@allmodels/dsh-speech 0.1.0 → 0.1.2

package/README.md

@@ -33,7 +33,7 @@ Or create a portable tarball:
 
 ```sh
 pnpm pack
-dsh plugin --profile web add ./allmodels-dsh-speech-0.1.0.tgz
+dsh plugin --profile web add ./allmodels-dsh-speech-0.1.2.tgz
 ```
 
 Restart the profile after adding or removing the bundle.
@@ -54,9 +54,9 @@ See the [AllModels agentic setup documentation](https://docs.allmodels.io/agenti
 1. Click the microphone next to Send.
 2. Speak while the amplitude bar and partial transcription update.
 3. While recording, choose **System default** or another input from the microphone selector beside TTFT. On a new chat, the selector appears beside the mode selector in the hero controls row.
-4. Click Stop. The plugin waits briefly for the last final transcript, unlocks the composer, and leaves the text as an editable draft.
+4. Click Stop to finalize and leave the text as an editable draft, or click the Send arrow to finalize and send in one action. The Send arrow becomes available after speech text first appears.
 
-Voice input never sends the message automatically. Existing draft text is preserved and the transcript is appended. Composer editing and Send are disabled while the microphone is active.
+Existing draft text is preserved and the transcript is appended. Composer editing remains locked while the microphone is active. Stop never sends; the recording Send arrow submits exactly once after finalization.
 
 Only one microphone may be active at a time. Leaving the originating session immediately stops capture, closes the speech socket, clears the transient composer lock, and preserves the latest draft text.
 
@@ -96,7 +96,7 @@ The bundle row may be given composition-level defaults in a Harness patch:
 
 - Audio is resampled in the browser and relayed through the local plugin host only while recording. It is never written to disk or retained for retry.
 - API keys remain host-side in the Harness credential service and are inserted only into authenticated AllModels requests.
-- Transcripts exist only in the browser's existing unsent composer draft until the user sends them normally.
+- Transcripts remain in the browser's composer draft until the user uses the normal Send arrow or the recording Send arrow.
 - The preferred microphone device ID is kept only in plugin-namespaced browser-local storage. It is not written to Harness sessions.
 - The host plugin does not inject Harness session services, import session packages, read session files, or write session files. A build check fails if a session/filesystem persistence dependency is introduced.
 - The plugin's custom HTTP and WebSocket routes accept only same-origin loopback requests and enforce bounded request/frame sizes.
@@ -117,14 +117,29 @@ The only durable plugin data is its namespaced settings, credential, and browser
 ```sh
 pnpm install
 pnpm check
+pnpm exec playwright install chromium
+pnpm test:e2e
 pnpm pack --dry-run
 ```
 
 `pnpm check` runs the no-session-access boundary, TypeScript, unit/protocol tests, and both production bundles. Tests cover catalog defaults, provider capabilities, targeted promotional balances, transcript sequencing/CJK spacing, route security, credential redaction behavior, and the AllModels streaming wire mapping.
 
+`pnpm test:e2e` builds and packs the plugin, installs that tarball into a disposable DeepSeek Harness `0.1.1-rc.2` web profile, launches a local AllModels HTTP/WebSocket mock, and runs the Chromium desktop and mobile suite with a generated virtual microphone. It covers first-run account UI, API-key connection, settings layout and persistence, balance/top-up, microphone placement, preparing/recording/finishing states, live partials and finals, waveform motion, composer locking, Stop-without-send, finish-and-send, Cancel, microphone selection, dark-mode menus, and mobile layout. The disposable profile and workspace are created below the operating system's temporary directory; the runner never opens or modifies the user's Harness profile or sessions.
+
+For the optional live smoke test, put the following in a local `.secrets` file (which is gitignored), or export the variables normally:
+
+```dotenv
+ALLMODELS_API_KEY=...
+DEEPSEEK_API_KEY=...
+```
+
+Then run `pnpm test:e2e:live`. The live test checks the environment-managed AllModels account, starts and stops a real streaming STT connection using the virtual microphone, sends a minimal DeepSeek request, and verifies microphone placement in both new-chat and existing-session layouts. It consumes a small amount of API balance. The coordinator removes both credentials from its ambient environment before starting build, package-manager, and Playwright children; only the isolated Harness process receives them. Live Harness output and Playwright traces, screenshots, videos, HTML reports, and artifact uploads are disabled so credentials cannot be reflected into CI diagnostics.
+
 ## Release
 
-Pull requests run the full check and package dry run. A `v*` GitHub release tag runs the same gates and publishes with npm provenance after the package owner configures this repository as an npm trusted publisher.
+Every branch push and pull request to `main` runs the static/unit/package gates followed by the deterministic Playwright suite. A push to `main` additionally runs the live browser smoke test with the `ALLMODELS_API_KEY` and `DEEPSEEK_API_KEY` repository secrets. Secrets are never exposed to pull-request jobs.
+
+After all jobs for a `main` push pass, the Release workflow checks out the exact tested commit and publishes its package version to npm with provenance if that version is not already present. It then tags the commit and creates the matching GitHub release. Before enabling this gate, configure both repository secrets and configure `allmodels-io/dsh-speech`'s `release.yml` workflow as the npm trusted publisher for `@allmodels/dsh-speech`. Increment `package.json` before merging a version intended for publication; already-published versions are safely skipped.
 
 ## License
 

package/lib/client.js

@@ -330,7 +330,8 @@ var SpeechController = class {
 		amplitude: 0,
 		metadataLoading: false,
 		microphones: [],
-		switchingMicrophone: false
+		switchingMicrophone: false,
+		hasTranscript: false
 	};
 	listeners = /* @__PURE__ */ new Set();
 	blocks;
@@ -408,6 +409,18 @@ var SpeechController = class {
 		}, 5e3);
 		await stoppingCapture;
 	}
+	async stopAndSend() {
+		const active = this.active;
+		if (active === void 0 || active.finished || !this.snapshot.hasTranscript) return;
+		active.sendAfterFinish = true;
+		await this.stop();
+	}
+	cancel() {
+		const active = this.active;
+		if (active === void 0 || active.finished) return;
+		active.inputActions.setDraft(active.draft);
+		this.finish(active);
+	}
 	async switchMicrophone(deviceId) {
 		const active = this.active;
 		if (active === void 0 || active.finished || this.snapshot.phase !== "recording") return;
@@ -467,7 +480,8 @@ var SpeechController = class {
 			microphones: [],
 			activeMicrophoneId: void 0,
 			switchingMicrophone: false,
-			microphoneError: void 0
+			microphoneError: void 0,
+			hasTranscript: false
 		});
 		try {
 			await new Promise((resolve, reject) => {
@@ -549,6 +563,7 @@ var SpeechController = class {
 						...typeof message.languageCode === "string" ? { languageCode: message.languageCode } : {}
 					});
 					active.inputActions.setDraft(transcriptText(active.transcript));
+					if (message.text.trim().length > 0 && !this.snapshot.hasTranscript) this.publish({ hasTranscript: true });
 					if (message.type === "final" && this.snapshot.phase === "finalizing") {
 						if (active.finalTimer !== void 0) clearTimeout(active.finalTimer);
 						active.finalTimer = setTimeout(() => {
@@ -589,8 +604,10 @@ var SpeechController = class {
 			activeMicrophoneId: void 0,
 			switchingMicrophone: false,
 			microphoneError: void 0,
+			hasTranscript: false,
 			...error === void 0 ? { error: void 0 } : { error }
 		});
+		if (error === void 0 && active.sendAfterFinish === true) active.inputActions.submit();
 		this.ensureMetadata(true);
 	}
 	publish(patch) {
@@ -660,7 +677,10 @@ const styles = {
 	recordingTakeover: "dsh-speech-recording-takeover",
 	recordingTrack: "dsh-speech-recording-track",
 	recordingCanvas: "dsh-speech-recording-canvas",
+	recordingCancel: "dsh-speech-recording-cancel",
+	recordingActions: "dsh-speech-recording-actions",
 	recordingStop: "dsh-speech-recording-stop",
+	recordingSend: "dsh-speech-recording-send",
 	recordingProgress: "dsh-speech-recording-progress",
 	deviceDock: "dsh-speech-device-dock",
 	deviceMarker: "dsh-speech-device-marker",
@@ -692,7 +712,7 @@ const STYLE_TEXT = String.raw`
 .dsh-speech-primary,.dsh-speech-secondary,.dsh-speech-checkout{border:1px solid transparent;border-radius:8px;padding:8px 12px;font:inherit;font-size:13px;font-weight:600;cursor:pointer;text-decoration:none;text-align:center}.dsh-speech-primary{background:#e8e8e8;color:#151515}.dsh-speech-secondary,.dsh-speech-checkout{border-color:var(--color-border,#4a4a4a);background:transparent;color:inherit}.dsh-speech-primary:disabled,.dsh-speech-secondary:disabled{cursor:default;opacity:.45}.dsh-speech-good{color:#65c88a}.dsh-speech-warning{margin:0;color:#efb85c}.dsh-speech-danger,.dsh-speech-dock-error{margin:0;color:#f07878}
 .dsh-speech-top-up{justify-content:flex-start;align-items:end;flex-wrap:wrap}.dsh-speech-top-up .dsh-speech-field{max-width:180px}
 .dsh-speech-mic-wrap{position:relative;display:inline-flex;align-items:center;gap:5px;order:1}[data-slot="conversation.input.right"]:has(>.dsh-speech-mic-wrap)~button:last-child{order:2}.dsh-speech-mic{width:30px;height:30px;display:inline-grid;place-items:center;padding:0;border:0;border-radius:7px;background:transparent;color:var(--color-text-secondary,#a8a8a8);cursor:pointer}.dsh-speech-mic:hover:not(:disabled),.dsh-speech-mic:focus-visible{background:var(--color-background-hover,rgba(255,255,255,.08));color:var(--color-text,#eee)}.dsh-speech-mic[aria-disabled="true"]{opacity:.48}.dsh-speech-mic:disabled{cursor:default}.dsh-speech-mic svg{width:18px;height:18px;fill:currentColor}.dsh-speech-mic-active{color:#ef6666;background:rgba(239,102,102,.12)}.dsh-speech-mic-tooltip{position:absolute;right:0;bottom:calc(100% + 8px);z-index:40;width:max-content;max-width:260px;padding:6px 9px;border:1px solid var(--color-border,#343434);border-radius:7px;background:var(--color-background-elevated,#202020);color:var(--color-text,#eee);font-size:12px;line-height:1.35;box-shadow:0 6px 18px rgba(0,0,0,.24);opacity:0;visibility:hidden;transform:translateY(2px);pointer-events:none;transition:opacity .12s ease,transform .12s ease,visibility .12s}.dsh-speech-mic-wrap:hover .dsh-speech-mic-tooltip,.dsh-speech-mic-wrap:focus-within .dsh-speech-mic-tooltip,.dsh-speech-mic-wrap[data-explanation-open="true"] .dsh-speech-mic-tooltip{opacity:1;visibility:visible;transform:translateY(0)}
-:where(div):has(>:where(div)>[data-slot="conversation.input.right"]>.dsh-speech-recording-takeover){position:relative}:where(div):has(>:where(div)>[data-slot="conversation.input.right"]>.dsh-speech-recording-takeover)>*{visibility:hidden}.dsh-speech-recording-takeover{position:absolute;inset:0;z-index:8;visibility:visible;display:flex;align-items:center;gap:10px;padding:4px 8px;box-sizing:border-box;color:var(--color-text-secondary,#a8a8a8)}.dsh-speech-recording-track{height:30px;min-width:0;flex:1;display:flex;align-items:center}.dsh-speech-recording-canvas{display:block;width:100%;height:30px;color:var(--color-text-secondary,#a8a8a8)}.dsh-speech-recording-stop{flex:none;border-radius:999px;color:#ef6666;background:rgba(239,102,102,.12)}.dsh-speech-recording-stop:hover:not(:disabled),.dsh-speech-recording-stop:focus-visible{color:#ff7777;background:rgba(239,102,102,.2)}.dsh-speech-recording-progress{display:flex;align-items:center;justify-content:flex-end;gap:8px;width:100%;color:var(--color-text-secondary,#a8a8a8);font-size:12px}.dsh-speech-recording-takeover[data-phase="starting"] .dsh-speech-recording-progress{justify-content:center}.dsh-speech-recording-progress i{width:13px;height:13px;border:2px solid currentColor;border-right-color:transparent;border-radius:50%;animation:dsh-speech-spin .8s linear infinite}.dsh-speech-sr-only{position:absolute;width:1px;height:1px;padding:0;margin:-1px;overflow:hidden;clip:rect(0,0,0,0);white-space:nowrap;border:0}@keyframes dsh-speech-spin{to{transform:rotate(360deg)}}
+:where(div):has(>:where(div)>[data-slot="conversation.input.right"]>.dsh-speech-recording-takeover){position:relative}:where(div):has(>:where(div)>[data-slot="conversation.input.right"]>.dsh-speech-recording-takeover)>*{visibility:hidden}.dsh-speech-recording-takeover{position:absolute;inset:0;z-index:8;visibility:visible;display:flex;align-items:center;gap:10px;padding:4px 8px;box-sizing:border-box;color:var(--color-text-secondary,#a8a8a8)}.dsh-speech-recording-track{height:30px;min-width:0;flex:1;display:flex;align-items:center}.dsh-speech-recording-canvas{display:block;width:100%;height:30px;color:var(--color-text-secondary,#a8a8a8)}.dsh-speech-recording-cancel,.dsh-speech-recording-stop{flex:none;border-radius:999px}.dsh-speech-recording-cancel{color:var(--color-text-secondary,#a8a8a8);background:transparent}.dsh-speech-recording-cancel:hover:not(:disabled),.dsh-speech-recording-cancel:focus-visible{background:var(--color-background-hover,rgba(255,255,255,.08));color:var(--color-text,#eee)}.dsh-speech-recording-actions{display:flex;align-items:center;gap:6px;flex:none}.dsh-speech-recording-stop{color:#ef6666;background:rgba(239,102,102,.12)}.dsh-speech-recording-stop:hover:not(:disabled),.dsh-speech-recording-stop:focus-visible{color:#ff7777;background:rgba(239,102,102,.2)}.dsh-speech-recording-send{width:34px;height:34px;display:inline-grid;place-items:center;flex:none;padding:0;border:0;border-radius:999px;background:var(--dsw-alias-button-info-fill,#679efe);color:#fff;cursor:pointer}.dsh-speech-recording-send:hover:not(:disabled),.dsh-speech-recording-send:focus-visible{outline:0;background:var(--dsw-alias-button-info-hover,#4176e6)}.dsh-speech-recording-send:disabled{cursor:default;opacity:.4}.dsh-speech-recording-send svg{width:16px;height:16px}.dsh-speech-recording-progress{display:flex;align-items:center;justify-content:flex-end;gap:8px;width:100%;color:var(--color-text-secondary,#a8a8a8);font-size:12px}.dsh-speech-recording-takeover[data-phase="starting"] .dsh-speech-recording-progress{justify-content:center}.dsh-speech-recording-progress i{width:13px;height:13px;border:2px solid currentColor;border-right-color:transparent;border-radius:50%;animation:dsh-speech-spin .8s linear infinite}.dsh-speech-sr-only{position:absolute;width:1px;height:1px;padding:0;margin:-1px;overflow:hidden;clip:rect(0,0,0,0);white-space:nowrap;border:0}@keyframes dsh-speech-spin{to{transform:rotate(360deg)}}
 .dsh-speech-device-marker{display:none}[data-slot="conversation.composer.dock"]>:has(>.dsh-speech-device-dock){overflow:visible}.dsh-speech-device-fallback{min-height:22px;display:flex;align-items:center;justify-content:flex-start;padding:0 8px;overflow:visible}.dsh-speech-device-fallback .dsh-speech-device-dock{margin-left:0}.dsh-speech-device-separator{margin-left:4px;color:var(--dsw-alias-label-tertiary,var(--color-text-secondary,#888))}.dsh-speech-device-dock{position:relative;display:inline-flex;vertical-align:middle;margin-left:5px;color:var(--dsw-alias-label-secondary,var(--color-text-secondary,#a8a8a8));font:inherit}.dsh-speech-device-trigger{max-width:126px;height:20px;display:inline-flex;align-items:center;gap:4px;padding:0 5px;border:0;border-radius:6px;background:transparent;color:inherit;font:inherit;font-size:11px;line-height:20px;cursor:pointer}.dsh-speech-device-trigger:hover,.dsh-speech-device-trigger:focus-visible,.dsh-speech-device-trigger[aria-expanded="true"]{outline:0;background:var(--dsw-alias-interactive-bg-hover,rgba(255,255,255,.08));color:var(--dsw-alias-label-primary,var(--color-text,#eee))}.dsh-speech-device-trigger:disabled{cursor:wait;opacity:.6}.dsh-speech-device-dock[data-error="true"] .dsh-speech-device-trigger{color:var(--dsw-alias-state-error-primary,#f07878)}.dsh-speech-device-icon,.dsh-speech-device-chevron{display:inline-flex;flex:none}.dsh-speech-device-icon svg{width:12px;height:12px}.dsh-speech-device-chevron svg{width:10px;height:10px}.dsh-speech-device-label{min-width:0;overflow:hidden;text-overflow:ellipsis;white-space:nowrap}.dsh-speech-device-menu{position:absolute;right:0;bottom:calc(100% + 6px);z-index:80;min-width:240px;max-width:min(320px,calc(100vw - 32px));display:grid;box-sizing:border-box;padding:4px;border:1px solid var(--dsw-alias-border-l2-darkmode-thin,rgba(255,255,255,.06));border-radius:12px;background:var(--dsw-alias-bg-layer-3,#353638);color:var(--dsw-alias-label-primary,#f9fafb);box-shadow:0 0 1px rgba(0,0,0,.2),0 0 4px rgba(0,0,0,.02),0 12px 32px rgba(0,0,0,.08);font:13px/20px -apple-system,BlinkMacSystemFont,"Segoe UI",sans-serif}.dsh-speech-device-menu-item{width:100%;min-width:0;height:38px;display:flex;align-items:center;justify-content:space-between;gap:12px;padding:0 10px;border:0;border-radius:10px;background:transparent;color:inherit;font:inherit;text-align:left;cursor:pointer}.dsh-speech-device-menu-item:hover,.dsh-speech-device-menu-item:focus-visible{outline:0;background:var(--dsw-alias-interactive-bg-hover,rgba(255,255,255,.08))}.dsh-speech-device-menu-item>span:first-child{min-width:0;overflow:hidden;text-overflow:ellipsis;white-space:nowrap}.dsh-speech-device-check{width:14px;height:14px;display:inline-flex;flex:none;color:var(--dsw-alias-brand-primary,var(--dsw-alias-label-primary,#f9fafb))}.dsh-speech-device-check svg{width:14px;height:14px}.dsh-speech-device-dock[data-variant="hero"]{margin-left:0;color:var(--dsw-alias-label-primary,var(--color-text,#f9fafb));font:500 13px/20px -apple-system,BlinkMacSystemFont,"Segoe UI","PingFang SC","Hiragino Sans GB","Microsoft YaHei","Helvetica Neue",Helvetica,Arial,sans-serif}.dsh-speech-device-dock[data-variant="hero"] .dsh-speech-device-trigger{max-width:220px;height:28px;padding:0 8px;gap:4px;border-radius:16px;font:inherit;line-height:20px}.dsh-speech-device-dock[data-variant="hero"] .dsh-speech-device-icon svg{width:16px;height:16px}.dsh-speech-device-dock[data-variant="hero"] .dsh-speech-device-chevron svg{width:14px;height:14px}.dsh-speech-device-dock[data-variant="hero"] .dsh-speech-device-menu{left:0;right:auto;top:calc(100% + 6px);bottom:auto;min-width:218px}
 [data-dsh-speech-nav]>svg{display:none}[data-dsh-speech-nav]::before{content:"";width:16px;height:16px;flex:none;background:currentColor;mask:url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 20 20'%3E%3Crect x='7' y='2' width='6' height='11' rx='3' fill='none' stroke='black' stroke-width='1.7'/%3E%3Cpath d='M4.5 9.5a5.5 5.5 0 0 0 11 0M10 15v3M7 18h6' fill='none' stroke='black' stroke-width='1.7' stroke-linecap='round'/%3E%3C/svg%3E") center/contain no-repeat}
 .dsh-speech-dock,.dsh-speech-dock-error{min-height:22px;display:flex;align-items:center;gap:9px;padding:4px 8px 0;font-size:12px}.dsh-speech-dock-detail{color:var(--color-text-secondary,#8d8d8d)}
@@ -762,6 +782,29 @@ function CheckIcon() {
 		})
 	});
 }
+function CloseIcon() {
+	return /* @__PURE__ */ (0, react_jsx_runtime.jsx)("svg", {
+		viewBox: "0 0 20 20",
+		"aria-hidden": "true",
+		children: /* @__PURE__ */ (0, react_jsx_runtime.jsx)("path", {
+			d: "m5.5 5.5 9 9m0-9-9 9",
+			fill: "none",
+			stroke: "currentColor",
+			strokeWidth: "1.7",
+			strokeLinecap: "round"
+		})
+	});
+}
+function SendIcon() {
+	return /* @__PURE__ */ (0, react_jsx_runtime.jsx)("svg", {
+		viewBox: "0 0 16 16",
+		"aria-hidden": "true",
+		children: /* @__PURE__ */ (0, react_jsx_runtime.jsx)("path", {
+			d: "M8.3125.980183c.35517.072917.66652.223997.9502.452147.22454.18064.46759.4256.71679.67481L14.707 6.83468l-1.414 1.41406L9 3.95577V15.0417H7V3.95577L2.70703 8.24874 1.29297 6.83468l4.72754-4.72754c.2492-.24921.49226-.49417.71679-.67481.23932-.19247.54715-.38831.9502-.452147.2098-.033177.4156-.025023.625 0Z",
+			fill: "currentColor"
+		})
+	});
+}
 const WAVEFORM_PITCH = 4;
 const WAVEFORM_BASELINE = .025;
 function RecordingWaveform({ amplitude }) {
@@ -831,12 +874,21 @@ function RecordingWaveform({ amplitude }) {
 		"aria-hidden": "true"
 	});
 }
-function RecordingTakeover({ amplitude, disabled, phase, label, onStop, stopLabel }) {
+function RecordingTakeover({ amplitude, cancelLabel, disabled, phase, label, onCancel, onSend, onStop, sendDisabled, sendLabel, stopLabel }) {
 	const recording = phase === "recording";
 	return /* @__PURE__ */ (0, react_jsx_runtime.jsxs)("div", {
 		className: styles.recordingTakeover,
 		"data-phase": phase,
 		children: [
+			recording ? /* @__PURE__ */ (0, react_jsx_runtime.jsx)("button", {
+				type: "button",
+				className: `${styles.mic} ${styles.recordingCancel}`,
+				disabled,
+				"aria-label": cancelLabel,
+				title: cancelLabel,
+				onClick: onCancel,
+				children: /* @__PURE__ */ (0, react_jsx_runtime.jsx)(CloseIcon, {})
+			}) : null,
 			/* @__PURE__ */ (0, react_jsx_runtime.jsx)("div", {
 				className: styles.recordingTrack,
 				children: recording ? /* @__PURE__ */ (0, react_jsx_runtime.jsx)(RecordingWaveform, { amplitude }) : /* @__PURE__ */ (0, react_jsx_runtime.jsxs)("span", {
@@ -846,14 +898,25 @@ function RecordingTakeover({ amplitude, disabled, phase, label, onStop, stopLabe
 					children: [/* @__PURE__ */ (0, react_jsx_runtime.jsx)("i", { "aria-hidden": "true" }), label]
 				})
 			}),
-			recording ? /* @__PURE__ */ (0, react_jsx_runtime.jsx)("button", {
-				type: "button",
-				className: `${styles.mic} ${styles.recordingStop}`,
-				disabled,
-				"aria-label": stopLabel,
-				title: stopLabel,
-				onClick: onStop,
-				children: /* @__PURE__ */ (0, react_jsx_runtime.jsx)(MicIcon, { stopped: true })
+			recording ? /* @__PURE__ */ (0, react_jsx_runtime.jsxs)("div", {
+				className: styles.recordingActions,
+				children: [/* @__PURE__ */ (0, react_jsx_runtime.jsx)("button", {
+					type: "button",
+					className: `${styles.mic} ${styles.recordingStop}`,
+					disabled,
+					"aria-label": stopLabel,
+					title: stopLabel,
+					onClick: onStop,
+					children: /* @__PURE__ */ (0, react_jsx_runtime.jsx)(MicIcon, { stopped: true })
+				}), /* @__PURE__ */ (0, react_jsx_runtime.jsx)("button", {
+					type: "button",
+					className: styles.recordingSend,
+					disabled: sendDisabled,
+					"aria-label": sendLabel,
+					title: sendLabel,
+					onClick: onSend,
+					children: /* @__PURE__ */ (0, react_jsx_runtime.jsx)(SendIcon, {})
+				})]
 			}) : null,
 			/* @__PURE__ */ (0, react_jsx_runtime.jsx)("span", {
 				className: styles.srOnly,
@@ -907,10 +970,21 @@ function SpeechMic({ controller, getLocale, sessionId, input, inputActions, t })
 		const takeoverPhase = state.phase === "starting" || state.phase === "finalizing" ? state.phase : "recording";
 		return /* @__PURE__ */ (0, react_jsx_runtime.jsx)(RecordingTakeover, {
 			amplitude: state.amplitude,
+			cancelLabel: t("micCancel"),
 			disabled,
 			phase: takeoverPhase,
 			label: state.phase === "starting" ? t("starting") : state.phase === "finalizing" ? t("finalizing") : t("listening"),
+			onCancel: () => {
+				controller.cancel();
+			},
+			onSend: () => {
+				controller.stopAndSend().catch((error) => {
+					controller.reportError(error instanceof Error ? error.message : "Unable to send voice input");
+				});
+			},
 			onStop: toggle,
+			sendDisabled: disabled || !state.hasTranscript,
+			sendLabel: t("micSend"),
 			stopLabel: title
 		});
 	}
@@ -1040,18 +1114,47 @@ function MicrophoneDockPortal({ controller, state, t }) {
 	const markerRef = (0, react.useRef)(null);
 	const [target, setTarget] = (0, react.useState)(null);
 	(0, react.useLayoutEffect)(() => {
-		const marker = markerRef.current;
-		const slot = marker?.closest("[data-slot=\"conversation.composer.dock\"]");
-		if (!(slot instanceof HTMLElement)) return;
+		let disposed = false;
+		let frame;
+		let observedSlot = null;
+		const observer = new MutationObserver(() => {
+			findTarget();
+		});
+		const retry = () => {
+			if (disposed || frame !== void 0) return;
+			frame = requestAnimationFrame(() => {
+				frame = void 0;
+				findTarget();
+			});
+		};
 		const findTarget = () => {
+			if (disposed) return;
+			const marker = markerRef.current;
+			const slot = marker?.closest("[data-slot=\"conversation.composer.dock\"]");
+			if (!(slot instanceof HTMLElement)) {
+				setTarget(null);
+				retry();
+				return;
+			}
+			if (observedSlot !== slot) {
+				observer.disconnect();
+				observedSlot = slot;
+				observer.observe(slot, {
+					childList: true,
+					subtree: true,
+					characterData: true
+				});
+			}
 			const metrics = Array.from(slot.children).find((child) => child !== marker && child.textContent?.includes("TTFT"));
-			setTarget(metrics instanceof HTMLElement ? metrics : null);
+			const nextTarget = metrics instanceof HTMLElement ? metrics : null;
+			setTarget((current) => current === nextTarget ? current : nextTarget);
+			if (nextTarget === null) retry();
 		};
 		findTarget();
-		const observer = new MutationObserver(findTarget);
-		observer.observe(slot, { childList: true });
 		return () => {
+			disposed = true;
 			observer.disconnect();
+			if (frame !== void 0) cancelAnimationFrame(frame);
 		};
 	}, []);
 	return /* @__PURE__ */ (0, react_jsx_runtime.jsxs)(react_jsx_runtime.Fragment, { children: [/* @__PURE__ */ (0, react_jsx_runtime.jsx)("span", {
@@ -1619,6 +1722,8 @@ const en = {
 	saved: "Saved",
 	micStart: "Start voice input",
 	micStop: "Stop voice input",
+	micSend: "Finish voice input and send",
+	micCancel: "Cancel voice input",
 	starting: "Preparing microphone…",
 	listening: "Listening…",
 	finalizing: "Finishing transcription…",
@@ -1680,6 +1785,8 @@ const zh = {
 	saved: "已保存",
 	micStart: "开始语音输入",
 	micStop: "停止语音输入",
+	micSend: "完成语音输入并发送",
+	micCancel: "取消语音输入",
 	starting: "正在准备麦克风…",
 	listening: "正在聆听…",
 	finalizing: "正在完成转写…",

package/lib/index.js

@@ -184,7 +184,10 @@ async function requestJson(baseURL, path, init = {}, apiKey) {
 			signal: controller.signal
 		});
 		const body = await response.json().catch(() => void 0);
-		if (!response.ok) throw new AllModelsError(response.status, `HTTP_${String(response.status)}`, errorMessage(body, "AllModels request failed"));
+		if (!response.ok) {
+			const message = apiKey === void 0 ? errorMessage(body, "AllModels request failed") : "AllModels rejected the authenticated request";
+			throw new AllModelsError(response.status, `HTTP_${String(response.status)}`, message);
+		}
 		return body;
 	} catch (error) {
 		if (error instanceof AllModelsError) throw error;
@@ -333,7 +336,7 @@ function upstreamUrl(settings, binding) {
 	if (binding.contextSupported && settings.context !== void 0 && settings.context.trim() !== "") base.searchParams.set("context", settings.context.trim());
 	return base;
 }
-function normalizedEvent(data) {
+function normalizedEvent(data, credential) {
 	try {
 		const text = typeof data === "string" ? data : data.toString("utf8");
 		const value = JSON.parse(text);
@@ -355,11 +358,15 @@ function normalizedEvent(data) {
 			text: event.text,
 			...languageCode === void 0 ? {} : { languageCode }
 		};
-		if (eventType === "stt.error") return {
-			type: "error",
-			code: typeof event.code === "string" ? event.code : "UPSTREAM_ERROR",
-			message: typeof event.message === "string" ? event.message : "Speech recognition failed"
-		};
+		if (eventType === "stt.error") {
+			const code = typeof event.code === "string" && /^[A-Za-z0-9_.-]{1,64}$/u.test(event.code) ? event.code : "UPSTREAM_ERROR";
+			const upstreamMessage = typeof event.message === "string" && event.message.length <= 500 ? event.message : "Speech recognition failed";
+			return {
+				type: "error",
+				code,
+				message: credential !== void 0 && credential.length > 0 && upstreamMessage.includes(credential) ? "Speech recognition failed" : upstreamMessage
+			};
+		}
 		if (eventType === "stt.session.ended" || eventType === "stt.ended") return { type: "ended" };
 		return;
 	} catch {
@@ -483,7 +490,7 @@ var SttProxy = class {
 			});
 		});
 		upstream.on("message", (data) => {
-			const event = normalizedEvent(data);
+			const event = normalizedEvent(data, credential.value);
 			if (event !== void 0) send(client, event);
 		});
 		upstream.on("error", () => {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@allmodels/dsh-speech",
-  "version": "0.1.0",
+  "version": "0.1.2",
   "description": "Streaming speech-to-text for DeepSeek Harness using AllModels.io.",
   "keywords": [
     "dsh-plugin",
@@ -61,6 +61,8 @@
     "clean": "rm -rf lib coverage",
     "test": "vitest run",
     "test:coverage": "vitest run --coverage",
+    "test:e2e": "node scripts/run-e2e.mjs mock",
+    "test:e2e:live": "node scripts/run-e2e.mjs live",
     "typecheck": "tsc --noEmit",
     "check:sessions": "node scripts/check-no-session-access.mjs",
     "check": "pnpm check:sessions && pnpm typecheck && pnpm test && pnpm build",
@@ -95,6 +97,7 @@
     "@deepseek-ai/dsh-settings": "0.1.1-rc.2",
     "@deepseek-ai/schemastery": "3.18.1",
     "@testing-library/react": "^16.3.0",
+    "@playwright/test": "^1.62.1",
     "@types/node": "^22.18.0",
     "@types/react": "^18.3.24",
     "@types/react-dom": "^18.3.7",