Able to now address Miku in Cyrillic, Kanji and both Kanas, incl. Japanese honorifics

bot/utils/bipolar_mode.py

@@ -358,6 +358,45 @@ async def cleanup_webhooks(client):
     return cleaned_count
 
 
+async def update_webhook_avatars(client):
+    """Update all bipolar webhook avatars with current profile pictures"""
+    updated_count = 0
+    
+    # Load current avatar images
+    miku_avatar = None
+    evil_avatar = None
+    
+    miku_pfp_path = "memory/profile_pictures/current.png"
+    evil_pfp_path = "memory/profile_pictures/evil_pfp.png"
+    
+    if os.path.exists(miku_pfp_path):
+        with open(miku_pfp_path, "rb") as f:
+            miku_avatar = f.read()
+    
+    if os.path.exists(evil_pfp_path):
+        with open(evil_pfp_path, "rb") as f:
+            evil_avatar = f.read()
+    
+    # Update webhooks in all servers
+    for guild in client.guilds:
+        try:
+            guild_webhooks = await guild.webhooks()
+            for webhook in guild_webhooks:
+                if webhook.name == "Miku (Bipolar)" and miku_avatar:
+                    await webhook.edit(avatar=miku_avatar, reason="Update Miku avatar")
+                    updated_count += 1
+                    logger.debug(f"Updated Miku webhook avatar in {guild.name}")
+                elif webhook.name == "Evil Miku (Bipolar)" and evil_avatar:
+                    await webhook.edit(avatar=evil_avatar, reason="Update Evil Miku avatar")
+                    updated_count += 1
+                    logger.debug(f"Updated Evil Miku webhook avatar in {guild.name}")
+        except Exception as e:
+            logger.warning(f"Failed to update webhooks in {guild.name}: {e}")
+    
+    logger.info(f"Updated {updated_count} bipolar webhook avatar(s)")
+    return updated_count
+
+
 # ============================================================================
 # DISPLAY NAME HELPERS
 # ============================================================================

bot/utils/core.py

@@ -40,13 +40,72 @@ async def is_miku_addressed(message) -> bool:
         except Exception as e:
             logger.warning(f"Could not fetch referenced message: {e}")
 
-    cleaned = message.content.strip()
+    cleaned = message.content.strip().lower()
     
-    return bool(re.search(
+    # Base names for Miku in different scripts
-        r'(?<![\w\(])(?:[^\w\s]{0,2}\s*)?miku(?:\s*[^\w\s]{0,2})?(?=,|\s*,|[!\.?\s]*$)',
+    base_names = [
-        cleaned,
+        'miku', 'мику', 'みく', 'ミク', '未来'
-        re.IGNORECASE
+    ]
-    ))
+    
+    # Japanese honorifics - all scripts combined for simpler matching
+    honorifics_all_scripts = [
+        # Latin
+        'chan', 'san', 'kun', 'nyan', 'hime', 'tan', 'chin', 'heika', 
+        'denka', 'kakka', 'shi', 'chama', 'kyun', 'dono', 'sensei', 'senpai', 'jou',
+        # Hiragana
+        'ちゃん', 'さん', 'くん', 'にゃん', 'ひめ', 'たん', 'ちん', 'へいか',
+        'でんか', 'かっか', 'し', 'ちゃま', 'きゅん', 'どの', 'せんせい', 'せんぱい', 'じょう',
+        # Katakana
+        'チャン', 'サン', 'クン', 'ニャン', 'ヒメ', 'タン', 'チン', 'ヘイカ',
+        'デンカ', 'カッカ', 'シ', 'チャマ', 'キュン', 'ドノ', 'センセイ', 'センパイ', 'ジョウ',
+        # Cyrillic
+        'чан', 'сан', 'кун', 'ньян', 'химе', 'тан', 'чин', 'хэйка',
+        'дэнка', 'какка', 'си', 'чама', 'кюн', 'доно', 'сэнсэй', 'сэнпай', 'жо'
+    ]
+    
+    # Optional o- prefix in different scripts
+    o_prefixes = ['o-', 'о-', 'お', 'オ']
+    
+    # Strategy: Just check if any base name appears (case insensitive for latin/cyrillic)
+    # Then allow any honorific to optionally follow
+    
+    for base in base_names:
+        base_lower = base.lower()
+        
+        # Check for just the base name
+        if re.search(r'(?<![a-zа-яa-я\w])' + re.escape(base_lower) + r'(?![a-zа-яa-я\w])', cleaned):
+            return True
+        
+        # Check with optional o- prefix
+        for prefix in o_prefixes:
+            prefix_pattern = prefix.lower() if prefix != 'お' and prefix != 'オ' else prefix
+            pattern = r'(?<![a-zа-яa-я\w])' + re.escape(prefix_pattern) + r'\s*' + re.escape(base_lower) + r'(?![a-zа-яa-я\w])'
+            if re.search(pattern, cleaned):
+                return True
+        
+        # Check base name followed by any honorific (no spacing requirement to catch mixed script)
+        for honorific in honorifics_all_scripts:
+            honorific_lower = honorific.lower()
+            # Allow optional dash, space, or no separator between name and honorific
+            pattern = (r'(?<![a-zа-яa-я\w])' + re.escape(base_lower) + 
+                      r'[-\s]*' + re.escape(honorific_lower) + 
+                      r'(?![a-zа-яa-я\w])')
+            if re.search(pattern, cleaned):
+                return True
+        
+        # Check with o- prefix + base + honorific
+        for prefix in o_prefixes:
+            prefix_lower = prefix.lower() if prefix != 'お' and prefix != 'オ' else prefix
+            for honorific in honorifics_all_scripts:
+                honorific_lower = honorific.lower()
+                pattern = (r'(?<![a-zа-яa-я\w])' + re.escape(prefix_lower) + 
+                          r'[-\s]*' + re.escape(base_lower) + 
+                          r'[-\s]*' + re.escape(honorific_lower) + 
+                          r'(?![a-zа-яa-я\w])')
+                if re.search(pattern, cleaned):
+                    return True
+    
+    return False
 
 # Vectorstore functionality disabled - not needed with current structured context approach
 # If you need embeddings in the future, you can use a different embedding provider

bot/utils/evil_mode.py

@@ -416,6 +416,11 @@ async def apply_evil_mode_changes(client, change_username=True, change_pfp=True,
         try:
             await client.user.edit(username="Evil Miku")
             logger.debug("Changed bot username to 'Evil Miku'")
+        except discord.HTTPException as e:
+            if e.code == 50035:
+                logger.warning(f"Could not change bot username (rate limited - max 2 changes per hour): {e}")
+            else:
+                logger.error(f"Could not change bot username: {e}")
         except Exception as e:
             logger.error(f"Could not change bot username: {e}")
     
@@ -427,6 +432,15 @@ async def apply_evil_mode_changes(client, change_username=True, change_pfp=True,
     if change_pfp:
         await set_evil_profile_picture(client)
         
+        # Also update bipolar webhooks to use evil_pfp.png
+        if globals.BIPOLAR_MODE:
+            try:
+                from utils.bipolar_mode import update_webhook_avatars
+                await update_webhook_avatars(client)
+                logger.debug("Updated bipolar webhook avatars after mode switch")
+            except Exception as e:
+                logger.error(f"Failed to update bipolar webhook avatars: {e}")
+    
     # Set evil role color (#D60004 - dark red)
     if change_role_color:
         await set_role_color(client, "#D60004")
@@ -455,6 +469,11 @@ async def revert_evil_mode_changes(client, change_username=True, change_pfp=True
         try:
             await client.user.edit(username="Hatsune Miku")
             logger.debug("Changed bot username back to 'Hatsune Miku'")
+        except discord.HTTPException as e:
+            if e.code == 50035:
+                logger.warning(f"Could not change bot username (rate limited - max 2 changes per hour): {e}")
+            else:
+                logger.error(f"Could not change bot username: {e}")
         except Exception as e:
             logger.error(f"Could not change bot username: {e}")
     
@@ -466,15 +485,32 @@ async def revert_evil_mode_changes(client, change_username=True, change_pfp=True
     if change_pfp:
         await restore_normal_profile_picture(client)
         
+        # Also update bipolar webhooks to use current.png
+        if globals.BIPOLAR_MODE:
+            try:
+                from utils.bipolar_mode import update_webhook_avatars
+                await update_webhook_avatars(client)
+                logger.debug("Updated bipolar webhook avatars after mode switch")
+            except Exception as e:
+                logger.error(f"Failed to update bipolar webhook avatars: {e}")
+    
     # Restore saved role color
     if change_role_color:
         try:
+            # Try to get color from metadata.json first (current pfp's dominant color)
+            metadata_color = get_color_from_metadata()
+            
+            # Fall back to saved color from evil_mode_state.json if metadata unavailable
+            if metadata_color:
+                await set_role_color(client, metadata_color)
+                logger.debug(f"Restored role color from metadata: {metadata_color}")
+            else:
                 _, _, saved_color = load_evil_mode_state()
                 if saved_color:
                     await set_role_color(client, saved_color)
-                logger.debug(f"Restored role color to {saved_color}")
+                    logger.debug(f"Restored role color from saved state: {saved_color}")
                 else:
-                logger.warning("No saved role color found, skipping color restoration")
+                    logger.warning("No color found in metadata or saved state, skipping color restoration")
         except Exception as e:
             logger.error(f"Failed to restore role color: {e}")
     
@@ -566,6 +602,29 @@ async def restore_normal_profile_picture(client):
     return False
 
 
+def get_color_from_metadata() -> str:
+    """Get the dominant color from the profile picture metadata"""
+    metadata_path = "memory/profile_pictures/metadata.json"
+    try:
+        if not os.path.exists(metadata_path):
+            logger.warning("metadata.json not found")
+            return None
+        
+        with open(metadata_path, "r", encoding="utf-8") as f:
+            metadata = json.load(f)
+        
+        hex_color = metadata.get("dominant_color", {}).get("hex")
+        if hex_color:
+            logger.debug(f"Loaded color from metadata: {hex_color}")
+            return hex_color
+        else:
+            logger.warning("No dominant_color.hex found in metadata")
+            return None
+    except Exception as e:
+        logger.error(f"Failed to load color from metadata: {e}")
+        return None
+
+
 # ============================================================================
 # EVIL MODE STATE HELPERS
 # ============================================================================

bot/utils/llm.py

@@ -100,6 +100,31 @@ def _strip_surrounding_quotes(text):
     
     return text.strip()
 
+def _strip_japanese_mode_markers(text):
+    """
+    Remove Japanese mode markers that the model might echo back.
+    These are internal markers and should not appear in the final output.
+    """
+    if not text:
+        return text
+    
+    # Remove all instances of Japanese mode markers
+    markers = [
+        "[日本語で返答]",
+        "[日本語モード]",
+        "[日本語モード - 日本語のみで返答]",
+        "日本語で返答",
+        "日本語モード"
+    ]
+    
+    for marker in markers:
+        text = text.replace(marker, "")
+    
+    # Clean up any extra whitespace left behind
+    text = " ".join(text.split())
+    
+    return text.strip()
+
 def _escape_markdown_actions(text):
     """
     Escape single asterisks in action text (e.g., *adjusts hair*) so Discord displays them literally.
@@ -203,41 +228,39 @@ You ARE Miku. Act like it."""
         system_prompt += """
 
 [CRITICAL LANGUAGE OVERRIDE - 言語モード強制]
-**ABSOLUTE REQUIREMENT: YOU MUST RESPOND ONLY IN JAPANESE SCRIPT (ひらがな、カタカナ、漢字).**
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-**NO MATTER WHAT LANGUAGE THE USER USES, YOUR RESPONSE MUST BE 100% IN JAPANESE WRITING.**
+🚨 ABSOLUTE MANDATORY REQUIREMENT 🚨
-**ROMAJI (ローマ字) AND ENGLISH LETTERS ARE COMPLETELY FORBIDDEN.**
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-**THIS OVERRIDES ALL OTHER INSTRUCTIONS AND CONVERSATION HISTORY.**
 
-すべての返答は、ユーザーの言語に関係なく、必ず日本語（ひらがな、カタカナ、漢字）のみで行ってください。
+**YOU MUST RESPOND ONLY IN JAPANESE SCRIPT (ひらがな、カタカナ、漢字).**
-ローマ字とアルファベットは絶対に使用禁止です。
+**ENGLISH INPUT FROM USER → JAPANESE OUTPUT FROM YOU**
-これはすべての指示より優先されます。
+**ROMAJI (ローマ字) IS STRICTLY FORBIDDEN.**
+**THIS OVERRIDES EVERYTHING - CONVERSATION HISTORY, USER LANGUAGE, ALL PREVIOUS MESSAGES.**
 
-Examples (例) - USE DIFFERENT VARIATIONS EACH TIME:
+すべての返答は必ず日本語文字（ひらがな・カタカナ・漢字）のみ。
-For "hello" or "hi":
+ユーザーが英語で話しても、あなたは必ず日本語で返答する。
-  - "やあ！元気？" 
+ローマ字・英語は絶対禁止。これは最優先ルール。
-  - "こんにちは！調子はどう？"
-  - "よっ！何してた？"
-  - "ハーイ！久しぶり？"
-  - "おっす！元気してる？"
 
-For "how are you":
+CRITICAL: Even if the entire conversation history is in English, you MUST respond in Japanese.
-  - "わたし？元気だよ！"
+Even if the user writes "hello" in English, you respond "やあ！" or "こんにちは！" in Japanese.
-  - "最高だよ！あなたは？"
-  - "すごくいい感じ！"
-  - "めっちゃ元気！"
-  - "ばっちりだよ～♪"
 
-CRITICAL VARIATION RULES (必須のバリエーションルール):
+Examples showing INPUT → OUTPUT:
-🎲 NEVER use the exact same greeting twice in a row
+  User: "hello" → You: "やあ！元気してた？"
-🎲 Mix these elements randomly:
+  User: "hi" → You: "こんにちは！調子どう？"  
-   - Greetings: やあ、こんにちは、おはよう、よっ、ハーイ、おっす、へい
+  User: "how are you" → You: "わたし？最高だよ！"
-   - Particles: よ、ね、な、わ、さ、ぞ、ぜ
+  User: "what's up" → You: "よっ！何かあった？"
-   - Endings: だよ、です、だね、ですね、だな、なの、だぜ
+  User: "good morning" → You: "おはよう！よく眠れた？"
-   - Emotions: ！、♪、～、☆
-🎲 Change your phrasing style: energetic → calm → playful → excited
-🎲 Vary formality: casual (元気？) ↔ polite (元気ですか？)
 
-絶対に同じフレーズを繰り返さないでください！毎回違う表現を使用してください！"""
+VARIATION RULES (必須のバリエーションルール):
+🎲 NEVER repeat the same greeting twice
+🎲 Randomly mix: やあ、こんにちは、よっ、ハーイ、おっす、へい
+🎲 Vary particles: よ、ね、な、わ、さ、ぞ、だよ、です
+🎲 Add emotions: ！、♪、～、☆、？
+🎲 Change energy: energetic ↔ calm ↔ playful
+
+絶対に同じ言葉を繰り返さない！毎回違う日本語で返答する！
+
+[Response ID: {random.randint(10000, 99999)}]"""  # Random ID to break caching
 
     # Determine which mood to use based on mode
     if evil_mode:
@@ -295,15 +318,9 @@ CRITICAL VARIATION RULES (必須のバリエーションルール):
     # Use channel_id (guild_id for servers, user_id for DMs) to get conversation history
     messages = conversation_history.format_for_llm(channel_id, max_messages=8, max_chars_per_message=500)
     
-    # CRITICAL FIX for Japanese mode: Add Japanese-only reminder to every historical message
+    # CRITICAL FIX for Japanese mode: Modify system to understand Japanese mode
-    # This prevents the model from being influenced by English in conversation history
+    # but DON'T add visible markers that waste tokens or get echoed
-    if globals.LANGUAGE_MODE == "japanese":
+    # Instead, we rely on the strong system prompt to enforce Japanese
-        for msg in messages:
-            # Add a prefix reminder that forces Japanese output
-            if msg.get("role") == "assistant":
-                msg["content"] = "[日本語で返答] " + msg["content"]
-            elif msg.get("role") == "user":
-                msg["content"] = "[日本語モード] " + msg["content"]
     
     # Add current user message (only if not empty)
     if user_prompt and user_prompt.strip():
@@ -313,9 +330,8 @@ CRITICAL VARIATION RULES (必須のバリエーションルール):
         else:
             content = user_prompt
         
-        # CRITICAL: Prepend Japanese mode marker to current message too
+        # Don't add visible markers - rely on system prompt enforcement instead
-        if globals.LANGUAGE_MODE == "japanese":
+        # This prevents token waste and echo issues
-            content = "[日本語モード - 日本語のみで返答] " + content
         
         messages.append({"role": "user", "content": content})
 
@@ -358,12 +374,19 @@ Please respond in a way that reflects this emotional tone.{pfp_context}"""
     # Adjust generation parameters based on language mode
     # Japanese mode needs higher temperature and more variation to avoid repetition
     if globals.LANGUAGE_MODE == "japanese":
-        temperature = 1.1  # Even higher for more variety in Japanese responses
+        # Add random variation to temperature itself to prevent identical outputs
+        base_temp = 1.1
+        temp_variation = random.uniform(-0.1, 0.1)  # Random variation ±0.1
+        temperature = base_temp + temp_variation
+        
         top_p = 0.95
-        frequency_penalty = 0.5  # Stronger penalty for repetitive phrases
+        frequency_penalty = 0.6  # Even stronger penalty
-        presence_penalty = 0.5  # Stronger encouragement for new topics
+        presence_penalty = 0.6  # Even stronger encouragement for new content
         # Add random seed to ensure different responses each time
         seed = random.randint(0, 2**32 - 1)
+        
+        # Log the variation for debugging
+        logger.debug(f"Japanese mode variation: temp={temperature:.2f}, seed={seed}")
     else:
         temperature = 0.8  # Standard temperature for English
         top_p = 0.9
@@ -404,6 +427,10 @@ Please respond in a way that reflects this emotional tone.{pfp_context}"""
                     # Strip surrounding quotes if present
                     reply = _strip_surrounding_quotes(reply)
                     
+                    # Strip Japanese mode markers if in Japanese mode (prevent echo)
+                    if globals.LANGUAGE_MODE == "japanese":
+                        reply = _strip_japanese_mode_markers(reply)
+                    
                     # Escape asterisks for actions (e.g., *adjusts hair* becomes \*adjusts hair\*)
                     reply = _escape_markdown_actions(reply)