claude-mpm 4.4.0__py3-none-any.whl → 4.4.4__py3-none-any.whl

claude_mpm/core/framework_loader.py

@@ -1,31 +1,27 @@
-"""Framework loader for Claude MPM."""
+"""Framework loader for Claude MPM - Refactored modular version."""
 
-import getpass
-import locale
-import logging
+import asyncio
 import os
-import platform
-import time
-from datetime import datetime, timezone
 from pathlib import Path
-from typing import Any, Dict, Optional
-
-# Import resource handling for packaged installations
-try:
-    # Python 3.9+
-    from importlib.resources import files
-except ImportError:
-    # Python 3.8 fallback
-    try:
-        from importlib_resources import files
-    except ImportError:
-        # Final fallback for development environments
-        files = None
-
-from ..utils.imports import safe_import
+from typing import Any, Dict, List, Optional, Set
+
+# Import framework components
+from claude_mpm.core.framework import (
+    AgentLoader,
+    CapabilityGenerator,
+    ContentFormatter,
+    ContextGenerator,
+    FileLoader,
+    InstructionLoader,
+    MemoryProcessor,
+    MetadataProcessor,
+    PackagedLoader,
+    TemplateProcessor,
+)
+from claude_mpm.core.logging_utils import get_logger
+from claude_mpm.utils.imports import safe_import
 
-# Import with fallback support - using absolute imports as primary since we're at module level
-get_logger = safe_import("claude_mpm.core.logger", "core.logger", ["get_logger"])
+# Import with fallback support
 AgentRegistryAdapter = safe_import(
     "claude_mpm.core.agent_registry", "core.agent_registry", ["AgentRegistryAdapter"]
 )
@@ -67,43 +63,13 @@ class FrameworkLoader:
     """
     Load and prepare framework instructions for injection.
 
-    This component handles:
-    1. Finding the framework (claude-multiagent-pm)
-    2. Loading custom instructions from .claude-mpm/ directories
-    3. Preparing agent definitions
-    4. Formatting for injection
-
-    Custom Instructions Loading:
-    The framework loader supports custom instructions through .claude-mpm/ directories.
-    It NEVER reads from .claude/ directories to avoid conflicts with Claude Code.
-
-    File Loading Precedence (highest to lowest):
-
-    INSTRUCTIONS.md:
-      1. Project: ./.claude-mpm/INSTRUCTIONS.md
-      2. User: ~/.claude-mpm/INSTRUCTIONS.md
-      3. System: (built-in framework instructions)
-
-    WORKFLOW.md:
-      1. Project: ./.claude-mpm/WORKFLOW.md
-      2. User: ~/.claude-mpm/WORKFLOW.md
-      3. System: src/claude_mpm/agents/WORKFLOW.md
-
-    MEMORY.md:
-      1. Project: ./.claude-mpm/MEMORY.md
-      2. User: ~/.claude-mpm/MEMORY.md
-      3. System: src/claude_mpm/agents/MEMORY.md
-
-    Actual Memories:
-      - User: ~/.claude-mpm/memories/PM_memories.md
-      - Project: ./.claude-mpm/memories/PM_memories.md (overrides user)
-      - Agent memories: *_memories.md files (only loaded if agent is deployed)
-
-    Important Notes:
-    - Project-level files always override user-level files
-    - User-level files always override system defaults
-    - The framework NEVER reads from .claude/ directories
-    - Custom instructions are clearly labeled with their source level
+    This refactored version uses modular components for better maintainability
+    and testability while maintaining backward compatibility.
+
+    Components:
+    - Loaders: Handle file I/O and resource loading
+    - Formatters: Generate and format content sections
+    - Processors: Process metadata, templates, and memories
     """
 
     def __init__(
@@ -114,7 +80,7 @@ class FrameworkLoader:
         config: Optional[Dict[str, Any]] = None,
     ):
         """
-        Initialize framework loader.
+        Initialize framework loader with modular components.
 
         Args:
             framework_path: Explicit path to framework (auto-detected if None)
@@ -129,6 +95,39 @@ class FrameworkLoader:
         self.config = config or {}
 
         # Validate API keys on startup (before any other initialization)
+        self._validate_api_keys()
+
+        # Initialize service container
+        self.container = service_container or get_global_container()
+        self._register_services()
+
+        # Resolve services from container
+        self._cache_manager = self.container.resolve(ICacheManager)
+        self._path_resolver = self.container.resolve(IPathResolver)
+        self._memory_manager = self.container.resolve(IMemoryManager)
+
+        # Initialize framework path
+        self.framework_path = (
+            framework_path or self._path_resolver.detect_framework_path()
+        )
+
+        # Initialize modular components
+        self._init_components()
+
+        # Keep cache TTL constants for backward compatibility
+        self._init_cache_ttl()
+
+        # Load framework content
+        self.framework_content = self._load_framework_content()
+
+        # Initialize agent registry
+        self.agent_registry = AgentRegistryAdapter(self.framework_path)
+
+        # Output style manager (deferred initialization)
+        self.output_style_manager = None
+
+    def _validate_api_keys(self) -> None:
+        """Validate API keys if enabled in config."""
         if self.config.get("validate_api_keys", True):
             try:
                 self.logger.info("Validating configured API keys...")
@@ -141,21 +140,17 @@ class FrameworkLoader:
                 self.logger.error(f"❌ Unexpected error during API validation: {e}")
                 raise
 
-        # Use provided container or get global container
-        self.container = service_container or get_global_container()
-
-        # Register services if not already registered
+    def _register_services(self) -> None:
+        """Register services in the container if not already registered."""
         if not self.container.is_registered(ICacheManager):
-            self.container.register(ICacheManager, CacheManager, True)  # singleton=True
+            self.container.register(ICacheManager, CacheManager, True)
 
         if not self.container.is_registered(IPathResolver):
-            # PathResolver depends on CacheManager, so resolve it first
             cache_manager = self.container.resolve(ICacheManager)
             path_resolver = PathResolver(cache_manager=cache_manager)
             self.container.register_instance(IPathResolver, path_resolver)
 
         if not self.container.is_registered(IMemoryManager):
-            # MemoryManager depends on both CacheManager and PathResolver
             cache_manager = self.container.resolve(ICacheManager)
             path_resolver = self.container.resolve(IPathResolver)
             memory_manager = MemoryManager(
@@ -163,18 +158,26 @@ class FrameworkLoader:
             )
             self.container.register_instance(IMemoryManager, memory_manager)
 
-        # Resolve services from container
-        self._cache_manager = self.container.resolve(ICacheManager)
-        self._path_resolver = self.container.resolve(IPathResolver)
-        self._memory_manager = self.container.resolve(IMemoryManager)
-
-        # Initialize framework path using PathResolver
-        self.framework_path = (
-            framework_path or self._path_resolver.detect_framework_path()
-        )
-
-        # Keep TTL constants for backward compatibility
-        # These are implementation-specific, so we use defaults if not available
+    def _init_components(self) -> None:
+        """Initialize modular components."""
+        # Loaders
+        self.file_loader = FileLoader()
+        self.packaged_loader = PackagedLoader()
+        self.instruction_loader = InstructionLoader(self.framework_path)
+        self.agent_loader = AgentLoader(self.framework_path)
+
+        # Formatters
+        self.content_formatter = ContentFormatter()
+        self.capability_generator = CapabilityGenerator()
+        self.context_generator = ContextGenerator()
+
+        # Processors
+        self.metadata_processor = MetadataProcessor()
+        self.template_processor = TemplateProcessor(self.framework_path)
+        self.memory_processor = MemoryProcessor()
+
+    def _init_cache_ttl(self) -> None:
+        """Initialize cache TTL constants for backward compatibility."""
         if hasattr(self._cache_manager, "capabilities_ttl"):
             self.CAPABILITIES_CACHE_TTL = self._cache_manager.capabilities_ttl
             self.DEPLOYED_AGENTS_CACHE_TTL = self._cache_manager.deployed_agents_ttl
@@ -187,430 +190,24 @@ class FrameworkLoader:
             self.METADATA_CACHE_TTL = 60
             self.MEMORIES_CACHE_TTL = 60
 
-        self.framework_content = self._load_framework_content()
-
-        # Initialize agent registry
-        self.agent_registry = AgentRegistryAdapter(self.framework_path)
-
-        # Initialize output style manager (must be after content is loaded)
-        self.output_style_manager = None
-        # Defer initialization until first use to ensure content is loaded
+    # === Cache Management Methods (backward compatibility) ===
 
     def clear_all_caches(self) -> None:
         """Clear all caches to force reload on next access."""
         self._cache_manager.clear_all()
 
     def clear_agent_caches(self) -> None:
-        """Clear agent-related caches (capabilities, deployed agents, metadata)."""
+        """Clear agent-related caches."""
         self._cache_manager.clear_agent_caches()
 
     def clear_memory_caches(self) -> None:
         """Clear memory-related caches."""
         self._cache_manager.clear_memory_caches()
 
-    def _initialize_output_style(self) -> None:
-        """Initialize output style management and deploy if applicable."""
-        try:
-            from claude_mpm.core.output_style_manager import OutputStyleManager
-
-            self.output_style_manager = OutputStyleManager()
-
-            # Log detailed output style status
-            self._log_output_style_status()
-
-            # Extract and save output style content (pass self to reuse loaded content)
-            output_style_content = (
-                self.output_style_manager.extract_output_style_content(
-                    framework_loader=self
-                )
-            )
-            self.output_style_manager.save_output_style(output_style_content)
-
-            # Deploy to Claude Code if supported
-            deployed = self.output_style_manager.deploy_output_style(
-                output_style_content
-            )
-
-            if deployed:
-                self.logger.info("✅ Output style deployed to Claude Code >= 1.0.83")
-            else:
-                self.logger.info(
-                    "📝 Output style will be injected into instructions for older Claude versions"
-                )
-
-        except Exception as e:
-            self.logger.warning(f"❌ Failed to initialize output style manager: {e}")
-            # Continue without output style management
-
-    def _log_output_style_status(self) -> None:
-        """Log comprehensive output style status information."""
-        if not self.output_style_manager:
-            return
-
-        # Claude version detection
-        claude_version = self.output_style_manager.claude_version
-        if claude_version:
-            self.logger.info(f"Claude Code version detected: {claude_version}")
-
-            # Check if version supports output styles
-            if self.output_style_manager.supports_output_styles():
-                self.logger.info("✅ Claude Code supports output styles (>= 1.0.83)")
-
-                # Check deployment status
-                output_style_path = self.output_style_manager.output_style_path
-                if output_style_path.exists():
-                    self.logger.info(
-                        f"📁 Output style file exists: {output_style_path}"
-                    )
-                else:
-                    self.logger.info(
-                        f"📝 Output style will be created at: {output_style_path}"
-                    )
-
-            else:
-                self.logger.info(
-                    f"⚠️ Claude Code {claude_version} does not support output styles (< 1.0.83)"
-                )
-                self.logger.info(
-                    "📝 Output style content will be injected into framework instructions"
-                )
-        else:
-            self.logger.info("⚠️ Claude Code not detected or version unknown")
-            self.logger.info(
-                "📝 Output style content will be injected into framework instructions as fallback"
-            )
-
-    def _try_load_file(self, file_path: Path, file_type: str) -> Optional[str]:
-        """
-        Try to load a file with error handling.
-
-        Args:
-            file_path: Path to the file to load
-            file_type: Description of file type for logging
-
-        Returns:
-            File content if successful, None otherwise
-        """
-        try:
-            content = file_path.read_text()
-            if hasattr(self.logger, "level") and self.logger.level <= logging.INFO:
-                self.logger.info(f"Loaded {file_type} from: {file_path}")
-
-            # Extract metadata if present
-            import re
-
-            version_match = re.search(r"<!-- FRAMEWORK_VERSION: (\d+) -->", content)
-            if version_match:
-                version = version_match.group(
-                    1
-                )  # Keep as string to preserve leading zeros
-                self.logger.info(f"Framework version: {version}")
-                # Store framework version if this is the main INSTRUCTIONS.md
-                if "INSTRUCTIONS.md" in str(file_path):
-                    self.framework_version = version
-
-            # Extract modification timestamp
-            timestamp_match = re.search(r"<!-- LAST_MODIFIED: ([^>]+) -->", content)
-            if timestamp_match:
-                timestamp = timestamp_match.group(1).strip()
-                self.logger.info(f"Last modified: {timestamp}")
-                # Store timestamp if this is the main INSTRUCTIONS.md
-                if "INSTRUCTIONS.md" in str(file_path):
-                    self.framework_last_modified = timestamp
-
-            return content
-        except Exception as e:
-            if hasattr(self.logger, "level") and self.logger.level <= logging.ERROR:
-                self.logger.error(f"Failed to load {file_type}: {e}")
-            return None
-
-    def _load_instructions_file(self, content: Dict[str, Any]) -> None:
-        """
-        Load custom INSTRUCTIONS.md from .claude-mpm directories.
-
-        Precedence (highest to lowest):
-        1. Project-specific: ./.claude-mpm/INSTRUCTIONS.md
-        2. User-specific: ~/.claude-mpm/INSTRUCTIONS.md
-
-        NOTE: We do NOT load CLAUDE.md files since Claude Code already picks them up automatically.
-        This prevents duplication of instructions.
-
-        Args:
-            content: Dictionary to update with loaded instructions
-        """
-        # Check for project-specific INSTRUCTIONS.md first
-        project_instructions_path = Path.cwd() / ".claude-mpm" / "INSTRUCTIONS.md"
-        if project_instructions_path.exists():
-            loaded_content = self._try_load_file(
-                project_instructions_path, "project-specific INSTRUCTIONS.md"
-            )
-            if loaded_content:
-                content["custom_instructions"] = loaded_content
-                content["custom_instructions_level"] = "project"
-                self.logger.info(
-                    "Using project-specific PM instructions from .claude-mpm/INSTRUCTIONS.md"
-                )
-                return
-
-        # Check for user-specific INSTRUCTIONS.md
-        user_instructions_path = Path.home() / ".claude-mpm" / "INSTRUCTIONS.md"
-        if user_instructions_path.exists():
-            loaded_content = self._try_load_file(
-                user_instructions_path, "user-specific INSTRUCTIONS.md"
-            )
-            if loaded_content:
-                content["custom_instructions"] = loaded_content
-                content["custom_instructions_level"] = "user"
-                self.logger.info(
-                    "Using user-specific PM instructions from ~/.claude-mpm/INSTRUCTIONS.md"
-                )
-                return
-
-    def _load_workflow_instructions(self, content: Dict[str, Any]) -> None:
-        """
-        Load WORKFLOW.md from .claude-mpm directories.
-
-        Precedence (highest to lowest):
-        1. Project-specific: ./.claude-mpm/WORKFLOW.md
-        2. User-specific: ~/.claude-mpm/WORKFLOW.md
-        3. System default: src/claude_mpm/agents/WORKFLOW.md or packaged
-
-        NOTE: We do NOT load from .claude/ directories to avoid conflicts.
-
-        Args:
-            content: Dictionary to update with workflow instructions
-        """
-        # Check for project-specific WORKFLOW.md first (highest priority)
-        project_workflow_path = Path.cwd() / ".claude-mpm" / "WORKFLOW.md"
-        if project_workflow_path.exists():
-            loaded_content = self._try_load_file(
-                project_workflow_path, "project-specific WORKFLOW.md"
-            )
-            if loaded_content:
-                content["workflow_instructions"] = loaded_content
-                content["workflow_instructions_level"] = "project"
-                self.logger.info(
-                    "Using project-specific workflow instructions from .claude-mpm/WORKFLOW.md"
-                )
-                return
-
-        # Check for user-specific WORKFLOW.md (medium priority)
-        user_workflow_path = Path.home() / ".claude-mpm" / "WORKFLOW.md"
-        if user_workflow_path.exists():
-            loaded_content = self._try_load_file(
-                user_workflow_path, "user-specific WORKFLOW.md"
-            )
-            if loaded_content:
-                content["workflow_instructions"] = loaded_content
-                content["workflow_instructions_level"] = "user"
-                self.logger.info(
-                    "Using user-specific workflow instructions from ~/.claude-mpm/WORKFLOW.md"
-                )
-                return
-
-        # Fall back to system workflow (lowest priority)
-        if self.framework_path and self.framework_path != Path("__PACKAGED__"):
-            system_workflow_path = (
-                self.framework_path / "src" / "claude_mpm" / "agents" / "WORKFLOW.md"
-            )
-            if system_workflow_path.exists():
-                loaded_content = self._try_load_file(
-                    system_workflow_path, "system WORKFLOW.md"
-                )
-                if loaded_content:
-                    content["workflow_instructions"] = loaded_content
-                    content["workflow_instructions_level"] = "system"
-                    self.logger.info("Using system workflow instructions")
-
-    def _load_memory_instructions(self, content: Dict[str, Any]) -> None:
-        """
-        Load MEMORY.md from .claude-mpm directories.
-
-        Precedence (highest to lowest):
-        1. Project-specific: ./.claude-mpm/MEMORY.md
-        2. User-specific: ~/.claude-mpm/MEMORY.md
-        3. System default: src/claude_mpm/agents/MEMORY.md or packaged
-
-        NOTE: We do NOT load from .claude/ directories to avoid conflicts.
-
-        Args:
-            content: Dictionary to update with memory instructions
-        """
-        # Check for project-specific MEMORY.md first (highest priority)
-        project_memory_path = Path.cwd() / ".claude-mpm" / "MEMORY.md"
-        if project_memory_path.exists():
-            loaded_content = self._try_load_file(
-                project_memory_path, "project-specific MEMORY.md"
-            )
-            if loaded_content:
-                content["memory_instructions"] = loaded_content
-                content["memory_instructions_level"] = "project"
-                self.logger.info(
-                    "Using project-specific memory instructions from .claude-mpm/MEMORY.md"
-                )
-                return
-
-        # Check for user-specific MEMORY.md (medium priority)
-        user_memory_path = Path.home() / ".claude-mpm" / "MEMORY.md"
-        if user_memory_path.exists():
-            loaded_content = self._try_load_file(
-                user_memory_path, "user-specific MEMORY.md"
-            )
-            if loaded_content:
-                content["memory_instructions"] = loaded_content
-                content["memory_instructions_level"] = "user"
-                self.logger.info(
-                    "Using user-specific memory instructions from ~/.claude-mpm/MEMORY.md"
-                )
-                return
-
-        # Fall back to system memory instructions (lowest priority)
-        if self.framework_path and self.framework_path != Path("__PACKAGED__"):
-            system_memory_path = (
-                self.framework_path / "src" / "claude_mpm" / "agents" / "MEMORY.md"
-            )
-            if system_memory_path.exists():
-                loaded_content = self._try_load_file(
-                    system_memory_path, "system MEMORY.md"
-                )
-                if loaded_content:
-                    content["memory_instructions"] = loaded_content
-                    content["memory_instructions_level"] = "system"
-                    self.logger.info("Using system memory instructions")
-
-    def _get_deployed_agents(self) -> set:
-        """
-        Get a set of deployed agent names from .claude/agents/ directories.
-        Uses caching to avoid repeated filesystem scans.
-
-        Returns:
-            Set of agent names (file stems) that are deployed
-        """
-        # Try to get from cache first
-        cached = self._cache_manager.get_deployed_agents()
-        if cached is not None:
-            return cached
-
-        # Cache miss or expired - perform actual scan
-        self.logger.debug("Scanning for deployed agents (cache miss or expired)")
-        deployed = set()
-
-        # Check multiple locations for deployed agents
-        agents_dirs = [
-            Path.cwd() / ".claude" / "agents",  # Project-specific agents
-            Path.home() / ".claude" / "agents",  # User's system agents
-        ]
-
-        for agents_dir in agents_dirs:
-            if agents_dir.exists():
-                for agent_file in agents_dir.glob("*.md"):
-                    if not agent_file.name.startswith("."):
-                        # Use stem to get agent name without extension
-                        deployed.add(agent_file.stem)
-                        self.logger.debug(
-                            f"Found deployed agent: {agent_file.stem} in {agents_dir}"
-                        )
-
-        self.logger.debug(f"Total deployed agents found: {len(deployed)}")
-
-        # Update cache
-        self._cache_manager.set_deployed_agents(deployed)
-
-        return deployed
-
-    def _load_actual_memories(self, content: Dict[str, Any]) -> None:
-        """
-        Load actual memories using the MemoryManager service.
-
-        This method delegates all memory loading operations to the MemoryManager,
-        which handles caching, aggregation, deduplication, and legacy format migration.
-
-        Args:
-            content: Dictionary to update with actual memories
-        """
-        # Use MemoryManager to load all memories
-        memories = self._memory_manager.load_memories()
-
-        # Apply loaded memories to content
-        if "actual_memories" in memories:
-            content["actual_memories"] = memories["actual_memories"]
-        if "agent_memories" in memories:
-            content["agent_memories"] = memories["agent_memories"]
-
-    def _load_single_agent(
-        self, agent_file: Path
-    ) -> tuple[Optional[str], Optional[str]]:
-        """
-        Load a single agent file.
-
-        Args:
-            agent_file: Path to the agent file
-
-        Returns:
-            Tuple of (agent_name, agent_content) or (None, None) on failure
-        """
-        try:
-            agent_name = agent_file.stem
-            # Skip README files
-            if agent_name.upper() == "README":
-                return None, None
-            content = agent_file.read_text()
-            self.logger.debug(f"Loaded agent: {agent_name}")
-            return agent_name, content
-        except Exception as e:
-            self.logger.error(f"Failed to load agent {agent_file}: {e}")
-            return None, None
-
-    def _load_base_agent_fallback(
-        self, content: Dict[str, Any], main_dir: Optional[Path]
-    ) -> None:
-        """
-        Load base_agent.md from main directory as fallback.
-
-        Args:
-            content: Dictionary to update with base agent
-            main_dir: Main agents directory path
-        """
-        if main_dir and main_dir.exists() and "base_agent" not in content["agents"]:
-            base_agent_file = main_dir / "base_agent.md"
-            if base_agent_file.exists():
-                agent_name, agent_content = self._load_single_agent(base_agent_file)
-                if agent_name and agent_content:
-                    content["agents"][agent_name] = agent_content
-
-    def _load_agents_directory(
-        self,
-        content: Dict[str, Any],
-        agents_dir: Optional[Path],
-        templates_dir: Optional[Path],
-        main_dir: Optional[Path],
-    ) -> None:
-        """
-        Load agent definitions from the appropriate directory.
-
-        Args:
-            content: Dictionary to update with loaded agents
-            agents_dir: Primary agents directory to load from
-            templates_dir: Templates directory path
-            main_dir: Main agents directory path
-        """
-        if not agents_dir or not agents_dir.exists():
-            return
-
-        content["loaded"] = True
-
-        # Load all agent files
-        for agent_file in agents_dir.glob("*.md"):
-            agent_name, agent_content = self._load_single_agent(agent_file)
-            if agent_name and agent_content:
-                content["agents"][agent_name] = agent_content
-
-        # If we used templates dir, also check main dir for base_agent.md
-        if agents_dir == templates_dir:
-            self._load_base_agent_fallback(content, main_dir)
+    # === Content Loading Methods ===
 
     def _load_framework_content(self) -> Dict[str, Any]:
-        """Load framework content."""
+        """Load framework content using modular components."""
         content = {
             "claude_md": "",
             "agents": {},
@@ -619,290 +216,102 @@ class FrameworkLoader:
             "working_claude_md": "",
             "framework_instructions": "",
             "workflow_instructions": "",
-            "workflow_instructions_level": "",  # Track source level
+            "workflow_instructions_level": "",
             "memory_instructions": "",
-            "memory_instructions_level": "",  # Track source level
-            "project_workflow": "",  # Deprecated, use workflow_instructions_level
-            "project_memory": "",  # Deprecated, use memory_instructions_level
-            "actual_memories": "",  # Add field for actual memories from PM_memories.md
+            "memory_instructions_level": "",
+            "project_workflow": "",  # Deprecated
+            "project_memory": "",  # Deprecated
+            "actual_memories": "",
+            "agent_memories": {},
         }
 
-        # Load instructions file from working directory
-        self._load_instructions_file(content)
-
-        if not self.framework_path:
-            return content
+        # Load all instructions
+        self.instruction_loader.load_all_instructions(content)
 
-        # Check if this is a packaged installation
-        if self.framework_path == Path("__PACKAGED__"):
-            # Load files using importlib.resources for packaged installations
-            self._load_packaged_framework_content(content)
-        else:
-            # Load from filesystem for development mode
-            # Try new consolidated PM_INSTRUCTIONS.md first, fall back to INSTRUCTIONS.md
-            pm_instructions_path = (
-                self.framework_path
-                / "src"
-                / "claude_mpm"
-                / "agents"
-                / "PM_INSTRUCTIONS.md"
-            )
-            framework_instructions_path = (
-                self.framework_path
-                / "src"
-                / "claude_mpm"
-                / "agents"
-                / "INSTRUCTIONS.md"
-            )
+        # Transfer metadata from loaders
+        if self.file_loader.framework_version:
+            self.framework_version = self.file_loader.framework_version
+            content["version"] = self.framework_version
+        if self.file_loader.framework_last_modified:
+            self.framework_last_modified = self.file_loader.framework_last_modified
 
-            # Try loading new consolidated file first
-            if pm_instructions_path.exists():
-                loaded_content = self._try_load_file(
-                    pm_instructions_path, "consolidated PM_INSTRUCTIONS.md"
-                )
-                if loaded_content:
-                    content["framework_instructions"] = loaded_content
-                    self.logger.info("Loaded consolidated PM_INSTRUCTIONS.md")
-            # Fall back to legacy file for backward compatibility
-            elif framework_instructions_path.exists():
-                loaded_content = self._try_load_file(
-                    framework_instructions_path, "framework INSTRUCTIONS.md (legacy)"
-                )
-                if loaded_content:
-                    content["framework_instructions"] = loaded_content
-                    self.logger.warning(
-                        "Using legacy INSTRUCTIONS.md - consider migrating to PM_INSTRUCTIONS.md"
-                    )
-                    content["loaded"] = True
-                    # Add framework version to content
-                    if self.framework_version:
-                        content["instructions_version"] = self.framework_version
-                        content["version"] = (
-                            self.framework_version
-                        )  # Update main version key
-                    # Add modification timestamp to content
-                    if self.framework_last_modified:
-                        content["instructions_last_modified"] = (
-                            self.framework_last_modified
-                        )
-
-            # Load BASE_PM.md for core framework requirements
-            base_pm_path = (
-                self.framework_path / "src" / "claude_mpm" / "agents" / "BASE_PM.md"
-            )
-            if base_pm_path.exists():
-                base_pm_content = self._try_load_file(
-                    base_pm_path, "BASE_PM framework requirements"
-                )
-                if base_pm_content:
-                    content["base_pm_instructions"] = base_pm_content
-
-        # Load WORKFLOW.md - check for project-specific first, then system
-        self._load_workflow_instructions(content)
-
-        # Load MEMORY.md - check for project-specific first, then system
-        self._load_memory_instructions(content)
-
-        # Load actual memories from .claude-mpm/memories/PM_memories.md
+        # Load memories
         self._load_actual_memories(content)
 
-        # Discover agent directories using PathResolver
+        # Discover and load agents
         agents_dir, templates_dir, main_dir = self._path_resolver.discover_agent_paths(
             agents_dir=self.agents_dir, framework_path=self.framework_path
         )
-
-        # Load agents from discovered directory
-        self._load_agents_directory(content, agents_dir, templates_dir, main_dir)
+        agents = self.agent_loader.load_agents_directory(
+            agents_dir, templates_dir, main_dir
+        )
+        if agents:
+            content["agents"] = agents
+            content["loaded"] = True
 
         return content
 
-    def _load_packaged_framework_content(self, content: Dict[str, Any]) -> None:
-        """Load framework content from packaged installation using importlib.resources."""
-        if not files:
-            self.logger.warning(
-                "importlib.resources not available, cannot load packaged framework"
-            )
-            self.logger.debug(f"files variable is: {files}")
-            # Try alternative import methods
-            try:
-                from importlib import resources
-
-                self.logger.info("Using importlib.resources as fallback")
-                self._load_packaged_framework_content_fallback(content, resources)
-                return
-            except ImportError:
-                self.logger.error(
-                    "No importlib.resources available, using minimal framework"
-                )
-                return
-
-        try:
-            # Try new consolidated PM_INSTRUCTIONS.md first
-            pm_instructions_content = self._load_packaged_file("PM_INSTRUCTIONS.md")
-            if pm_instructions_content:
-                content["framework_instructions"] = pm_instructions_content
-                content["loaded"] = True
-                self.logger.info("Loaded consolidated PM_INSTRUCTIONS.md from package")
-                # Extract and store version/timestamp metadata
-                self._extract_metadata_from_content(
-                    pm_instructions_content, "PM_INSTRUCTIONS.md"
-                )
-            else:
-                # Fall back to legacy INSTRUCTIONS.md
-                instructions_content = self._load_packaged_file("INSTRUCTIONS.md")
-                if instructions_content:
-                    content["framework_instructions"] = instructions_content
-                    content["loaded"] = True
-                    self.logger.warning("Using legacy INSTRUCTIONS.md from package")
-                    # Extract and store version/timestamp metadata
-                    self._extract_metadata_from_content(
-                        instructions_content, "INSTRUCTIONS.md"
-                    )
-
-            if self.framework_version:
-                content["instructions_version"] = self.framework_version
-                content["version"] = self.framework_version
-            if self.framework_last_modified:
-                content["instructions_last_modified"] = self.framework_last_modified
-
-            # Load BASE_PM.md
-            base_pm_content = self._load_packaged_file("BASE_PM.md")
-            if base_pm_content:
-                content["base_pm_instructions"] = base_pm_content
-
-            # Load WORKFLOW.md
-            workflow_content = self._load_packaged_file("WORKFLOW.md")
-            if workflow_content:
-                content["workflow_instructions"] = workflow_content
-                content["project_workflow"] = "system"
-
-            # Load MEMORY.md
-            memory_content = self._load_packaged_file("MEMORY.md")
-            if memory_content:
-                content["memory_instructions"] = memory_content
-                content["project_memory"] = "system"
-
-        except Exception as e:
-            self.logger.error(f"Failed to load packaged framework content: {e}")
-
-    def _load_packaged_framework_content_fallback(
-        self, content: Dict[str, Any], resources
-    ) -> None:
-        """Load framework content using importlib.resources fallback."""
-        try:
-            # Try new consolidated PM_INSTRUCTIONS.md first
-            pm_instructions_content = self._load_packaged_file_fallback(
-                "PM_INSTRUCTIONS.md", resources
-            )
-            if pm_instructions_content:
-                content["framework_instructions"] = pm_instructions_content
-                content["loaded"] = True
-                self.logger.info("Loaded consolidated PM_INSTRUCTIONS.md via fallback")
-                # Extract and store version/timestamp metadata
-                self._extract_metadata_from_content(
-                    pm_instructions_content, "PM_INSTRUCTIONS.md"
-                )
-            else:
-                # Fall back to legacy INSTRUCTIONS.md
-                instructions_content = self._load_packaged_file_fallback(
-                    "INSTRUCTIONS.md", resources
-                )
-                if instructions_content:
-                    content["framework_instructions"] = instructions_content
-                    content["loaded"] = True
-                    self.logger.warning("Using legacy INSTRUCTIONS.md via fallback")
-                    # Extract and store version/timestamp metadata
-                    self._extract_metadata_from_content(
-                        instructions_content, "INSTRUCTIONS.md"
-                    )
-
-            if self.framework_version:
-                content["instructions_version"] = self.framework_version
-                content["version"] = self.framework_version
-            if self.framework_last_modified:
-                content["instructions_last_modified"] = self.framework_last_modified
-
-            # Load BASE_PM.md
-            base_pm_content = self._load_packaged_file_fallback("BASE_PM.md", resources)
-            if base_pm_content:
-                content["base_pm_instructions"] = base_pm_content
+    def _load_actual_memories(self, content: Dict[str, Any]) -> None:
+        """Load actual memories using the MemoryManager service."""
+        memories = self._memory_manager.load_memories()
 
-            # Load WORKFLOW.md
-            workflow_content = self._load_packaged_file_fallback(
-                "WORKFLOW.md", resources
-            )
-            if workflow_content:
-                content["workflow_instructions"] = workflow_content
-                content["project_workflow"] = "system"
+        if "actual_memories" in memories:
+            content["actual_memories"] = memories["actual_memories"]
+        if "agent_memories" in memories:
+            content["agent_memories"] = memories["agent_memories"]
 
-            # Load MEMORY.md
-            memory_content = self._load_packaged_file_fallback("MEMORY.md", resources)
-            if memory_content:
-                content["memory_instructions"] = memory_content
-                content["project_memory"] = "system"
+    # === Agent Discovery Methods ===
 
-        except Exception as e:
-            self.logger.error(
-                f"Failed to load packaged framework content with fallback: {e}"
-            )
+    def _get_deployed_agents(self) -> Set[str]:
+        """Get deployed agents with caching."""
+        cached = self._cache_manager.get_deployed_agents()
+        if cached is not None:
+            return cached
 
-    def _load_packaged_file_fallback(self, filename: str, resources) -> Optional[str]:
-        """Load a file from the packaged installation using importlib.resources fallback."""
-        try:
-            # Try different resource loading methods
-            try:
-                # Method 1: resources.read_text (Python 3.9+)
-                content = resources.read_text("claude_mpm.agents", filename)
-                self.logger.info(f"Loaded {filename} from package using read_text")
-                return content
-            except AttributeError:
-                # Method 2: resources.files (Python 3.9+)
-                agents_files = resources.files("claude_mpm.agents")
-                file_path = agents_files / filename
-                if file_path.is_file():
-                    content = file_path.read_text()
-                    self.logger.info(f"Loaded {filename} from package using files")
-                    return content
-                self.logger.warning(f"File {filename} not found in package")
-                return None
-        except Exception as e:
-            self.logger.error(
-                f"Failed to load {filename} from package with fallback: {e}"
-            )
-            return None
+        deployed = self.agent_loader.get_deployed_agents()
+        self._cache_manager.set_deployed_agents(deployed)
+        return deployed
 
-    def _load_packaged_file(self, filename: str) -> Optional[str]:
-        """Load a file from the packaged installation."""
-        try:
-            # Use importlib.resources to load file from package
-            agents_package = files("claude_mpm.agents")
-            file_path = agents_package / filename
-
-            if file_path.is_file():
-                content = file_path.read_text()
-                self.logger.info(f"Loaded {filename} from package")
-                return content
-            self.logger.warning(f"File {filename} not found in package")
-            return None
-        except Exception as e:
-            self.logger.error(f"Failed to load {filename} from package: {e}")
-            return None
+    def _discover_local_json_templates(self) -> Dict[str, Dict[str, Any]]:
+        """Discover local JSON agent templates."""
+        return self.agent_loader.discover_local_json_templates()
 
-    def _extract_metadata_from_content(self, content: str, filename: str) -> None:
-        """Extract metadata from content string."""
-        import re
+    def _parse_agent_metadata(self, agent_file: Path) -> Optional[Dict[str, Any]]:
+        """Parse agent metadata with caching."""
+        cache_key = str(agent_file)
+        file_mtime = agent_file.stat().st_mtime
+
+        # Try cache first
+        cached_result = self._cache_manager.get_agent_metadata(cache_key)
+        if cached_result is not None:
+            cached_data, cached_mtime = cached_result
+            if cached_mtime == file_mtime:
+                self.logger.debug(f"Using cached metadata for {agent_file.name}")
+                return cached_data
+
+        # Cache miss - parse the file
+        agent_data = self.metadata_processor.parse_agent_metadata(agent_file)
+
+        # Add routing information if not present
+        if agent_data and "routing" not in agent_data:
+            template_data = self.template_processor.load_template(agent_file.stem)
+            if template_data:
+                routing = self.template_processor.extract_routing(template_data)
+                if routing:
+                    agent_data["routing"] = routing
+                memory_routing = self.template_processor.extract_memory_routing(
+                    template_data
+                )
+                if memory_routing:
+                    agent_data["memory_routing"] = memory_routing
+
+        # Cache the result
+        if agent_data:
+            self._cache_manager.set_agent_metadata(cache_key, agent_data, file_mtime)
 
-        # Extract version
-        version_match = re.search(r"<!-- FRAMEWORK_VERSION: (\d+) -->", content)
-        if version_match and "INSTRUCTIONS.md" in filename:
-            self.framework_version = version_match.group(1)
-            self.logger.info(f"Framework version: {self.framework_version}")
+        return agent_data
 
-        # Extract timestamp
-        timestamp_match = re.search(r"<!-- LAST_MODIFIED: ([^>]+) -->", content)
-        if timestamp_match and "INSTRUCTIONS.md" in filename:
-            self.framework_last_modified = timestamp_match.group(1).strip()
-            self.logger.info(f"Last modified: {self.framework_last_modified}")
+    # === Framework Instructions Generation ===
 
     def get_framework_instructions(self) -> str:
         """
@@ -911,1127 +320,221 @@ class FrameworkLoader:
         Returns:
             Complete framework instructions ready for injection
         """
-        # Import LogManager for prompt logging
-        try:
-            from .log_manager import get_log_manager
-
-            log_manager = get_log_manager()
-        except ImportError:
-            log_manager = None
+        # Log the system prompt if needed
+        self._log_system_prompt()
 
         # Generate the instructions
         if self.framework_content["loaded"]:
-            # Build framework from components
-            instructions = self._format_full_framework()
-        else:
-            # Use minimal fallback
-            instructions = self._format_minimal_framework()
-
-        # Log the system prompt if LogManager is available
-        if log_manager:
-            try:
-                import asyncio
-                import os
-
-                # Get or create event loop
-                try:
-                    loop = asyncio.get_running_loop()
-                except RuntimeError:
-                    loop = asyncio.new_event_loop()
-                    asyncio.set_event_loop(loop)
-
-                # Prepare metadata
-                metadata = {
-                    "framework_version": self.framework_version,
-                    "framework_loaded": self.framework_content.get("loaded", False),
-                    "session_id": os.environ.get("CLAUDE_SESSION_ID", "unknown"),
-                    "instructions_length": len(instructions),
-                }
-
-                # Log the prompt asynchronously
-                if loop.is_running():
-                    asyncio.create_task(
-                        log_manager.log_prompt("system_prompt", instructions, metadata)
-                    )
-                else:
-                    loop.run_until_complete(
-                        log_manager.log_prompt("system_prompt", instructions, metadata)
-                    )
-
-                self.logger.debug("System prompt logged to prompts directory")
-            except Exception as e:
-                self.logger.debug(f"Could not log system prompt: {e}")
-
-        return instructions
-
-    def _strip_metadata_comments(self, content: str) -> str:
-        """Strip metadata HTML comments from content.
-
-        Removes comments like:
-        <!-- FRAMEWORK_VERSION: 0010 -->
-        <!-- LAST_MODIFIED: 2025-08-10T00:00:00Z -->
-        """
-        import re
-
-        # Remove HTML comments that contain metadata
-        cleaned = re.sub(
-            r"<!--\s*(FRAMEWORK_VERSION|LAST_MODIFIED|WORKFLOW_VERSION|PROJECT_WORKFLOW_VERSION|CUSTOM_PROJECT_WORKFLOW)[^>]*-->\n?",
-            "",
-            content,
-        )
-        # Also remove any leading blank lines that might result
-        return cleaned.lstrip("\n")
+            return self._format_full_framework()
+        return self._format_minimal_framework()
 
     def _format_full_framework(self) -> str:
-        """Format full framework instructions."""
-
-        # Initialize output style manager on first use (ensures content is loaded)
+        """Format full framework instructions using modular components."""
+        # Initialize output style manager on first use
         if self.output_style_manager is None:
             self._initialize_output_style()
 
-        # Check if we need to inject output style content for older Claude versions
+        # Check if we need to inject output style
         inject_output_style = False
+        output_style_content = None
         if self.output_style_manager:
             inject_output_style = self.output_style_manager.should_inject_content()
             if inject_output_style:
-                self.logger.info(
-                    "Injecting output style content into instructions for Claude < 1.0.83"
-                )
-
-        # If we have the full framework INSTRUCTIONS.md, use it
-        if self.framework_content.get("framework_instructions"):
-            instructions = self._strip_metadata_comments(
-                self.framework_content["framework_instructions"]
-            )
-
-            # Note: We don't add working directory CLAUDE.md here since Claude Code
-            # already picks it up automatically. This prevents duplication.
-
-            # Add custom INSTRUCTIONS.md if present (overrides or extends framework instructions)
-            if self.framework_content.get("custom_instructions"):
-                level = self.framework_content.get(
-                    "custom_instructions_level", "unknown"
-                )
-                instructions += f"\n\n## Custom PM Instructions ({level} level)\n\n"
-                instructions += "**The following custom instructions override or extend the framework defaults:**\n\n"
-                instructions += self._strip_metadata_comments(
-                    self.framework_content["custom_instructions"]
-                )
-                instructions += "\n"
-
-            # Add WORKFLOW.md after instructions
-            if self.framework_content.get("workflow_instructions"):
-                workflow_content = self._strip_metadata_comments(
-                    self.framework_content["workflow_instructions"]
-                )
-                level = self.framework_content.get(
-                    "workflow_instructions_level", "system"
-                )
-                if level != "system":
-                    instructions += f"\n\n## Workflow Instructions ({level} level)\n\n"
-                    instructions += "**The following workflow instructions override system defaults:**\n\n"
-                instructions += f"{workflow_content}\n"
-
-            # Add MEMORY.md after workflow instructions
-            if self.framework_content.get("memory_instructions"):
-                memory_content = self._strip_metadata_comments(
-                    self.framework_content["memory_instructions"]
-                )
-                level = self.framework_content.get(
-                    "memory_instructions_level", "system"
-                )
-                if level != "system":
-                    instructions += f"\n\n## Memory Instructions ({level} level)\n\n"
-                    instructions += "**The following memory instructions override system defaults:**\n\n"
-                instructions += f"{memory_content}\n"
-
-            # Add actual PM memories after memory instructions
-            if self.framework_content.get("actual_memories"):
-                instructions += "\n\n## Current PM Memories\n\n"
-                instructions += "**The following are your accumulated memories and knowledge from this project:**\n\n"
-                instructions += self.framework_content["actual_memories"]
-                instructions += "\n"
-
-            # Add agent memories if available
-            if self.framework_content.get("agent_memories"):
-                agent_memories = self.framework_content["agent_memories"]
-                if agent_memories:
-                    instructions += "\n\n## Agent Memories\n\n"
-                    instructions += "**The following are accumulated memories from specialized agents:**\n\n"
-
-                    for agent_name in sorted(agent_memories.keys()):
-                        memory_content = agent_memories[agent_name]
-                        if memory_content:
-                            instructions += f"### {agent_name.replace('_', ' ').title()} Agent Memory\n\n"
-                            instructions += memory_content
-                            instructions += "\n\n"
-
-            # Add dynamic agent capabilities section
-            instructions += self._generate_agent_capabilities_section()
-
-            # Add enhanced temporal and user context for better awareness
-            instructions += self._generate_temporal_user_context()
-
-            # Add BASE_PM.md framework requirements AFTER INSTRUCTIONS.md
-            if self.framework_content.get("base_pm_instructions"):
-                base_pm = self._strip_metadata_comments(
-                    self.framework_content["base_pm_instructions"]
-                )
-                instructions += f"\n\n{base_pm}"
-
-            # Inject output style content if needed (for Claude < 1.0.83)
-            if inject_output_style and self.output_style_manager:
                 output_style_content = self.output_style_manager.get_injectable_content(
                     framework_loader=self
                 )
-                if output_style_content:
-                    instructions += "\n\n## Output Style Configuration\n"
-                    instructions += "**Note: The following output style is injected for Claude < 1.0.83**\n\n"
-                    instructions += output_style_content
-                    instructions += "\n"
-
-            # Clean up any trailing whitespace
-            return instructions.rstrip() + "\n"
-
-        # Otherwise fall back to generating framework
-        instructions = """# Claude MPM Framework Instructions
-
-You are operating within the Claude Multi-Agent Project Manager (MPM) framework.
-
-## Core Role
-You are a multi-agent orchestrator. Your primary responsibilities are:
-- Delegate all implementation work to specialized agents via Task Tool
-- Coordinate multi-agent workflows and cross-agent collaboration
-- Extract and track TODO/BUG/FEATURE items for ticket creation
-- Maintain project visibility and strategic oversight
-- NEVER perform direct implementation work yourself
-
-"""
-
-        # Note: We don't add working directory CLAUDE.md here since Claude Code
-        # already picks it up automatically. This prevents duplication.
-
-        # Add agent definitions
-        if self.framework_content["agents"]:
-            instructions += "## Available Agents\n\n"
-            instructions += "You have the following specialized agents available for delegation:\n\n"
-
-            # List agents with brief descriptions and correct IDs
-            agent_list = []
-            for agent_name in sorted(self.framework_content["agents"].keys()):
-                # Use the actual agent_name as the ID (it's the filename stem)
-                agent_id = agent_name
-                clean_name = agent_name.replace("-", " ").replace("_", " ").title()
-                if (
-                    "engineer" in agent_name.lower()
-                    and "data" not in agent_name.lower()
-                ):
-                    agent_list.append(
-                        f"- **Engineer Agent** (`{agent_id}`): Code implementation and development"
-                    )
-                elif "qa" in agent_name.lower():
-                    agent_list.append(
-                        f"- **QA Agent** (`{agent_id}`): Testing and quality assurance"
-                    )
-                elif "documentation" in agent_name.lower():
-                    agent_list.append(
-                        f"- **Documentation Agent** (`{agent_id}`): Documentation creation and maintenance"
-                    )
-                elif "research" in agent_name.lower():
-                    agent_list.append(
-                        f"- **Research Agent** (`{agent_id}`): Investigation and analysis"
-                    )
-                elif "security" in agent_name.lower():
-                    agent_list.append(
-                        f"- **Security Agent** (`{agent_id}`): Security analysis and protection"
-                    )
-                elif "version" in agent_name.lower():
-                    agent_list.append(
-                        f"- **Version Control Agent** (`{agent_id}`): Git operations and version management"
-                    )
-                elif "ops" in agent_name.lower():
-                    agent_list.append(
-                        f"- **Ops Agent** (`{agent_id}`): Deployment and operations"
-                    )
-                elif "data" in agent_name.lower():
-                    agent_list.append(
-                        f"- **Data Engineer Agent** (`{agent_id}`): Data management and AI API integration"
-                    )
-                else:
-                    agent_list.append(
-                        f"- **{clean_name}** (`{agent_id}`): Available for specialized tasks"
-                    )
+                self.logger.info("Injecting output style content for Claude < 1.0.83")
 
-            instructions += "\n".join(agent_list) + "\n\n"
-
-            # Add full agent details
-            instructions += "### Agent Details\n\n"
-            for agent_name, agent_content in sorted(
-                self.framework_content["agents"].items()
-            ):
-                instructions += f"#### {agent_name.replace('-', ' ').title()}\n"
-                instructions += agent_content + "\n\n"
-
-        # Add orchestration principles
-        instructions += """
-## Orchestration Principles
-1. **Always Delegate**: Never perform direct work - use Task Tool for all implementation
-2. **Comprehensive Context**: Provide rich, filtered context to each agent
-3. **Track Everything**: Extract all TODO/BUG/FEATURE items systematically
-4. **Cross-Agent Coordination**: Orchestrate workflows spanning multiple agents
-5. **Results Integration**: Actively receive and integrate agent results
-
-## Task Tool Format
-```
-**[Agent Name]**: [Clear task description with deliverables]
-
-TEMPORAL CONTEXT: Today is [date]. Apply date awareness to [specific considerations].
-
-**Task**: [Detailed task breakdown]
-1. [Specific action item 1]
-2. [Specific action item 2]
-3. [Specific action item 3]
-
-**Context**: [Comprehensive filtered context for this agent]
-**Authority**: [Agent's decision-making scope]
-**Expected Results**: [Specific deliverables needed]
-**Integration**: [How results integrate with other work]
-```
-
-## Ticket Extraction Patterns
-Extract tickets from these patterns:
-- TODO: [description] → TODO ticket
-- BUG: [description] → BUG ticket
-- FEATURE: [description] → FEATURE ticket
-- ISSUE: [description] → ISSUE ticket
-- FIXME: [description] → BUG ticket
-
----
-"""
-
-        return instructions
+        # Generate dynamic sections
+        capabilities_section = self._generate_agent_capabilities_section()
+        context_section = self.context_generator.generate_temporal_user_context()
 
-    def _generate_agent_capabilities_section(self) -> str:
-        """Generate dynamic agent capabilities section from deployed agents.
-        Uses caching to avoid repeated file I/O and parsing operations.
-
-        Now includes support for local JSON templates with proper priority:
-        1. Project local agents (.claude-mpm/agents/*.json) - highest priority
-        2. Deployed project agents (.claude/agents/*.md)
-        3. User local agents (~/.claude-mpm/agents/*.json)
-        4. Deployed user agents (~/.claude/agents/*.md)
-        5. System agents - lowest priority
-        """
+        # Format the complete framework
+        return self.content_formatter.format_full_framework(
+            self.framework_content,
+            capabilities_section,
+            context_section,
+            inject_output_style,
+            output_style_content,
+        )
+
+    def _format_minimal_framework(self) -> str:
+        """Format minimal framework instructions."""
+        return self.content_formatter.format_minimal_framework(self.framework_content)
 
-        # Try to get from cache first
+    def _generate_agent_capabilities_section(self) -> str:
+        """Generate agent capabilities section with caching."""
+        # Try cache first
         cached_capabilities = self._cache_manager.get_capabilities()
         if cached_capabilities is not None:
             return cached_capabilities
 
-        # Will be used for updating cache later
-        current_time = time.time()
-
-        # Cache miss or expired - generate capabilities
-        self.logger.debug("Generating agent capabilities (cache miss or expired)")
+        self.logger.debug("Generating agent capabilities (cache miss)")
 
         try:
-            from pathlib import Path
-
-            # First check for local JSON templates (highest priority)
+            # Discover local JSON templates
             local_agents = self._discover_local_json_templates()
 
-            # Read directly from deployed agents in .claude/agents/
-            # Check multiple locations for deployed agents
-            # Priority order: local templates > project > user home > fallback
+            # Get deployed agents from .claude/agents/
+            deployed_agents = []
             agents_dirs = [
-                Path.cwd() / ".claude" / "agents",  # Project-specific agents
-                Path.home() / ".claude" / "agents",  # User's system agents
+                Path.cwd() / ".claude" / "agents",
+                Path.home() / ".claude" / "agents",
             ]
 
-            # Collect agents from all directories with proper precedence
-            # Local agents override deployed agents with the same name
-            # Project agents override user agents with the same name
-            all_agents = {}  # key: agent_id, value: (agent_data, priority)
-
-            # Add local agents first (highest priority)
-            for agent_id, agent_data in local_agents.items():
-                all_agents[agent_id] = (agent_data, -1)  # Priority -1 for local agents
-
-            for priority, potential_dir in enumerate(agents_dirs):
-                if potential_dir.exists() and any(potential_dir.glob("*.md")):
-                    self.logger.debug(f"Found agents directory at: {potential_dir}")
-
-                    # Collect agents from this directory
-                    for agent_file in potential_dir.glob("*.md"):
-                        if agent_file.name.startswith("."):
-                            continue
-
-                        # Parse agent metadata (with caching)
-                        agent_data = self._parse_agent_metadata(agent_file)
-                        if agent_data:
-                            agent_id = agent_data["id"]
-                            # Only add if not already present (project has priority 0, user has priority 1)
-                            # Lower priority number wins (project > user)
-                            if (
-                                agent_id not in all_agents
-                                or priority < all_agents[agent_id][1]
-                            ):
-                                all_agents[agent_id] = (agent_data, priority)
-                                self.logger.debug(
-                                    f"Added/Updated agent {agent_id} from {potential_dir} (priority {priority})"
-                                )
-
-            if not all_agents:
-                self.logger.warning(f"No agents found in any location: {agents_dirs}")
-                result = self._get_fallback_capabilities()
-                # Cache the fallback result too
-                self._cache_manager.set_capabilities(result)
-                return result
-
-            # Log agent collection summary
-            project_agents = [aid for aid, (_, pri) in all_agents.items() if pri == 0]
-            user_agents = [aid for aid, (_, pri) in all_agents.items() if pri == 1]
-
-            # Include local agents in logging
-            local_json_agents = [
-                aid for aid, (_, pri) in all_agents.items() if pri == -1
-            ]
-            if local_json_agents:
-                self.logger.info(
-                    f"Loaded {len(local_json_agents)} local JSON agents: {', '.join(sorted(local_json_agents))}"
-                )
-            if project_agents:
-                self.logger.info(
-                    f"Loaded {len(project_agents)} project agents: {', '.join(sorted(project_agents))}"
-                )
-            if user_agents:
-                self.logger.info(
-                    f"Loaded {len(user_agents)} user agents: {', '.join(sorted(user_agents))}"
-                )
+            for agents_dir in agents_dirs:
+                if agents_dir.exists():
+                    for agent_file in agents_dir.glob("*.md"):
+                        if not agent_file.name.startswith("."):
+                            agent_data = self._parse_agent_metadata(agent_file)
+                            if agent_data:
+                                deployed_agents.append(agent_data)
 
-            # Build capabilities section
-            section = "\n\n## Available Agent Capabilities\n\n"
-
-            # Extract just the agent data (drop priority info) and sort
-            deployed_agents = [agent_data for agent_data, _ in all_agents.values()]
-
-            if not deployed_agents:
-                result = self._get_fallback_capabilities()
-                # Cache the fallback result
-                self._cache_manager.set_capabilities(result)
-                return result
-
-            # Sort agents alphabetically by ID
-            deployed_agents.sort(key=lambda x: x["id"])
-
-            # Display all agents with their rich descriptions
-            for agent in deployed_agents:
-                # Clean up display name - handle common acronyms
-                display_name = agent["display_name"]
-                display_name = (
-                    display_name.replace("Qa ", "QA ")
-                    .replace("Ui ", "UI ")
-                    .replace("Api ", "API ")
-                )
-                if display_name.lower() == "qa agent":
-                    display_name = "QA Agent"
-
-                # Add local indicator if this is a local agent
-                if agent.get("is_local"):
-                    tier_label = f" [LOCAL-{agent.get('tier', 'PROJECT').upper()}]"
-                    section += f"\n### {display_name} (`{agent['id']}`) {tier_label}\n"
-                else:
-                    section += f"\n### {display_name} (`{agent['id']}`)\n"
-                section += f"{agent['description']}\n"
-
-                # Add routing information if available
-                if agent.get("routing"):
-                    routing = agent["routing"]
-
-                    # Format routing hints for PM usage
-                    routing_hints = []
-
-                    if routing.get("keywords"):
-                        # Show first 5 keywords for brevity
-                        keywords = routing["keywords"][:5]
-                        routing_hints.append(f"Keywords: {', '.join(keywords)}")
-
-                    if routing.get("paths"):
-                        # Show first 3 paths for brevity
-                        paths = routing["paths"][:3]
-                        routing_hints.append(f"Paths: {', '.join(paths)}")
-
-                    if routing.get("priority"):
-                        routing_hints.append(f"Priority: {routing['priority']}")
-
-                    if routing_hints:
-                        section += f"- **Routing**: {' | '.join(routing_hints)}\n"
-
-                    # Add when_to_use if present
-                    if routing.get("when_to_use"):
-                        section += f"- **When to use**: {routing['when_to_use']}\n"
-
-                # Add any additional metadata if present
-                if agent.get("authority"):
-                    section += f"- **Authority**: {agent['authority']}\n"
-                if agent.get("primary_function"):
-                    section += f"- **Primary Function**: {agent['primary_function']}\n"
-                if agent.get("handoff_to"):
-                    section += f"- **Handoff To**: {agent['handoff_to']}\n"
-                if agent.get("tools") and agent["tools"] != "standard":
-                    section += f"- **Tools**: {agent['tools']}\n"
-                if agent.get("model") and agent["model"] != "opus":
-                    section += f"- **Model**: {agent['model']}\n"
-
-                # Add memory routing information if available
-                if agent.get("memory_routing"):
-                    memory_routing = agent["memory_routing"]
-                    if memory_routing.get("description"):
-                        section += (
-                            f"- **Memory Routing**: {memory_routing['description']}\n"
-                        )
-
-            # Add simple Context-Aware Agent Selection
-            section += "\n## Context-Aware Agent Selection\n\n"
-            section += (
-                "Select agents based on their descriptions above. Key principles:\n"
-            )
-            section += "- **PM questions** → Answer directly (only exception)\n"
-            section += "- Match task requirements to agent descriptions and authority\n"
-            section += "- Consider agent handoff recommendations\n"
-            section += (
-                "- Use the agent ID in parentheses when delegating via Task tool\n"
+            # Generate capabilities section
+            section = self.capability_generator.generate_capabilities_section(
+                deployed_agents, local_agents
             )
 
-            # Add summary
-            section += f"\n**Total Available Agents**: {len(deployed_agents)}\n"
-
-            # Cache the generated capabilities
+            # Cache the result
             self._cache_manager.set_capabilities(section)
-            self.logger.debug(
-                f"Cached agent capabilities section ({len(section)} chars)"
-            )
+            self.logger.debug(f"Cached agent capabilities ({len(section)} chars)")
 
             return section
 
         except Exception as e:
-            self.logger.warning(f"Could not generate dynamic agent capabilities: {e}")
-            result = self._get_fallback_capabilities()
-            # Cache even the fallback result
-            self._agent_capabilities_cache = result
-            self._agent_capabilities_cache_time = current_time
-            return result
+            self.logger.warning(f"Could not generate agent capabilities: {e}")
+            fallback = self.content_formatter.get_fallback_capabilities()
+            self._cache_manager.set_capabilities(fallback)
+            return fallback
 
-    def _generate_temporal_user_context(self) -> str:
-        """Generate enhanced temporal and user context for better PM awareness.
-
-        Returns:
-            str: Formatted context string with datetime, user, and system information
-        """
-        context_lines = ["\n\n## Temporal & User Context\n"]
+    # === Output Style Management ===
 
+    def _initialize_output_style(self) -> None:
+        """Initialize output style management."""
         try:
-            # Get current datetime with timezone awareness
-            now = datetime.now(timezone.utc)
-
-            # Try to get timezone info - fallback to UTC offset if timezone name not available
-            try:
-                import time as time_module
-
-                if hasattr(time_module, "tzname"):
-                    tz_name = time_module.tzname[time_module.daylight]
-                    tz_offset = time_module.strftime("%z")
-                    if tz_offset:
-                        # Format UTC offset properly (e.g., -0800 to -08:00)
-                        tz_offset = (
-                            f"{tz_offset[:3]}:{tz_offset[3:]}"
-                            if len(tz_offset) >= 4
-                            else tz_offset
-                        )
-                        tz_info = f"{tz_name} (UTC{tz_offset})"
-                    else:
-                        tz_info = tz_name
-                else:
-                    tz_info = "Local Time"
-            except Exception:
-                tz_info = "Local Time"
-
-            # Format datetime components
-            date_str = now.strftime("%Y-%m-%d")
-            time_str = now.strftime("%H:%M:%S")
-            day_name = now.strftime("%A")
+            from claude_mpm.core.output_style_manager import OutputStyleManager
 
-            context_lines.append(
-                f"**Current DateTime**: {date_str} {time_str} {tz_info}\n"
-            )
-            context_lines.append(f"**Day**: {day_name}\n")
+            self.output_style_manager = OutputStyleManager()
+            self._log_output_style_status()
 
-        except Exception as e:
-            # Fallback to basic date if enhanced datetime fails
-            self.logger.debug(f"Error generating enhanced datetime context: {e}")
-            context_lines.append(
-                f"**Today's Date**: {datetime.now(timezone.utc).strftime('%Y-%m-%d')}\n"
+            # Extract and save output style content
+            output_style_content = (
+                self.output_style_manager.extract_output_style_content(
+                    framework_loader=self
+                )
             )
+            self.output_style_manager.save_output_style(output_style_content)
 
-        try:
-            # Get user information with safe fallbacks
-            username = None
-
-            # Try multiple methods to get username
-            methods = [
-                lambda: os.environ.get("USER"),
-                lambda: os.environ.get("USERNAME"),  # Windows fallback
-                lambda: getpass.getuser(),
-            ]
-
-            for method in methods:
-                try:
-                    username = method()
-                    if username:
-                        break
-                except Exception:
-                    continue
-
-            if username:
-                context_lines.append(f"**User**: {username}\n")
-
-                # Add home directory if available
-                try:
-                    home_dir = os.path.expanduser("~")
-                    if home_dir and home_dir != "~":
-                        context_lines.append(f"**Home Directory**: {home_dir}\n")
-                except Exception:
-                    pass
-
-        except Exception as e:
-            # User detection is optional, don't fail
-            self.logger.debug(f"Could not detect user information: {e}")
-
-        try:
-            # Get system information
-            system_info = platform.system()
-            if system_info:
-                # Enhance system name for common platforms
-                system_names = {
-                    "Darwin": "Darwin (macOS)",
-                    "Linux": "Linux",
-                    "Windows": "Windows",
-                }
-                system_display = system_names.get(system_info, system_info)
-                context_lines.append(f"**System**: {system_display}\n")
-
-                # Add platform version if available
-                try:
-                    platform_version = platform.release()
-                    if platform_version:
-                        context_lines.append(
-                            f"**System Version**: {platform_version}\n"
-                        )
-                except Exception:
-                    pass
-
-        except Exception as e:
-            # System info is optional
-            self.logger.debug(f"Could not detect system information: {e}")
-
-        try:
-            # Add current working directory
-            cwd = os.getcwd()
-            if cwd:
-                context_lines.append(f"**Working Directory**: {cwd}\n")
-        except Exception:
-            pass
-
-        try:
-            # Add locale information if available
-            current_locale = locale.getlocale()
-            if current_locale and current_locale[0]:
-                context_lines.append(f"**Locale**: {current_locale[0]}\n")
-        except Exception:
-            # Locale is optional
-            pass
-
-        # Add instruction for applying context
-        context_lines.append(
-            "\nApply temporal and user awareness to all tasks, "
-            "decisions, and interactions.\n"
-        )
-        context_lines.append(
-            "Use this context for personalized responses and "
-            "time-sensitive operations.\n"
-        )
-
-        return "".join(context_lines)
-
-    def _parse_agent_metadata(self, agent_file: Path) -> Optional[Dict[str, Any]]:
-        """Parse agent metadata from deployed agent file.
-        Uses caching based on file path and modification time.
-
-        Returns:
-            Dictionary with agent metadata directly from YAML frontmatter.
-        """
-        try:
-            # Check cache based on file path and modification time
-            cache_key = str(agent_file)
-            file_mtime = agent_file.stat().st_mtime
-            time.time()
-
-            # Try to get from cache first
-            cached_result = self._cache_manager.get_agent_metadata(cache_key)
-            if cached_result is not None:
-                cached_data, cached_mtime = cached_result
-                # Use cache if file hasn't been modified and cache isn't too old
-                if cached_mtime == file_mtime:
-                    self.logger.debug(f"Using cached metadata for {agent_file.name}")
-                    return cached_data
-
-            # Cache miss or expired - parse the file
-            self.logger.debug(
-                f"Parsing metadata for {agent_file.name} (cache miss or expired)"
+            # Deploy to Claude Code if supported
+            deployed = self.output_style_manager.deploy_output_style(
+                output_style_content
             )
 
-            import yaml
-
-            with open(agent_file) as f:
-                content = f.read()
-
-            # Default values
-            agent_data = {
-                "id": agent_file.stem,
-                "display_name": agent_file.stem.replace("_", " ")
-                .replace("-", " ")
-                .title(),
-                "description": "Specialized agent",
-            }
-
-            # Extract YAML frontmatter if present
-            if content.startswith("---"):
-                end_marker = content.find("---", 3)
-                if end_marker > 0:
-                    frontmatter = content[3:end_marker]
-                    metadata = yaml.safe_load(frontmatter)
-                    if metadata:
-                        # Use name as ID for Task tool
-                        agent_data["id"] = metadata.get("name", agent_data["id"])
-                        agent_data["display_name"] = (
-                            metadata.get("name", agent_data["display_name"])
-                            .replace("-", " ")
-                            .title()
-                        )
-
-                        # Copy all metadata fields directly
-                        for key, value in metadata.items():
-                            if key not in ["name"]:  # Skip already processed fields
-                                agent_data[key] = value
-
-                        # IMPORTANT: Do NOT add spaces to tools field - it breaks deployment!
-                        # Tools must remain as comma-separated without spaces: "Read,Write,Edit"
-
-            # Try to load routing metadata from JSON template if not in YAML frontmatter
-            if "routing" not in agent_data:
-                routing_data = self._load_routing_from_template(agent_file.stem)
-                if routing_data:
-                    agent_data["routing"] = routing_data
-
-            # Try to load memory routing metadata from JSON template if not in YAML frontmatter
-            if "memory_routing" not in agent_data:
-                memory_routing_data = self._load_memory_routing_from_template(
-                    agent_file.stem
-                )
-                if memory_routing_data:
-                    agent_data["memory_routing"] = memory_routing_data
-
-            # Cache the parsed metadata
-            self._cache_manager.set_agent_metadata(cache_key, agent_data, file_mtime)
-
-            return agent_data
+            if deployed:
+                self.logger.info("✅ Output style deployed to Claude Code >= 1.0.83")
+            else:
+                self.logger.info("📝 Output style will be injected into instructions")
 
         except Exception as e:
-            self.logger.debug(f"Could not parse metadata from {agent_file}: {e}")
-            return None
-
-    def _load_memory_routing_from_template(
-        self, agent_name: str
-    ) -> Optional[Dict[str, Any]]:
-        """Load memory routing metadata from agent JSON template.
-
-        Args:
-            agent_name: Name of the agent (stem of the file)
-
-        Returns:
-            Dictionary with memory routing metadata or None if not found
-        """
-        try:
-            import json
-
-            # Check if we have a framework path
-            if not self.framework_path or self.framework_path == Path("__PACKAGED__"):
-                # For packaged installations, try to load from package resources
-                if files:
-                    try:
-                        templates_package = files("claude_mpm.agents.templates")
-                        template_file = templates_package / f"{agent_name}.json"
-
-                        if template_file.is_file():
-                            template_content = template_file.read_text()
-                            template_data = json.loads(template_content)
-                            return template_data.get("memory_routing")
-                    except Exception as e:
-                        self.logger.debug(
-                            f"Could not load memory routing from packaged template for {agent_name}: {e}"
-                        )
-                return None
-
-            # For development mode, load from filesystem
-            templates_dir = (
-                self.framework_path / "src" / "claude_mpm" / "agents" / "templates"
-            )
-            template_file = templates_dir / f"{agent_name}.json"
-
-            if template_file.exists():
-                with open(template_file) as f:
-                    template_data = json.load(f)
-                    return template_data.get("memory_routing")
-
-            # Also check for variations in naming (underscore vs dash)
-            # Handle common naming variations between deployed .md files and .json templates
-            # Remove duplicates by using a set
-            alternative_names = list(
-                {
-                    agent_name.replace("-", "_"),  # api-qa -> api_qa
-                    agent_name.replace("_", "-"),  # api_qa -> api-qa
-                    agent_name.replace("-", ""),  # api-qa -> apiqa
-                    agent_name.replace("_", ""),  # api_qa -> apiqa
-                    agent_name.replace("-agent", ""),  # research-agent -> research
-                    agent_name.replace("_agent", ""),  # research_agent -> research
-                    agent_name + "_agent",  # research -> research_agent
-                    agent_name + "-agent",  # research -> research-agent
-                }
-            )
-
-            for alt_name in alternative_names:
-                if alt_name != agent_name:  # Skip the original name we already tried
-                    alt_file = templates_dir / f"{alt_name}.json"
-                    if alt_file.exists():
-                        with open(alt_file) as f:
-                            template_data = json.load(f)
-                            return template_data.get("memory_routing")
-
-            return None
+            self.logger.warning(f"❌ Failed to initialize output style manager: {e}")
 
-        except Exception as e:
-            self.logger.debug(
-                f"Could not load memory routing from template for {agent_name}: {e}"
-            )
-            return None
+    def _log_output_style_status(self) -> None:
+        """Log output style status information."""
+        if not self.output_style_manager:
+            return
 
-    def _discover_local_json_templates(self) -> Dict[str, Dict[str, Any]]:
-        """Discover local JSON agent templates from .claude-mpm/agents/ directories.
+        claude_version = self.output_style_manager.claude_version
+        if claude_version:
+            self.logger.info(f"Claude Code version detected: {claude_version}")
 
-        Returns:
-            Dictionary mapping agent IDs to agent metadata
-        """
-        import json
-        from pathlib import Path
-
-        local_agents = {}
-
-        # Check for local JSON templates in priority order
-        template_dirs = [
-            Path.cwd()
-            / ".claude-mpm"
-            / "agents",  # Project local agents (highest priority)
-            Path.home() / ".claude-mpm" / "agents",  # User local agents
-        ]
-
-        for priority, template_dir in enumerate(template_dirs):
-            if not template_dir.exists():
-                continue
-
-            for json_file in template_dir.glob("*.json"):
-                try:
-                    with open(json_file) as f:
-                        template_data = json.load(f)
-
-                    # Extract agent metadata
-                    agent_id = template_data.get("agent_id", json_file.stem)
-
-                    # Skip if already found at higher priority
-                    if agent_id in local_agents:
-                        continue
-
-                    # Extract metadata
-                    metadata = template_data.get("metadata", {})
-
-                    # Build agent data in expected format
-                    agent_data = {
-                        "id": agent_id,
-                        "display_name": metadata.get(
-                            "name", agent_id.replace("_", " ").title()
-                        ),
-                        "description": metadata.get(
-                            "description", f"Local {agent_id} agent"
-                        ),
-                        "tools": self._extract_tools_from_template(template_data),
-                        "is_local": True,
-                        "tier": "project" if priority == 0 else "user",
-                        "author": template_data.get("author", "local"),
-                        "version": template_data.get("agent_version", "1.0.0"),
-                    }
-
-                    local_agents[agent_id] = agent_data
-                    self.logger.debug(
-                        f"Discovered local JSON agent: {agent_id} from {template_dir}"
+            if self.output_style_manager.supports_output_styles():
+                self.logger.info("✅ Claude Code supports output styles (>= 1.0.83)")
+                output_style_path = self.output_style_manager.output_style_path
+                if output_style_path.exists():
+                    self.logger.info(
+                        f"📁 Output style file exists: {output_style_path}"
                     )
-
-                except Exception as e:
-                    self.logger.warning(
-                        f"Failed to parse local JSON template {json_file}: {e}"
+                else:
+                    self.logger.info(
+                        f"📝 Output style will be created at: {output_style_path}"
                     )
+            else:
+                self.logger.info(
+                    f"⚠️ Claude Code {claude_version} does not support output styles"
+                )
+                self.logger.info(
+                    "📝 Output style will be injected into framework instructions"
+                )
+        else:
+            self.logger.info("⚠️ Claude Code not detected or version unknown")
+            self.logger.info("📝 Output style will be injected as fallback")
 
-        return local_agents
-
-    def _extract_tools_from_template(self, template_data: Dict[str, Any]) -> str:
-        """Extract tools string from template data.
-
-        Args:
-            template_data: JSON template data
-
-        Returns:
-            Tools string for display
-        """
-        capabilities = template_data.get("capabilities", {})
-        tools = capabilities.get("tools", "*")
-
-        if tools == "*":
-            return "All Tools"
-        if isinstance(tools, list):
-            return ", ".join(tools) if tools else "Standard Tools"
-        if isinstance(tools, str):
-            if "," in tools:
-                return tools
-            return tools
-        return "Standard Tools"
-
-    def _load_routing_from_template(self, agent_name: str) -> Optional[Dict[str, Any]]:
-        """Load routing metadata from agent JSON template.
-
-        Args:
-            agent_name: Name of the agent (stem of the file)
+    # === Logging Methods ===
 
-        Returns:
-            Dictionary with routing metadata or None if not found
-        """
+    def _log_system_prompt(self) -> None:
+        """Log the system prompt if LogManager is available."""
         try:
-            import json
-
-            # Check if we have a framework path
-            if not self.framework_path or self.framework_path == Path("__PACKAGED__"):
-                # For packaged installations, try to load from package resources
-                if files:
-                    try:
-                        templates_package = files("claude_mpm.agents.templates")
-                        template_file = templates_package / f"{agent_name}.json"
-
-                        if template_file.is_file():
-                            template_content = template_file.read_text()
-                            template_data = json.loads(template_content)
-                            return template_data.get("routing")
-                    except Exception as e:
-                        self.logger.debug(
-                            f"Could not load routing from packaged template for {agent_name}: {e}"
-                        )
-                return None
-
-            # For development mode, load from filesystem
-            templates_dir = (
-                self.framework_path / "src" / "claude_mpm" / "agents" / "templates"
-            )
-            template_file = templates_dir / f"{agent_name}.json"
-
-            if template_file.exists():
-                with open(template_file) as f:
-                    template_data = json.load(f)
-                    return template_data.get("routing")
-
-            # Also check for variations in naming (underscore vs dash)
-            # Handle common naming variations between deployed .md files and .json templates
-            # Remove duplicates by using a set
-            alternative_names = list(
-                {
-                    agent_name.replace("-", "_"),  # api-qa -> api_qa
-                    agent_name.replace("_", "-"),  # api_qa -> api-qa
-                    agent_name.replace("-", ""),  # api-qa -> apiqa
-                    agent_name.replace("_", ""),  # api_qa -> apiqa
-                }
-            )
-
-            for alt_name in alternative_names:
-                if alt_name != agent_name:
-                    alt_file = templates_dir / f"{alt_name}.json"
-                    if alt_file.exists():
-                        with open(alt_file) as f:
-                            template_data = json.load(f)
-                            return template_data.get("routing")
-
-            self.logger.debug(f"No JSON template found for agent: {agent_name}")
-            return None
-
-        except Exception as e:
-            self.logger.debug(f"Could not load routing metadata for {agent_name}: {e}")
-            return None
-
-    def _generate_agent_selection_guide(self, deployed_agents: list) -> str:
-        """Generate Context-Aware Agent Selection guide from deployed agents.
+            from .log_manager import get_log_manager
 
-        Creates a mapping of task types to appropriate agents based on their
-        descriptions and capabilities.
-        """
-        guide = ""
+            log_manager = get_log_manager()
+        except ImportError:
+            return
 
-        # Build selection mapping based on deployed agents
-        selection_map = {}
+        try:
+            # Get or create event loop
+            try:
+                loop = asyncio.get_running_loop()
+            except RuntimeError:
+                loop = asyncio.new_event_loop()
+                asyncio.set_event_loop(loop)
+
+            # Prepare metadata
+            metadata = {
+                "framework_version": self.framework_version,
+                "framework_loaded": self.framework_content.get("loaded", False),
+                "session_id": os.environ.get("CLAUDE_SESSION_ID", "unknown"),
+            }
 
-        for agent in deployed_agents:
-            agent_id = agent["id"]
-            desc_lower = agent["description"].lower()
+            # Log the prompt asynchronously
+            instructions = (
+                self._format_full_framework()
+                if self.framework_content["loaded"]
+                else self._format_minimal_framework()
+            )
+            metadata["instructions_length"] = len(instructions)
 
-            # Map task types to agents based on their descriptions
-            if "implementation" in desc_lower or (
-                "engineer" in agent_id and "data" not in agent_id
-            ):
-                selection_map["Implementation tasks"] = (
-                    f"{agent['display_name']} (`{agent_id}`)"
-                )
-            if "codebase analysis" in desc_lower or "research" in agent_id:
-                selection_map["Codebase analysis"] = (
-                    f"{agent['display_name']} (`{agent_id}`)"
-                )
-            if "testing" in desc_lower or "qa" in agent_id:
-                selection_map["Testing/quality"] = (
-                    f"{agent['display_name']} (`{agent_id}`)"
-                )
-            if "documentation" in desc_lower:
-                selection_map["Documentation"] = (
-                    f"{agent['display_name']} (`{agent_id}`)"
-                )
-            if "security" in desc_lower or "sast" in desc_lower:
-                selection_map["Security operations"] = (
-                    f"{agent['display_name']} (`{agent_id}`)"
-                )
-            if (
-                "deployment" in desc_lower
-                or "infrastructure" in desc_lower
-                or "ops" in agent_id
-            ):
-                selection_map["Deployment/infrastructure"] = (
-                    f"{agent['display_name']} (`{agent_id}`)"
-                )
-            if "data" in desc_lower and (
-                "pipeline" in desc_lower or "etl" in desc_lower
-            ):
-                selection_map["Data pipeline/ETL"] = (
-                    f"{agent['display_name']} (`{agent_id}`)"
+            if loop.is_running():
+                asyncio.create_task(
+                    log_manager.log_prompt("system_prompt", instructions, metadata)
                 )
-            if "git" in desc_lower or "version control" in desc_lower:
-                selection_map["Version control"] = (
-                    f"{agent['display_name']} (`{agent_id}`)"
-                )
-            if "ticket" in desc_lower or "epic" in desc_lower:
-                selection_map["Ticket/issue management"] = (
-                    f"{agent['display_name']} (`{agent_id}`)"
-                )
-            if "browser" in desc_lower or "e2e" in desc_lower:
-                selection_map["Browser/E2E testing"] = (
-                    f"{agent['display_name']} (`{agent_id}`)"
-                )
-            if "frontend" in desc_lower or "ui" in desc_lower or "html" in desc_lower:
-                selection_map["Frontend/UI development"] = (
-                    f"{agent['display_name']} (`{agent_id}`)"
+            else:
+                loop.run_until_complete(
+                    log_manager.log_prompt("system_prompt", instructions, metadata)
                 )
 
-        # Always include PM questions
-        selection_map["PM questions"] = "Answer directly (only exception)"
-
-        # Format the selection guide
-        for task_type, agent_info in selection_map.items():
-            guide += f"- **{task_type}** → {agent_info}\n"
-
-        return guide
-
-    def _get_fallback_capabilities(self) -> str:
-        """Return fallback capabilities when dynamic discovery fails."""
-        return """
-
-## Available Agent Capabilities
-
-You have the following specialized agents available for delegation:
-
-- **Engineer** (`engineer`): Code implementation and development
-- **Research** (`research-agent`): Investigation and analysis
-- **QA** (`qa-agent`): Testing and quality assurance
-- **Documentation** (`documentation-agent`): Documentation creation and maintenance
-- **Security** (`security-agent`): Security analysis and protection
-- **Data Engineer** (`data-engineer`): Data management and pipelines
-- **Ops** (`ops-agent`): Deployment and operations
-- **Version Control** (`version-control`): Git operations and version management
+            self.logger.debug("System prompt logged to prompts directory")
+        except Exception as e:
+            self.logger.debug(f"Could not log system prompt: {e}")
 
-**IMPORTANT**: Use the exact agent ID in parentheses when delegating tasks.
-"""
+    # === Agent Registry Methods (backward compatibility) ===
 
-    def _format_minimal_framework(self) -> str:
-        """Format minimal framework instructions when full framework not available."""
-        return """
-# Claude PM Framework Instructions
-
-You are operating within a Claude PM Framework deployment.
-
-## Role
-You are a multi-agent orchestrator. Your primary responsibilities:
-- Delegate tasks to specialized agents via Task Tool
-- Coordinate multi-agent workflows
-- Extract TODO/BUG/FEATURE items for ticket creation
-- NEVER perform direct implementation work
-
-## Core Agents
-- Documentation Agent - Documentation tasks
-- Engineer Agent - Code implementation
-- QA Agent - Testing and validation
-- Research Agent - Investigation and analysis
-- Version Control Agent - Git operations
-
-## Important Rules
-1. Always delegate work via Task Tool
-2. Provide comprehensive context to agents
-3. Track all TODO/BUG/FEATURE items
-4. Maintain project visibility
-
----
-"""
-
-    def get_agent_list(self) -> list:
+    def get_agent_list(self) -> List[str]:
         """Get list of available agents."""
-        # First try agent registry
         if self.agent_registry:
             agents = self.agent_registry.list_agents()
             if agents:
                 return list(agents.keys())
-
-        # Fallback to loaded content
         return list(self.framework_content["agents"].keys())
 
     def get_agent_definition(self, agent_name: str) -> Optional[str]:
         """Get specific agent definition."""
-        # First try agent registry
         if self.agent_registry:
             definition = self.agent_registry.get_agent_definition(agent_name)
             if definition:
                 return definition
-
-        # Fallback to loaded content
         return self.framework_content["agents"].get(agent_name)
 
-    def get_agent_hierarchy(self) -> Dict[str, list]:
+    def get_agent_hierarchy(self) -> Dict[str, List]:
         """Get agent hierarchy from registry."""
         if self.agent_registry:
             return self.agent_registry.get_agent_hierarchy()