claude-mpm 5.1.9__py3-none-any.whl → 5.4.48__py3-none-any.whl

claude_mpm/services/skills/skill_to_agent_mapper.py

@@ -0,0 +1,406 @@
+"""Service for mapping skills to agents based on YAML configuration.
+
+WHY: Progressive skills discovery requires knowing which agents need which skills.
+This service uses a YAML configuration to map skill paths to agent IDs, enabling
+selective skill deployment based on agent requirements.
+
+DESIGN DECISIONS:
+- Load YAML configuration with skill_path -> [agent_ids] mappings
+- Handle ALL_AGENTS marker expansion from YAML anchor
+- Build inverse index (agent_id -> [skill_paths]) for efficient lookup
+- Support pattern-based inference for unmatched skill paths
+- Cache configuration to avoid repeated file I/O
+
+YAML Configuration Format:
+    skill_mappings:
+      toolchains/python/frameworks/django:
+        - python-engineer
+        - data-engineer
+        - engineer
+
+      universal/collaboration/git-workflow: *all_agents
+
+    inference_rules:
+      language_patterns:
+        python: [python-engineer, data-engineer, engineer]
+      framework_patterns:
+        django: [python-engineer, engineer]
+
+    all_agents_list:
+      - engineer
+      - python-engineer
+      - typescript-engineer
+      ...
+
+References:
+- Feature: Progressive skills discovery (#117)
+- Research: docs/research/skill-path-to-agent-mapping-2025-12-16.md
+- Config: src/claude_mpm/config/skill_to_agent_mapping.yaml
+"""
+
+from pathlib import Path
+from typing import Any, Dict, List, Optional, Set
+
+import yaml
+
+from claude_mpm.core.logging_config import get_logger
+
+logger = get_logger(__name__)
+
+
+class SkillToAgentMapper:
+    """Maps skills to agents using YAML configuration.
+
+    This service provides bidirectional mapping between skill paths and agent IDs:
+    - Forward: skill_path -> [agent_ids]
+    - Inverse: agent_id -> [skill_paths]
+
+    The service uses a YAML configuration file with explicit mappings and
+    pattern-based inference rules for skill paths not explicitly mapped.
+
+    Example:
+        >>> mapper = SkillToAgentMapper()
+        >>> agents = mapper.get_agents_for_skill('toolchains/python/frameworks/django')
+        >>> print(agents)
+        ['python-engineer', 'data-engineer', 'engineer', 'api-qa']
+
+        >>> skills = mapper.get_skills_for_agent('python-engineer')
+        >>> print(f"Found {len(skills)} skills for python-engineer")
+    """
+
+    # Default configuration path (relative to package root)
+    DEFAULT_CONFIG_PATH = (
+        Path(__file__).parent.parent.parent / "config" / "skill_to_agent_mapping.yaml"
+    )
+
+    def __init__(self, config_path: Optional[Path] = None):
+        """Initialize skill-to-agent mapper.
+
+        Args:
+            config_path: Optional path to YAML config file.
+                        If None, uses default config from package.
+
+        Raises:
+            FileNotFoundError: If config file not found
+            yaml.YAMLError: If config file is invalid YAML
+            ValueError: If config file is missing required sections
+        """
+        self.config_path = config_path or self.DEFAULT_CONFIG_PATH
+        self.logger = get_logger(__name__)
+
+        # Load and validate configuration
+        self._config = self._load_config()
+
+        # Build forward and inverse indexes
+        self._skill_to_agents: Dict[str, List[str]] = {}
+        self._agent_to_skills: Dict[str, List[str]] = {}
+        self._build_indexes()
+
+        self.logger.info(
+            f"SkillToAgentMapper initialized: {len(self._skill_to_agents)} skill mappings, "
+            f"{len(self._agent_to_skills)} agents"
+        )
+
+    def _load_config(self) -> Dict[str, Any]:
+        """Load and validate YAML configuration.
+
+        Returns:
+            Parsed YAML configuration
+
+        Raises:
+            FileNotFoundError: If config file not found
+            yaml.YAMLError: If config file is invalid YAML
+            ValueError: If config file is missing required sections
+        """
+        if not self.config_path.exists():
+            raise FileNotFoundError(f"Configuration file not found: {self.config_path}")
+
+        try:
+            with open(self.config_path, encoding="utf-8") as f:
+                config = yaml.safe_load(f)
+        except yaml.YAMLError as e:
+            raise yaml.YAMLError(f"Invalid YAML in {self.config_path}: {e}") from e
+
+        # Validate required sections
+        if not isinstance(config, dict):
+            raise ValueError("Configuration must be a YAML dictionary")
+
+        if "skill_mappings" not in config:
+            raise ValueError("Configuration missing required section: skill_mappings")
+
+        if "all_agents_list" not in config:
+            raise ValueError("Configuration missing required section: all_agents_list")
+
+        self.logger.debug(f"Loaded configuration from {self.config_path}")
+        return config
+
+    def _build_indexes(self) -> None:
+        """Build forward and inverse mapping indexes.
+
+        Processes skill_mappings from config and expands ALL_AGENTS markers.
+        Builds bidirectional indexes for efficient lookup.
+
+        Index Structure:
+            _skill_to_agents: {"skill/path": ["agent1", "agent2", ...]}
+            _agent_to_skills: {"agent1": ["skill/path1", "skill/path2", ...]}
+        """
+        skill_mappings = self._config["skill_mappings"]
+        all_agents = self._config["all_agents_list"]
+
+        for skill_path, agent_list in skill_mappings.items():
+            # Handle ALL_AGENTS marker expansion
+            if (
+                isinstance(agent_list, list)
+                and len(agent_list) == 1
+                and agent_list[0] == "ALL_AGENTS"
+            ):
+                expanded_agents = all_agents.copy()
+                self.logger.debug(
+                    f"Expanded ALL_AGENTS for {skill_path}: {len(expanded_agents)} agents"
+                )
+            else:
+                expanded_agents = agent_list
+
+            # Ensure agent_list is actually a list
+            if not isinstance(expanded_agents, list):
+                self.logger.warning(
+                    f"Invalid agent list for {skill_path}: {type(expanded_agents)}. Skipping."
+                )
+                continue
+
+            # Build forward index: skill -> agents
+            self._skill_to_agents[skill_path] = expanded_agents
+
+            # Build inverse index: agent -> skills
+            for agent_id in expanded_agents:
+                if agent_id not in self._agent_to_skills:
+                    self._agent_to_skills[agent_id] = []
+                self._agent_to_skills[agent_id].append(skill_path)
+
+        self.logger.debug(
+            f"Built indexes: {len(self._skill_to_agents)} skills, {len(self._agent_to_skills)} agents"
+        )
+
+    def get_agents_for_skill(self, skill_path: str) -> List[str]:
+        """Get list of agent IDs for a skill path.
+
+        Looks up skill path in configuration. If not found, attempts to infer
+        agents using pattern-based rules.
+
+        Args:
+            skill_path: Skill path (e.g., "toolchains/python/frameworks/django")
+
+        Returns:
+            List of agent IDs that should receive this skill.
+            Empty list if no mapping found and inference fails.
+
+        Example:
+            >>> agents = mapper.get_agents_for_skill('toolchains/python/frameworks/django')
+            >>> print(agents)
+            ['python-engineer', 'data-engineer', 'engineer', 'api-qa']
+
+            >>> # Fallback to inference
+            >>> agents = mapper.get_agents_for_skill('toolchains/python/new-framework')
+            >>> print(agents)
+            ['python-engineer', 'data-engineer', 'engineer']
+        """
+        # Try exact match first
+        if skill_path in self._skill_to_agents:
+            return self._skill_to_agents[skill_path].copy()
+
+        # Fallback to pattern-based inference
+        inferred_agents = self.infer_agents_from_pattern(skill_path)
+        if inferred_agents:
+            self.logger.debug(
+                f"Inferred {len(inferred_agents)} agents for unmapped skill: {skill_path}"
+            )
+            return inferred_agents
+
+        # No mapping or inference available
+        self.logger.debug(f"No mapping or inference available for skill: {skill_path}")
+        return []
+
+    def get_skills_for_agent(self, agent_id: str) -> List[str]:
+        """Get list of skill paths for an agent (inverse lookup).
+
+        Args:
+            agent_id: Agent identifier (e.g., "python-engineer")
+
+        Returns:
+            List of skill paths assigned to this agent.
+            Empty list if agent not found in configuration.
+
+        Example:
+            >>> skills = mapper.get_skills_for_agent('python-engineer')
+            >>> print(f"Found {len(skills)} skills")
+            >>> for skill in skills[:5]:
+            ...     print(f"  - {skill}")
+        """
+        if agent_id not in self._agent_to_skills:
+            self.logger.debug(f"No skills found for agent: {agent_id}")
+            return []
+
+        return self._agent_to_skills[agent_id].copy()
+
+    def infer_agents_from_pattern(self, skill_path: str) -> List[str]:
+        """Infer agents for a skill path using pattern matching.
+
+        Uses inference_rules from configuration to match skill paths against
+        language, framework, and domain patterns.
+
+        Pattern Matching Algorithm:
+        1. Extract path components (language, framework, domain)
+        2. Match against language_patterns (e.g., "python" -> python-engineer)
+        3. Match against framework_patterns (e.g., "django" -> django agents)
+        4. Match against domain_patterns (e.g., "testing" -> qa agents)
+        5. Combine and deduplicate results
+
+        Args:
+            skill_path: Skill path to infer agents for
+
+        Returns:
+            List of inferred agent IDs, or empty list if no patterns match
+
+        Example:
+            >>> # Infer from language pattern
+            >>> agents = mapper.infer_agents_from_pattern('toolchains/python/new-lib')
+            >>> 'python-engineer' in agents
+            True
+
+            >>> # Infer from framework pattern
+            >>> agents = mapper.infer_agents_from_pattern('toolchains/typescript/frameworks/nextjs-advanced')
+            >>> 'nextjs-engineer' in agents
+            True
+        """
+        if "inference_rules" not in self._config:
+            return []
+
+        inference_rules = self._config["inference_rules"]
+        inferred_agents: Set[str] = set()
+
+        # Normalize skill path for matching (lowercase, split on /)
+        path_parts = skill_path.lower().split("/")
+
+        # Match language patterns
+        if "language_patterns" in inference_rules:
+            for language, agents in inference_rules["language_patterns"].items():
+                if language in path_parts:
+                    inferred_agents.update(agents)
+                    self.logger.debug(
+                        f"Matched language pattern '{language}' in {skill_path}"
+                    )
+
+        # Match framework patterns
+        if "framework_patterns" in inference_rules:
+            for framework, agents in inference_rules["framework_patterns"].items():
+                # Match framework name anywhere in path (e.g., "nextjs" in path)
+                if any(framework in part for part in path_parts):
+                    inferred_agents.update(agents)
+                    self.logger.debug(
+                        f"Matched framework pattern '{framework}' in {skill_path}"
+                    )
+
+        # Match domain patterns
+        if "domain_patterns" in inference_rules:
+            for domain, agents in inference_rules["domain_patterns"].items():
+                if domain in path_parts:
+                    inferred_agents.update(agents)
+                    self.logger.debug(
+                        f"Matched domain pattern '{domain}' in {skill_path}"
+                    )
+
+        return sorted(inferred_agents)
+
+    def get_all_mapped_skills(self) -> List[str]:
+        """Get all skill paths with explicit mappings.
+
+        Returns:
+            List of all skill paths in configuration (sorted)
+
+        Example:
+            >>> skills = mapper.get_all_mapped_skills()
+            >>> print(f"Total mapped skills: {len(skills)}")
+        """
+        return sorted(self._skill_to_agents.keys())
+
+    def get_all_agents(self) -> List[str]:
+        """Get all agent IDs referenced in mappings.
+
+        Returns:
+            List of all agent IDs in configuration (sorted)
+
+        Example:
+            >>> agents = mapper.get_all_agents()
+            >>> print(f"Total agents: {len(agents)}")
+        """
+        return sorted(self._agent_to_skills.keys())
+
+    def is_skill_mapped(self, skill_path: str) -> bool:
+        """Check if skill path has an explicit mapping.
+
+        Args:
+            skill_path: Skill path to check
+
+        Returns:
+            True if skill has explicit mapping, False otherwise
+
+        Example:
+            >>> mapper.is_skill_mapped('toolchains/python/frameworks/django')
+            True
+            >>> mapper.is_skill_mapped('toolchains/python/unknown')
+            False
+        """
+        return skill_path in self._skill_to_agents
+
+    def get_mapping_stats(self) -> Dict[str, Any]:
+        """Get statistics about skill-to-agent mappings.
+
+        Returns:
+            Dictionary with mapping statistics:
+            {
+                "total_skills": int,
+                "total_agents": int,
+                "avg_agents_per_skill": float,
+                "avg_skills_per_agent": float,
+                "config_path": str,
+                "config_version": str
+            }
+
+        Example:
+            >>> stats = mapper.get_mapping_stats()
+            >>> print(f"Total skills: {stats['total_skills']}")
+            >>> print(f"Total agents: {stats['total_agents']}")
+        """
+        total_skills = len(self._skill_to_agents)
+        total_agents = len(self._agent_to_skills)
+
+        # Calculate averages
+        avg_agents_per_skill = (
+            sum(len(agents) for agents in self._skill_to_agents.values()) / total_skills
+            if total_skills > 0
+            else 0.0
+        )
+
+        avg_skills_per_agent = (
+            sum(len(skills) for skills in self._agent_to_skills.values()) / total_agents
+            if total_agents > 0
+            else 0.0
+        )
+
+        return {
+            "total_skills": total_skills,
+            "total_agents": total_agents,
+            "avg_agents_per_skill": round(avg_agents_per_skill, 2),
+            "avg_skills_per_agent": round(avg_skills_per_agent, 2),
+            "config_path": str(self.config_path),
+            "config_version": self._config.get("metadata", {}).get(
+                "version", "unknown"
+            ),
+        }
+
+    def __repr__(self) -> str:
+        """Return string representation."""
+        return (
+            f"SkillToAgentMapper(skills={len(self._skill_to_agents)}, "
+            f"agents={len(self._agent_to_skills)})"
+        )

claude_mpm/services/skills_deployer.py

@@ -82,6 +82,8 @@ class SkillsDeployerService(LoggerMixin):
         toolchain: Optional[List[str]] = None,
         categories: Optional[List[str]] = None,
         force: bool = False,
+        selective: bool = True,
+        project_root: Optional[Path] = None,
     ) -> Dict:
         """Deploy skills from GitHub repository.
 
@@ -89,14 +91,17 @@ class SkillsDeployerService(LoggerMixin):
         1. Downloads skills from GitHub collection
         2. Parses manifest for metadata
         3. Filters by toolchain and categories
-        4. Deploys to ~/.claude/skills/
-        5. Warns about Claude Code restart
+        4. (If selective=True) Filters to only agent-referenced skills
+        5. Deploys to ~/.claude/skills/
+        6. Warns about Claude Code restart
 
         Args:
             collection: Collection name to deploy from (default: uses default collection)
             toolchain: Filter by toolchain (e.g., ['python', 'javascript'])
             categories: Filter by categories (e.g., ['testing', 'debugging'])
             force: Overwrite existing skills
+            selective: If True, only deploy skills referenced by agents (default)
+            project_root: Project root directory (for finding agents, auto-detected if None)
 
         Returns:
             Dict containing:
@@ -107,10 +112,14 @@ class SkillsDeployerService(LoggerMixin):
             - restart_required: True if Claude Code needs restart
             - restart_instructions: Message about restarting
             - collection: Collection name used for deployment
+            - selective_mode: True if selective deployment was used
+            - total_available: Total skills available before filtering
 
         Example:
             >>> result = deployer.deploy_skills(collection="obra-superpowers")
             >>> result = deployer.deploy_skills(toolchain=['python'])  # Uses default
+            >>> # Deploy all skills (not just agent-referenced)
+            >>> result = deployer.deploy_skills(selective=False)
             >>> if result['restart_required']:
             >>>     print(result['restart_instructions'])
         """
@@ -152,7 +161,7 @@ class SkillsDeployerService(LoggerMixin):
 
         self.logger.info(f"Found {len(skills)} skills in repository")
 
-        # Step 3: Filter skills
+        # Step 3: Filter skills by toolchain and categories
         filtered_skills = self._filter_skills(skills, toolchain, categories)
 
         self.logger.info(
@@ -160,11 +169,68 @@ class SkillsDeployerService(LoggerMixin):
             f" (toolchain={toolchain}, categories={categories})"
         )
 
+        # Step 3.5: Apply selective filtering (only agent-referenced skills)
+        total_available = len(filtered_skills)
+        if selective:
+            # Auto-detect project root if not provided
+            if project_root is None:
+                # Try to find project root by looking for .claude directory
+                # Start from current directory and walk up
+                current = Path.cwd()
+                while current != current.parent:
+                    if (current / ".claude").exists():
+                        project_root = current
+                        break
+                    current = current.parent
+
+            if project_root:
+                agents_dir = Path(project_root) / ".claude" / "agents"
+            else:
+                # Fallback to current directory's .claude/agents
+                agents_dir = Path.cwd() / ".claude" / "agents"
+
+            from claude_mpm.services.skills.selective_skill_deployer import (
+                get_required_skills_from_agents,
+            )
+
+            required_skill_names = get_required_skills_from_agents(agents_dir)
+
+            if required_skill_names:
+                # Filter to only required skills
+                # Match on either 'name' or 'skill_id' field
+                filtered_skills = [
+                    s
+                    for s in filtered_skills
+                    if s.get("name") in required_skill_names
+                    or s.get("skill_id") in required_skill_names
+                ]
+
+                self.logger.info(
+                    f"Selective deployment: {len(filtered_skills)}/{total_available} skills "
+                    f"(agent-referenced only)"
+                )
+            else:
+                self.logger.warning(
+                    f"No skills found in agent frontmatter at {agents_dir}. "
+                    f"Deploying all {total_available} skills."
+                )
+        else:
+            self.logger.info(
+                f"Selective mode disabled: deploying all {total_available} skills"
+            )
+
         # Step 4: Deploy skills
         deployed = []
         skipped = []
         errors = []
 
+        # Extract skill names for cleanup (needed regardless of deployment outcome)
+        filtered_skills_names = [
+            skill["name"]
+            for skill in filtered_skills
+            if isinstance(skill, dict) and "name" in skill
+        ]
+
         for skill in filtered_skills:
             try:
                 # Validate skill is a dictionary
@@ -173,7 +239,9 @@ class SkillsDeployerService(LoggerMixin):
                     errors.append(f"Invalid skill format: {skill}")
                     continue
 
-                result = self._deploy_skill(skill, skills_data["temp_dir"], force=force)
+                result = self._deploy_skill(
+                    skill, skills_data["temp_dir"], collection_name, force=force
+                )
                 if result["deployed"]:
                     deployed.append(skill["name"])
                 elif result["skipped"]:
@@ -189,10 +257,34 @@ class SkillsDeployerService(LoggerMixin):
                 self.logger.error(f"Failed to deploy {skill_name}: {e}")
                 errors.append(f"{skill_name}: {e}")
 
-        # Step 5: Cleanup
+        # Step 5: Cleanup orphaned skills (always run in selective mode)
+        cleanup_result = {"removed_count": 0, "removed_skills": []}
+        if selective:
+            # Get the set of skills that should remain deployed
+            # This is the union of what we just deployed and what was already there
+            try:
+                from claude_mpm.services.skills.selective_skill_deployer import (
+                    cleanup_orphan_skills,
+                )
+
+                # Cleanup orphaned skills not referenced by agents
+                # This runs even if nothing new was deployed to remove stale skills
+                cleanup_result = cleanup_orphan_skills(
+                    self.CLAUDE_SKILLS_DIR, set(filtered_skills_names)
+                )
+
+                if cleanup_result["removed_count"] > 0:
+                    self.logger.info(
+                        f"Removed {cleanup_result['removed_count']} orphaned skills: "
+                        f"{', '.join(cleanup_result['removed_skills'])}"
+                    )
+            except Exception as e:
+                self.logger.warning(f"Failed to cleanup orphaned skills: {e}")
+
+        # Step 6: Cleanup temp directory
         self._cleanup(skills_data["temp_dir"])
 
-        # Step 6: Check if Claude Code restart needed
+        # Step 7: Check if Claude Code restart needed
         restart_required = len(deployed) > 0
         restart_instructions = ""
 
@@ -216,7 +308,8 @@ class SkillsDeployerService(LoggerMixin):
 
         self.logger.info(
             f"Deployment complete: {len(deployed)} deployed, "
-            f"{len(skipped)} skipped, {len(errors)} errors"
+            f"{len(skipped)} skipped, {len(errors)} errors, "
+            f"{cleanup_result['removed_count']} orphaned skills removed"
         )
 
         return {
@@ -228,6 +321,9 @@ class SkillsDeployerService(LoggerMixin):
             "restart_required": restart_required,
             "restart_instructions": restart_instructions,
             "collection": collection_name,
+            "selective_mode": selective,
+            "total_available": total_available,
+            "cleanup": cleanup_result,
         }
 
     def list_available_skills(self, collection: Optional[str] = None) -> Dict:
@@ -412,6 +508,13 @@ class SkillsDeployerService(LoggerMixin):
                 removed.append(skill_name)
                 self.logger.info(f"Removed skill: {skill_name}")
 
+                # Untrack skill from deployment index
+                from claude_mpm.services.skills.selective_skill_deployer import (
+                    untrack_skill,
+                )
+
+                untrack_skill(self.CLAUDE_SKILLS_DIR, skill_name)
+
             except Exception as e:
                 self.logger.error(f"Failed to remove {skill_name}: {e}")
                 errors.append(f"{skill_name}: {e}")
@@ -677,17 +780,25 @@ class SkillsDeployerService(LoggerMixin):
         return filtered
 
     def _deploy_skill(
-        self, skill: Dict, collection_dir: Path, force: bool = False
+        self,
+        skill: Dict,
+        collection_dir: Path,
+        collection_name: str,
+        force: bool = False,
     ) -> Dict:
-        """Deploy a single skill to ~/.claude/skills/.
+        """Deploy a single skill to ~/.claude/skills/ and track deployment.
 
         NOTE: With multi-collection support, skills are now stored in collection
         subdirectories. This method creates symlinks or copies to maintain the
         flat structure that Claude Code expects in ~/.claude/skills/.
 
+        Additionally tracks deployed skills in .mpm-deployed-skills.json index
+        for orphan cleanup functionality.
+
         Args:
             skill: Skill metadata dict
             collection_dir: Collection directory containing skills
+            collection_name: Name of collection (for tracking)
             force: Overwrite if already exists
 
         Returns:
@@ -777,6 +888,13 @@ class SkillsDeployerService(LoggerMixin):
             # NOTE: We use copy instead of symlink to maintain Claude Code compatibility
             shutil.copytree(source_dir, target_dir)
 
+            # Track deployment in index
+            from claude_mpm.services.skills.selective_skill_deployer import (
+                track_deployed_skill,
+            )
+
+            track_deployed_skill(self.CLAUDE_SKILLS_DIR, skill_name, collection_name)
+
             self.logger.debug(
                 f"Deployed {skill_name} from {source_dir} to {target_dir}"
             )

claude_mpm/services/socketio/dashboard_server.py

@@ -152,6 +152,7 @@ class DashboardServer(SocketIOServiceInterface):
 
         # Register handlers for all events we want to relay from monitor to dashboard
         relay_events = [
+            "claude_event",  # Tool events from Claude Code hooks
             "session_started",
             "session_ended",
             "claude_status",