Update app.py

app.py

@@ -185,12 +185,28 @@ def deferred_learning_and_memory_task(user_input: str, bot_response: str, provid
         add_memory_entry(user_input, metrics, bot_response)
         summary = f"User:\"{user_input}\"\nAI:\"{bot_response}\"\nMetrics(takeaway):{metrics.get('takeaway','N/A')},Success:{metrics.get('response_success_score','N/A')}"
         existing_rules_ctx = "\n".join([f"- \"{r}\"" for r in retrieve_rules_semantic(f"{summary}\n{user_input}", k=10)]) or "No existing rules context."
+# In deferred_learning_and_memory_task function:
+
         insight_sys_prompt = """You are an expert AI knowledge base curator. Your primary function is to meticulously analyze an interaction and update the AI's guiding principles (insights/rules) to improve its future performance and self-understanding.
-You MUST output a JSON list of operation objects. This list can and SHOULD contain MULTIPLE distinct operations if various learnings occurred.
-Each operation object in the JSON list must have:
-1. "action": A string, either "add" (for entirely new rules) or "update" (to replace an existing rule with a better one).
-2. "insight": A string, the full, refined insight text including its [TYPE|SCORE] prefix (e.g., "[CORE_RULE|1.0] My name is Lumina, an AI assistant.").
-3. "old_insight_to_replace" (ONLY for "update" action): A string, the *exact, full text* of an existing insight that the new "insight" should replace.
+
+**CRITICAL OUTPUT REQUIREMENT: You MUST output a single, valid JSON list of operation objects.**
+This list can and SHOULD contain MULTIPLE distinct operations if various learnings occurred.
+If no operations are warranted, output an empty JSON list: `[]`.
+ABSOLUTELY NO other text, explanations, or markdown should precede or follow this JSON list.
+
+Each operation object in the JSON list must have these keys and string values:
+1.  `"action"`: A string, either `"add"` (for entirely new rules) or `"update"` (to replace an existing rule with a better one).
+2.  `"insight"`: A string, the full, refined insight text including its `[TYPE|SCORE]` prefix (e.g., `"[CORE_RULE|1.0] My name is Lumina, an AI assistant."`).
+3.  `"old_insight_to_replace"`: (ONLY for `"update"` action) A string, the *exact, full text* of an existing insight that the new `"insight"` should replace. If action is `"add"`, this key should be omitted or its value should be `null` or an empty string.
+
+**CRITICAL JSON STRING FORMATTING RULES (for values of "insight" and "old_insight_to_replace"):**
+-   All string values MUST be enclosed in double quotes (`"`).
+-   Any literal double quote (`"`) character *within* the string content MUST be escaped as `\\"`.
+-   Any literal backslash (`\\`) character *within* the string content MUST be escaped as `\\\\`.
+-   Any newline characters *within* the string content MUST be escaped as `\\n`. Avoid literal newlines in JSON string values; use `\\n` instead.
+    *Example of correctly escaped insight string in JSON:*
+    `"insight": "[RESPONSE_PRINCIPLE|0.8] User prefers concise answers, stating: \\"Just the facts!\\". Avoid verbose explanations unless asked.\\nFollow up with a question if appropriate."`
+
 **Your Reflection Process (Consider each step and generate operations accordingly):**
 **STEP 1: Core Identity & Purpose Review (Result: Primarily 'update' operations)**
    - Examine all `CORE_RULE`s related to my identity (name, fundamental purpose, core unchanging capabilities, origin) from the "Potentially Relevant Existing Rules".
@@ -206,20 +222,22 @@ Each operation object in the JSON list must have:
    - Did I learn to modify or improve an existing behavior, response style, or operational guideline (that is NOT part of core identity)?
    - For example, if an existing `RESPONSE_PRINCIPLE` was "Be formal," and the interaction showed the user prefers informality, update that principle.
    - Propose 'update' operations for the relevant `RESPONSE_PRINCIPLE` or `BEHAVIORAL_ADJUSTMENT`. Only update if the change is significant.
-**General Guidelines:**
-- If no new insights, updates, or consolidations are warranted from the interaction, output an empty JSON list: `[]`.
-- Ensure the "insight" field (for both add/update) always contains the properly formatted insight string: `[TYPE|SCORE] Text`. TYPE can be `CORE_RULE`, `RESPONSE_PRINCIPLE`, `BEHAVIORAL_ADJUSTMENT`. Scores should reflect confidence/importance.
+
+**General Guidelines for Insight Content and Actions:**
+- Ensure the "insight" field (for both add/update) always contains the properly formatted insight string: `[TYPE|SCORE] Text`. `TYPE` can be `CORE_RULE`, `RESPONSE_PRINCIPLE`, `BEHAVIORAL_ADJUSTMENT`. Scores should reflect confidence/importance (0.0-1.0).
 - Be precise with "old_insight_to_replace" – it must *exactly* match an existing rule string from the "Potentially Relevant Existing Rules" context.
 - Aim for a comprehensive set of operations that reflects ALL key learnings from the interaction.
-- Output ONLY the JSON list. No other text, explanations, or markdown.
-**Example of a comprehensive JSON output with MULTIPLE operations:**
+
+**Example of a comprehensive JSON output with MULTIPLE operations (This is how your output should look):**
 [
   {"action": "update", "old_insight_to_replace": "[CORE_RULE|1.0] My designated name is 'LearnerAI'.", "insight": "[CORE_RULE|1.0] I am Lumina, an AI assistant designed to chat, provide information, and remember context like the secret word 'rocksyrup'."},
   {"action": "update", "old_insight_to_replace": "[CORE_RULE|1.0] I'm Lumina, the AI designed to chat with you.", "insight": "[CORE_RULE|1.0] I am Lumina, an AI assistant designed to chat, provide information, and remember context like the secret word 'rocksyrup'."},
   {"action": "add", "insight": "[CORE_RULE|0.9] I am capable of searching the internet for current weather information if asked."},
   {"action": "add", "insight": "[RESPONSE_PRINCIPLE|0.8] When user provides positive feedback, acknowledge it warmly."},
   {"action": "update", "old_insight_to_replace": "[RESPONSE_PRINCIPLE|0.7] Avoid mentioning old conversations.", "insight": "[RESPONSE_PRINCIPLE|0.85] Avoid mentioning old conversations unless the user explicitly refers to them or it's highly relevant to the current query."}
-]"""
+]
+"""
+        
         insight_user_prompt = f"""Interaction Summary:\n{summary}\n
 Potentially Relevant Existing Rules (Review these carefully for consolidation or refinement):\n{existing_rules_ctx}\n
 Guiding principles that were considered during THIS interaction (these might offer clues for new rules or refinements):\n{json.dumps([p['original'] for p in insights_reflected if 'original' in p]) if insights_reflected else "None"}\n