@ouro.bot/cli 0.1.0-alpha.891 → 0.1.0-alpha.892

package/changelog.json

@@ -1,6 +1,12 @@
 {
   "_note": "This changelog is maintained as part of the PR/version-bump workflow. Agent-curated, not auto-generated. Agents read this file directly via read_file to understand what changed between versions.",
   "versions": [
+    {
+      "version": "0.1.0-alpha.892",
+      "changes": [
+        "Local audio joins play the agent's voice through ffmpeg instead of sox, because sox's CoreAudio output can deadlock and stay silent when the process runs at lowered priority. An outbound call is treated as the job's friend only when the dialed number is on that friend's record. Phone greetings now name the agent: they had introduced themselves as ChatGPT in half of the live test calls."
+      ]
+    },
     {
       "version": "0.1.0-alpha.891",
       "changes": [

package/deploy/unraid/sanctuary.ouro/bundle-meta.json

@@ -1,5 +1,5 @@
 {
-  "runtimeVersion": "0.1.0-alpha.891",
+  "runtimeVersion": "0.1.0-alpha.892",
   "bundleSchemaVersion": 3,
   "lastUpdated": "2026-09-03T00:00:00.000Z"
 }

package/deploy/unraid/sanctuary.xml

@@ -1,7 +1,7 @@
 <?xml version="1.0"?>
 <Container version="2">
   <Name>ouro-butler</Name>
-  <Repository>ghcr.io/ourostack/ouroboros-butler:0.1.0-alpha.891</Repository>
+  <Repository>ghcr.io/ourostack/ouroboros-butler:0.1.0-alpha.892</Repository>
   <Registry>https://github.com/ourostack/ouroboros/pkgs/container/ouroboros-butler</Registry>
   <Network>host</Network>
   <Shell>sh</Shell>

package/dist/senses/voice/local-audio-devices.js

@@ -35,7 +35,10 @@ var __importStar = (this && this.__importStar) || (function () {
 Object.defineProperty(exports, "__esModule", { value: true });
 exports.LocalAudioDevices = exports.KILL_GRACE_MS = exports.defaultProcessHost = exports.PS_ENV = exports.PLAYBACK_DEVICE_NAME = exports.CAPTURE_DEVICE_NAME = void 0;
 exports.soxCaptureArgs = soxCaptureArgs;
-exports.soxPlaybackArgs = soxPlaybackArgs;
+exports.ffmpegPlaybackArgs = ffmpegPlaybackArgs;
+exports.audioToolboxListArgs = audioToolboxListArgs;
+exports.findAudioToolboxDeviceIndex = findAudioToolboxDeviceIndex;
+exports.defaultListAudioToolboxDevices = defaultListAudioToolboxDevices;
 exports.fileCaptureArgs = fileCaptureArgs;
 exports.filePlaybackArgs = filePlaybackArgs;
 exports.defaultKillGroup = defaultKillGroup;
@@ -52,6 +55,7 @@ const runtime_1 = require("../../nerves/runtime");
 const local_audio_1 = require("./local-audio");
 const local_audio_mute_1 = require("./local-audio-mute");
 const RAW_MULAW = ["-t", "raw", "-r", "8000", "-e", "mu-law", "-b", "8", "-c", "1"];
+const AUDIO_PROCESS_NAMES = new Set(["sox", "ffmpeg"]);
 exports.CAPTURE_DEVICE_NAME = "BlackHole 16ch";
 exports.PLAYBACK_DEVICE_NAME = "BlackHole 2ch";
 /** Capture a CoreAudio device as mono 8 kHz mu-law on stdout (channels 1 and 2 mixed). */
@@ -62,9 +66,50 @@ function soxCaptureArgs(device) {
     // `--buffer 8192` delayed everything the agent heard by about a second.
     return ["-q", "--buffer", "1024", "--input-buffer", "8192", "-t", "coreaudio", device, ...RAW_MULAW, "-", "remix", "1,2"];
 }
-/** Play raw mono 8 kHz mu-law from stdin to a CoreAudio device. */
-function soxPlaybackArgs(device) {
-    return ["-q", "--buffer", "160", ...RAW_MULAW, "-", "-t", "coreaudio", device];
+/**
+ * Play raw mono 8 kHz mu-law from stdin to an AudioToolbox output device, chosen by index.
+ *
+ * sox's CoreAudio output can deadlock when the process runs at lowered priority (a zsh background
+ * job is niced by default, and so is a launchd Background job): its main thread holds a lock inside
+ * `AudioDeviceStart` while the device IO thread waits for that lock, so nothing is ever played.
+ * Observed live: 5 of 5 niced joins were silent with sox and 2 of 2 spoke with ffmpeg. The probe
+ * flags keep ffmpeg from buffering input before it starts playing.
+ */
+function ffmpegPlaybackArgs(deviceIndex) {
+    return [
+        "-hide_banner", "-loglevel", "error", "-nostdin",
+        "-probesize", "32", "-analyzeduration", "0", "-fflags", "nobuffer",
+        "-f", "mulaw", "-ar", "8000", "-ac", "1", "-i", "pipe:0",
+        "-f", "audiotoolbox", "-audio_device_index", String(deviceIndex), "-",
+    ];
+}
+/** ffmpeg prints the AudioToolbox devices to its log; `-t 0` means nothing is played. */
+function audioToolboxListArgs() {
+    return ["-hide_banner", "-f", "lavfi", "-i", "anullsrc", "-t", "0", "-f", "audiotoolbox", "-list_devices", "true", "-"];
+}
+// "[AudioToolbox @ 0x7b514003c0] [1]                  BlackHole 2ch, BlackHole2ch_UID"
+const AUDIOTOOLBOX_DEVICE_LINE = /\[(\d+)\]\s+(.+),\s*[^,]*$/;
+/** The index ffmpeg uses for a device with exactly this name, or undefined when it is not listed. */
+function findAudioToolboxDeviceIndex(listing, deviceName) {
+    for (const line of listing.split("\n")) {
+        const match = AUDIOTOOLBOX_DEVICE_LINE.exec(line.trim());
+        if (match && match[2].trim() === deviceName)
+            return Number(match[1]);
+    }
+    return undefined;
+}
+/** Runs ffmpeg's device listing and returns its log. Rejects only when ffmpeg itself cannot run. */
+function defaultListAudioToolboxDevices(ffmpegPath = "ffmpeg", run = (file, args, callback) => {
+    (0, child_process_1.execFile)(file, args, { encoding: "utf8", timeout: 5000 }, (error, stdout, stderr) => callback(error, stdout, stderr));
+}) {
+    return new Promise((resolve, reject) => {
+        run(ffmpegPath, audioToolboxListArgs(), (error, _stdout, stderr) => {
+            if (error && typeof error.code !== "number")
+                reject(error);
+            else
+                resolve(stderr);
+        });
+    });
 }
 /** Decode any audio file to the capture stream format (stdout). The caller paces it. */
 function fileCaptureArgs(inputPath) {
@@ -145,7 +190,7 @@ class LocalAudioDevices {
         this.stop();
         const command = this.options.soxPath ?? "sox";
         const capture = this.spawnTracked(command, this.options.captureArgs);
-        const playback = this.spawnTracked(command, this.options.playbackArgs);
+        const playback = this.spawnTracked(this.options.playbackCommand ?? command, this.options.playbackArgs);
         this.writePidFile();
         this.recordStartTimes([...this.running]);
         this.watchParentSignals();
@@ -254,10 +299,10 @@ class LocalAudioDevices {
         });
         return killed;
     }
-    /** Only a process that is still sox AND started when we recorded it is ours to kill (pids get reused). */
+    /** Only a process that is still one of our audio tools AND started when we recorded it is ours to kill (pids get reused). */
     isRecordedSox(pid, recordedStart) {
         const info = this.inspectProcess(pid);
-        if (!info || path.basename(info.command) !== "sox")
+        if (!info || !AUDIO_PROCESS_NAMES.has(path.basename(info.command)))
             return false;
         return recordedStart !== "" && info.startTime === recordedStart;
     }

package/dist/senses/voice/local-audio-transport.js

@@ -137,6 +137,23 @@ class LocalAudioDeviceTransport {
     async leave(reason = "left") {
         await this.finish(reason);
     }
+    /** Live playback goes through ffmpeg, which names devices by index: find BlackHole 2ch's index now. */
+    async resolveLivePlayback() {
+        const command = this.deps.ffmpegPath ?? "ffmpeg";
+        const list = this.deps.listAudioToolboxDevices ?? (() => (0, local_audio_devices_1.defaultListAudioToolboxDevices)(command));
+        let listing;
+        try {
+            listing = await list();
+        }
+        catch (error) {
+            throw new Error(`Local audio routing is not ready:\n- Live playback needs ffmpeg (${errorText(error)}): brew install ffmpeg`);
+        }
+        const index = (0, local_audio_devices_1.findAudioToolboxDeviceIndex)(listing, local_audio_devices_1.PLAYBACK_DEVICE_NAME);
+        if (index === undefined) {
+            throw new Error(`Local audio routing is not ready:\n- ffmpeg does not list the playback device "${local_audio_devices_1.PLAYBACK_DEVICE_NAME}". Install it with brew install --cask blackhole-2ch, then restart macOS audio or reboot.`);
+        }
+        return { command, args: (0, local_audio_devices_1.ffmpegPlaybackArgs)(index) };
+    }
     async startUnchecked() {
         const { request, deps } = this;
         if (!deps.bridgeOptions.openaiRealtime?.apiKey?.trim()) {
@@ -144,10 +161,14 @@ class LocalAudioDeviceTransport {
         }
         const files = request.files;
         const soxPath = deps.soxPath;
+        const livePlayback = files ? undefined : await this.resolveLivePlayback();
+        if (this.finished)
+            return;
         this.devices = new local_audio_devices_1.LocalAudioDevices({
             spawner: deps.spawner,
             captureArgs: files ? (0, local_audio_devices_1.fileCaptureArgs)(files.inputPath) : (0, local_audio_devices_1.soxCaptureArgs)(local_audio_devices_1.CAPTURE_DEVICE_NAME),
-            playbackArgs: files ? (0, local_audio_devices_1.filePlaybackArgs)(files.outputPath) : (0, local_audio_devices_1.soxPlaybackArgs)(local_audio_devices_1.PLAYBACK_DEVICE_NAME),
+            playbackArgs: livePlayback ? livePlayback.args : (0, local_audio_devices_1.filePlaybackArgs)(files.outputPath),
+            ...(livePlayback ? { playbackCommand: livePlayback.command } : {}),
             pidFile: deps.pidFile,
             soxPath,
             killGroup: deps.killGroup,

package/dist/senses/voice/twilio-phone.js

@@ -417,13 +417,19 @@ function isSafeVoiceFriendId(id) {
  * (the store's display-name fallback is never used for voice). Everything else resolves from the
  * phone number, or from the call itself when there is none.
  */
+function friendHasPhone(friend, phone) {
+    return friend.externalIds.some((externalId) => (0, phone_1.normalizeTwilioE164PhoneNumber)(externalId.externalId) === phone);
+}
 async function resolveVoiceFriendContext(options, input) {
     const agentRoot = resolveTwilioPhoneAgentRoot(options);
     const friendStore = new friends_1.FileFriendStore(path.join(agentRoot, "friends"));
     const explicitFriendId = input.friendId?.trim();
+    const remotePhone = (0, phone_1.normalizeTwilioE164PhoneNumber)(input.remotePhone);
     if (explicitFriendId && isSafeVoiceFriendId(explicitFriendId)) {
         const existing = await friendStore.get(explicitFriendId);
-        if (existing && existing.id === explicitFriendId) {
+        // An outbound job names a friend, but the person who answers is whoever has the dialed number:
+        // a number that is not on the friend's record gets resolved like any other caller.
+        if (existing && existing.id === explicitFriendId && (!remotePhone || friendHasPhone(existing, remotePhone))) {
             return {
                 friendId: existing.id,
                 friendStore,
@@ -431,7 +437,6 @@ async function resolveVoiceFriendContext(options, input) {
             };
         }
     }
-    const remotePhone = (0, phone_1.normalizeTwilioE164PhoneNumber)(input.remotePhone);
     const provider = remotePhone ? "imessage-handle" : "local";
     const externalId = remotePhone || friendIdFromCaller("", input.callSid);
     const resolver = new friends_1.FriendResolver(friendStore, {
@@ -481,6 +486,14 @@ function twilioPhoneVoiceSessionKey(options) {
         return `twilio-phone-line-${lineSegment}`;
     return `twilio-phone-${safeSegment(options.callSid ?? "incoming")}`;
 }
+/**
+ * A Realtime `response.create` that carries `instructions` replaces the session instructions for
+ * that response, so a greeting sent that way knows nothing about who the agent is. Without this
+ * line the model introduced itself as "ChatGPT" in 3 of 6 live calls.
+ */
+function realtimeGreetingInstructions(agentName, prompt) {
+    return `You are ${agentName}. If you introduce yourself, use that name; never call yourself ChatGPT or an OpenAI assistant.\n${prompt}`;
+}
 function callConnectedPrompt(params) {
     const from = params.From?.trim();
     const to = params.To?.trim();
@@ -2098,7 +2111,7 @@ class TwilioOpenAIRealtimeMediaStreamSession {
                 createdAt: new Date().toISOString(),
             }, { From: this.to, To: this.from })
             : callConnectedPrompt({ From: this.from, To: this.to });
-        this.requestRealtimeResponse({ instructions: promptText });
+        this.requestRealtimeResponse({ instructions: realtimeGreetingInstructions(this.options.agentName, promptText) });
     }
     handleMedia(media) {
         const payload = stringField(media?.payload);
@@ -3350,7 +3363,7 @@ class OpenAISipPhoneSession {
             return;
         this.initialGreetingSent = true;
         this.requestRealtimeResponse({
-            instructions: openAISipCallConnectedPrompt(this.metadata, this.options.openaiRealtime?.voiceStyle),
+            instructions: realtimeGreetingInstructions(this.options.agentName, openAISipCallConnectedPrompt(this.metadata, this.options.openaiRealtime?.voiceStyle)),
         });
     }
     handleOpenAIMessage(raw) {

package/npm-shrinkwrap.json

@@ -1,12 +1,12 @@
 {
   "name": "@ouro.bot/cli",
-  "version": "0.1.0-alpha.891",
+  "version": "0.1.0-alpha.892",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "@ouro.bot/cli",
-      "version": "0.1.0-alpha.891",
+      "version": "0.1.0-alpha.892",
       "dependencies": {
         "@anthropic-ai/sdk": "^0.78.0",
         "@azure/identity": "^4.13.0",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@ouro.bot/cli",
-  "version": "0.1.0-alpha.891",
+  "version": "0.1.0-alpha.892",
   "engines": {
     "node": ">=22"
   },