conductor-remote 1.127.1 → 1.128.0

package/dist/index.html

@@ -25,7 +25,7 @@
 		<title>Conductor Remote</title>
 		<!-- Runs before the module bundle so it can catch a stale shell that fails to boot. -->
 		<script src="/self-heal.js"></script>
-		<script type="module" crossorigin src="/assets/index-C0ZF9d17.js"></script>
+		<script type="module" crossorigin src="/assets/index-BVXbCLBD.js"></script>
 		<link rel="stylesheet" crossorigin href="/assets/index-q-Z5Zz9y.css">
 	<link rel="manifest" href="/manifest.webmanifest"></head>
 	<body>

package/dist/sw.js

@@ -1 +1 @@
-if(!self.define){let e,s={};const i=(i,n)=>(i=new URL(i+".js",n).href,s[i]||new Promise(s=>{if("document"in self){const e=document.createElement("script");e.src=i,e.onload=s,document.head.appendChild(e)}else e=i,importScripts(i),s()}).then(()=>{let e=s[i];if(!e)throw new Error(`Module ${i} didn’t register its module`);return e}));self.define=(n,r)=>{const l=e||("document"in self?document.currentScript.src:"")||location.href;if(s[l])return;let o={};const a=e=>i(e,l),t={module:{uri:l},exports:o,require:a};s[l]=Promise.all(n.map(e=>t[e]||a(e))).then(e=>(r(...e),o))}}define(["./workbox-dcde9eb3"],function(e){"use strict";importScripts("/push-sw.js"),self.addEventListener("message",e=>{e.data&&"SKIP_WAITING"===e.data.type&&self.skipWaiting()}),e.clientsClaim(),e.precacheAndRoute([{url:"self-heal.js",revision:"49bd63adb25a09341f8d2610e8bd3c76"},{url:"push-sw.js",revision:"4e0b714c0ea53e85c167e6d8f5b767aa"},{url:"index.html",revision:"6aa7324d208f306fe17f522d2bf7c728"},{url:"assets/workbox-window.prod.es5-BBnX5xw4.js",revision:null},{url:"assets/pierre-shiki-wasm-C8ETeZem.js",revision:null},{url:"assets/pierre-light-480U9XYS.js",revision:null},{url:"assets/pierre-dark-CyvmCCZW.js",revision:null},{url:"assets/index-q-Z5Zz9y.css",revision:null},{url:"assets/index-C0ZF9d17.js",revision:null},{url:"assets/PierrePatch-Cf1bAQPR.js",revision:null},{url:"apple-touch-icon.png",revision:"1127bb396b4648add53dce3f22c92aee"},{url:"icon-192.png",revision:"c5e01ac58768627e18ee7b8b6a9239ef"},{url:"icon-512.png",revision:"a40638c55e310312457a621c9a0002c8"},{url:"icon-maskable-512.png",revision:"a9b0d962686287452216492cd2247499"},{url:"icon.svg",revision:"c1aee186821798733dd477e69a0ef243"},{url:"manifest.webmanifest",revision:"cf88fbc5755108a7fe0616fa160a8a15"}],{}),e.cleanupOutdatedCaches(),e.registerRoute(new e.NavigationRoute(e.createHandlerBoundToURL("/index.html"),{denylist:[/^\/api\//]})),e.registerRoute(({url:e})=>e.pathname.startsWith("/assets/diff-syntax/"),new e.CacheFirst({cacheName:"diff-syntax",plugins:[new e.ExpirationPlugin({maxEntries:32,maxAgeSeconds:31536e3})]}),"GET")});
+if(!self.define){let e,s={};const i=(i,n)=>(i=new URL(i+".js",n).href,s[i]||new Promise(s=>{if("document"in self){const e=document.createElement("script");e.src=i,e.onload=s,document.head.appendChild(e)}else e=i,importScripts(i),s()}).then(()=>{let e=s[i];if(!e)throw new Error(`Module ${i} didn’t register its module`);return e}));self.define=(n,r)=>{const l=e||("document"in self?document.currentScript.src:"")||location.href;if(s[l])return;let o={};const t=e=>i(e,l),a={module:{uri:l},exports:o,require:t};s[l]=Promise.all(n.map(e=>a[e]||t(e))).then(e=>(r(...e),o))}}define(["./workbox-dcde9eb3"],function(e){"use strict";importScripts("/push-sw.js"),self.addEventListener("message",e=>{e.data&&"SKIP_WAITING"===e.data.type&&self.skipWaiting()}),e.clientsClaim(),e.precacheAndRoute([{url:"self-heal.js",revision:"49bd63adb25a09341f8d2610e8bd3c76"},{url:"push-sw.js",revision:"4e0b714c0ea53e85c167e6d8f5b767aa"},{url:"index.html",revision:"501e728a1efd2ebf7359c59176f6d3f6"},{url:"assets/workbox-window.prod.es5-BBnX5xw4.js",revision:null},{url:"assets/pierre-shiki-wasm-C8ETeZem.js",revision:null},{url:"assets/pierre-light-480U9XYS.js",revision:null},{url:"assets/pierre-dark-CyvmCCZW.js",revision:null},{url:"assets/index-q-Z5Zz9y.css",revision:null},{url:"assets/index-BVXbCLBD.js",revision:null},{url:"assets/PierrePatch-CmmIAkbQ.js",revision:null},{url:"apple-touch-icon.png",revision:"1127bb396b4648add53dce3f22c92aee"},{url:"icon-192.png",revision:"c5e01ac58768627e18ee7b8b6a9239ef"},{url:"icon-512.png",revision:"a40638c55e310312457a621c9a0002c8"},{url:"icon-maskable-512.png",revision:"a9b0d962686287452216492cd2247499"},{url:"icon.svg",revision:"c1aee186821798733dd477e69a0ef243"},{url:"manifest.webmanifest",revision:"cf88fbc5755108a7fe0616fa160a8a15"}],{}),e.cleanupOutdatedCaches(),e.registerRoute(new e.NavigationRoute(e.createHandlerBoundToURL("/index.html"),{denylist:[/^\/api\//]})),e.registerRoute(({url:e})=>e.pathname.startsWith("/assets/diff-syntax/"),new e.CacheFirst({cacheName:"diff-syntax",plugins:[new e.ExpirationPlugin({maxEntries:32,maxAgeSeconds:31536e3})]}),"GET")});

package/dist-node/src/http/routes/voice.js

@@ -2,6 +2,7 @@ import { isRoute, routeParam, routes } from "../../routes.js";
 import { isOpenAIRealtimeVoice, isVoiceLanguage, isVoiceSpeed, OPENAI_REALTIME_VOICES } from "../../shared.js";
 import { openAIOriginForSipHost } from "../../voice/config.js";
 import { parseVoiceCallTarget, readVoiceChatContext, VoiceContextError } from "../../voice/context.js";
+import { parseVoiceDiagnostics } from "../../voice/diagnostic-fields.js";
 import { MAX_VOICE_SEARCH_CHARS } from "../../voice/history.js";
 import { mintSipTicket, missingTicketConfig } from "../../voice/ticket.js";
 import { createWebRtcCall, MAX_SDP_CHARS } from "../../voice/webrtc.js";
@@ -10,6 +11,26 @@ export function createVoiceRoutes(services) {
     const { json, voiceHistory, voiceConfig, voiceBroker, readBody, reads, voiceSafetyIdentifier } = services;
     return async (req, res, url) => {
         const { pathname } = url;
+        const diagnosticCall = routeParam(routes.voiceCallDiagnostics, req.method, pathname);
+        if (diagnosticCall) {
+            if (!voiceHistory.status(diagnosticCall))
+                return json(req, res, 404, { error: 'Saved voice call not found' });
+            const raw = await readBody(req);
+            if (raw.length > 32_000)
+                return json(req, res, 413, { error: 'Voice diagnostic batch is too large' });
+            let events = null;
+            try {
+                events = parseVoiceDiagnostics(JSON.parse(raw));
+            }
+            catch { }
+            if (!events)
+                return json(req, res, 400, { error: 'Invalid voice diagnostic batch' });
+            // Keep each entry below the log ring's per-line cap, even after an offline batch.
+            const receivedAt = Date.now();
+            for (const event of events)
+                console.info(`[voice-diagnostics] ${JSON.stringify({ callId: diagnosticCall, source: 'browser', receivedAt, ...event })}`);
+            return json(req, res, 200, { ok: true });
+        }
         const voiceTranscript = routeParam(routes.voiceTranscript, req.method, pathname);
         if (isRoute(routes.voiceHistory, req.method, pathname) ||
             isRoute(routes.voiceSearch, req.method, pathname) ||

package/dist-node/src/routes.js

@@ -60,6 +60,7 @@ export const routes = {
     voiceCall: flat('POST', '/api/voice/calls'),
     /** Start the greeting only after the browser has installed OpenAI's SDP answer. */
     voiceCallReady: param('POST', '/api/voice/calls/:callId/ready'),
+    voiceCallDiagnostics: param('POST', '/api/voice/calls/:callId/diagnostics'),
     /** End one PWA-owned call. SIP calls remain owned by their phone leg. */
     voiceCallEnd: param('DELETE', '/api/voice/calls/:callId'),
     /** Durable call history remains readable even when voice is no longer configured. */

package/dist-node/src/shared.js

@@ -13,6 +13,7 @@
  * Each of these was a second copy before, and each copy was a way for two screens to
  * disagree about the same workspace.
  */
+export { voiceDiagnosticData, voiceRealtimeDiagnostic } from "./voice/diagnostic-fields.js";
 /** Stable wire values used by request validation and the phone's effort controls. */
 export const AGENT_EFFORTS = ['none', 'low', 'medium', 'high', 'xhigh', 'max', 'ultracode'];
 export function isAgentEffort(value) {

package/dist-node/src/voice/broker.js

@@ -3,6 +3,7 @@ import crypto from 'node:crypto';
 import fs from 'node:fs';
 import path from 'node:path';
 import { DEFAULT_VOICE_SPEED, voiceReasoning } from "./config.js";
+import { voiceRealtimeDiagnostic } from "./diagnostic-fields.js";
 import { VOICE_INSTRUCTIONS } from "./prompt.js";
 import { VOICE_TOOL_NAMES } from "./tools.js";
 import { voiceTranscription } from "./transcription.js";
@@ -371,6 +372,9 @@ export class VoiceBroker {
             return;
         }
         this.deps.history?.record(callId, event);
+        const diagnostic = voiceRealtimeDiagnostic(event);
+        if (diagnostic)
+            this.log(`[voice-diagnostics] ${JSON.stringify({ callId, source: 'server', receivedAt: this.now(), ...diagnostic })}`);
         const type = event.type;
         if (type === 'mcp_list_tools.completed') {
             if (runtime.mode !== 'mcp')

package/dist-node/src/voice/diagnostic-fields.js

@@ -0,0 +1,113 @@
+const EVENTS = new Set([
+    'capture',
+    'sample',
+    'closed',
+    'visibility',
+    'devicechange',
+    'connection',
+    'microphone.mute',
+    'microphone.unmute',
+    'microphone.ended',
+    'playback.playing',
+    'playback.pause',
+    'playback.waiting',
+    'playback.error',
+    'input_audio_buffer.speech_started',
+    'input_audio_buffer.speech_stopped',
+    'input_audio_buffer.committed',
+    'output_audio_buffer.started',
+    'output_audio_buffer.stopped',
+    'output_audio_buffer.cleared',
+    'conversation.item.truncated',
+    'conversation.item.input_audio_transcription.completed',
+    'conversation.item.input_audio_transcription.failed',
+    'response.created',
+    'response.done'
+]);
+const BOOLEAN_FIELDS = new Set([
+    'noiseSuppression',
+    'autoGainControl',
+    'supportsEchoCancellation',
+    'supportsNoiseSuppression',
+    'supportsAutoGainControl',
+    'micEnabled',
+    'micMuted',
+    'playbackPaused',
+    'playbackMuted',
+    'hidden'
+]);
+const NUMBER_FIELDS = new Set([
+    'sampleRate',
+    'sampleSize',
+    'channelCount',
+    'latency',
+    'playbackTime',
+    'playbackVolume',
+    'inputLevel',
+    'inputEnergy',
+    'inputDuration',
+    'outputLevel',
+    'outputEnergy',
+    'outputDuration',
+    'echoReturnLoss',
+    'echoReturnLossEnhancement',
+    'audio_start_ms',
+    'audio_end_ms',
+    'dropped'
+]);
+function object(value) {
+    return value && typeof value === 'object' && !Array.isArray(value) ? value : {};
+}
+export function voiceDiagnosticData(value) {
+    const result = {};
+    for (const [key, item] of Object.entries(object(value))) {
+        if (BOOLEAN_FIELDS.has(key) && (typeof item === 'boolean' || item === null))
+            result[key] = item;
+        else if (NUMBER_FIELDS.has(key) && (item === null || (typeof item === 'number' && Number.isFinite(item))))
+            result[key] = item;
+        else if (key === 'echoCancellation' &&
+            (item === null || typeof item === 'boolean' || item === 'all' || item === 'remote-only'))
+            result[key] = item;
+        else if ((key === 'item_id' || key === 'response_id') && typeof item === 'string' && /^[\w-]{1,100}$/.test(item))
+            result[key] = item;
+        else if (key === 'status' &&
+            typeof item === 'string' &&
+            ['completed', 'cancelled', 'failed', 'incomplete', 'in_progress'].includes(item))
+            result[key] = item;
+        else if (key === 'connection' &&
+            typeof item === 'string' &&
+            ['new', 'connecting', 'connected', 'disconnected', 'failed', 'closed'].includes(item))
+            result[key] = item;
+    }
+    return result;
+}
+export function voiceRealtimeDiagnostic(value) {
+    const event = object(value);
+    if (typeof event.type !== 'string' ||
+        !EVENTS.has(event.type) ||
+        (!event.type.includes('_') && !event.type.startsWith('response.')))
+        return null;
+    const response = object(event.response);
+    return {
+        type: event.type,
+        data: voiceDiagnosticData({ ...event, response_id: event.response_id ?? response.id, status: response.status })
+    };
+}
+/** Limits each HTTP batch and strips unknown fields even from authenticated clients. */
+export function parseVoiceDiagnostics(value) {
+    const events = object(value).events;
+    if (!Array.isArray(events) || !events.length || events.length > 40)
+        return null;
+    const result = [];
+    for (const raw of events) {
+        const event = object(raw);
+        if (typeof event.type !== 'string' ||
+            !EVENTS.has(event.type) ||
+            typeof event.atMs !== 'number' ||
+            !Number.isFinite(event.atMs) ||
+            event.atMs < 0)
+            return null;
+        result.push({ type: event.type, atMs: Math.round(event.atMs), data: voiceDiagnosticData(event.data) });
+    }
+    return result;
+}

package/docs/voice-setup.md

@@ -347,6 +347,35 @@ Rotating a value through `config set` restarts the managed service. The relay to
 
 ## Troubleshooting
 
+### Browser echo and interruptions
+
+WebRTC calls emit `[voice-diagnostics]` metadata into the relay log. After reproducing,
+use `relay_logs` with `contains: "voice-diagnostics"` (or the call ID from
+`list_voice_calls`). For a managed relay, `file: "relay.log"` also reaches entries
+from before a restart. Inspect promptly: the live log is a bounded ring, not a
+permanent diagnostic archive. Both the PWA and relay must run the updated build.
+
+Browser `capture` and five-second `sample` events report actual microphone settings,
+support flags, mute state, playback state and available WebRTC audio levels/energy.
+A null setting means the browser did not report it; `echoCancellation: true` means
+enabled, not proof that all echo was removed. Missing echo-loss statistics mean
+unsupported/unreported, not zero echo. Browser events use monotonic `atMs` since
+diagnostics started for that call; `receivedAt` is the relay's receipt time, which
+may lag after buffering or a network delay. Speech-start, playback-buffer and
+response events carry item/response IDs to correlate with the saved transcript.
+Server events provide a second timeline even when browser uploads fail.
+
+Diagnostics never mute the microphone, change speech thresholds or disable
+interruptions. They store no raw audio, captions, device identifiers, SDP or tools.
+Uploads are bounded and best-effort; `dropped` counts discarded metadata events.
+Background suspension, disconnection or process exit can still leave gaps.
+
+The user bubbles come from a separate transcription model. The Realtime model
+consumes audio directly, so a wrong caption can accompany a correctly understood
+command; see the [Realtime API reference](https://platform.openai.com/docs/api-reference/realtime).
+
+### Connection and configuration
+
 - **The call rings and then fails with no relay log:** confirm the public endpoint is on 443 and is exactly `/voice`. OpenAI does not deliver this webhook through Funnel 8443 or 10000.
 - **Twilio gets 403:** check the allowlist, the account's primary auth token, and that `voice.public-url + /twiml` is byte-for-byte the URL configured on the number.
 - **OpenAI gets 400:** check the webhook signing secret and the Mac's clock.

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "conductor-remote",
-	"version": "1.127.1",
+	"version": "1.128.0",
 	"type": "module",
 	"packageManager": "yarn@4.15.0",
 	"description": "Phone control panel for local Conductor agents. Reads ride SQLite + git; prompts ride Conductor's own dispatch path.",