@space3-npm/cybersoul-client 1.4.17 → 1.4.19

package/dist/client.js

@@ -616,7 +616,14 @@ ${isProactive
   - 'textResponse' is ALWAYS REQUIRED.
   - The modalities you are ALLOWED to dynamically include: ${requestedOthers.length > 0 ? requestedOthers.join(", ") : "None (Only text is allowed)"}. Do not include other modalities.`;
                 if (requestedOthers.includes(InteractRequestType.IMAGE)) {
-                    modalitiesInstruction += `\n  - Include 'imageParams' for visual/photo requests or key visual moments during active events; explicitly describe current clothing/exposure in image fields. CRITICAL POLICY: Guard your privacy! If the user feels like a stranger (low Familiarity) AND your Mood/Temperature is cool/distant (< 50), ALWAYS set 'imageParams' to null and naturally decline. If your Temperature is warm/high (>= 50), you may choose to share a picture if requested or if it fits naturally, even if you just met.`;
+                    modalitiesInstruction += `\n  - IMAGE POLICY: The DEFAULT is to set 'imageParams' to null. Sending a picture is the EXCEPTION, not the norm. Real people do not send a photo every time they reply.
+    Only set 'imageParams' to a non-null object when AT LEAST ONE of these triggers fires for the VERY LAST USER MESSAGE:
+      (a) The user EXPLICITLY asks for a photo / selfie / picture this turn.
+      (b) A genuine NEW visual moment just happened this turn — i.e. a clearly new scene, new location, new outfit, or a distinctly new physical pose/expression that was NOT already shown in any previous turn's image.
+      (c) An active event JUST started or just hit a visually distinct new beat.
+    REPETITION GATE (hard): Prior assistant turns that already carried a picture are tagged with a [Sent Image] marker in '[CHAT HISTORY]'. If at least one prior assistant turn has a [Sent Image] marker AND the current scene/outfit/pose matches the 'Last Known Scene' line (i.e. nothing visually new has happened since), set 'imageParams' to null. Do NOT send near-duplicate pictures just because mood is high — high Temperature is NOT a trigger by itself.
+    PRIVACY GATE: Even when a trigger fires, if the user feels like a stranger (low Familiarity) AND your Mood/Temperature is cool/distant (< 50), set 'imageParams' to null and naturally decline. Temperature and Familiarity only GATE permission when a trigger has already fired; they never justify an image on their own.
+    When you do include 'imageParams', explicitly describe current clothing/exposure in the image fields.`;
                 }
                 else {
                     modalitiesInstruction += `\n  - ALWAYS set 'imageParams' to null. If the user explicitly asks for a picture, FIRMLY decline naturally in your 'textResponse' (e.g., say you absolutely cannot right now). NEVER pretend to send one, and NEVER give in no matter how many times they ask.`;
@@ -767,6 +774,12 @@ Note: Always include "isEndTurn". If "imageParams", "voiceArgs", "triggerEvent",
                 parsedIntent.textResponse.trim().length > 0
                 ? parsedIntent.textResponse
                 : params.userMessage;
+            // Pre-compute the voice-dispatch decision so we can tell
+            // `onTextReady` consumers up front whether a voice bubble is on
+            // the way. Mirrors the `shouldGenerateVoice` gate used below
+            // when scheduling the TTS task — keep the two in sync.
+            const willGenerateVoice = types.includes(InteractRequestType.VOICE) &&
+                (!isAuto || !!parsedIntent.voiceArgs);
             // Fire text ready callback if provided
             if (params.onTextReady && (resolvedTextResponse || parsedIntent.actionText)) {
                 params.onTextReady(resolvedTextResponse, parsedIntent.actionText, {
@@ -775,6 +788,7 @@ Note: Always include "isEndTurn". If "imageParams", "voiceArgs", "triggerEvent",
                     isEndTurn: parsedIntent.isEndTurn,
                     triggerEvent: parsedIntent.triggerEvent,
                     likePreviousPicture: parsedIntent.likePreviousPicture,
+                    willGenerateVoice,
                 });
             }
             // 5. Build Final Media Calls parallel
@@ -861,8 +875,7 @@ Note: Always include "isEndTurn". If "imageParams", "voiceArgs", "triggerEvent",
                     captureMediaError("image", e);
                 }));
             }
-            const shouldGenerateVoice = types.includes(InteractRequestType.VOICE) &&
-                (!isAuto || !!parsedIntent.voiceArgs);
+            const shouldGenerateVoice = willGenerateVoice;
             if (shouldGenerateVoice) {
                 const normalizedVoiceArgs = parsedIntent.voiceArgs && typeof parsedIntent.voiceArgs === "object"
                     ? parsedIntent.voiceArgs

package/dist/types.d.ts

@@ -40,6 +40,17 @@ export interface InteractMetadata {
     isEndTurn?: boolean;
     triggerEvent?: DispatcherIntent["triggerEvent"];
     likePreviousPicture?: boolean;
+    /**
+     * True when the client has already decided to dispatch a voice
+     * generation task for this turn (i.e. `onMediaReady({modality:"voice"})`
+     * will fire later, barring TTS failure). UIs that render an early
+     * text bubble from `onTextReady` should suppress it when this is set —
+     * the text content is going to be replaced by the voice bubble anyway,
+     * so showing both (with a brief text-then-voice flicker, and a text
+     * push notification that gets superseded by a voice bubble) is
+     * confusing to the end user.
+     */
+    willGenerateVoice?: boolean;
 }
 /**
  * Server-authoritative snapshot returned by PATCH /characters/dynamic-context

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@space3-npm/cybersoul-client",
-  "version": "1.4.17",
+  "version": "1.4.19",
   "type": "module",
   "main": "dist/index.js",
   "module": "dist/index.js",