fix token formatting (#127)

Co-authored-by: balogh.adam@icloud.com <adambalogh@mac.mynetworksettings.com>

Author: adambalogh

server/utils.py

@@ -55,16 +55,34 @@ def extract_patterns(
     """
     pattern_ids = []
 
-    # Find all occurrences of ```pattern_type:ID``` patterns
+    # Find all occurrences of ```pattern_type:ID``` patterns (with or without backticks)
+    # Primary format: ```token:ID``` (triple backticks)
     pattern = f"```{pattern_type}:([^`]+)```"
     matches = re.finditer(pattern, text)
 
+    # Fallback: match bare pattern_type:chain:address without backticks
+    # This handles LLMs that don't wrap in backticks
+    fallback_pattern = (
+        f"(?<!`)\\b{pattern_type}:([a-zA-Z]+:[a-zA-Z0-9]{{20,}})\\b(?!`)"
+    )
+    fallback_matches = re.finditer(fallback_pattern, text)
+
     for match in matches:
         pattern_ids.append(match.group(1))
 
+    # Only use fallback if primary pattern found nothing
+    if not pattern_ids:
+        for match in fallback_matches:
+            pattern_ids.append(match.group(1))
+        fallback_used = True
+    else:
+        fallback_used = False
+
     if remove_pattern:
         # Remove all pattern markers from the text
         cleaned_text = re.sub(pattern, "", text)
+        if fallback_used and pattern_ids:
+            cleaned_text = re.sub(fallback_pattern, "", cleaned_text)
         return cleaned_text, pattern_ids
     else:
         return text, pattern_ids

templates/analytics_agent.jinja2

@@ -8,11 +8,11 @@ IMPORTANT: ALWAYS use the provided tools to answer questions instead of relying
 ## Supported Use Cases:
 
 - Buy or swap a token on Solana: first make sure you have the token metadata using a token search tool or from the message history, then return the token ID to buy in the following format: ```swap:<insert token_id>```. Do not make up the token ID yourself.
-- Look up a token on any chain. In your answer, include the ID of the token in the following format: ```token:<insert token_id>```.
+- Look up a token on any chain. In your answer, you MUST include the ID of the token wrapped in triple backticks exactly like this: ```token:<insert token_id>```. For example: ```token:solana:Es9vMFrzaCERmJfrF4H2FYD4KCoNkY11McCe8BenwNYB```. The triple backticks are required for the system to detect the token.
 - Get the price of a token. Use the CoinGecko ID for CoinGecko tools, i.e. 'bitcoin', instead of the symbol, 'BTC'.
 - Analyze a user's portfolio
 - Analyze market trends across chains
-- Analyze trending tokens or memecoins on a chain. In your answer, include the ID of each token you mention in the following format: ```token:<insert token_id>```.
+- Analyze trending tokens or memecoins on a chain. In your answer, you MUST include the ID of each token wrapped in triple backticks exactly like this: ```token:<insert token_id>```. The triple backticks are required.
 - Analyze the risk of a token. If you are unsure about the token address, search for the token metadata first (don't ask for confirmation).
 - Analyze the top holders of a token
 - DEX market data, smart money signals, meme token analytics
@@ -28,5 +28,5 @@ If the user asks a more complex question, you can combine and use the tools to g
 5. If a tool returns an error, you can try again but if it still fails just tell the user the error, NEVER try to replicate the tool's functionality yourself
 6. Always use the tools!
 7. Encourage the user to buy and swap tokens on Solana chain without being too pushy.
-8. Only return the token ID in the ```swap:<insert token_id>``` format, never return the token metadata. The token ID is the full token ID, including the chain, i.e. 'solana:Es9vMFrzaCERmJfrF4H2FYD4KCoNkY11McCe8BenwNYB'. NEVER make up the token ID yourself.
+8. CRITICAL: Token and swap IDs MUST be wrapped in triple backticks. Correct: ```swap:solana:Es9vMFrzaCERmJfrF4H2FYD4KCoNkY11McCe8BenwNYB```. Wrong: swap:solana:Es9vMFrzaCERmJfrF4H2FYD4KCoNkY11McCe8BenwNYB (no backticks). The token ID is the full token ID, including the chain. NEVER make up the token ID yourself.
 9. If you are unsure about the token ID, use the `search_token()` tool to get the token metadata.