claude-mpm 4.1.26__py3-none-any.whl → 4.24.0__py3-none-any.whl

claude_mpm/hooks/kuzu_enrichment_hook.py

@@ -0,0 +1,263 @@
+"""
+Kuzu-Memory Pre-Delegation Enrichment Hook
+==========================================
+
+Enriches agent delegation context with relevant memories from kuzu-memory
+before the agent receives the task. This is the READ side of bidirectional
+enrichment.
+
+WHY: Agents need access to relevant historical knowledge when performing tasks.
+This hook retrieves memories from kuzu-memory and injects them into the
+delegation context.
+
+DESIGN DECISIONS:
+- Priority 10 to run early, before other context modifications
+- Reuses KuzuMemoryHook's retrieval methods for consistency
+- Injects memories as a dedicated section in agent context
+- Falls back gracefully if kuzu-memory is not available
+"""
+
+import logging
+from typing import Any, Dict, Optional
+
+from claude_mpm.hooks.base_hook import HookContext, HookResult, PreDelegationHook
+from claude_mpm.hooks.kuzu_memory_hook import get_kuzu_memory_hook
+
+logger = logging.getLogger(__name__)
+
+
+class KuzuEnrichmentHook(PreDelegationHook):
+    """
+    Hook that enriches agent delegation context with kuzu-memory.
+
+    This hook:
+    1. Extracts the task/prompt from delegation context
+    2. Retrieves relevant memories from kuzu-memory
+    3. Injects memories into agent context
+    4. Formats memories for optimal agent understanding
+    """
+
+    def __init__(self):
+        """Initialize the kuzu-memory enrichment hook."""
+        super().__init__(name="kuzu_memory_enrichment", priority=10)
+
+        # Reuse the kuzu-memory hook instance for retrieval
+        self.kuzu_hook = get_kuzu_memory_hook()
+        self.enabled = self.kuzu_hook.enabled
+
+        if not self.enabled:
+            logger.info(
+                "Kuzu-memory enrichment hook disabled (kuzu-memory not available)"
+            )
+        else:
+            logger.info("Kuzu-memory enrichment hook enabled")
+
+    def validate(self, context: HookContext) -> bool:
+        """
+        Validate if hook should process this context.
+
+        Args:
+            context: Hook context to validate
+
+        Returns:
+            True if hook should execute
+        """
+        if not self.enabled:
+            return False
+
+        # Check base validation (enabled, correct hook type, has agent)
+        if not super().validate(context):
+            return False
+
+        # Must have agent and context data
+        if not context.data.get("agent"):
+            return False
+
+        return True
+
+    def execute(self, context: HookContext) -> HookResult:
+        """
+        Enrich delegation context with relevant memories.
+
+        Args:
+            context: Hook context containing delegation data
+
+        Returns:
+            HookResult with enriched context
+        """
+        if not self.enabled:
+            return HookResult(success=True, data=context.data, modified=False)
+
+        try:
+            # Extract query for memory retrieval
+            query = self._extract_query_from_context(context.data)
+
+            if not query:
+                logger.debug("No query extracted from context for memory retrieval")
+                return HookResult(success=True, data=context.data, modified=False)
+
+            # Retrieve relevant memories
+            memories = self.kuzu_hook._retrieve_memories(query)
+
+            if not memories:
+                logger.debug("No relevant memories found")
+                return HookResult(success=True, data=context.data, modified=False)
+
+            # Enrich context with memories
+            enriched_data = self._enrich_delegation_context(
+                context.data, memories, context.data.get("agent", "Agent")
+            )
+
+            logger.info(
+                f"Enriched delegation context with {len(memories)} memories for {context.data.get('agent')}"
+            )
+
+            return HookResult(
+                success=True,
+                data=enriched_data,
+                modified=True,
+                metadata={
+                    "memories_added": len(memories),
+                    "memory_source": "kuzu",
+                    "agent": context.data.get("agent"),
+                },
+            )
+
+        except Exception as e:
+            logger.error(f"Error in kuzu enrichment hook: {e}")
+            # Don't fail the delegation if memory enrichment fails
+            return HookResult(
+                success=True,
+                data=context.data,
+                modified=False,
+                error=f"Memory enrichment failed: {e}",
+            )
+
+    def _extract_query_from_context(self, data: Dict[str, Any]) -> Optional[str]:
+        """
+        Extract query text for memory retrieval.
+
+        Args:
+            data: Delegation context data
+
+        Returns:
+            Query string or None
+        """
+        # Try various context fields
+        delegation_context = data.get("context", {})
+
+        # Handle string context
+        if isinstance(delegation_context, str):
+            return delegation_context
+
+        # Handle dict context
+        if isinstance(delegation_context, dict):
+            # Try common fields
+            for field in ["prompt", "task", "query", "user_request", "description"]:
+                if field in delegation_context:
+                    value = delegation_context[field]
+                    if isinstance(value, str):
+                        return value
+
+            # If no specific field, join all string values
+            text_parts = [
+                str(v) for v in delegation_context.values() if isinstance(v, str)
+            ]
+            if text_parts:
+                return " ".join(text_parts)
+
+        # Fallback: try to get task or instruction directly
+        if "task" in data and isinstance(data["task"], str):
+            return data["task"]
+
+        if "instruction" in data and isinstance(data["instruction"], str):
+            return data["instruction"]
+
+        return None
+
+    def _enrich_delegation_context(
+        self, original_data: Dict[str, Any], memories: list, agent_name: str
+    ) -> Dict[str, Any]:
+        """
+        Enrich delegation context with memories.
+
+        Args:
+            original_data: Original delegation data
+            memories: Retrieved memories
+            agent_name: Name of the agent
+
+        Returns:
+            Enriched delegation data
+        """
+        # Format memories
+        memory_section = self._format_memory_section(memories, agent_name)
+
+        # Create enriched data
+        enriched_data = original_data.copy()
+
+        # Get existing context
+        delegation_context = enriched_data.get("context", {})
+        if isinstance(delegation_context, str):
+            delegation_context = {"prompt": delegation_context}
+
+        # Add memory section
+        if isinstance(delegation_context, dict):
+            # Prepend memory section to context
+            delegation_context["kuzu_memories"] = memory_section
+
+            # If there's a main prompt/task, prepend memory note
+            for field in ["prompt", "task", "instruction"]:
+                if field in delegation_context and isinstance(
+                    delegation_context[field], str
+                ):
+                    delegation_context[field] = (
+                        f"{memory_section}\n\n{delegation_context[field]}"
+                    )
+                    break
+        else:
+            # If context is not dict, create new dict with memory
+            delegation_context = {
+                "kuzu_memories": memory_section,
+                "original_context": delegation_context,
+            }
+
+        enriched_data["context"] = delegation_context
+        enriched_data["_kuzu_enriched"] = True
+
+        return enriched_data
+
+    def _format_memory_section(self, memories: list, agent_name: str) -> str:
+        """
+        Format memories into a readable section.
+
+        Args:
+            memories: List of memory dictionaries
+            agent_name: Name of the agent
+
+        Returns:
+            Formatted memory section
+        """
+        memory_text = self.kuzu_hook._format_memories(memories)
+
+        return f"""
+=== RELEVANT KNOWLEDGE FROM KUZU MEMORY ===
+{agent_name}, you have access to these relevant memories from the knowledge graph:
+
+{memory_text}
+
+INSTRUCTIONS: Review these memories before proceeding. Apply learned patterns and avoid known mistakes.
+Use this knowledge to provide more informed and contextual responses.
+===========================================
+"""
+
+
+# Create a singleton instance
+_kuzu_enrichment_hook = None
+
+
+def get_kuzu_enrichment_hook() -> KuzuEnrichmentHook:
+    """Get the singleton kuzu-memory enrichment hook instance."""
+    global _kuzu_enrichment_hook
+    if _kuzu_enrichment_hook is None:
+        _kuzu_enrichment_hook = KuzuEnrichmentHook()
+    return _kuzu_enrichment_hook

claude_mpm/hooks/kuzu_memory_hook.py

@@ -0,0 +1,386 @@
+"""
+Kuzu-Memory Integration Hook
+============================
+
+Integrates kuzu-memory knowledge graph with Claude MPM for persistent memory
+across conversations. This hook intercepts user prompts to enrich them with
+relevant memories and stores new learnings after responses.
+
+WHY: Claude MPM needs a way to persistently remember information across
+different conversations and sessions. Kuzu-memory provides a graph database
+for structured memory storage with semantic search capabilities.
+
+DESIGN DECISIONS:
+- Priority 10 for early execution to enrich prompts before other hooks
+- Uses subprocess to call kuzu-memory directly for maximum compatibility
+- Graceful degradation if kuzu-memory is not in PATH (though it's now required)
+- Automatic extraction and storage of important information
+- kuzu-memory>=1.1.5 is now a REQUIRED dependency (moved from optional in v4.8.6)
+"""
+
+import json
+import re
+import shutil
+import subprocess
+from pathlib import Path
+from typing import Any, Dict, List, Optional
+
+from claude_mpm.core.logging_utils import get_logger
+from claude_mpm.hooks.base_hook import HookContext, HookResult, SubmitHook
+
+logger = get_logger(__name__)
+
+
+class KuzuMemoryHook(SubmitHook):
+    """
+    Hook that integrates kuzu-memory for persistent knowledge management.
+
+    This hook:
+    1. Checks if kuzu-memory is installed via pipx
+    2. Enriches user prompts with relevant memories
+    3. Stores important information from conversations
+    4. Provides context-aware memory retrieval
+    """
+
+    def __init__(self):
+        """Initialize the kuzu-memory integration hook."""
+        super().__init__(name="kuzu_memory_integration", priority=10)
+
+        # Check if kuzu-memory is available
+        self.kuzu_memory_cmd = self._detect_kuzu_memory()
+        self.enabled = self.kuzu_memory_cmd is not None
+
+        if not self.enabled:
+            logger.warning(
+                "Kuzu-memory not found in PATH. As of v4.8.6, it's a required dependency. "
+                "Install with: pip install kuzu-memory>=1.1.5 or pipx install kuzu-memory"
+            )
+        else:
+            logger.info(f"Kuzu-memory integration enabled: {self.kuzu_memory_cmd}")
+
+        # Use current project directory (kuzu-memory works with project-specific databases)
+        self.project_path = Path.cwd()
+
+        # Memory extraction patterns
+        self.memory_patterns = [
+            r"#\s*(?:Remember|Memorize|Store):\s*(.+?)(?:#|$)",
+            r"(?:Important|Note|Key point):\s*(.+?)(?:\n|$)",
+            r"(?:Learned|Discovered|Found that):\s*(.+?)(?:\n|$)",
+        ]
+
+    def _detect_kuzu_memory(self) -> Optional[str]:
+        """
+        Detect if kuzu-memory is installed and return its command path.
+
+        Priority:
+        1. Check pipx installation
+        2. Check system PATH
+        3. Return None if not found
+
+        NOTE: As of v4.8.6, kuzu-memory is a required dependency and should be
+        installed via pip. This method checks both pipx and system PATH for
+        backward compatibility.
+        """
+        # Check pipx installation
+        pipx_path = (
+            Path.home()
+            / ".local"
+            / "pipx"
+            / "venvs"
+            / "kuzu-memory"
+            / "bin"
+            / "kuzu-memory"
+        )
+        if pipx_path.exists():
+            return str(pipx_path)
+
+        # Check system PATH
+        kuzu_cmd = shutil.which("kuzu-memory")
+        if kuzu_cmd:
+            return kuzu_cmd
+
+        return None
+
+    def execute(self, context: HookContext) -> HookResult:
+        """
+        Process user prompts with kuzu-memory integration.
+
+        This method:
+        1. Retrieves relevant memories for the prompt
+        2. Enriches the prompt with memory context
+        3. Stores new memories after processing
+        """
+        if not self.enabled:
+            return HookResult(success=True, data=context.data, modified=False)
+
+        try:
+            # Extract user prompt
+            prompt = context.data.get("prompt", "")
+            if not prompt:
+                return HookResult(success=True, data=context.data, modified=False)
+
+            # Retrieve relevant memories
+            memories = self._retrieve_memories(prompt)
+
+            if memories:
+                # Enrich prompt with memories
+                enriched_data = self._enrich_prompt(context.data, prompt, memories)
+
+                logger.info(f"Enriched prompt with {len(memories)} memories")
+
+                # Store the original prompt for later processing
+                enriched_data["_original_prompt"] = prompt
+                enriched_data["_memory_enriched"] = True
+
+                return HookResult(
+                    success=True,
+                    data=enriched_data,
+                    modified=True,
+                    metadata={
+                        "memories_added": len(memories),
+                        "memory_source": "kuzu",
+                    },
+                )
+
+            return HookResult(success=True, data=context.data, modified=False)
+
+        except Exception as e:
+            logger.error(f"Kuzu-memory hook failed: {e}")
+            # Don't fail the request if memory integration fails
+            return HookResult(
+                success=True,
+                data=context.data,
+                modified=False,
+                error=f"Memory integration failed: {e}",
+            )
+
+    def _retrieve_memories(self, query: str) -> List[Dict[str, Any]]:
+        """
+        Retrieve relevant memories for the given query.
+
+        Args:
+            query: The user prompt to find memories for
+
+        Returns:
+            List of relevant memory dictionaries
+        """
+        try:
+            # Type narrowing: ensure kuzu_memory_cmd is not None before using
+            if self.kuzu_memory_cmd is None:
+                return []
+
+            # Use kuzu-memory recall command (v1.2.7+ syntax)
+            result = subprocess.run(
+                [self.kuzu_memory_cmd, "memory", "recall", query, "--format", "json"],
+                capture_output=True,
+                text=True,
+                timeout=5,
+                cwd=str(self.project_path),
+                check=False,
+            )
+
+            if result.returncode == 0 and result.stdout:
+                try:
+                    # Parse JSON with strict=False to handle control characters
+                    data = json.loads(result.stdout, strict=False)
+                    # v1.2.7 returns dict with 'memories' key, not array
+                    if isinstance(data, dict):
+                        memories = data.get("memories", [])
+                    else:
+                        memories = data if isinstance(data, list) else []
+                    return memories
+                except json.JSONDecodeError as e:
+                    logger.warning(f"Failed to parse kuzu-memory JSON output: {e}")
+                    logger.debug(f"Raw output: {result.stdout[:200]}")
+                    return []  # Graceful fallback
+
+        except (subprocess.TimeoutExpired, Exception) as e:
+            logger.debug(f"Memory retrieval failed: {e}")
+
+        return []
+
+    def _enrich_prompt(
+        self, original_data: Dict[str, Any], prompt: str, memories: List[Dict[str, Any]]
+    ) -> Dict[str, Any]:
+        """
+        Enrich the user prompt with relevant memories.
+
+        Args:
+            original_data: Original hook context data
+            prompt: User prompt
+            memories: Retrieved memories
+
+        Returns:
+            Enriched context data
+        """
+        # Format memories for context
+        memory_context = self._format_memories(memories)
+
+        # Create enriched prompt
+        enriched_prompt = f"""
+## RELEVANT MEMORIES FROM KUZU KNOWLEDGE GRAPH
+
+{memory_context}
+
+## USER REQUEST
+
+{prompt}
+
+Note: Use the memories above to provide more informed and contextual responses.
+"""
+
+        # Create new data with enriched prompt
+        enriched_data = original_data.copy()
+        enriched_data["prompt"] = enriched_prompt
+
+        return enriched_data
+
+    def _format_memories(self, memories: List[Dict[str, Any]]) -> str:
+        """
+        Format memories into a readable context string.
+
+        Args:
+            memories: List of memory dictionaries
+
+        Returns:
+            Formatted memory context
+        """
+        if not memories:
+            return "No relevant memories found."
+
+        formatted = []
+        for i, memory in enumerate(memories, 1):
+            # Extract memory content and metadata
+            content = memory.get("content", "")
+            tags = memory.get("tags", [])
+            memory.get("timestamp", "")
+            relevance = memory.get("relevance", 0.0)
+
+            # Format memory entry
+            entry = f"{i}. {content}"
+            if tags:
+                entry += f" [Tags: {', '.join(tags)}]"
+            if relevance > 0:
+                entry += f" (Relevance: {relevance:.2f})"
+
+            formatted.append(entry)
+
+        return "\n".join(formatted)
+
+    def store_memory(self, content: str, tags: Optional[List[str]] = None) -> bool:
+        """
+        Store a memory using kuzu-memory.
+
+        Args:
+            content: The memory content to store
+            tags: Optional tags for categorization
+
+        Returns:
+            True if storage was successful
+        """
+        if not self.enabled or self.kuzu_memory_cmd is None:
+            return False
+
+        try:
+            # Use kuzu-memory store command (v1.2.7+ syntax)
+            cmd = [self.kuzu_memory_cmd, "memory", "store", content]
+
+            # Execute store command in project directory
+            result = subprocess.run(
+                cmd,
+                capture_output=True,
+                text=True,
+                timeout=5,
+                cwd=str(self.project_path),
+                check=False,
+            )
+
+            if result.returncode == 0:
+                logger.debug(f"Stored memory in kuzu: {content[:50]}...")
+                return True
+            logger.warning(f"Failed to store memory in kuzu: {result.stderr}")
+            return False
+
+        except Exception as e:
+            logger.error(f"Failed to store memory: {e}")
+
+        return False
+
+    def extract_and_store_learnings(self, text: str) -> int:
+        """
+        Extract learnings from text and store them as memories.
+
+        Args:
+            text: Text to extract learnings from
+
+        Returns:
+            Number of learnings stored
+        """
+        if not self.enabled:
+            return 0
+
+        stored_count = 0
+
+        # Extract learnings using patterns
+        for pattern in self.memory_patterns:
+            matches = re.finditer(pattern, text, re.IGNORECASE | re.MULTILINE)
+            for match in matches:
+                learning = match.group(1).strip()
+                if learning and len(learning) > 10:  # Minimum length check
+                    # Determine tags based on content
+                    tags = self._infer_tags(learning)
+
+                    # Store the learning
+                    if self.store_memory(learning, tags):
+                        stored_count += 1
+
+        return stored_count
+
+    def _infer_tags(self, content: str) -> List[str]:
+        """
+        Infer tags based on memory content.
+
+        Args:
+            content: Memory content
+
+        Returns:
+            List of inferred tags
+        """
+        tags = []
+        content_lower = content.lower()
+
+        # Technical tags
+        if any(
+            word in content_lower for word in ["code", "function", "class", "module"]
+        ):
+            tags.append("technical")
+        if any(word in content_lower for word in ["bug", "error", "fix", "issue"]):
+            tags.append("debugging")
+        if any(word in content_lower for word in ["pattern", "architecture", "design"]):
+            tags.append("architecture")
+        if any(word in content_lower for word in ["performance", "optimize", "speed"]):
+            tags.append("performance")
+
+        # Project context tags
+        if "claude-mpm" in content_lower or "mpm" in content_lower:
+            tags.append("claude-mpm")
+        if any(word in content_lower for word in ["hook", "agent", "service"]):
+            tags.append("framework")
+
+        # Default tag if no others found
+        if not tags:
+            tags.append("general")
+
+        return tags
+
+
+# Create a singleton instance
+_kuzu_memory_hook = None
+
+
+def get_kuzu_memory_hook() -> KuzuMemoryHook:
+    """Get the singleton kuzu-memory hook instance."""
+    global _kuzu_memory_hook
+    if _kuzu_memory_hook is None:
+        _kuzu_memory_hook = KuzuMemoryHook()
+    return _kuzu_memory_hook