@agentprojectcontext/apx 1.51.0 → 1.52.0

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@agentprojectcontext/apx",
-  "version": "1.51.0",
+  "version": "1.52.0",
   "description": "APX — unified CLI + daemon for the Agent Project Context (APC) standard.",
   "publishConfig": {
     "access": "public"

package/src/core/voice/emotions.js

@@ -62,7 +62,10 @@ export function emotionConfigFor(globalConfig, providerId) {
  * the intent of selectTtsEngine without the async availability probes:
  *   - explicit provider arg wins
  *   - single mode → voice.tts.provider
- *   - chain mode → first enabled engine with emotions on, else first enabled
+ *   - chain mode → the FIRST enabled engine in order (what selectTtsEngine
+ *     would speak with). We deliberately do NOT prefer an emotion-capable
+ *     engine here: the guide must reflect the engine that will actually speak,
+ *     otherwise the agent emits tags a different engine never asked for.
  */
 export function resolveSpeakingProvider(globalConfig, provider) {
   if (provider && provider !== "auto") return provider;
@@ -72,8 +75,7 @@ export function resolveSpeakingProvider(globalConfig, provider) {
   const order = resolveChainOrder(cfg).filter(
     (id) => id !== "mock" && enabledOf(cfg, id)
   );
-  const withEmotion = order.find((id) => emotionConfigFor(globalConfig, id).enabled);
-  return withEmotion || order[0] || cfg.provider || undefined;
+  return order[0] || cfg.provider || undefined;
 }
 
 /**

package/src/host/daemon/api/voice.js

@@ -22,6 +22,7 @@ import fs from "node:fs";
 import path from "node:path";
 import { readConfig } from "#core/config/index.js";
 import { synthesize } from "#core/voice/tts.js";
+import { stripEmotionTags } from "#core/voice/emotions.js";
 import { transcribe } from "#core/voice/transcription.js";
 import { decodeAudioInput } from "#core/voice/audio-decode.js";
 import { runSuperAgent, isSuperAgentEnabled } from "#core/agent/super-agent.js";
@@ -174,6 +175,12 @@ export function register(app, { projects, plugins, registries }) {
         replyText = userText;
       }
 
+      // Emotion tags ([excited], [whisper], …) are a TTS-only signal: keep them
+      // for synthesis (engines that support them split on them), but strip them
+      // from everything the user reads — the chat bubble, history, RAG.
+      const audioText = replyText;
+      replyText = stripEmotionTags(replyText);
+
       // Persist the turn to the cross-channel store (feeds RAG index,
       // search_messages, and the "active threads" block). channelCtx.channel is
       // the resolved surface ("deck"/"desktop"). Best-effort.
@@ -190,7 +197,7 @@ export function register(app, { projects, plugins, registries }) {
       if (replyText) {
         try {
           tts = await synthesize({
-            text: replyText,
+            text: audioText,
             voice: body.voice,
             language: body.language,
             provider: body.provider,