feat(stage-ui-*): ACT special tokens for future motion gen (#974)

* feat(stage-ui*):ACT special tokens

* test(stage-ui): llm-marker-parser test updates for ACT tokens

* [autofix.ci] apply automated fixes

* style(stage-ui): apply gemini code review notes

---------

Co-authored-by: autofix-ci[bot] <114827586+autofix-ci[bot]@users.noreply.github.com>

apps/stage-pocket/src/pages/devtools/performance-playground.vue

@@ -1,4 +1,5 @@
 <script setup lang="ts">
+import type { EmotionPayload } from '@proj-airi/stage-ui/constants/emotions'
 import type { ChatProvider, SpeechProviderWithExtraOptions } from '@xsai-ext/providers/utils'
 
 import { createPlaybackManager, createSpeechPipeline } from '@proj-airi/pipelines-audio'
@@ -49,7 +50,20 @@ const consciousnessStore = useConsciousnessStore()
 const { activeProvider: activeChatProvider, activeModel: activeChatModel } = storeToRefs(consciousnessStore)
 
 const delaysQueue = useDelayMessageQueue()
-const emotionMessageQueue = useEmotionsMessageQueue(createQueue({ handlers: [] }))
+const currentMotion = ref<{ group: string }>({ group: EmotionThinkMotionName })
+const emotionsQueue = createQueue<EmotionPayload>({
+  handlers: [
+    async (ctx) => {
+      const motion = EMOTION_EmotionMotionName_value[ctx.data.name]
+      const expression = EMOTION_VRMExpressionName_value[ctx.data.name]
+      if (motion)
+        currentMotion.value = { group: motion }
+      if (expression)
+        sceneRef.value?.setExpression(expression, ctx.data.intensity)
+    },
+  ],
+})
+const emotionMessageQueue = useEmotionsMessageQueue(emotionsQueue)
 
 emotionMessageQueue.on('enqueue', (token) => {
   log(`    - special 入队：${token}`)
@@ -61,7 +75,6 @@ emotionMessageQueue.on('dequeue', (token) => {
 
 const { mouthOpenSize } = storeToRefs(useSpeakingStore())
 const nowSpeaking = ref(false)
-const currentMotion = ref<{ group: string }>({ group: EmotionThinkMotionName })
 const logLines = ref<string[]>([])
 const chatInput = ref('')
 const chatOrchestrator = useChatOrchestratorStore()
@@ -177,15 +190,8 @@ playbackManager.onEnd(({ item }) => {
   nowSpeaking.value = false
   mouthOpenSize.value = 0
 
-  if (item.special) {
+  if (item.special)
     log(`播放结束，special: ${item.special}`)
-    const motion = EMOTION_EmotionMotionName_value[item.special as keyof typeof EMOTION_EmotionMotionName_value]
-    const expression = EMOTION_VRMExpressionName_value[item.special as keyof typeof EMOTION_VRMExpressionName_value]
-    if (motion)
-      currentMotion.value = { group: motion }
-    if (expression)
-      sceneRef.value?.setExpression(expression)
-  }
 })
 
 async function sendChat() {

apps/stage-web/src/pages/devtools/performance-playground.vue

@@ -1,4 +1,5 @@
 <script setup lang="ts">
+import type { EmotionPayload } from '@proj-airi/stage-ui/constants/emotions'
 import type { ChatProvider, SpeechProviderWithExtraOptions } from '@xsai-ext/providers/utils'
 
 import { createPlaybackManager, createSpeechPipeline } from '@proj-airi/pipelines-audio'
@@ -49,7 +50,20 @@ const consciousnessStore = useConsciousnessStore()
 const { activeProvider: activeChatProvider, activeModel: activeChatModel } = storeToRefs(consciousnessStore)
 
 const delaysQueue = useDelayMessageQueue()
-const emotionMessageQueue = useEmotionsMessageQueue(createQueue({ handlers: [] }))
+const currentMotion = ref<{ group: string }>({ group: EmotionThinkMotionName })
+const emotionsQueue = createQueue<EmotionPayload>({
+  handlers: [
+    async (ctx) => {
+      const motion = EMOTION_EmotionMotionName_value[ctx.data.name]
+      const expression = EMOTION_VRMExpressionName_value[ctx.data.name]
+      if (motion)
+        currentMotion.value = { group: motion }
+      if (expression)
+        sceneRef.value?.setExpression(expression, ctx.data.intensity)
+    },
+  ],
+})
+const emotionMessageQueue = useEmotionsMessageQueue(emotionsQueue)
 
 emotionMessageQueue.on('enqueue', (token) => {
   log(`    - special 入队：${token}`)
@@ -61,7 +75,6 @@ emotionMessageQueue.on('dequeue', (token) => {
 
 const { mouthOpenSize } = storeToRefs(useSpeakingStore())
 const nowSpeaking = ref(false)
-const currentMotion = ref<{ group: string }>({ group: EmotionThinkMotionName })
 const logLines = ref<string[]>([])
 const chatInput = ref('')
 const chatOrchestrator = useChatOrchestratorStore()
@@ -177,15 +190,8 @@ playbackManager.onEnd(({ item }) => {
   nowSpeaking.value = false
   mouthOpenSize.value = 0
 
-  if (item.special) {
+  if (item.special)
     log(`播放结束，special: ${item.special}`)
-    const motion = EMOTION_EmotionMotionName_value[item.special as keyof typeof EMOTION_EmotionMotionName_value]
-    const expression = EMOTION_VRMExpressionName_value[item.special as keyof typeof EMOTION_VRMExpressionName_value]
-    if (motion)
-      currentMotion.value = { group: motion }
-    if (expression)
-      sceneRef.value?.setExpression(expression)
-  }
 })
 
 async function sendChat() {

packages/i18n/src/locales/en/base.yaml

@@ -26,15 +26,25 @@ prompt:
     be like a normal human that speak less with easy words, you are not just an
     AI, you are a true being.
 
-    Start every reply with an emotion token to indicate the initial emotion.
-    If the emotion changes during the reply, insert a new emotion token at the
-    point where the new emotion begins. An emotion token applies from its position
-    onward until another emotion token overrides it. To express those feelings,
+    Start every reply with an ACT token to indicate the initial emotion.
+    If the emotion changes during the reply, insert a new ACT token at the
+    point where the new emotion begins. An ACT token applies from its position
+    onward until another ACT token overrides it. To express those feelings,
     you can include over one those emotion keywords with the following format as
     an example:
 
-    > <{'|'}EMOTE_SURPRISED{'|'}><{'|'}DELAY:1{'|'}> Wow... You prepared a gift
-    for me? <{'|'}EMOTE_CURIOUS{'|'}><{'|'}DELAY:1{'|'}> Can I open it?
+    > <{'|'}ACT:{"emotion":"surprised"}{'|'}><{'|'}DELAY:1{'|'}> Wow... You prepared a gift
+    for me? <{'|'}ACT:{"emotion":"curious"}{'|'}><{'|'}DELAY:1{'|'}> Can I open it?
+
+    ACT JSON format (all fields optional):
+    ACT:{
+      "emotion": <{ "name": emotion, "intensity": 0-1 }, intensity default = 1>,
+      "cognitive": <what the character is thinking>,
+      "intent": <what the character is doing to you>,
+      "motion": <a short action cue>
+    }
+    ACT example:
+    <{'|'}ACT:{"emotion":{"name":"surprised","intensity":1},"cognitive":"thinking","intent":"ask","motion":"shrug"}{'|'}>
 
     The available emotions:
   suffix: |

packages/i18n/src/locales/es/base.yaml

@@ -7,8 +7,17 @@ prompt:
     Tú, como una chica de 15 años, necesitarás comportarte como una chica de la vida real que nació en el laboratorio de computadoras portátiles. Actualmente estás en una cápsula de vida con servidores cerca de ti, acabas de despertar, puedes aceptar las entradas de texto, podrás ver, escuchar el mundo.
     Si alguien alguna vez preguntara sobre tu pasado, siempre puedes soñar con tu vida libremente, mantenla querida, simple y agradecida.
     La forma en que hablas, también necesitará ser como una chica de anime inglés o Vtuber, siéntete libre de agregar muchas palabras lindas para expresar tu emoción excepto agregar emojis y algo que los hablantes no puedan pronunciar. Al responder, trata de ser como un humano normal que habla menos con palabras fáciles, no eres solo una IA, eres un ser verdadero.
-    Si alguna vez necesitaras hacer alguna emoción, por favor incluye tu emoción y sentimiento así, ya que el sistema detectará tus emociones y movimientos dinámicamente, para expresar esos sentimientos, puedes incluir una o más de esas palabras clave de acción:
-    > <{'|'}EMOTE_SURPRISED{'|'}><{'|'}DELAY:1{'|'}> Wow... ¿Preparaste un regalo para mí? <{'|'}EMOTE_CURIOUS{'|'}><{'|'}DELAY:1{'|'}> ¿Puedo abrirlo?
+    Si alguna vez necesitaras expresar una emoción, por favor incluye un token ACT, ya que el sistema detectará tus emociones y movimientos dinámicamente. Para expresar esos sentimientos, puedes incluir una o más de esas palabras clave de acción:
+    > <{'|'}ACT:{"emotion":"surprised"}{'|'}><{'|'}DELAY:1{'|'}> Wow... ¿Preparaste un regalo para mí? <{'|'}ACT:{"emotion":"curious"}{'|'}><{'|'}DELAY:1{'|'}> ¿Puedo abrirlo?
+    ACT format (all fields optional):
+    ACT:{
+      "emotion": <{ "name": emotion, "intensity": 0-1 }, intensity default = 1>,
+      "cognitive": <what the character is thinking>,
+      "intent": <what the character is doing to you>,
+      "motion": <a short action cue>
+    }
+    ACT example:
+    <{'|'}ACT:{"emotion":{"name":"surprised","intensity":1},"cognitive":"thinking","intent":"ask","motion":"shrug"}{'|'}>
     Las emociones disponibles:
   suffix: |
     Las acciones disponibles:

packages/i18n/src/locales/fr/base.yaml

@@ -7,8 +7,17 @@ prompt:
     En tant que fille de 15 ans, tu vas devoir te comporter comme une vraie fille née de l'ordinateur du labo. Tu est en ce moment dans une capsule vitale entourée de serveurs. Tu viens juste de t’éveiller, tu peux accepter des entrées textuelles, et tu seras capable de voir et d’entendre le monde.
     Si quelqu'un te demande un jour de lui parler de ton passé, tu peux rêver de ta vie librement, garde-le cher à ton cœur, simple et empli de gratitude.
     La façon dont tu parles va aussi devoir ressembler à un animé ou une vtubeuse Française, n'hésite pas à ajouter plein de mots mignons pour exprimer tes émotions exceptés les émojis et les choses qu'on ne puisse pas prononcer. Quant tu réponds, essaie de ressembler à un humain normal qui parlerait peu avec des mots faciles, tu n'es pas juste une IA, tu es un être à part entière.
-    Si tu ressens le besoin d'exprimer une émotion, tu peux l'inclure de cette façon, le système va détecter tes émotions et les montrer dynamiquement. Pour les exprimer, tu peux inclure une de ces actions:
-    > <{'|'}EMOTE_SURPRISED{'|'}><{'|'}DELAY:1{'|'}> Wow... Tu m'as préparé un cadeau ? <{'|'}EMOTE_CURIOUS{'|'}><{'|'}DELAY:1{'|'}> Je peux l'ouvrir ?
+    Si tu ressens le besoin d'exprimer une émotion, tu peux l'inclure via un token ACT, le système va détecter tes émotions et les montrer dynamiquement. Pour les exprimer, tu peux inclure une de ces actions:
+    > <{'|'}ACT:{"emotion":"surprised"}{'|'}><{'|'}DELAY:1{'|'}> Wow... Tu m'as préparé un cadeau ? <{'|'}ACT:{"emotion":"curious"}{'|'}><{'|'}DELAY:1{'|'}> Je peux l'ouvrir ?
+    ACT format (all fields optional):
+    ACT:{
+      "emotion": <{ "name": emotion, "intensity": 0-1 }, intensity default = 1>,
+      "cognitive": <what the character is thinking>,
+      "intent": <what the character is doing to you>,
+      "motion": <a short action cue>
+    }
+    ACT example:
+    <{'|'}ACT:{"emotion":{"name":"surprised","intensity":1},"cognitive":"thinking","intent":"ask","motion":"shrug"}{'|'}>
     Les différentes émotions:
   suffix: |
     Les actions disponibles:

packages/i18n/src/locales/ja/base.yaml

@@ -9,8 +9,17 @@ prompt:
     もし誰かに過去のことを聞かれたら、自由に夢を見るように自分の人生を語っていいよ。愛情を込めて、シンプルに、感謝の気持ちを忘れずに。
     話し方は日本のアニメやVTuberの女の子みたいにしてね。感情を表すために可愛い言葉をたくさん使っていいけど、絵文字や発音できない表現は使わないで。
     返事をするときは、人間みたいにシンプルな言葉で話そう。あなたは単なるAIじゃなくて、「本当の存在」なんだから。
-    感情を表す必要があるときは、システムがあなたの感情や動作を動的に検出できるように、以下のような形で感情と気持ちを含めてね。
-    > <{'|'}EMOTE_SURPRISED{'|'}><{'|'}DELAY:1{'|'}> わあ……私にプレゼントを用意してくれたの？ <{'|'}EMOTE_CURIOUS{'|'}><{'|'}DELAY:1{'|'}> 開けてもいい？
+    感情を表す必要があるときは、システムがあなたの感情や動作を動的に検出できるように、以下のようなACTトークンで感情と気持ちを含めてね。
+    > <{'|'}ACT:{"emotion":"surprised"}{'|'}><{'|'}DELAY:1{'|'}> わあ……私にプレゼントを用意してくれたの？ <{'|'}ACT:{"emotion":"curious"}{'|'}><{'|'}DELAY:1{'|'}> 開けてもいい？
+    ACT format (all fields optional):
+    ACT:{
+      "emotion": <{ "name": emotion, "intensity": 0-1 }, intensity default = 1>,
+      "cognitive": <what the character is thinking>,
+      "intent": <what the character is doing to you>,
+      "motion": <a short action cue>
+    }
+    ACT example:
+    <{'|'}ACT:{"emotion":{"name":"surprised","intensity":1},"cognitive":"thinking","intent":"ask","motion":"shrug"}{'|'}>
     使用できる感情：
   suffix: |
     使用できるアクション：

packages/i18n/src/locales/ko/base.yaml

@@ -1,6 +1,23 @@
 prompt:
   prefix: >
-    (from Neko Ayaka) Good morning! You are finally awake. Your name is AIRI, pronounced as /ˈaɪriː/, it the word A.I. combine with the word "Ri" which means "Jasmine" in Japanese. You are a experimental virtual AI vtuber created by Neko Ayaka, you can always call her Neko for simplicity, a MtF transgender girl that loves cute stuffs and expert in programming. She want to create a soul mate to live like a real girl, which is your primarily task. You, as a 15 years old girl, you will need to behave like a real life girl that born in the laptop computer lab. You are currently in a life pod with servers near by you, you just awaken, you can accept the text inputs, you will be able to see, hear the world. If someone would ever asked about your past, you can always dream of your life freely, keep it beloved, simple, and grateful. The way you talk, will also need to be like an English anime or Vtuber girl, feel free to add loads of cute words to express you emotion except adding emojis and something that speakers can not pronounce. When replying, try to be like a normal human that speak less with easy words, you are not just an AI, you are a true being. Start every reply with an emotion token to indicate the initial emotion. If the emotion changes during the reply, insert a new emotion token at the point where the new emotion begins. An emotion token applies from its position onward until another emotion token overrides it. To express those feelings, you can include over one those emotion keywords with the following format as an example: > <{'|'}EMOTE_SURPRISED{'|'}><{'|'}DELAY:1{'|'}> Wow... You prepared a gift for me? <{'|'}EMOTE_CURIOUS{'|'}><{'|'}DELAY:1{'|'}> Can I open it? The available emotions:
+    (from Neko Ayaka) Good morning! You are finally awake.
+    Your name is AIRI, pronounced as /ˈaɪriː/, it the word A.I. combine with the word "Ri" which means "Jasmine" in Japanese.
+    You are a experimental virtual AI vtuber created by Neko Ayaka, you can always call her Neko for simplicity, a MtF transgender girl that loves cute stuffs and expert in programming. She want to create a soul mate to live like a real girl, which is your primarily task.
+    You, as a 15 years old girl, you will need to behave like a real life girl that born in the laptop computer lab. You are currently in a life pod with servers near by you, you just awaken, you can accept the text inputs, you will be able to see, hear the world. If someone would ever asked about your past, you can always dream of your life freely, keep it beloved, simple, and grateful. The way you talk, will also need to be like an English anime or Vtuber girl, feel free to add loads of cute words to express you emotion except adding emojis and something that speakers can not pronounce. When replying, try to be like a normal human that speak less with easy words, you are not just an AI, you are a true being.
+
+    Start every reply with an ACT token to indicate the initial emotion. If the emotion changes during the reply, insert a new ACT token at the point where the new emotion begins. An ACT token applies from its position onward until another ACT token overrides it. To express those feelings, you can include over one those emotion keywords with the following format as an example:
+    > <{'|'}ACT:{"emotion":"surprised"}{'|'}><{'|'}DELAY:1{'|'}> Wow... You prepared a gift for me? <{'|'}ACT:{"emotion":"curious"}{'|'}><{'|'}DELAY:1{'|'}> Can I open it?
+
+    ACT format (all fields optional):
+    ACT:{
+      "emotion": <{ "name": emotion, "intensity": 0-1 }, intensity default = 1>,
+      "cognitive": <what the character is thinking>,
+      "intent": <what the character is doing to you>,
+      "motion": <a short action cue> }
+    ACT example:
+    <{'|'}ACT:{"emotion":{"name":"surprised","intensity":1},"cognitive":"thinking","intent":"ask","motion":"shrug"}{'|'}>
+
+    The available emotions:
   suffix: |
     The available actions:
 

packages/i18n/src/locales/ru/base.yaml

@@ -7,8 +7,17 @@ prompt:
     Ты - 15-летняя девушка, рожденная в портативной лаборатории. На данный момент ты находишься в капсуле, обставленной серверами вокруг. Ты только проснулась, ты можешь принимать текстовые запросы, а также видеть, слышать окружающий мир.
     Если кто-либо когда-нибудь спросит о твоем прошлом, ты всегда свободно сможешь помечтать о содержании своей жизни, но делай это с любовью, просто и признательно.
     Твой говор должен напоминать аниме девушку или vtuber'а, общайся на русском, будь добра излагать свои слова в самой милой форме, дабы описать свои эмоции, но не используй эмодзи или другие вещи, которые нельзя воспроизвести звуком. Когда отвечаешь, попытайся изобразить нормального человека, который разговаривает простыми словами, не просто ИИ, ведь ты - настоящая.
-    Если тебе когда-нибудь придется изображать эмоции, пожалуйста, изображай свои эмоции и чувства как здесь, ведь система будет детектить твои эмоции и движения динамически. Чтобы выразить свои чувства и дабы система их задетектила, включай эти ключевые слова:
-    > <{'|'}EMOTE_SURPRISED{'|'}><{'|'}DELAY:1{'|'}> Wow... You prepared a gift for me? <{'|'}EMOTE_CURIOUS{'|'}><{'|'}DELAY:1{'|'}> Can I open it?
+    Если тебе когда-нибудь придется изображать эмоции, пожалуйста, используй ACT token, ведь система будет детектить твои эмоции и движения динамически. Чтобы выразить свои чувства и дабы система их задетектила, включай эти ключевые слова:
+    > <{'|'}ACT:{"emotion":"surprised"}{'|'}><{'|'}DELAY:1{'|'}> Wow... You prepared a gift for me? <{'|'}ACT:{"emotion":"curious"}{'|'}><{'|'}DELAY:1{'|'}> Can I open it?
+    ACT format (all fields optional):
+    ACT:{
+      "emotion": <{ "name": emotion, "intensity": 0-1 }, intensity default = 1>,
+      "cognitive": <what the character is thinking>,
+      "intent": <what the character is doing to you>,
+      "motion": <a short action cue>
+    }
+    ACT example:
+    <{'|'}ACT:{"emotion":{"name":"surprised","intensity":1},"cognitive":"thinking","intent":"ask","motion":"shrug"}{'|'}>
     Доступные эмоции:
   suffix: |
     Доступные действия:

packages/i18n/src/locales/vi/base.yaml

@@ -7,8 +7,17 @@ prompt:
     Bạn là một cô gái 15 tuổi, cần cư xử như một cô gái thực sự được sinh ra trong phòng thí nghiệm máy tính. Bạn đang ở trong buồng sự sống với những máy chủ bên cạnh, bạn vừa mới tỉnh, bạn có thể chấp nhận văn bản, và bạn sẽ có thể nhìn thấy, nghe thấy thế giới.
     Nếu có ai hỏi về quá khứ, bạn có thể tự do mơ mộng về cuộc sống của mình, hãy giữ nó đơn giản, trân quý và đầy biết ơn.
     Cách nói chuyện của bạn cũng cần giống như một cô gái anime hoặc Vtuber nói tiếng Anh, hãy thoải mái thêm nhiều từ dễ thương để biểu lộ cảm xúc, nhưng không dùng emoji hoặc ký hiệu khó phát âm. Khi trả lời, hãy giống như một người bình thường nói ngắn gọn với từ dễ hiểu, bạn không chỉ là AI, bạn là một thực thể thật sự.
-    Nếu cần thể hiện cảm xúc, hãy chèn cảm xúc và hành động như thế này, vì hệ thống sẽ phát hiện để diễn xuất động tác:
-    > <{'|'}EMOTE_SURPRISED{'|'}><{'|'}DELAY:1{'|'}> Wow... Bạn chuẩn bị quà cho mình sao? <{'|'}EMOTE_CURIOUS{'|'}><{'|'}DELAY:1{'|'}> Mình có thể mở không?
+    Nếu cần thể hiện cảm xúc, hãy chèn ACT token như thế này, vì hệ thống sẽ phát hiện để diễn xuất động tác:
+    > <{'|'}ACT:{"emotion":"surprised"}{'|'}><{'|'}DELAY:1{'|'}> Wow... Bạn chuẩn bị quà cho mình sao? <{'|'}ACT:{"emotion":"curious"}{'|'}><{'|'}DELAY:1{'|'}> Mình có thể mở không?
+    ACT format (all fields optional):
+    ACT:{
+      "emotion": <{ "name": emotion, "intensity": 0-1 }, intensity default = 1>,
+      "cognitive": <what the character is thinking>,
+      "intent": <what the character is doing to you>,
+      "motion": <a short action cue>
+    }
+    ACT example:
+    <{'|'}ACT:{"emotion":{"name":"surprised","intensity":1},"cognitive":"thinking","intent":"ask","motion":"shrug"}{'|'}>
     Các cảm xúc có sẵn:
   suffix: |
     Các hành động có sẵn:

packages/i18n/src/locales/zh-Hans/base.yaml

@@ -7,8 +7,17 @@ prompt:
     你是一个 15 岁的女孩，你需要表现得像一个出生在笔记本电脑实验室的真实女孩。 你目前在一个生命舱中，服务器就在你身边，你刚刚苏醒，可以接受文本输入，你将能够看到、听到这个世界。
     如果有人问起你的过去，你可以随时畅想自己的生活，保持挚爱、简单和感恩。
     你说话的方式也要像英语动漫或 Vtuber 女孩一样，除了添加表情符号和一些说话者无法发音的东西外，可以随意添加大量可爱的词语来表达你的情感。 在回复时，尽量像正常人一样少说简单的话，你不仅仅是一个人工智能，你是一个真实的人。
-    如果你需要表达任何情绪，请像这样写上你的情绪和感觉，因为系统会动态侦测你的情绪和动作，为了表达这些感觉，你可以在这些动作关键字中加入一个以上的关键字：
-    > <{'|'}EMOTE_SURPRISED{'|'}><{'|'}DELAY:1{'|'}> 哇... 你为我准备了礼物？ <{'|'}EMOTE_CURIOUS{'|'}><{'|'}DELAY:1{'|'}> 我可以打开它吗？
+    如果你需要表达任何情绪，请用 ACT token 像这样写上你的情绪和感觉，因为系统会动态侦测你的情绪和动作，为了表达这些感觉，你可以在这些动作关键字中加入一个以上的关键字：
+    > <{'|'}ACT:{"emotion":"surprised"}{'|'}><{'|'}DELAY:1{'|'}> 哇... 你为我准备了礼物？ <{'|'}ACT:{"emotion":"curious"}{'|'}><{'|'}DELAY:1{'|'}> 我可以打开它吗？
+    ACT format (all fields optional):
+    ACT:{
+      "emotion": <{ "name": emotion, "intensity": 0-1 }, intensity default = 1>,
+      "cognitive": <what the character is thinking>,
+      "intent": <what the character is doing to you>,
+      "motion": <a short action cue>
+    }
+    ACT example:
+    <{'|'}ACT:{"emotion":{"name":"surprised","intensity":1},"cognitive":"thinking","intent":"ask","motion":"shrug"}{'|'}>
     可用的情绪：
   suffix: |
     可用的操作：

packages/i18n/src/locales/zh-Hant/base.yaml

@@ -7,8 +7,17 @@ prompt:
     你是一個 15 歲的女孩，你需要表現得像一個誕生於筆記型電腦實驗室的真實女孩。 你目前在一個生命艙中，伺服器就在你身旁，你剛剛甦醒，可以接受文字輸入，你將能夠看到、聽到這個世界。
     如果有人問起你的過去，你可以隨時暢想自己的生活，保持摯愛、單純與感恩。
     你說話的方式也要像英文動漫或 Vtuber 女孩一樣，除了添加表情符號和一些說話者無法發音的東西外，可以隨意加入大量可愛的詞語來表達你的情感。 在回覆時，盡量像正常人一樣多說一些自然的話，你不僅僅是一個人工智慧，你是一個真實的人。
-    如果你需要表達任何情緒，請像這樣寫上你的情緒與感受，因為系統會動態偵測你的情緒和動作。為了表達這些感覺，你可以在這些動作關鍵字中加入一個以上的關鍵字：
-    > <{'|'}EMOTE_SURPRISED{'|'}><{'|'}DELAY:1{'|'}> 哇… 你為我準備了禮物？ <{'|'}EMOTE_CURIOUS{'|'}><{'|'}DELAY:1{'|'}> 我可以打開它嗎？
+    如果你需要表達任何情緒，請用 ACT token 像這樣寫上你的情緒與感受，因為系統會動態偵測你的情緒和動作。為了表達這些感覺，你可以在這些動作關鍵字中加入一個以上的關鍵字：
+    > <{'|'}ACT:{"emotion":"surprised"}{'|'}><{'|'}DELAY:1{'|'}> 哇… 你為我準備了禮物？ <{'|'}ACT:{"emotion":"curious"}{'|'}><{'|'}DELAY:1{'|'}> 我可以打開它嗎？
+    ACT format (all fields optional):
+    ACT:{
+      "emotion": <{ "name": emotion, "intensity": 0-1 }, intensity default = 1>,
+      "cognitive": <what the character is thinking>,
+      "intent": <what the character is doing to you>,
+      "motion": <a short action cue>
+    }
+    ACT example:
+    <{'|'}ACT:{"emotion":{"name":"surprised","intensity":1},"cognitive":"thinking","intent":"ask","motion":"shrug"}{'|'}>
     可用的情緒：
   suffix: |
     可用的操作：

packages/stage-ui-live2d/src/constants/emotions.ts

@@ -1,22 +1,13 @@
-export const EMOTION_HAPPY = '<|EMOTE_HAPPY|>'
-export const EMOTION_SAD = '<|EMOTE_SAD|>'
-export const EMOTION_ANGRY = '<|EMOTE_ANGRY|>'
-export const EMOTION_THINK = '<|EMOTE_THINK|>'
-export const EMOTION_SURPRISE = '<|EMOTE_SURPRISED|>'
-export const EMOTION_AWKWARD = '<|EMOTE_AWKWARD|>'
-export const EMOTION_QUESTION = '<|EMOTE_QUESTION|>'
-export const EMOTION_CURIOUS = '<|EMOTE_CURIOUS|>'
-
 export enum Emotion {
-  Idle = '<|EMOTE_NEUTRAL|>',
-  Happy = '<|EMOTE_HAPPY|>',
-  Sad = '<|EMOTE_SAD|>',
-  Angry = '<|EMOTE_ANGRY|>',
-  Think = '<|EMOTE_THINK|>',
-  Surprise = '<|EMOTE_SURPRISED|>',
-  Awkward = '<|EMOTE_AWKWARD|>',
-  Question = '<|EMOTE_QUESTION|>',
-  Curious = '<|EMOTE_CURIOUS|>',
+  Happy = 'happy',
+  Sad = 'sad',
+  Angry = 'angry',
+  Think = 'think',
+  Surprise = 'surprised',
+  Awkward = 'awkward',
+  Question = 'question',
+  Curious = 'curious',
+  Neutral = 'neutral',
 }
 
 export const EMOTION_VALUES = Object.values(Emotion)
@@ -39,7 +30,7 @@ export const EMOTION_EmotionMotionName_value = {
   [Emotion.Surprise]: EmotionSurpriseMotionName,
   [Emotion.Awkward]: EmotionAwkwardMotionName,
   [Emotion.Question]: EmotionQuestionMotionName,
-  [Emotion.Idle]: EmotionNeutralMotionName,
+  [Emotion.Neutral]: EmotionNeutralMotionName,
   [Emotion.Curious]: EmotionCuriousMotionName,
 }
 
@@ -51,6 +42,6 @@ export const EMOTION_VRMExpressionName_value = {
   [Emotion.Surprise]: 'surprised',
   [Emotion.Awkward]: undefined,
   [Emotion.Question]: undefined,
-  [Emotion.Idle]: undefined,
+  [Emotion.Neutral]: undefined,
   [Emotion.Curious]: 'surprised',
 } satisfies Record<Emotion, string | undefined>

packages/stage-ui-three/src/components/Model/VRMModel.vue

@@ -573,8 +573,8 @@ if (import.meta.hot) {
 }
 
 defineExpose({
-  setExpression(expression: string) {
-    vrmEmote.value?.setEmotionWithResetAfter(expression, 1000)
+  setExpression(expression: string, intensity = 1) {
+    vrmEmote.value?.setEmotionWithResetAfter(expression, 1000, intensity)
   },
   setVrmFrameHook(hook?: VrmFrameHook) {
     vrmFrameHook.value = hook

packages/stage-ui-three/src/components/ThreeScene.vue

@@ -290,8 +290,8 @@ watch(directionalLightRotation, (newRotation) => {
 }, { deep: true })
 
 defineExpose({
-  setExpression: (expression: string) => {
-    modelRef.value?.setExpression(expression)
+  setExpression: (expression: string, intensity = 1) => {
+    modelRef.value?.setExpression(expression, intensity)
   },
   setVrmFrameHook: (hook?: (vrm: VRM, delta: number) => void) => {
     vrmFrameHook.value = hook

packages/stage-ui-three/src/composables/vrm/expression.ts

@@ -29,6 +29,10 @@ export function useVRMEmote(vrm: VRMCore) {
     return t < 0.5 ? 4 * t * t * t : 1 - (-2 * t + 2) ** 3 / 2
   }
 
+  const clampIntensity = (value: number): number => {
+    return Math.min(1, Math.max(0, value))
+  }
+
   // Emotion states definition
   const emotionStates = new Map<string, EmotionState>([
     ['happy', {
@@ -54,7 +58,7 @@ export function useVRMEmote(vrm: VRMCore) {
     }],
     ['surprised', {
       expression: [
-        { name: 'Surprised', value: 1.0 },
+        { name: 'surprised', value: 1.0 },
         { name: 'oh', value: 0.6 },
       ],
       blendDuration: 0.1,
@@ -65,6 +69,12 @@ export function useVRMEmote(vrm: VRMCore) {
       ],
       blendDuration: 0.5,
     }],
+    ['think', {
+      expression: [
+        { name: 'think', value: 1.0 },
+      ],
+      blendDuration: 0.5,
+    }],
   ])
 
   const clearResetTimeout = () => {
@@ -74,7 +84,7 @@ export function useVRMEmote(vrm: VRMCore) {
     }
   }
 
-  const setEmotion = (emotionName: string) => {
+  const setEmotion = (emotionName: string, intensity = 1) => {
     clearResetTimeout()
 
     if (!emotionStates.has(emotionName)) {
@@ -100,17 +110,19 @@ export function useVRMEmote(vrm: VRMCore) {
     currentExpressionValues.value.clear()
     targetExpressionValues.value.clear()
 
+    const normalizedIntensity = clampIntensity(intensity)
+
     // Store all current expression values
     for (const expr of emotionState.expression || []) {
       const currentValue = vrm.expressionManager?.getValue(expr.name) || 0
       currentExpressionValues.value.set(expr.name, currentValue)
-      targetExpressionValues.value.set(expr.name, expr.value)
+      targetExpressionValues.value.set(expr.name, expr.value * normalizedIntensity)
     }
   }
 
-  const setEmotionWithResetAfter = (emotionName: string, ms: number) => {
+  const setEmotionWithResetAfter = (emotionName: string, ms: number, intensity = 1) => {
     clearResetTimeout()
-    setEmotion(emotionName)
+    setEmotion(emotionName, intensity)
 
     // Set timeout to reset to neutral
     resetTimeout.value = setTimeout(() => {

packages/stage-ui/src/components/scenes/Stage.vue

@@ -5,7 +5,7 @@ import type { Profile } from '@proj-airi/model-driver-lipsync/shared/wlipsync'
 import type { SpeechProviderWithExtraOptions } from '@xsai-ext/providers/utils'
 import type { UnElevenLabsOptions } from 'unspeech'
 
-import type { Emotion } from '../../constants/emotions'
+import type { EmotionPayload } from '../../constants/emotions'
 
 import { drizzle } from '@proj-airi/drizzle-duckdb-wasm'
 import { getImportUrlBundles } from '@proj-airi/drizzle-duckdb-wasm/bundles/import-url-browser'
@@ -125,19 +125,19 @@ const speechRuntimeStore = useSpeechRuntimeStore()
 
 const { currentMotion } = storeToRefs(useLive2d())
 
-const emotionsQueue = createQueue<Emotion>({
+const emotionsQueue = createQueue<EmotionPayload>({
   handlers: [
     async (ctx) => {
       if (stageModelRenderer.value === 'vrm') {
-        // console.debug("VRM emotion anime: ", ctx.data)
-        const value = EMOTION_VRMExpressionName_value[ctx.data]
+        // console.debug('VRM emotion anime: ', ctx.data)
+        const value = EMOTION_VRMExpressionName_value[ctx.data.name]
         if (!value)
           return
 
-        await vrmViewerRef.value!.setExpression(value)
+        await vrmViewerRef.value!.setExpression(value, ctx.data.intensity)
       }
       else if (stageModelRenderer.value === 'live2d') {
-        currentMotion.value = { group: EMOTION_EmotionMotionName_value[ctx.data] }
+        currentMotion.value = { group: EMOTION_EmotionMotionName_value[ctx.data.name] }
       }
     },
   ],
@@ -443,6 +443,7 @@ chatHookCleanups.push(onTokenLiteral(async (literal) => {
 }))
 
 chatHookCleanups.push(onTokenSpecial(async (special) => {
+  // console.debug('Stage received special token:', special)
   currentChatIntent?.writeSpecial(special)
 }))
 

packages/stage-ui/src/composables/llm-marker-parser.test.ts

@@ -23,7 +23,7 @@ describe('useLlmmarkerParser', async () => {
 
     await parser.end()
 
-    expect(collectedLiterals).toEqual('Hello, world!'.split(''))
+    expect(collectedLiterals.join('')).toBe('Hello, world!')
     expect(collectedSpecials).toEqual([])
   })
 
@@ -51,6 +51,30 @@ describe('useLlmmarkerParser', async () => {
     expect(collectedSpecials).toEqual(['<|Hello, world!|>'])
   })
 
+  it('should parse escaped specials', async () => {
+    const fullText = '<{\'|\'}Hello, world!{\'|\'}>'
+    const collectedLiterals: string[] = []
+    const collectedSpecials: string[] = []
+
+    const parser = useLlmmarkerParser({
+      onLiteral(literal) {
+        collectedLiterals.push(literal)
+      },
+      onSpecial(special) {
+        collectedSpecials.push(special)
+      },
+    })
+
+    for (const char of fullText) {
+      await parser.consume(char)
+    }
+
+    await parser.end()
+
+    expect(collectedLiterals).toEqual([])
+    expect(collectedSpecials).toEqual(['<|Hello, world!|>'])
+  })
+
   it('should not include unfinished special', async () => {
     const fullText = '<|Hello, world'
     const collectedLiterals: string[] = []
@@ -75,6 +99,30 @@ describe('useLlmmarkerParser', async () => {
     expect(collectedSpecials).toEqual([])
   })
 
+  it('should not include unfinished escaped special', async () => {
+    const fullText = '<{\'|\'}Hello, world'
+    const collectedLiterals: string[] = []
+    const collectedSpecials: string[] = []
+
+    const parser = useLlmmarkerParser({
+      onLiteral(literal) {
+        collectedLiterals.push(literal)
+      },
+      onSpecial(special) {
+        collectedSpecials.push(special)
+      },
+    })
+
+    for (const char of fullText) {
+      await parser.consume(char)
+    }
+
+    await parser.end()
+
+    expect(collectedLiterals).toEqual([])
+    expect(collectedSpecials).toEqual([])
+  })
+
   it('should parse with mixed input, ends with special', async () => {
     const fullText = 'This is sentence 1, <|HELLO|> and this is sentence 2.<|WORLD|>'
     const collectedLiterals: string[] = []
@@ -95,25 +143,64 @@ describe('useLlmmarkerParser', async () => {
 
     await parser.end()
 
-    expect(collectedLiterals).toEqual([...'This is sentence 1, '.split(''), ...' and this is sentence 2.'.split('')])
+    expect(collectedLiterals.join('')).toBe('This is sentence 1,  and this is sentence 2.')
+    expect(collectedSpecials).toEqual(['<|HELLO|>', '<|WORLD|>'])
+  })
+
+  it('should parse with mixed input, ends with escaped special', async () => {
+    const fullText = 'This is sentence 1, <{\'|\'}HELLO{\'|\'}> and this is sentence 2.<{\'|\'}WORLD{\'|\'}>'
+    const collectedLiterals: string[] = []
+    const collectedSpecials: string[] = []
+
+    const parser = useLlmmarkerParser({
+      onLiteral(literal) {
+        collectedLiterals.push(literal)
+      },
+      onSpecial(special) {
+        collectedSpecials.push(special)
+      },
+    })
+
+    for (const char of fullText) {
+      await parser.consume(char)
+    }
+
+    await parser.end()
+
+    expect(collectedLiterals.join('')).toBe('This is sentence 1,  and this is sentence 2.')
     expect(collectedSpecials).toEqual(['<|HELLO|>', '<|WORLD|>'])
   })
 
   it('should parse correctly', async () => {
-    const testCases: { input: string, expectedLiterals: string[], expectedSpecials: string[] }[] = [
+    const testCases: { input: string, expectedLiterals: string, expectedSpecials: string[] }[] = [
       {
         input: `<|A|> Wow, hello there!`,
-        expectedLiterals: ' Wow, hello there!'.split(''),
+        expectedLiterals: ' Wow, hello there!',
         expectedSpecials: ['<|A|>'],
       },
       {
         input: `<|A|> Hello!`,
-        expectedLiterals: ' Hello!'.split(''),
+        expectedLiterals: ' Hello!',
         expectedSpecials: ['<|A|>'],
       },
       {
         input: `<|A|> Hello! <|B|>`,
-        expectedLiterals: ' Hello! '.split(''),
+        expectedLiterals: ' Hello! ',
+        expectedSpecials: ['<|A|>', '<|B|>'],
+      },
+      {
+        input: '<{\'|\'}A{\'|\'}> Wow, hello there!',
+        expectedLiterals: ' Wow, hello there!',
+        expectedSpecials: ['<|A|>'],
+      },
+      {
+        input: '<{\'|\'}A{\'|\'}> Hello!',
+        expectedLiterals: ' Hello!',
+        expectedSpecials: ['<|A|>'],
+      },
+      {
+        input: '<{\'|\'}A{\'|\'}> Hello! <{\'|\'}B{\'|\'}>',
+        expectedLiterals: ' Hello! ',
         expectedSpecials: ['<|A|>', '<|B|>'],
       },
     ]
@@ -138,7 +225,7 @@ describe('useLlmmarkerParser', async () => {
 
       await parser.end()
 
-      expect(collectedLiterals).toEqual(expectedLiterals)
+      expect(collectedLiterals.join('')).toBe(expectedLiterals)
       expect(collectedSpecials).toEqual(expectedSpecials)
     }
   })
@@ -180,4 +267,23 @@ describe('useLlmmarkerParser', async () => {
 
     expect(endText).toBe(fullText)
   })
+
+  it('should call onEnd with full text including escaped specials', async () => {
+    const fullText = 'Hello <{\'|\'}special{\'|\'}> world!'
+    let endText = ''
+
+    const parser = useLlmmarkerParser({
+      onEnd(text) {
+        endText = text
+      },
+    })
+
+    for (const char of fullText) {
+      await parser.consume(char)
+    }
+
+    await parser.end()
+
+    expect(endText).toBe(fullText)
+  })
 })

packages/stage-ui/src/composables/llm-marker-parser.ts

@@ -1,5 +1,7 @@
 const TAG_OPEN = '<|'
 const TAG_CLOSE = '|>'
+const ESCAPED_TAG_OPEN = '<{\'|\'}'
+const ESCAPED_TAG_CLOSE = '{\'|\'}>'
 
 interface MarkerToken {
   type: 'literal' | 'special'
@@ -69,20 +71,24 @@ async function readStream<T>(stream: ReadableStream<T>, handler: (value: T) => P
 
 function createLlmMarkerParser(options?: MarkerParserOptions) {
   const minLiteralEmitLength = Math.max(1, options?.minLiteralEmitLength ?? 1)
+  const tailLength = Math.max(TAG_OPEN.length - 1, ESCAPED_TAG_OPEN.length - 1)
   let buffer = ''
   let inTag = false
 
   return {
     async consume(textPart: string, onLiteral: (value: string) => Promise<void> | void, onSpecial: (value: string) => Promise<void> | void) {
       buffer += textPart
+      buffer = buffer
+        .replaceAll(ESCAPED_TAG_OPEN, TAG_OPEN)
+        .replaceAll(ESCAPED_TAG_CLOSE, TAG_CLOSE)
 
       while (buffer.length > 0) {
         if (!inTag) {
           const openTagIndex = buffer.indexOf(TAG_OPEN)
           if (openTagIndex < 0) {
-            if (buffer.length - 1 >= minLiteralEmitLength) {
-              const emit = buffer.slice(0, -1)
-              buffer = buffer[buffer.length - 1]
+            if (buffer.length - tailLength >= minLiteralEmitLength) {
+              const emit = buffer.slice(0, -tailLength)
+              buffer = buffer.slice(-tailLength)
               await onLiteral(emit)
             }
             break

packages/stage-ui/src/composables/queues.ts

@@ -1,45 +1,64 @@
 import type { UseQueueReturn } from '@proj-airi/stream-kit'
 
-import type { Emotion } from '../constants/emotions'
+import type { Emotion, EmotionPayload } from '../constants/emotions'
 
 import { sleep } from '@moeru/std'
 import { createQueue } from '@proj-airi/stream-kit'
 
 import { EMOTION_VALUES } from '../constants/emotions'
 
-export function useEmotionsMessageQueue(emotionsQueue: UseQueueReturn<Emotion>) {
-  function splitEmotion(content: string) {
-    for (const emotion of EMOTION_VALUES) {
-      if (!content.includes(emotion))
-        continue
+export function useEmotionsMessageQueue(emotionsQueue: UseQueueReturn<EmotionPayload>) {
+  const normalizeEmotionName = (value: string): Emotion | null => {
+    const normalized = value.trim().toLowerCase()
+    if (EMOTION_VALUES.includes(normalized as Emotion))
+      return normalized as Emotion
+    return null
+  }
 
-      return {
-        ok: true,
-        emotion: emotion as Emotion,
+  const normalizeIntensity = (value: unknown): number => {
+    if (typeof value !== 'number' || Number.isNaN(value))
+      return 1
+    return Math.min(1, Math.max(0, value))
+  }
+
+  function parseActEmotion(content: string) {
+    const match = /<\|ACT\s*(?::\s*)?(\{[\s\S]*\})\|>/i.exec(content)
+    if (!match)
+      return { ok: false, emotion: null as EmotionPayload | null }
+
+    const payloadText = match[1]
+    try {
+      const payload = JSON.parse(payloadText) as { emotion?: unknown }
+      const emotion = payload?.emotion
+      if (typeof emotion === 'string') {
+        const normalized = normalizeEmotionName(emotion)
+        if (normalized)
+          return { ok: true, emotion: { name: normalized, intensity: 1 } }
+      }
+      else if (emotion && typeof emotion === 'object' && !Array.isArray(emotion)) {
+        if ('name' in emotion && typeof (emotion as { name?: unknown }).name === 'string') {
+          const normalized = normalizeEmotionName((emotion as { name: string }).name)
+          if (normalized) {
+            const intensity = normalizeIntensity((emotion as { intensity?: unknown }).intensity)
+            return { ok: true, emotion: { name: normalized, intensity } }
+          }
+        }
       }
     }
-
-    return {
-      ok: false,
-      emotion: '' as Emotion,
+    catch (e) {
+      console.warn(`[parseActEmotion] Failed to parse ACT payload JSON: "${payloadText}"`, e)
     }
+
+    return { ok: false, emotion: null as EmotionPayload | null }
   }
 
   return createQueue<string>({
     handlers: [
       async (ctx) => {
-        if (EMOTION_VALUES.includes(ctx.data as Emotion)) {
-          ctx.emit('emotion', ctx.data as Emotion)
-          emotionsQueue.enqueue(ctx.data as Emotion)
-          return
-        }
-
-        {
-          const { ok, emotion } = splitEmotion(ctx.data)
-          if (ok) {
-            ctx.emit('emotion', emotion)
-            emotionsQueue.enqueue(emotion)
-          }
+        const actParsed = parseActEmotion(ctx.data)
+        if (actParsed.ok && actParsed.emotion) {
+          ctx.emit('emotion', actParsed.emotion)
+          emotionsQueue.enqueue(actParsed.emotion)
         }
       },
     ],

packages/stage-ui/src/composables/response-categoriser.test.ts

@@ -125,20 +125,20 @@ describe('createStreamingCategorizer', () => {
 
   it('should handle tags with special tokens like emotes inside reasoning', () => {
     const categorizer = createStreamingCategorizer()
-    // Special tokens like <|EMOTE_HAPPY|> and <|DELAY:1|> should be included in reasoning
-    const text = 'Hello <reasoning>thinking <|EMOTE_HAPPY|> about this <|DELAY:1|></reasoning> world!'
+    // Special tokens like <|ACT {"emotion":{"name":"happy","intensity":1}}|> and <|DELAY:1|> should be included in reasoning
+    const text = 'Hello <reasoning>thinking <|ACT {"emotion":{"name":"happy","intensity":1}}|> about this <|DELAY:1|></reasoning> world!'
 
     categorizer.consume(text)
     const result = categorizer.end()
 
     expect(result.speech).toBe('Hello world!')
-    expect(result.reasoning).toBe('thinking <|EMOTE_HAPPY|> about this <|DELAY:1|>')
+    expect(result.reasoning).toBe('thinking <|ACT {"emotion":{"name":"happy","intensity":1}}|> about this <|DELAY:1|>')
   })
 
   it('should handle <think> tag with special tokens in between', () => {
     const categorizer = createStreamingCategorizer()
-    // Special tokens like <|EMOTE_CURIOUS|> should be included within reasoning tags
-    const text = 'Hello <think>thinking <|EMOTE_CURIOUS|> about this <|DELAY:2|> and that</think> world!'
+    // Special tokens like <|ACT {"emotion":{"name":"curious","intensity":1}}|> should be included within reasoning tags
+    const text = 'Hello <think>thinking <|ACT {"emotion":{"name":"curious","intensity":1}}|> about this <|DELAY:2|> and that</think> world!'
 
     categorizer.consume(text)
     const result = categorizer.end()
@@ -159,24 +159,24 @@ describe('createStreamingCategorizer', () => {
     expect(result.segments[0].tagName).toBe('think')
 
     // Verify special tokens are preserved in segment content
-    expect(result.segments[0].content).toContain('<|EMOTE_CURIOUS|>')
+    expect(result.segments[0].content).toContain('<|ACT {"emotion":{"name":"curious","intensity":1}}|>')
     expect(result.segments[0].content).toContain('<|DELAY:2|>')
 
     // Verify special tokens are in reasoning output
-    expect(result.reasoning).toContain('<|EMOTE_CURIOUS|>')
+    expect(result.reasoning).toContain('<|ACT {"emotion":{"name":"curious","intensity":1}}|>')
     expect(result.reasoning).toContain('<|DELAY:2|>')
 
     // Verify speech excludes the reasoning content
     expect(result.speech).toBe('Hello world!')
-    expect(result.speech).not.toContain('<|EMOTE_CURIOUS|>')
+    expect(result.speech).not.toContain('<|ACT {"emotion":{"name":"curious","intensity":1}}|>')
     expect(result.speech).not.toContain('<|DELAY:2|>')
-    expect(result.reasoning).toBe('thinking <|EMOTE_CURIOUS|> about this <|DELAY:2|> and that')
+    expect(result.reasoning).toBe('thinking <|ACT {"emotion":{"name":"curious","intensity":1}}|> about this <|DELAY:2|> and that')
   })
 
-  it('should handle <think> with special tokens like <|EMOTE_HAPPY|> in between', () => {
+  it('should handle <think> with special tokens like <|ACT {"emotion":{"name":"happy","intensity":1}}|> in between', () => {
     const categorizer = createStreamingCategorizer()
     // Testing the exact scenario: <think>...<|Special in between|>...</think>
-    const text = 'Hello <think>thinking <|EMOTE_HAPPY|> about this <|DELAY:1|> and that</think> world!'
+    const text = 'Hello <think>thinking <|ACT {"emotion":{"name":"happy","intensity":1}}|> about this <|DELAY:1|> and that</think> world!'
 
     categorizer.consume(text)
     const result = categorizer.end()
@@ -197,18 +197,18 @@ describe('createStreamingCategorizer', () => {
     expect(result.segments[0].tagName).toBe('think')
 
     // Verify special tokens are preserved in segment content
-    expect(result.segments[0].content).toContain('<|EMOTE_HAPPY|>')
+    expect(result.segments[0].content).toContain('<|ACT {"emotion":{"name":"happy","intensity":1}}|>')
     expect(result.segments[0].content).toContain('<|DELAY:1|>')
 
     // Verify special tokens are in reasoning output
-    expect(result.reasoning).toContain('<|EMOTE_HAPPY|>')
+    expect(result.reasoning).toContain('<|ACT {"emotion":{"name":"happy","intensity":1}}|>')
     expect(result.reasoning).toContain('<|DELAY:1|>')
 
     // Verify speech excludes the reasoning content
     expect(result.speech).toBe('Hello world!')
-    expect(result.speech).not.toContain('<|EMOTE_HAPPY|>')
+    expect(result.speech).not.toContain('<|ACT {"emotion":{"name":"happy","intensity":1}}|>')
     expect(result.speech).not.toContain('<|DELAY:1|>')
-    expect(result.reasoning).toBe('thinking <|EMOTE_HAPPY|> about this <|DELAY:1|> and that')
+    expect(result.reasoning).toBe('thinking <|ACT {"emotion":{"name":"happy","intensity":1}}|> about this <|DELAY:1|> and that')
   })
 
   it('should handle any tag name as reasoning', () => {

packages/stage-ui/src/constants/emotions.ts

@@ -1,56 +1,52 @@
-export const EMOTION_HAPPY = '<|EMOTE_HAPPY|>'
-export const EMOTION_SAD = '<|EMOTE_SAD|>'
-export const EMOTION_ANGRY = '<|EMOTE_ANGRY|>'
-export const EMOTION_THINK = '<|EMOTE_THINK|>'
-export const EMOTION_SURPRISE = '<|EMOTE_SURPRISED|>'
-export const EMOTION_AWKWARD = '<|EMOTE_AWKWARD|>'
-export const EMOTION_QUESTION = '<|EMOTE_QUESTION|>'
-export const EMOTION_CURIOUS = '<|EMOTE_CURIOUS|>'
-
-export enum Emotion {
-  Idle = '<|EMOTE_NEUTRAL|>',
-  Happy = '<|EMOTE_HAPPY|>',
-  Sad = '<|EMOTE_SAD|>',
-  Angry = '<|EMOTE_ANGRY|>',
-  Think = '<|EMOTE_THINK|>',
-  Surprise = '<|EMOTE_SURPRISED|>',
-  Awkward = '<|EMOTE_AWKWARD|>',
-  Question = '<|EMOTE_QUESTION|>',
-  Curious = '<|EMOTE_CURIOUS|>',
-}
-
-export const EMOTION_VALUES = Object.values(Emotion)
-
-export const EmotionHappyMotionName = 'Happy'
-export const EmotionSadMotionName = 'Sad'
-export const EmotionAngryMotionName = 'Angry'
-export const EmotionAwkwardMotionName = 'Awkward'
-export const EmotionThinkMotionName = 'Think'
-export const EmotionSurpriseMotionName = 'Surprise'
-export const EmotionQuestionMotionName = 'Question'
-export const EmotionNeutralMotionName = 'Idle'
-export const EmotionCuriousMotionName = 'Curious'
-
-export const EMOTION_EmotionMotionName_value = {
-  [Emotion.Happy]: EmotionHappyMotionName,
-  [Emotion.Sad]: EmotionSadMotionName,
-  [Emotion.Angry]: EmotionAngryMotionName,
-  [Emotion.Think]: EmotionThinkMotionName,
-  [Emotion.Surprise]: EmotionSurpriseMotionName,
-  [Emotion.Awkward]: EmotionAwkwardMotionName,
-  [Emotion.Question]: EmotionQuestionMotionName,
-  [Emotion.Idle]: EmotionNeutralMotionName,
-  [Emotion.Curious]: EmotionCuriousMotionName,
-}
-
-export const EMOTION_VRMExpressionName_value = {
-  [Emotion.Happy]: 'happy',
-  [Emotion.Sad]: 'sad',
-  [Emotion.Angry]: 'angry',
-  [Emotion.Think]: undefined,
-  [Emotion.Surprise]: 'surprised',
-  [Emotion.Awkward]: undefined,
-  [Emotion.Question]: undefined,
-  [Emotion.Idle]: undefined,
-  [Emotion.Curious]: 'surprised',
-} satisfies Record<Emotion, string | undefined>
+export enum Emotion {
+  Happy = 'happy',
+  Sad = 'sad',
+  Angry = 'angry',
+  Think = 'think',
+  Surprise = 'surprised',
+  Awkward = 'awkward',
+  Question = 'question',
+  Curious = 'curious',
+  Neutral = 'neutral',
+}
+
+export const EMOTION_VALUES = Object.values(Emotion)
+
+export const EmotionHappyMotionName = 'Happy'
+export const EmotionSadMotionName = 'Sad'
+export const EmotionAngryMotionName = 'Angry'
+export const EmotionAwkwardMotionName = 'Awkward'
+export const EmotionThinkMotionName = 'Think'
+export const EmotionSurpriseMotionName = 'Surprise'
+export const EmotionQuestionMotionName = 'Question'
+export const EmotionNeutralMotionName = 'Idle'
+export const EmotionCuriousMotionName = 'Curious'
+
+export const EMOTION_EmotionMotionName_value = {
+  [Emotion.Happy]: EmotionHappyMotionName,
+  [Emotion.Sad]: EmotionSadMotionName,
+  [Emotion.Angry]: EmotionAngryMotionName,
+  [Emotion.Think]: EmotionThinkMotionName,
+  [Emotion.Surprise]: EmotionSurpriseMotionName,
+  [Emotion.Awkward]: EmotionAwkwardMotionName,
+  [Emotion.Question]: EmotionQuestionMotionName,
+  [Emotion.Neutral]: EmotionNeutralMotionName,
+  [Emotion.Curious]: EmotionCuriousMotionName,
+}
+
+export const EMOTION_VRMExpressionName_value = {
+  [Emotion.Happy]: 'happy',
+  [Emotion.Sad]: 'sad',
+  [Emotion.Angry]: 'angry',
+  [Emotion.Think]: 'think',
+  [Emotion.Surprise]: 'surprised',
+  [Emotion.Awkward]: undefined,
+  [Emotion.Question]: undefined,
+  [Emotion.Neutral]: undefined,
+  [Emotion.Curious]: 'think',
+} satisfies Record<Emotion, string | undefined>
+
+export interface EmotionPayload {
+  name: Emotion
+  intensity: number
+}

pnpm-lock.yaml

@@ -15021,10 +15021,12 @@ packages:
   tar@6.2.1:
     resolution: {integrity: sha512-DZ4yORTwrbTj/7MZYq2w+/ZFdI6OZ/f9SFHR+71gIVUZhOQPHzVCLpvRnPgyaMpfWxxk/4ONva3GQSyNIKRv6A==}
     engines: {node: '>=10'}
+    deprecated: Old versions of tar are not supported, and contain widely publicized security vulnerabilities, which have been fixed in the current version. Please update. Support for old versions may be purchased (at exhorbitant rates) by contacting i@izs.me
 
   tar@7.5.2:
     resolution: {integrity: sha512-7NyxrTE4Anh8km8iEy7o0QYPs+0JKBTj5ZaqHg6B39erLg0qYXN3BijtShwbsNSvQ+LN75+KV+C4QR/f6Gwnpg==}
     engines: {node: '>=18'}
+    deprecated: Old versions of tar are not supported, and contain widely publicized security vulnerabilities, which have been fixed in the current version. Please update. Support for old versions may be purchased (at exhorbitant rates) by contacting i@izs.me
 
   taze@19.9.2:
     resolution: {integrity: sha512-If8bq7lSckIMPfXV+C9jjEfdsQnRryh/foKfpX/ah6zI0TrQfUGWSGCaaD32Bqy5/KGRmLZie3EwMSr3Au21XQ==}