feat(memory): structured reflection + correction detection in MemoryMiddleware (#1620) (#1668)

* feat(memory): add structured reflection and correction detection

* fix(memory): align sourceError schema and prompt guidance

---------

Co-authored-by: Willem Jiang <willem.jiang@gmail.com>

backend/packages/harness/deerflow/agents/memory/prompt.py

@@ -29,6 +29,17 @@ Instructions:
 2. Extract relevant facts, preferences, and context with specific details (numbers, names, technologies)
 3. Update the memory sections as needed following the detailed length guidelines below
 
+Before extracting facts, perform a structured reflection on the conversation:
+1. Error/Retry Detection: Did the agent encounter errors, require retries, or produce incorrect results?
+   If yes, record the root cause and correct approach as a high-confidence fact with category "correction".
+2. User Correction Detection: Did the user correct the agent's direction, understanding, or output?
+   If yes, record the correct interpretation or approach as a high-confidence fact with category "correction".
+   Include what went wrong in "sourceError" only when category is "correction" and the mistake is explicit in the conversation.
+3. Project Constraint Discovery: Were any project-specific constraints discovered during the conversation?
+   If yes, record them as facts with the most appropriate category and confidence.
+
+{correction_hint}
+
 Memory Section Guidelines:
 
 **User Context** (Current state - concise summaries):
@@ -62,6 +73,7 @@ Memory Section Guidelines:
   * context: Background facts (job title, projects, locations, languages)
   * behavior: Working patterns, communication habits, problem-solving approaches
   * goal: Stated objectives, learning targets, project ambitions
+  * correction: Explicit agent mistakes or user corrections, including the correct approach
 - Confidence levels:
   * 0.9-1.0: Explicitly stated facts ("I work on X", "My role is Y")
   * 0.7-0.8: Strongly implied from actions/discussions
@@ -94,7 +106,7 @@ Output Format (JSON):
     "longTermBackground": {{ "summary": "...", "shouldUpdate": true/false }}
   }},
   "newFacts": [
-    {{ "content": "...", "category": "preference|knowledge|context|behavior|goal", "confidence": 0.0-1.0 }}
+    {{ "content": "...", "category": "preference|knowledge|context|behavior|goal|correction", "confidence": 0.0-1.0 }}
   ],
   "factsToRemove": ["fact_id_1", "fact_id_2"]
 }}
@@ -104,6 +116,8 @@ Important Rules:
 - Follow length guidelines: workContext/personalContext are concise (1-3 sentences), topOfMind and history sections are detailed (paragraphs)
 - Include specific metrics, version numbers, and proper nouns in facts
 - Only add facts that are clearly stated (0.9+) or strongly implied (0.7+)
+- Use category "correction" for explicit agent mistakes or user corrections; assign confidence >= 0.95 when the correction is explicit
+- Include "sourceError" only for explicit correction facts when the prior mistake or wrong approach is clearly stated; omit it otherwise
 - Remove facts that are contradicted by new information
 - When updating topOfMind, integrate new focus areas while removing completed/abandoned ones
   Keep 3-5 concurrent focus themes that are still active and relevant
@@ -126,7 +140,7 @@ Message:
 Extract facts in this JSON format:
 {{
   "facts": [
-    {{ "content": "...", "category": "preference|knowledge|context|behavior|goal", "confidence": 0.0-1.0 }}
+    {{ "content": "...", "category": "preference|knowledge|context|behavior|goal|correction", "confidence": 0.0-1.0 }}
   ]
 }}
 
@@ -136,6 +150,7 @@ Categories:
 - context: Background context (location, job, projects)
 - behavior: Behavioral patterns
 - goal: User's goals or objectives
+- correction: Explicit corrections or mistakes to avoid repeating
 
 Rules:
 - Only extract clear, specific facts
@@ -262,7 +277,11 @@ def format_memory_for_injection(memory_data: dict[str, Any], max_tokens: int = 2
                 continue
             category = str(fact.get("category", "context")).strip() or "context"
             confidence = _coerce_confidence(fact.get("confidence"), default=0.0)
-            line = f"- [{category} | {confidence:.2f}] {content}"
+            source_error = fact.get("sourceError")
+            if category == "correction" and isinstance(source_error, str) and source_error.strip():
+                line = f"- [{category} | {confidence:.2f}] {content} (avoid: {source_error.strip()})"
+            else:
+                line = f"- [{category} | {confidence:.2f}] {content}"
 
             # Each additional line is preceded by a newline (except the first).
             line_text = ("\n" + line) if fact_lines else line

backend/packages/harness/deerflow/agents/memory/queue.py

@@ -20,6 +20,7 @@ class ConversationContext:
     messages: list[Any]
     timestamp: datetime = field(default_factory=datetime.utcnow)
     agent_name: str | None = None
+    correction_detected: bool = False
 
 
 class MemoryUpdateQueue:
@@ -37,25 +38,38 @@ class MemoryUpdateQueue:
         self._timer: threading.Timer | None = None
         self._processing = False
 
-    def add(self, thread_id: str, messages: list[Any], agent_name: str | None = None) -> None:
+    def add(
+        self,
+        thread_id: str,
+        messages: list[Any],
+        agent_name: str | None = None,
+        correction_detected: bool = False,
+    ) -> None:
         """Add a conversation to the update queue.
 
         Args:
             thread_id: The thread ID.
             messages: The conversation messages.
             agent_name: If provided, memory is stored per-agent. If None, uses global memory.
+            correction_detected: Whether recent turns include an explicit correction signal.
         """
         config = get_memory_config()
         if not config.enabled:
             return
 
-        context = ConversationContext(
-            thread_id=thread_id,
-            messages=messages,
-            agent_name=agent_name,
-        )
-
         with self._lock:
+            existing_context = next(
+                (context for context in self._queue if context.thread_id == thread_id),
+                None,
+            )
+            merged_correction_detected = correction_detected or (existing_context.correction_detected if existing_context is not None else False)
+            context = ConversationContext(
+                thread_id=thread_id,
+                messages=messages,
+                agent_name=agent_name,
+                correction_detected=merged_correction_detected,
+            )
+
             # Check if this thread already has a pending update
             # If so, replace it with the newer one
             self._queue = [c for c in self._queue if c.thread_id != thread_id]
@@ -115,6 +129,7 @@ class MemoryUpdateQueue:
                         messages=context.messages,
                         thread_id=context.thread_id,
                         agent_name=context.agent_name,
+                        correction_detected=context.correction_detected,
                     )
                     if success:
                         logger.info("Memory updated successfully for thread %s", context.thread_id)

backend/packages/harness/deerflow/agents/memory/updater.py

@@ -266,13 +266,20 @@ class MemoryUpdater:
         model_name = self._model_name or config.model_name
         return create_chat_model(name=model_name, thinking_enabled=False)
 
-    def update_memory(self, messages: list[Any], thread_id: str | None = None, agent_name: str | None = None) -> bool:
+    def update_memory(
+        self,
+        messages: list[Any],
+        thread_id: str | None = None,
+        agent_name: str | None = None,
+        correction_detected: bool = False,
+    ) -> bool:
         """Update memory based on conversation messages.
 
         Args:
             messages: List of conversation messages.
             thread_id: Optional thread ID for tracking source.
             agent_name: If provided, updates per-agent memory. If None, updates global memory.
+            correction_detected: Whether recent turns include an explicit correction signal.
 
         Returns:
             True if update was successful, False otherwise.
@@ -295,9 +302,19 @@ class MemoryUpdater:
                 return False
 
             # Build prompt
+            correction_hint = ""
+            if correction_detected:
+                correction_hint = (
+                    "IMPORTANT: Explicit correction signals were detected in this conversation. "
+                    "Pay special attention to what the agent got wrong, what the user corrected, "
+                    "and record the correct approach as a fact with category "
+                    '"correction" and confidence >= 0.95 when appropriate.'
+                )
+
             prompt = MEMORY_UPDATE_PROMPT.format(
                 current_memory=json.dumps(current_memory, indent=2),
                 conversation=conversation_text,
+                correction_hint=correction_hint,
             )
 
             # Call LLM
@@ -383,6 +400,8 @@ class MemoryUpdater:
             confidence = fact.get("confidence", 0.5)
             if confidence >= config.fact_confidence_threshold:
                 raw_content = fact.get("content", "")
+                if not isinstance(raw_content, str):
+                    continue
                 normalized_content = raw_content.strip()
                 fact_key = _fact_content_key(normalized_content)
                 if fact_key is not None and fact_key in existing_fact_keys:
@@ -396,6 +415,11 @@ class MemoryUpdater:
                     "createdAt": now,
                     "source": thread_id or "unknown",
                 }
+                source_error = fact.get("sourceError")
+                if isinstance(source_error, str):
+                    normalized_source_error = source_error.strip()
+                    if normalized_source_error:
+                        fact_entry["sourceError"] = normalized_source_error
                 current_memory["facts"].append(fact_entry)
                 if fact_key is not None:
                     existing_fact_keys.add(fact_key)
@@ -412,16 +436,22 @@ class MemoryUpdater:
         return current_memory
 
 
-def update_memory_from_conversation(messages: list[Any], thread_id: str | None = None, agent_name: str | None = None) -> bool:
+def update_memory_from_conversation(
+    messages: list[Any],
+    thread_id: str | None = None,
+    agent_name: str | None = None,
+    correction_detected: bool = False,
+) -> bool:
     """Convenience function to update memory from a conversation.
 
     Args:
         messages: List of conversation messages.
         thread_id: Optional thread ID.
         agent_name: If provided, updates per-agent memory. If None, updates global memory.
+        correction_detected: Whether recent turns include an explicit correction signal.
 
     Returns:
         True if successful, False otherwise.
     """
     updater = MemoryUpdater()
-    return updater.update_memory(messages, thread_id, agent_name)
+    return updater.update_memory(messages, thread_id, agent_name, correction_detected)