@stackbone/cli 0.9.0 → 0.9.1

package/CHANGELOG.md

@@ -7,6 +7,23 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [0.9.1] - 2026-11-10
+
+### Changed
+
+- **The thinking sound waits for a long reply.** The voice worker, the one
+  `stackbone dev` runs too, starts the thinking sound at once for a box tool
+  call, and for any other wait only once it passes the
+  `backgroundAudio.thinkingDelayMs` voice setting (default 2000 ms). Before,
+  it played on every reply, since every reply waits for the knowledge search
+  and the model's first token. Studio's Voice tab gets a **Thinking delay
+  (ms)** field. Deploy the worker on LiveKit Cloud before the boxes: an older
+  worker refuses a call whose settings carry the delay.
+- **The box's missing-credentials messages name the Voice provider screen.**
+  A call or a voice status check on a box with no readable LiveKit
+  credentials now sends you to Studio's **Voice provider** screen, where the
+  credentials moved (STA3-89).
+
 ## [0.9.0] - 2026-11-10
 
 ### Added

package/main.js

@@ -3673,13 +3673,21 @@ var init_studio_voice = __esmMin((() => {
 		ambientSound: z.enum(VOICE_AMBIENT_SOUNDS),
 		ambientVolume: volumeSchema,
 		thinkingSound: z.enum(VOICE_THINKING_SOUNDS),
-		thinkingVolume: volumeSchema
+		thinkingVolume: volumeSchema,
+		/**
+		* How long a reply's wait runs before the thinking sound starts, when no box tool
+		* call is open (a tool call starts it at once). Every turn waits for the knowledge
+		* search and the model's first token, about 1.4 s at the median on a voice box
+		* (2026-11-09), so a delay under that plays the sound on nearly every reply.
+		*/
+		thinkingDelayMs: millisecondsSchema
 	});
 	VOICE_BACKGROUND_AUDIO_DEFAULTS = Object.freeze({
 		ambientSound: "office-ambience",
 		ambientVolume: .7,
 		thinkingSound: "keyboard-typing",
-		thinkingVolume: .6
+		thinkingVolume: .6,
+		thinkingDelayMs: 2e3
 	});
 	effectiveVoiceSettingsObjectSchema = z.strictObject({
 		/** BCP-47-ish tag the STT is hinted with and the TTS speaks (`es`, `en-US`), or `multi`. */
@@ -3749,7 +3757,7 @@ var init_studio_voice = __esmMin((() => {
 		recording: true,
 		rag: true,
 		refusalPhrase: true
-	}).extend({ backgroundAudio: backgroundAudioSchema.default(() => ({ ...VOICE_BACKGROUND_AUDIO_DEFAULTS })) });
+	}).extend({ backgroundAudio: backgroundAudioSchema.extend({ thinkingDelayMs: millisecondsSchema.default(VOICE_BACKGROUND_AUDIO_DEFAULTS.thinkingDelayMs) }).default(() => ({ ...VOICE_BACKGROUND_AUDIO_DEFAULTS })) });
 	callVoiceSettingsSchema = callVoiceSettingsObjectSchema.superRefine(checkCrossFieldRules);
 	toCallVoiceSettings = ({ recording: _recording, rag: _rag, refusalPhrase: _refusalPhrase, ...call }) => ({
 		...call,
@@ -3779,6 +3787,7 @@ var init_studio_voice = __esmMin((() => {
 		"backgroundAudio.ambientVolume",
 		"backgroundAudio.thinkingSound",
 		"backgroundAudio.thinkingVolume",
+		"backgroundAudio.thinkingDelayMs",
 		"recording",
 		"rag.collection",
 		"rag.minScore",
@@ -22665,7 +22674,7 @@ var init_role = __esmMin((() => {
 var STACKBONE_CLI_BUILD_VERSION, EMULATOR_CAPABILITY_ROUTES, EMULATOR_CAPABILITIES, buildContractResponse;
 var init_contract = __esmMin((() => {
 	init_src$26();
-	STACKBONE_CLI_BUILD_VERSION = "0.9.0";
+	STACKBONE_CLI_BUILD_VERSION = "0.9.1";
 	EMULATOR_CAPABILITY_ROUTES = {
 		"database.postgres_direct": "/api/db",
 		"rag.basic": "/api/rag",
@@ -91331,9 +91340,9 @@ var lookUpLiveKitCredentials = async (opts) => {
 		credentials = await readLiveKitCredentials(opts);
 	} catch (err) {
 		if (!(err instanceof SecretUndecryptableError)) throw err;
-		return voiceNotConfigured("This box cannot read the saved LiveKit credentials: they were saved with another key. Open the Voice settings screen and save them again.");
+		return voiceNotConfigured("This box cannot read the saved LiveKit credentials: they were saved with another key. Open the Voice provider screen and save them again.");
 	}
-	if (!credentials) return voiceNotConfigured("This box has no LiveKit credentials. Save the URL, API key and API secret on the Voice settings screen.");
+	if (!credentials) return voiceNotConfigured("This box has no LiveKit credentials. Save the URL, API key and API secret on the Voice provider screen.");
 	return { credentials };
 };
 /**
@@ -104326,7 +104335,7 @@ var calls_workflow_ts_default = "import { z } from \"zod\";\nimport { streamDeep
 var plain_workflow_ts_default = "import { z } from \"zod\";\n\n/**\n * Workflow \"{{name}}\" — a durable pipeline.\n *\n * Each `\"use step\"` is a durable checkpoint: kill the runtime mid-run and it\n * resumes from the last completed step instead of replaying everything.\n */\n\n// THE contract — sibling exports next to the bare workflow function. The build\n// harvests these (by convention) into the manifest + a live validator. The\n// parameter type is DERIVED from `inputSchema` and the RETURN type from\n// `outputSchema` (both `z.infer`), so neither side can drift from the contract\n// the build harvests. Without the return annotation the schema is published to\n// Studio and the generated types and never compared to what the body returns\n// (PLA-1272).\nexport const inputSchema = z.object({\n  message: z.string().describe(\"The input to process\"),\n});\n\nexport const outputSchema = z.object({\n  message: z.string(),\n  processed: z.boolean(),\n});\n\ntype {{Name}}Input = z.infer<typeof inputSchema>;\n\nexport async function {{export}}(input: {{Name}}Input): Promise<z.infer<typeof outputSchema>> {\n  \"use workflow\";\n\n  const result = await processMessage(input); // step 1 — durable checkpoint\n  return result;\n}\n\nasync function processMessage(input: {{Name}}Input) {\n  \"use step\";\n  // Replace this with the deterministic work of your step (DB write, HTTP call,\n  // an LLM call, …). Returning a value records it as the step's checkpoint.\n  return { message: input.message, processed: true };\n}\n";
 //#endregion
 //#region src/services/scaffold/template-loader.ts
-var SDK_TEMPLATE_VERSION = "0.10.0";
+var SDK_TEMPLATE_VERSION = "0.10.1";
 var PREFIX = "./templates/";
 var templates = new Map(Object.entries(/* @__PURE__ */ Object.assign({
 	"./templates/deep-agents/index.ts.tmpl": index_ts_default,
@@ -106449,7 +106458,7 @@ var buildWorkspaceBundle = async (options) => {
 			const artifactPath = resolve(outDir, CONFIG_SCHEMA_ARTIFACT_PATH);
 			await mkdir(dirname(artifactPath), { recursive: true });
 			await writeFile(artifactPath, JSON.stringify(configSchema), "utf8");
-		} else logger?.warn(`This workspace declares ${CONFIG_SCHEMA_SOURCE}, but no config JSON Schema was produced for the bundle. Studio's config editor will fall back to a raw JSON field and writes will not be validated.`);
+		} else logger?.warn(`This workspace declares ${CONFIG_SCHEMA_SOURCE}, but no config JSON Schema was produced for the bundle. Studio shows no Dynamic config section (edit the values with \`stackbone config\`) and writes will not be validated.`);
 	}
 	const manifest = deriveWorkspaceManifest({
 		workspace,
@@ -106552,7 +106561,7 @@ var extractConfigSchema = async (projectRoot, opts = {}) => {
 	try {
 		loaded = await loader(projectRoot);
 	} catch (err) {
-		emit(`[config-schema] config schema extraction bundle failed; no config schema is available — the Config editor will fall back to a raw JSON field. Cause: ${errMessage$3(err)}`);
+		emit(`[config-schema] config schema extraction bundle failed; no config schema is available — Studio shows no Dynamic config section; edit the values with \`stackbone config\`. Cause: ${errMessage$3(err)}`);
 		return {
 			schema: null,
 			warnings
@@ -106568,7 +106577,7 @@ var extractConfigSchema = async (projectRoot, opts = {}) => {
 			warnings
 		};
 	} catch (err) {
-		emit(`[config-schema] config.schema.ts could not be converted to JSON Schema — the Config editor will fall back to a raw JSON field. Cause: ${errMessage$3(err)}`);
+		emit(`[config-schema] config.schema.ts could not be converted to JSON Schema — Studio shows no Dynamic config section; edit the values with \`stackbone config\`. Cause: ${errMessage$3(err)}`);
 		return {
 			schema: null,
 			warnings

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@stackbone/cli",
-  "version": "0.9.0",
+  "version": "0.9.1",
   "license": "UNLICENSED",
   "type": "module",
   "bin": {

package/voice-worker/main.js

@@ -995,13 +995,21 @@ var backgroundAudioSchema = z8.strictObject({
   ambientSound: z8.enum(VOICE_AMBIENT_SOUNDS),
   ambientVolume: volumeSchema,
   thinkingSound: z8.enum(VOICE_THINKING_SOUNDS),
-  thinkingVolume: volumeSchema
+  thinkingVolume: volumeSchema,
+  /**
+   * How long a reply's wait runs before the thinking sound starts, when no box tool
+   * call is open (a tool call starts it at once). Every turn waits for the knowledge
+   * search and the model's first token, about 1.4 s at the median on a voice box
+   * (2026-11-09), so a delay under that plays the sound on nearly every reply.
+   */
+  thinkingDelayMs: millisecondsSchema
 });
 var VOICE_BACKGROUND_AUDIO_DEFAULTS = Object.freeze({
   ambientSound: "office-ambience",
   ambientVolume: 0.7,
   thinkingSound: "keyboard-typing",
-  thinkingVolume: 0.6
+  thinkingVolume: 0.6,
+  thinkingDelayMs: 2e3
 });
 var effectiveVoiceSettingsObjectSchema = z8.strictObject({
   /** BCP-47-ish tag the STT is hinted with and the TTS speaks (`es`, `en-US`), or `multi`. */
@@ -1078,7 +1086,12 @@ var callVoiceSettingsObjectSchema = effectiveVoiceSettingsObjectSchema.omit({
   rag: true,
   refusalPhrase: true
 }).extend({
-  backgroundAudio: backgroundAudioSchema.default(() => ({ ...VOICE_BACKGROUND_AUDIO_DEFAULTS }))
+  // A box older than `thinkingDelayMs` dispatches the group without it.
+  backgroundAudio: backgroundAudioSchema.extend({
+    thinkingDelayMs: millisecondsSchema.default(
+      VOICE_BACKGROUND_AUDIO_DEFAULTS.thinkingDelayMs
+    )
+  }).default(() => ({ ...VOICE_BACKGROUND_AUDIO_DEFAULTS }))
 });
 var callVoiceSettingsSchema = callVoiceSettingsObjectSchema.superRefine(checkCrossFieldRules);
 var VOICE_SETTINGS_LEAF_PATHS = [
@@ -1105,6 +1118,7 @@ var VOICE_SETTINGS_LEAF_PATHS = [
   "backgroundAudio.ambientVolume",
   "backgroundAudio.thinkingSound",
   "backgroundAudio.thinkingVolume",
+  "backgroundAudio.thinkingDelayMs",
   "recording",
   "rag.collection",
   "rag.minScore",
@@ -2220,13 +2234,14 @@ var CALL_CLOSING_PHRASE = Object.freeze({
 });
 var MAX_FAILED_TURNS_IN_A_ROW = 3;
 var inCallLanguage = (phrase, language) => language.split("-")[0]?.toLowerCase() === "es" ? phrase.es : phrase.en;
-var thinkingCue = () => {
+var thinkingCue = (delayMs) => {
   let cue;
   let open = 0;
-  let thinking = false;
+  let waitedLong = false;
+  let wait;
   let playing = false;
   const sync = () => {
-    const play = thinking || open > 0;
+    const play = waitedLong || open > 0;
     if (play === playing || !cue) return;
     playing = play;
     if (play) cue.start();
@@ -2239,13 +2254,32 @@ var thinkingCue = () => {
       sync();
     },
     onAgentState: (state) => {
-      thinking = state === "thinking";
-      sync();
+      if (state !== "thinking") {
+        clearTimeout(wait);
+        wait = void 0;
+        waitedLong = false;
+        sync();
+        return;
+      }
+      if (wait !== void 0 || waitedLong) return;
+      wait = setTimeout(() => {
+        wait = void 0;
+        waitedLong = true;
+        sync();
+      }, delayMs);
     },
     onToolCall: (phase) => {
       if (phase === "start") open += 1;
       else if (open > 0) open -= 1;
       sync();
+    },
+    /** The call is ending: a pending delay must not start the sound in a closed room. */
+    close: () => {
+      clearTimeout(wait);
+      wait = void 0;
+      waitedLong = false;
+      open = 0;
+      sync();
     }
   };
 };
@@ -2290,7 +2324,7 @@ var runVoiceJob = async (ctx, deps) => {
     deps
   );
   const settings = dispatch.voiceSettings;
-  const cue = thinkingCue();
+  const cue = thinkingCue(settings.backgroundAudio.thinkingDelayMs);
   const fallback = inCallLanguage(TURN_FALLBACK_PHRASE, settings.language);
   let failedInARow = 0;
   let ending = false;
@@ -2391,6 +2425,7 @@ var runVoiceJob = async (ctx, deps) => {
   const finishCall = (outcome = "disconnected") => {
     if (completion) return completion;
     ending = true;
+    cue.close();
     completion = Promise.resolve().then(async () => {
       await hangUp.pendingResume();
       await session.close();