(r.length, i.length) >= p;
+}
+var h = new Set([
+	"uh",
+	"um",
+	"uhm",
+	"umm",
+	"er",
+	"erm",
+	"eh",
+	"ah",
+	"hm",
+	"hmm",
+	"hmmm",
+	"mm",
+	"mmm",
+	"mhm",
+	"huh",
+	"oh"
+]), g = /[\p{Script=Han}\p{Script=Hiragana}\p{Script=Katakana}\p{Script=Hangul}]/u;
+function _(e) {
+	let t = a(e);
+	if (t.length === 0 || t.every((e) => h.has(e))) return !1;
+	let n = [...t.join("")];
+	return n.length >= 2 ? !0 : n.length === 1 && g.test(n[0] ?? "");
+}
+function v(t) {
+	if (a(t.transcript).length === 0) return {
+		kind: "ignore",
+		reason: "empty"
+	};
+	if (m(t.transcript, t.submitted, t.sinceSubmitMs)) return {
+		kind: "ignore",
+		reason: "trailing"
+	};
+	let n = d(t.transcript, t.submitted);
+	return _(n) ? e(n, "en") === "affirm" ? {
+		kind: "ignore",
+		reason: "affirmation"
+	} : {
+		kind: "supersede",
+		text: n
+	} : {
+		kind: "ignore",
+		reason: "noise"
+	};
+}
+//#endregion
+export { n as DEFAULT_VOICE_GRACE_MS, f as TRAILING_RESULT_WINDOW_MS, t as VOICE_GRACE_CHOICES_MS, _ as isGenuineUserSpeech, m as isTrailingResultOfSubmitted, u as mergeFinalPhrase, i as planSubmitTimer, v as planThinkingSpeech, r as resolveVoiceGraceMs, a as speechWords, d as stripSubmittedPrefix };