@xagent-ai/cli 1.2.2 → 1.3.0

package/src/skill-invoker.ts

@@ -1,935 +1,960 @@
-import fs from 'fs/promises';
-import path from 'path';
-import os from 'os';
-import { getSkillLoader, SkillInfo, SkillLoader } from './skill-loader.js';
-import { getToolRegistry } from './tools.js';
-import { ExecutionMode, Tool } from './types.js';
-import { getConfigManager } from './config.js';
-
-// Re-export SkillInfo for other modules
-export type { SkillInfo };
-
-/**
- * Track skill execution history for tracking failures
- */
-export class SkillExecutionHistory {
-  private history: Map<string, number> = new Map();
-
-  /**
-   * Get failure count for a task
-   */
-  getFailureCount(taskKey: string): number {
-    return this.history.get(taskKey) || 0;
-  }
-
-  /**
-   * Increment failure count for a task
-   */
-  incrementFailure(taskKey: string): number {
-    const count = this.getFailureCount(taskKey) + 1;
-    this.history.set(taskKey, count);
-    return count;
-  }
-
-  /**
-   * Reset failure count for a task (e.g., after success)
-   */
-  reset(taskKey: string): void {
-    this.history.delete(taskKey);
-  }
-
-  /**
-   * Check if threshold reached
-   */
-  shouldUseFallback(taskKey: string, threshold: number = 2): boolean {
-    return this.getFailureCount(taskKey) >= threshold;
-  }
-}
-
-// Singleton execution history
-const executionHistory = new SkillExecutionHistory();
-
-export function getExecutionHistory(): SkillExecutionHistory {
-  return executionHistory;
-}
-
-export interface SkillExecutionParams {
-  skillId: string;
-  taskDescription: string;
-  inputFile?: string;
-  outputFile?: string;
-  options?: Record<string, any>;
-  /** Task ID for workspace directory naming */
-  taskId?: string;
-}
-
-/**
- * Execution step interface - tells Agent what to do next
- */
-export interface ExecutionStep {
-  step: number;
-  action: string;
-  description: string;
-  command?: string;
-  file?: string;
-  reason: string;
-}
-
-/**
- * Skill execution result - contains guidance and next actions
- */
-export interface SkillExecutionResult {
-  success: boolean;
-  output?: string;
-  error?: string;
-  files?: string[];
-  /** Tells Agent what to do next */
-  nextSteps?: ExecutionStep[];
-  /** Skill type for determining if manual execution is needed */
-  requiresManualExecution?: boolean;
-  /** Workspace directory used, for cleanup */
-  workspaceDir?: string;
-  /** Files to preserve (relative paths), skipped during cleanup */
-  preserveFiles?: string[];
-}
-
-export interface SkillMatcherResult {
-  skill: SkillInfo;
-  confidence: number;
-  matchedKeywords: string[];
-  category: string;
-}
-
-// ============================================================
-// Workspace Utility Functions
-// ============================================================
-
-/**
- * Get workspace directory path
- * @param taskId Task ID for creating unique workspace directory
- * @returns Absolute path to workspace directory
- */
-export function getWorkspaceDir(taskId: string): string {
-  // Try to get from config first
-  try {
-    const configManager = getConfigManager();
-    const config = configManager.getSettings?.();
-    if (config?.workspacePath) {
-      return path.join(config.workspacePath, taskId);
-    }
-  } catch {
-    // Config not available, use default
-  }
-
-  // Default to ~/.xagent/workspace
-  return path.join(os.homedir(), '.xagent', 'workspace', taskId);
-}
-
-/**
- * Get base workspace directory (without task-id)
- */
-export function getBaseWorkspaceDir(): string {
-  try {
-    const configManager = getConfigManager();
-    const config = configManager.getSettings?.();
-    if (config?.workspacePath) {
-      return config.workspacePath;
-    }
-  } catch {
-    // Config not available, use default
-  }
-
-  return path.join(os.homedir(), '.xagent', 'workspace');
-}
-
-/**
- * Get workspace directory description for AI
- * Returns the actual workspace path from config, or default path
- */
-export function getWorkspaceDescription(): string {
-  try {
-    const configManager = getConfigManager();
-    const config = configManager.getSettings?.();
-    if (config?.workspacePath) {
-      return config.workspacePath;
-    }
-  } catch {
-    // Config not available, use default
-  }
-
-  return path.join(os.homedir(), '.xagent', 'workspace');
-}
-
-/**
- * Ensure workspace directory exists
- * @param workspaceDir Workspace directory path
- */
-export async function ensureWorkspaceDir(workspaceDir: string): Promise<void> {
-  await fs.mkdir(workspaceDir, { recursive: true });
-}
-
-/**
- * Clean up workspace directory
- * @param workspaceDir Workspace directory path
- * @param preserveFiles Files to preserve (relative paths)
- */
-export async function cleanupWorkspace(workspaceDir: string, preserveFiles: string[] = []): Promise<void> {
-  try {
-    const entries = await fs.readdir(workspaceDir, { withFileTypes: true });
-
-    for (const entry of entries) {
-      const fullPath = path.join(workspaceDir, entry.name);
-
-      // Skip files to preserve
-      if (preserveFiles.includes(entry.name)) {
-        continue;
-      }
-
-      if (entry.isDirectory()) {
-        // Recursively delete subdirectories
-        await fs.rm(fullPath, { recursive: true, force: true });
-      } else {
-        // Delete files
-        await fs.unlink(fullPath);
-      }
-    }
-  } catch (error: any) {
-    if (error.code !== 'ENOENT') {
-      console.warn(`Workspace cleanup failed: ${error.message}`);
-    }
-  }
-}
-
-/**
- * Determine if workspace should be auto-cleaned based on ExecutionMode
- * @param executionMode Execution mode
- * @returns Whether auto-cleanup should happen
- */
-export function shouldAutoCleanup(executionMode: ExecutionMode): boolean {
-  // YOLO mode: fully automatic, clean up directly
-  if (executionMode === ExecutionMode.YOLO) {
-    return true;
-  }
-  // Other modes require user confirmation
-  return false;
-}
-
-/**
- * Generate cleanup prompt message
- * @param workspaceDir Workspace directory path
- */
-export async function getCleanupInfo(workspaceDir: string): Promise<{ files: string[]; totalSize: string }> {
-  try {
-    const entries = await fs.readdir(workspaceDir, { withFileTypes: true });
-    const files: string[] = [];
-
-    for (const entry of entries) {
-      files.push(entry.name);
-    }
-
-    // Calculate total size
-    let totalSize = 0;
-    for (const entry of entries) {
-      if (entry.isFile()) {
-        const stats = await fs.stat(path.join(workspaceDir, entry.name));
-        totalSize += stats.size;
-      }
-    }
-
-    const formatSize = (bytes: number): string => {
-      if (bytes < 1024) return `${bytes} B`;
-      if (bytes < 1024 * 1024) return `${(bytes / 1024).toFixed(2)} KB`;
-      return `${(bytes / (1024 * 1024)).toFixed(2)} MB`;
-    };
-
-    return { files, totalSize: formatSize(totalSize) };
-  } catch {
-    return { files: [], totalSize: '0 B' };
-  }
-}
-
-// ============================================================
-// Shared Content Extraction Utilities
-// ============================================================
-
-/**
- * Remove Markdown formatting (bold, italic, etc.)
- */
-export function stripMarkdown(text: string): string {
-  return text
-    .replace(/\*\*(.+?)\*\*/g, '$1')  // Remove bold **
-    .replace(/\*(.+?)\*/g, '$1')      // Remove italic *
-    .replace(/`(.+?)`/g, '$1')        // Remove inline code `
-    .trim();
-}
-
-/**
- * Extract content related to keywords (for SKILL.md content matching)
- * @param content SKILL.md full content
- * @param keywords Keyword list
- * @param maxLength Maximum return length
- * @returns Extracted relevant content
- */
-export function extractContent(content: string, keywords: string[], maxLength: number = 5000): string {
-  const lines = content.split('\n');
-  const relevantLines: string[] = [];
-  let inRelevantSection = false;
-  let sectionDepth = 0;
-  let found = false;
-
-  for (let i = 0; i < lines.length; i++) {
-    const line = lines[i];
-
-    // Detect headings
-    if (line.match(/^#{1,6}\s/)) {
-      const strippedLine = stripMarkdown(line);
-      const lowerLine = strippedLine.toLowerCase();
-
-      // Check if contains keywords
-      const hasKeyword = keywords.some(kw => lowerLine.includes(kw.toLowerCase()));
-
-      if (hasKeyword) {
-        inRelevantSection = true;
-        found = true;
-        sectionDepth = line.match(/^(#+)/)?.[1].length || 1;
-      } else if (inRelevantSection) {
-        // Check if same level or higher heading (end current section)
-        const currentDepth = line.match(/^(#+)/)?.[1].length || 1;
-        if (currentDepth <= sectionDepth) {
-          inRelevantSection = false;
-        }
-      }
-    }
-
-    if (inRelevantSection || found) {
-      relevantLines.push(line);
-    }
-
-    // Limit content length
-    if (relevantLines.join('\n').length > maxLength) {
-      relevantLines.push('\n...(content truncated for brevity)...');
-      break;
-    }
-  }
-
-  if (relevantLines.length > 0) {
-    return relevantLines.join('\n').trim();
-  }
-
-  // If still not found, return first 100 lines
-  return lines.slice(0, 100).join('\n').trim() + '\n\n...(See SKILL.md for full instructions)';
-}
-
-/**
- * Read SKILL.md and extract relevant content based on task
- */
-export async function readSkillContent(skillPath: string, keywords: string[], maxLength: number = 5000): Promise<string> {
-  const skillMdPath = path.join(skillPath, 'SKILL.md');
-  const content = await fs.readFile(skillMdPath, 'utf-8');
-  return extractContent(content, keywords, maxLength);
-}
-
-// ============================================================
-// SKILL Trigger Keywords Mapping
-// ============================================================
-
-// NOTE: SKILL_TRIGGERS is disabled for experiment purposes.
-// Let the LLM decide which skill to use based on system prompt information.
-
-// interface SkillTrigger {
-//   skillId: string;
-//   keywords: string[];
-//   category: string;
-// }
-
-// export const SKILL_TRIGGERS: Record<string, SkillTrigger> = {
-//   docx: {
-//     skillId: 'docx',
-//     keywords: [
-//       'word document', 'docx', 'microsoft word', 'create word', 'edit word',
-//       'create .docx', '.docx file', 'word file', 'document creation',
-//       'word editing', 'tracked changes', 'comments'
-//     ],
-//     category: 'Document Processing'
-//   },
-//   pdf: {
-//     skillId: 'pdf',
-//     keywords: [
-//       'pdf', 'create pdf', 'edit pdf', 'pdf document', 'pdf file',
-//       'extract pdf', 'merge pdf', 'split pdf', 'pdf form', 'manipulate pdf'
-//     ],
-//     category: 'Document Processing'
-//   },
-//   pptx: {
-//     skillId: 'pptx',
-//     keywords: [
-//       'powerpoint', 'ppt', 'pptx', 'presentation', 'slide',
-//       'create presentation', 'edit powerpoint', 'create slides',
-//       'powerpoint file', 'presentation file'
-//     ],
-//     category: 'Document Processing'
-//   },
-//   xlsx: {
-//     skillId: 'xlsx',
-//     keywords: [
-//       'excel', 'spreadsheet', 'xlsx', 'create excel', 'edit spreadsheet',
-//       'excel file', 'spreadsheet file', 'formulas', 'data analysis'
-//     ],
-//     category: 'Spreadsheet & Data'
-//   },
-//   frontend_design: {
-//     skillId: 'frontend-design',
-//     keywords: [
-//       'web page', 'website', 'web app', 'frontend', 'ui', 'user interface',
-//       'create website', 'build website', 'web component', 'html css',
-//       'landing page', 'dashboard', 'react', 'vue', 'web interface'
-//     ],
-//     category: 'Frontend & Web Development'
-//   },
-//   web_artifacts_builder: {
-//     skillId: 'web-artifacts-builder',
-//     keywords: [
-//       'complex react', 'react artifact', 'stateful artifact', 'routing',
-//       'web artifact', 'interactive artifact', 'web-based tool'
-//     ],
-//     category: 'Frontend & Web Development'
-//   },
-//   webapp_testing: {
-//     skillId: 'webapp-testing',
-//     keywords: [
-//       'test web', 'web testing', 'browser test', 'playwright', 'e2e test',
-//       'frontend test', 'capture screenshot', 'verify web'
-//     ],
-//     category: 'Frontend & Web Development'
-//   },
-//   canvas_design: {
-//     skillId: 'canvas-design',
-//     keywords: [
-//       'poster', 'artwork', 'visual art', 'canvas', 'design art',
-//       'create poster', 'create artwork', 'visual design', 'graphic art'
-//     ],
-//     category: 'Visual & Creative Design'
-//   },
-//   algorithmic_art: {
-//     skillId: 'algorithmic-art',
-//     keywords: [
-//       'generative art', 'algorithmic art', 'p5.js', 'particle system',
-//       'flow field', 'creative coding', 'code art'
-//     ],
-//     category: 'Visual & Creative Design'
-//   },
-//   theme_factory: {
-//     skillId: 'theme-factory',
-//     keywords: [
-//       'theme', 'color scheme', 'font theme', 'styling theme',
-//       'consistent theme', 'apply theme', 'theme colors'
-//     ],
-//     category: 'Visual & Creative Design'
-//   },
-//   brand_guidelines: {
-//     skillId: 'brand-guidelines',
-//     keywords: [
-//       'brand colors', 'brand guidelines', 'anthropic brand',
-//       'official brand', 'brand styling'
-//     ],
-//     category: 'Visual & Creative Design'
-//   },
-//   slack_gif_creator: {
-//     skillId: 'slack-gif-creator',
-//     keywords: [
-//       'slack gif', 'animated gif', 'gif for slack', 'slack animation'
-//     ],
-//     category: 'Visual & Creative Design'
-//   },
-//   mcp_builder: {
-//     skillId: 'mcp-builder',
-//     keywords: [
-//       'mcp server', 'model context protocol', 'create mcp',
-//       'mcp integration', 'external api integration'
-//     ],
-//     category: 'Development & Integration'
-//   },
-//   skill_creator: {
-//     skillId: 'skill-creator',
-//     keywords: [
-//       'create skill', 'new skill', 'skill development',
-//       'extend capabilities', 'custom skill'
-//     ],
-//     category: 'Development & Integration'
-//   },
-//   doc_coauthoring: {
-//     skillId: 'doc-coauthoring',
-//     keywords: [
-//       'documentation', 'technical docs', 'write documentation',
-//       'coauthor', 'doc writing', 'technical writing'
-//     ],
-//     category: 'Communication & Documentation'
-//   }
-// };
-
-// ============================================================
-// SkillInvoker Main Class
-// ============================================================
-
-export class SkillInvoker {
-  private skillLoader: SkillLoader;
-  private initialized: boolean = false;
-  private skillCache: Map<string, SkillInfo> = new Map(); // Stores metadata only
-  private skillContentCache: Map<string, string> = new Map(); // Stores full SKILL.md content
-
-  constructor(skillLoader?: SkillLoader) {
-    this.skillLoader = skillLoader || getSkillLoader();
-  }
-
-  async initialize(): Promise<void> {
-    if (this.initialized) return;
-
-    // Use discoverSkills to only discover directories without loading full content
-    const skillIds = await this.skillLoader.discoverSkills();
-
-    // Create minimal SkillInfo objects with metadata only
-    for (const skillId of skillIds) {
-      const skillPath = this.skillLoader.getSkillDirectory?.(skillId) || '';
-      const skillInfo: SkillInfo = {
-        id: skillId,
-        name: skillId,
-        description: '', // Will be loaded lazily
-        license: 'Unknown',
-        version: '1.0.0',
-        author: 'Anonymous',
-        category: '',
-        markdown: '', // Full content loaded lazily
-        skillsPath: skillPath
-      };
-      this.skillCache.set(skillId, skillInfo);
-    }
-
-    this.initialized = true;
-  }
-
-  /**
-   * Load skill metadata (name, description, category) from SKILL.md frontmatter
-   * This is called lazily when skill details are needed
-   */
-  async loadSkillMetadata(skillId: string): Promise<void> {
-    const skill = this.skillCache.get(skillId);
-    if (!skill) return;
-
-    // Check if metadata already loaded
-    if (skill.description && skill.category) return;
-
-    const skillPath = skill.skillsPath;
-    const skillMdPath = path.join(skillPath, 'SKILL.md');
-
-    try {
-      const content = await fs.readFile(skillMdPath, 'utf-8');
-      const parsed = this.skillLoader.parseSkillMarkdown(content);
-
-      skill.name = parsed.name || skillId;
-      skill.description = parsed.description || '';
-      skill.license = parsed.license || 'Unknown';
-      skill.version = parsed.version || '1.0.0';
-      skill.author = parsed.author || 'Anonymous';
-
-      // Extract category from path
-      const pathParts = skillPath.split(path.sep);
-      const skillsIndex = pathParts.findIndex(p => p === 'skills');
-      if (skillsIndex >= 0 && pathParts.length > skillsIndex + 1) {
-        skill.category = pathParts[skillsIndex + 1];
-      }
-
-      skill.markdown = content; // Now we have the full content
-      this.skillContentCache.set(skillId, content);
-    } catch (error) {
-      console.warn(`[SkillInvoker] Failed to load metadata for skill ${skillId}:`, error);
-    }
-  }
-
-  /**
-   * Get list of all available skills (with metadata)
-   */
-  async listAvailableSkills(): Promise<SkillInfo[]> {
-    await this.initialize();
-
-    // Load metadata for all skills if not already loaded
-    for (const skillId of this.skillCache.keys()) {
-      await this.loadSkillMetadata(skillId);
-    }
-
-    return Array.from(this.skillCache.values());
-  }
-
-  /**
-   * Match the most relevant skill based on user input
-   * NOTE: SKILL_TRIGGERS disabled. Let LLM decide based on system prompt.
-   * Returns null to indicate no explicit match - LLM should use its own judgment.
-   */
-  async matchSkill(userInput: string): Promise<SkillMatcherResult | null> {
-    // SKILL_TRIGGERS is disabled for experiment purposes.
-    // The LLM should decide which skill to use based on system prompt information.
-    return null;
-  }
-
-  /**
-   * Get skill details
-   */
-  async getSkillDetails(skillId: string): Promise<SkillInfo | null> {
-    await this.initialize();
-    return this.skillLoader.getSkill(skillId) || null;
-  }
-
-  /**
-   * Execute skill
-   */
-  async executeSkill(params: SkillExecutionParams): Promise<SkillExecutionResult> {
-    // Ensure initialized
-    await this.initialize();
-
-    // Load skill metadata if not already loaded
-    await this.loadSkillMetadata(params.skillId);
-
-    const skill = this.skillCache.get(params.skillId);
-
-    if (!skill) {
-      return {
-        success: false,
-        error: `Skill not found: ${params.skillId}`
-      };
-    }
-
-    // Generate task ID (if not provided)
-    const taskId = params.taskId || `${params.skillId}-${Date.now()}`;
-
-    try {
-      // Execute based on skillId
-      const executor = this.getSkillExecutor(skill.id);
-      const result = await executor.execute(skill, { ...params, taskId });
-
-      // Add workspaceDir to result
-      if (result.success && result.nextSteps && result.nextSteps.length > 0) {
-        result.workspaceDir = getWorkspaceDir(taskId);
-        await ensureWorkspaceDir(result.workspaceDir);
-      }
-
-      return result;
-    } catch (error: any) {
-      return {
-        success: false,
-        error: error.message
-      };
-    }
-  }
-
-  /**
-   * Clean up workspace based on execution result
-   * @param result Skill execution result
-   * @param executionMode Execution mode
-   * @returns Whether cleanup was performed
-   */
-  async cleanupAfterExecution(result: SkillExecutionResult, executionMode: ExecutionMode): Promise<boolean> {
-    if (!result.workspaceDir) {
-      return false;
-    }
-
-    // YOLO mode: auto cleanup
-    if (executionMode === ExecutionMode.YOLO) {
-      await cleanupWorkspace(result.workspaceDir, result.preserveFiles || []);
-      return true;
-    }
-
-    // Other modes: don't auto cleanup, let user decide
-    return false;
-  }
-
-  /**
-   * Get cleanup prompt (for asking user)
-   */
-  async getCleanupPrompt(result: SkillExecutionResult): Promise<string | null> {
-    if (!result.workspaceDir) {
-      return null;
-    }
-
-    const info = await getCleanupInfo(result.workspaceDir);
-    if (info.files.length === 0) {
-      return null;
-    }
-
-    return `Task completed! Workspace directory contains the following files:\n` +
-      `📁 ${result.workspaceDir}\n` +
-      `Files: ${info.files.join(', ')}\n` +
-      `Size: ${info.totalSize}\n\n` +
-      `Do you want to clean up these temporary files?`;
-  }
-
-  /**
-   * Get executor for skill
-   * Unified dynamic approach - all skills use GenericSkillExecutor
-   */
-  private getSkillExecutor(skillId: string): SkillExecutor {
-    return new GenericSkillExecutor();
-  }
-
-  // ============================================================================
-  // Remote Mode Tool Support Methods
-  // ============================================================================
-
-  /**
-   * Check if it's a Skill tool
-   * Used for remote mode tool execution
-   */
-  isSkillTool(toolName: string): boolean {
-    // Check if it's a skill ID in cache
-    return this.skillCache.has(toolName);
-  }
-
-  /**
-   * Get all Skill definitions (for syncing to remote server)
-   * NOTE: triggers field is empty since SKILL_TRIGGERS is disabled
-   */
-  getAllSkillDefinitions(): Array<{
-    id: string;
-    name: string;
-    description: string;
-    category: string;
-    triggers: string[];
-  }> {
-    const definitions: Array<{
-      id: string;
-      name: string;
-      description: string;
-      category: string;
-      triggers: string[];
-    }> = [];
-
-    for (const skill of this.skillCache.values()) {
-      definitions.push({
-        id: skill.id,
-        name: skill.name,
-        description: skill.description,
-        category: skill.category,
-        triggers: [] // SKILL_TRIGGERS disabled - LLM decides based on description
-      });
-    }
-
-    return definitions;
-  }
-
-  /**
-   * Execute Skill tool (for remote mode tool execution)
-   * @param toolName - Tool name (skillId)
-   * @param params - Tool parameters
-   * @returns Execution result
-   */
-  async executeSkillTool(
-    toolName: string,
-    params: Record<string, any>
-  ): Promise<{ success: boolean; result?: any; error?: string }> {
-    // Check if skill exists in cache
-    if (!this.skillCache.has(toolName)) {
-      return { success: false, error: `Skill not found: ${toolName}` };
-    }
-
-    try {
-      const result = await this.executeSkill({
-        skillId: toolName,
-        taskDescription: params.taskDescription || params.description || '',
-        inputFile: params.inputFile,
-        outputFile: params.outputFile,
-        options: params.options || {}
-      });
-
-      return { success: result.success, result };
-    } catch (error: any) {
-      return { success: false, error: error.message };
-    }
-  }
-
-  /**
-   * Get all available Skill ID list
-   * NOTE: SKILL_TRIGGERS disabled - return all skill IDs from cache
-   */
-  getAvailableSkillIds(): string[] {
-    return Array.from(this.skillCache.keys());
-  }
-}
-
-// ============================================================
-// Skill Executor Interface and Implementation
-// ============================================================
-
-interface SkillExecutor {
-  execute(skill: SkillInfo, params: SkillExecutionParams): Promise<SkillExecutionResult>;
-}
-
-/**
- * Generic Skill Executor - Unified dynamic approach for all skills
- */
-class GenericSkillExecutor implements SkillExecutor {
-  async execute(skill: SkillInfo, params: SkillExecutionParams): Promise<SkillExecutionResult> {
-    const outputMessages: string[] = [];
-    const files: string[] = [];
-    const nextSteps: ExecutionStep[] = [];
-
-    outputMessages.push(`## ${skill.name} Skill - Execution Guide\n`);
-    outputMessages.push(`**Task**: ${params.taskDescription}\n`);
-
-    try {
-      // Generate task ID
-      const taskId = params.taskId || `${skill.id}-${Date.now()}`;
-
-      // Read complete skill documentation
-      const skillPath = skill.skillsPath;
-      const skillMdPath = path.join(skillPath, 'SKILL.md');
-      files.push(skillMdPath);
-
-      // Read SKILL.md content
-      const skillContent = await fs.readFile(skillMdPath, 'utf-8');
-
-      // Extract relevant content based on task type and generate execution steps
-      const taskContent = await this.extractRelevantContent(skill, params, skillContent, nextSteps, taskId);
-      outputMessages.push(taskContent);
-
-      // Add input/output files to list if they exist
-      if (params.inputFile) files.push(params.inputFile);
-      if (params.outputFile) files.push(params.outputFile);
-
-      return {
-        success: true,
-        output: outputMessages.join('\n'),
-        files: files,
-        nextSteps: nextSteps,
-        requiresManualExecution: true
-      };
-    } catch (error: any) {
-      return {
-        success: false,
-        error: error.message
-      };
-    }
-  }
-
-  /**
-   * Extract relevant skill content dynamically
-   */
-  private async extractRelevantContent(
-    skill: SkillInfo,
-    params: SkillExecutionParams,
-    fullContent: string,
-    nextSteps: ExecutionStep[],
-    taskId: string
-  ): Promise<string> {
-    const workspaceBase = getWorkspaceDescription();
-    const taskWorkspace = `${workspaceBase}/${taskId}`;
-
-    // Dynamically discover files in skill directory
-    const skillPath = skill.skillsPath;
-    let allFiles: string[] = [];
-
-    try {
-      const entries = await fs.readdir(skillPath, { withFileTypes: true });
-      for (const entry of entries) {
-        const fullPath = path.join(skillPath, entry.name);
-        if (entry.isFile()) {
-          allFiles.push(fullPath);
-        } else if (entry.isDirectory()) {
-          // Recursively list files in subdirectories (limited depth)
-          const subEntries = await fs.readdir(fullPath, { withFileTypes: true });
-          for (const subEntry of subEntries) {
-            if (subEntry.isFile()) {
-              allFiles.push(path.join(fullPath, subEntry.name));
-            }
-          }
-        }
-      }
-    } catch {
-      // Fallback to just SKILL.md if directory can't be read
-      allFiles = [path.join(skillPath, 'SKILL.md')];
-    }
-
-    // Step 1: Read SKILL.md first
-    const skillMdPathOnly = path.join(skillPath, 'SKILL.md');
-    nextSteps.push({
-      step: 1,
-      action: 'Read SKILL.md to understand the skill workflow',
-      description: `Read: ${skillMdPathOnly}`,
-      reason: 'Understand the skill workflow and best practices from the main documentation'
-    });
-
-    // Step 2: Explore skill directory and read reference files if needed (optional)
-    nextSteps.push({
-      step: 2,
-      action: 'Explore skill directory and read reference files (if needed)',
-      description: `Explore: ${skillPath}`,
-      reason: 'Discover available reference files and read them based on SKILL.md guidance'
-    });
-
-    nextSteps.push({
-      step: 3,
-      action: 'Analyze documentation, verify data/content completeness, and design approach',
-      description: `For content creation: ensure all info/materials collected. For info retrieval: ensure all required data retrieved. Then design execution plan for: ${taskWorkspace}`,
-      reason: 'Review requirements, verify data/content completeness, fill gaps if needed, then plan execution based on the documentation'
-    });
-
-    nextSteps.push({
-      step: 4,
-      action: 'Execute your plan',
-      description: 'Create workspace, write code, run scripts, verify output',
-      reason: 'Execute the task using your own understanding'
-    });
-
-    return `### Skill Execution\n\n` +
-           `**Your task**: ${params.taskDescription}\n\n` +
-           `**Step 1**: Read SKILL.md to understand the skill workflow\n` +
-           `  - read_file: ${skillMdPathOnly}\n\n` +
-           `**Step 2**: Explore skill directory and read reference files (if needed)\n` +
-           `  - ListDirectory(path="${skillPath}")\n` +
-           `  - read_file relevant .md and script files as needed\n\n` +
-           `Then analyze the documentation and create your own execution plan.\n\n` +
-           `**Workspace**: \`${taskWorkspace}\`\n\n` +
-           `**⚠️ Windows Path Execution**: Use absolute paths, NOT \`cd && command\`:\n` +
-           `  - ✅ Correct: \`node "${taskWorkspace}/script.js"\`\n` +
-           `  - ❌ Wrong: \`cd "${taskWorkspace}" && node script.js\` (fails in PowerShell 5.1)\n` +
-           `  - ✅ Correct: \`python "${taskWorkspace}/script.py"\`\n\n` +
-           `**📦 Dependency Management**:\n` +
-           `  - BashTool automatically sets NODE_PATH when executing, scripts can use require() directly\n` +
-           `  - ✅ Correct: Bash(command="node script.js", cwd="${taskWorkspace}")\n` +
-           `  - ❌ Wrong: cd "${taskWorkspace}" && node script.js (loses NODE_PATH!)\n` +
-           `  - If script needs to run by user manually, pass NODE_PATH in command:\n` +
-           `    Windows: set NODE_PATH=xAgent/node_modules/path && node script.js\n` +
-           `    Linux/Mac: NODE_PATH=xAgent/node_modules/path node script.js\n\n` +
-           `**🧹 Cleanup**: Delete all intermediate/temporary files when task completes:\n` +
-           `  - Remove: all files generated during the task\n` +
-           `  - Keep: Only the final output file (output.pptx/docx/xlsx/pdf)\n\n` +
-           `**Instructions**: read_file the documentation, understand the API, and create your own execution plan.\n` +
-           `**If you encounter issues**: Explain what went wrong and suggest a different approach.\n`;
-  }
-}
-// ============================================================
-// ============================================================
-
-/**
- * Execute skill - LLM analyzes SKILL.md and generates its own steps
- * @param skill Skill to execute
- * @param params Execution parameters
- * @returns Execution result with guidance
- */
-export async function executeSkill(
-  skill: SkillInfo,
-  params: SkillExecutionParams
-): Promise<SkillExecutionResult> {
-  const executor = new GenericSkillExecutor();
-  return executor.execute(skill, params);
-}
-
-// ============================================================
-// Singleton Instance and Exports
-// ============================================================
-
-let skillInvokerInstance: SkillInvoker | null = null;
-
-export function getSkillInvoker(): SkillInvoker {
-  if (!skillInvokerInstance) {
-    skillInvokerInstance = new SkillInvoker();
-  }
-  return skillInvokerInstance;
-}
+import fs from 'fs/promises';
+import path from 'path';
+import os from 'os';
+import { getSkillLoader, SkillInfo, SkillLoader } from './skill-loader.js';
+import { getToolRegistry } from './tools.js';
+import { ExecutionMode, Tool } from './types.js';
+import { getConfigManager } from './config.js';
+
+// Re-export SkillInfo for other modules
+export type { SkillInfo };
+
+/**
+ * Track skill execution history for tracking failures
+ */
+export class SkillExecutionHistory {
+  private history: Map<string, number> = new Map();
+
+  /**
+   * Get failure count for a task
+   */
+  getFailureCount(taskKey: string): number {
+    return this.history.get(taskKey) || 0;
+  }
+
+  /**
+   * Increment failure count for a task
+   */
+  incrementFailure(taskKey: string): number {
+    const count = this.getFailureCount(taskKey) + 1;
+    this.history.set(taskKey, count);
+    return count;
+  }
+
+  /**
+   * Reset failure count for a task (e.g., after success)
+   */
+  reset(taskKey: string): void {
+    this.history.delete(taskKey);
+  }
+
+  /**
+   * Check if threshold reached
+   */
+  shouldUseFallback(taskKey: string, threshold: number = 2): boolean {
+    return this.getFailureCount(taskKey) >= threshold;
+  }
+}
+
+// Singleton execution history
+const executionHistory = new SkillExecutionHistory();
+
+export function getExecutionHistory(): SkillExecutionHistory {
+  return executionHistory;
+}
+
+export interface SkillExecutionParams {
+  skillId: string;
+  taskDescription: string;
+  inputFile?: string;
+  outputFile?: string;
+  options?: Record<string, any>;
+  /** Task ID for workspace directory naming */
+  taskId?: string;
+}
+
+/**
+ * Execution step interface - tells Agent what to do next
+ */
+export interface ExecutionStep {
+  step: number;
+  action: string;
+  description: string;
+  command?: string;
+  file?: string;
+  reason: string;
+}
+
+/**
+ * Skill execution result - contains guidance and next actions
+ */
+export interface SkillExecutionResult {
+  success: boolean;
+  output?: string;
+  error?: string;
+  files?: string[];
+  /** Tells Agent what to do next */
+  nextSteps?: ExecutionStep[];
+  /** Skill type for determining if manual execution is needed */
+  requiresManualExecution?: boolean;
+  /** Workspace directory used, for cleanup */
+  workspaceDir?: string;
+  /** Files to preserve (relative paths), skipped during cleanup */
+  preserveFiles?: string[];
+  /** Skill directory path - for dependency management and file operations */
+  skillPath?: string;
+}
+
+export interface SkillMatcherResult {
+  skill: SkillInfo;
+  confidence: number;
+  matchedKeywords: string[];
+  category: string;
+}
+
+// ============================================================
+// Workspace Utility Functions
+// ============================================================
+
+/**
+ * Get workspace directory path
+ * @param taskId Task ID for creating unique workspace directory
+ * @returns Absolute path to workspace directory
+ */
+export function getWorkspaceDir(taskId: string): string {
+  // Try to get from config first
+  try {
+    const configManager = getConfigManager();
+    const config = configManager.getSettings?.();
+    if (config?.workspacePath) {
+      return path.join(config.workspacePath, taskId);
+    }
+  } catch {
+    // Config not available, use default
+  }
+
+  // Default to ~/.xagent/workspace
+  return path.join(os.homedir(), '.xagent', 'workspace', taskId);
+}
+
+/**
+ * Get base workspace directory (without task-id)
+ */
+export function getBaseWorkspaceDir(): string {
+  try {
+    const configManager = getConfigManager();
+    const config = configManager.getSettings?.();
+    if (config?.workspacePath) {
+      return config.workspacePath;
+    }
+  } catch {
+    // Config not available, use default
+  }
+
+  return path.join(os.homedir(), '.xagent', 'workspace');
+}
+
+/**
+ * Get workspace directory description for AI
+ * Returns the actual workspace path from config, or default path
+ */
+export function getWorkspaceDescription(): string {
+  try {
+    const configManager = getConfigManager();
+    const config = configManager.getSettings?.();
+    if (config?.workspacePath) {
+      return config.workspacePath;
+    }
+  } catch {
+    // Config not available, use default
+  }
+
+  return path.join(os.homedir(), '.xagent', 'workspace');
+}
+
+/**
+ * Ensure workspace directory exists
+ * @param workspaceDir Workspace directory path
+ */
+export async function ensureWorkspaceDir(workspaceDir: string): Promise<void> {
+  await fs.mkdir(workspaceDir, { recursive: true });
+}
+
+/**
+ * Clean up workspace directory
+ * @param workspaceDir Workspace directory path
+ * @param preserveFiles Files to preserve (relative paths)
+ */
+export async function cleanupWorkspace(workspaceDir: string, preserveFiles: string[] = []): Promise<void> {
+  try {
+    const entries = await fs.readdir(workspaceDir, { withFileTypes: true });
+
+    for (const entry of entries) {
+      const fullPath = path.join(workspaceDir, entry.name);
+
+      // Skip files to preserve
+      if (preserveFiles.includes(entry.name)) {
+        continue;
+      }
+
+      if (entry.isDirectory()) {
+        // Recursively delete subdirectories
+        await fs.rm(fullPath, { recursive: true, force: true });
+      } else {
+        // Delete files
+        await fs.unlink(fullPath);
+      }
+    }
+  } catch (error: any) {
+    if (error.code !== 'ENOENT') {
+      console.warn(`Workspace cleanup failed: ${error.message}`);
+    }
+  }
+}
+
+/**
+ * Determine if workspace should be auto-cleaned based on ExecutionMode
+ * @param executionMode Execution mode
+ * @returns Whether auto-cleanup should happen
+ */
+export function shouldAutoCleanup(executionMode: ExecutionMode): boolean {
+  // YOLO mode: fully automatic, clean up directly
+  if (executionMode === ExecutionMode.YOLO) {
+    return true;
+  }
+  // Other modes require user confirmation
+  return false;
+}
+
+/**
+ * Generate cleanup prompt message
+ * @param workspaceDir Workspace directory path
+ */
+export async function getCleanupInfo(workspaceDir: string): Promise<{ files: string[]; totalSize: string }> {
+  try {
+    const entries = await fs.readdir(workspaceDir, { withFileTypes: true });
+    const files: string[] = [];
+
+    for (const entry of entries) {
+      files.push(entry.name);
+    }
+
+    // Calculate total size
+    let totalSize = 0;
+    for (const entry of entries) {
+      if (entry.isFile()) {
+        const stats = await fs.stat(path.join(workspaceDir, entry.name));
+        totalSize += stats.size;
+      }
+    }
+
+    const formatSize = (bytes: number): string => {
+      if (bytes < 1024) return `${bytes} B`;
+      if (bytes < 1024 * 1024) return `${(bytes / 1024).toFixed(2)} KB`;
+      return `${(bytes / (1024 * 1024)).toFixed(2)} MB`;
+    };
+
+    return { files, totalSize: formatSize(totalSize) };
+  } catch {
+    return { files: [], totalSize: '0 B' };
+  }
+}
+
+// ============================================================
+// Shared Content Extraction Utilities
+// ============================================================
+
+/**
+ * Remove Markdown formatting (bold, italic, etc.)
+ */
+export function stripMarkdown(text: string): string {
+  return text
+    .replace(/\*\*(.+?)\*\*/g, '$1')  // Remove bold **
+    .replace(/\*(.+?)\*/g, '$1')      // Remove italic *
+    .replace(/`(.+?)`/g, '$1')        // Remove inline code `
+    .trim();
+}
+
+/**
+ * Extract content related to keywords (for SKILL.md content matching)
+ * @param content SKILL.md full content
+ * @param keywords Keyword list
+ * @param maxLength Maximum return length
+ * @returns Extracted relevant content
+ */
+export function extractContent(content: string, keywords: string[], maxLength: number = 5000): string {
+  const lines = content.split('\n');
+  const relevantLines: string[] = [];
+  let inRelevantSection = false;
+  let sectionDepth = 0;
+  let found = false;
+
+  for (let i = 0; i < lines.length; i++) {
+    const line = lines[i];
+
+    // Detect headings
+    if (line.match(/^#{1,6}\s/)) {
+      const strippedLine = stripMarkdown(line);
+      const lowerLine = strippedLine.toLowerCase();
+
+      // Check if contains keywords
+      const hasKeyword = keywords.some(kw => lowerLine.includes(kw.toLowerCase()));
+
+      if (hasKeyword) {
+        inRelevantSection = true;
+        found = true;
+        sectionDepth = line.match(/^(#+)/)?.[1].length || 1;
+      } else if (inRelevantSection) {
+        // Check if same level or higher heading (end current section)
+        const currentDepth = line.match(/^(#+)/)?.[1].length || 1;
+        if (currentDepth <= sectionDepth) {
+          inRelevantSection = false;
+        }
+      }
+    }
+
+    if (inRelevantSection || found) {
+      relevantLines.push(line);
+    }
+
+    // Limit content length
+    if (relevantLines.join('\n').length > maxLength) {
+      relevantLines.push('\n...(content truncated for brevity)...');
+      break;
+    }
+  }
+
+  if (relevantLines.length > 0) {
+    return relevantLines.join('\n').trim();
+  }
+
+  // If still not found, return first 100 lines
+  return lines.slice(0, 100).join('\n').trim() + '\n\n...(See SKILL.md for full instructions)';
+}
+
+/**
+ * Read SKILL.md and extract relevant content based on task
+ */
+export async function readSkillContent(skillPath: string, keywords: string[], maxLength: number = 5000): Promise<string> {
+  const skillMdPath = path.join(skillPath, 'SKILL.md');
+  const content = await fs.readFile(skillMdPath, 'utf-8');
+  return extractContent(content, keywords, maxLength);
+}
+
+// ============================================================
+// SKILL Trigger Keywords Mapping
+// ============================================================
+
+// NOTE: SKILL_TRIGGERS is disabled for experiment purposes.
+// Let the LLM decide which skill to use based on system prompt information.
+
+// interface SkillTrigger {
+//   skillId: string;
+//   keywords: string[];
+//   category: string;
+// }
+
+// export const SKILL_TRIGGERS: Record<string, SkillTrigger> = {
+//   docx: {
+//     skillId: 'docx',
+//     keywords: [
+//       'word document', 'docx', 'microsoft word', 'create word', 'edit word',
+//       'create .docx', '.docx file', 'word file', 'document creation',
+//       'word editing', 'tracked changes', 'comments'
+//     ],
+//     category: 'Document Processing'
+//   },
+//   pdf: {
+//     skillId: 'pdf',
+//     keywords: [
+//       'pdf', 'create pdf', 'edit pdf', 'pdf document', 'pdf file',
+//       'extract pdf', 'merge pdf', 'split pdf', 'pdf form', 'manipulate pdf'
+//     ],
+//     category: 'Document Processing'
+//   },
+//   pptx: {
+//     skillId: 'pptx',
+//     keywords: [
+//       'powerpoint', 'ppt', 'pptx', 'presentation', 'slide',
+//       'create presentation', 'edit powerpoint', 'create slides',
+//       'powerpoint file', 'presentation file'
+//     ],
+//     category: 'Document Processing'
+//   },
+//   xlsx: {
+//     skillId: 'xlsx',
+//     keywords: [
+//       'excel', 'spreadsheet', 'xlsx', 'create excel', 'edit spreadsheet',
+//       'excel file', 'spreadsheet file', 'formulas', 'data analysis'
+//     ],
+//     category: 'Spreadsheet & Data'
+//   },
+//   frontend_design: {
+//     skillId: 'frontend-design',
+//     keywords: [
+//       'web page', 'website', 'web app', 'frontend', 'ui', 'user interface',
+//       'create website', 'build website', 'web component', 'html css',
+//       'landing page', 'dashboard', 'react', 'vue', 'web interface'
+//     ],
+//     category: 'Frontend & Web Development'
+//   },
+//   web_artifacts_builder: {
+//     skillId: 'web-artifacts-builder',
+//     keywords: [
+//       'complex react', 'react artifact', 'stateful artifact', 'routing',
+//       'web artifact', 'interactive artifact', 'web-based tool'
+//     ],
+//     category: 'Frontend & Web Development'
+//   },
+//   webapp_testing: {
+//     skillId: 'webapp-testing',
+//     keywords: [
+//       'test web', 'web testing', 'browser test', 'playwright', 'e2e test',
+//       'frontend test', 'capture screenshot', 'verify web'
+//     ],
+//     category: 'Frontend & Web Development'
+//   },
+//   canvas_design: {
+//     skillId: 'canvas-design',
+//     keywords: [
+//       'poster', 'artwork', 'visual art', 'canvas', 'design art',
+//       'create poster', 'create artwork', 'visual design', 'graphic art'
+//     ],
+//     category: 'Visual & Creative Design'
+//   },
+//   algorithmic_art: {
+//     skillId: 'algorithmic-art',
+//     keywords: [
+//       'generative art', 'algorithmic art', 'p5.js', 'particle system',
+//       'flow field', 'creative coding', 'code art'
+//     ],
+//     category: 'Visual & Creative Design'
+//   },
+//   theme_factory: {
+//     skillId: 'theme-factory',
+//     keywords: [
+//       'theme', 'color scheme', 'font theme', 'styling theme',
+//       'consistent theme', 'apply theme', 'theme colors'
+//     ],
+//     category: 'Visual & Creative Design'
+//   },
+//   brand_guidelines: {
+//     skillId: 'brand-guidelines',
+//     keywords: [
+//       'brand colors', 'brand guidelines', 'anthropic brand',
+//       'official brand', 'brand styling'
+//     ],
+//     category: 'Visual & Creative Design'
+//   },
+//   slack_gif_creator: {
+//     skillId: 'slack-gif-creator',
+//     keywords: [
+//       'slack gif', 'animated gif', 'gif for slack', 'slack animation'
+//     ],
+//     category: 'Visual & Creative Design'
+//   },
+//   mcp_builder: {
+//     skillId: 'mcp-builder',
+//     keywords: [
+//       'mcp server', 'model context protocol', 'create mcp',
+//       'mcp integration', 'external api integration'
+//     ],
+//     category: 'Development & Integration'
+//   },
+//   skill_creator: {
+//     skillId: 'skill-creator',
+//     keywords: [
+//       'create skill', 'new skill', 'skill development',
+//       'extend capabilities', 'custom skill'
+//     ],
+//     category: 'Development & Integration'
+//   },
+//   doc_coauthoring: {
+//     skillId: 'doc-coauthoring',
+//     keywords: [
+//       'documentation', 'technical docs', 'write documentation',
+//       'coauthor', 'doc writing', 'technical writing'
+//     ],
+//     category: 'Communication & Documentation'
+//   }
+// };
+
+// ============================================================
+// SkillInvoker Main Class
+// ============================================================
+
+export class SkillInvoker {
+  private skillLoader: SkillLoader;
+  private initialized: boolean = false;
+  private skillCache: Map<string, SkillInfo> = new Map(); // Stores metadata only
+  // private skillContentCache: Map<string, string> = new Map(); // Stores full SKILL.md content - UNUSED
+
+  constructor(skillLoader?: SkillLoader) {
+    this.skillLoader = skillLoader || getSkillLoader();
+  }
+
+  async initialize(): Promise<void> {
+    if (this.initialized) return;
+
+    // Use discoverSkills to only discover directories without loading full content
+    const skillIds = await this.skillLoader.discoverSkills();
+
+    // Create minimal SkillInfo objects with metadata only
+    for (const skillId of skillIds) {
+      const skillPath = this.skillLoader.getSkillDirectory?.(skillId) || '';
+      const skillInfo: SkillInfo = {
+        id: skillId,
+        name: skillId,
+        description: '', // Will be loaded lazily
+        license: 'Unknown',
+        version: '1.0.0',
+        author: 'Anonymous',
+        category: '',
+        markdown: '', // Full content loaded lazily
+        skillsPath: skillPath
+      };
+      this.skillCache.set(skillId, skillInfo);
+    }
+
+    this.initialized = true;
+  }
+
+  /**
+   * Load skill metadata (name, description, category) from SKILL.md frontmatter
+   * This is called lazily when skill details are needed
+   */
+  async loadSkillMetadata(skillId: string): Promise<void> {
+    const skill = this.skillCache.get(skillId);
+    if (!skill) return;
+
+    // Check if metadata already loaded
+    if (skill.description && skill.category) return;
+
+    const skillPath = skill.skillsPath;
+    const skillMdPath = path.join(skillPath, 'SKILL.md');
+
+    try {
+      const content = await fs.readFile(skillMdPath, 'utf-8');
+      const parsed = this.skillLoader.parseSkillMarkdown(content);
+
+      skill.name = parsed.name || skillId;
+      skill.description = parsed.description || '';
+      skill.license = parsed.license || 'Unknown';
+      skill.version = parsed.version || '1.0.0';
+      skill.author = parsed.author || 'Anonymous';
+
+      // Extract category from path
+      const pathParts = skillPath.split(path.sep);
+      const skillsIndex = pathParts.findIndex(p => p === 'skills');
+      if (skillsIndex >= 0 && pathParts.length > skillsIndex + 1) {
+        skill.category = pathParts[skillsIndex + 1];
+      }
+
+      skill.markdown = content; // Now we have the full content
+      // this.skillContentCache.set(skillId, content); // UNUSED
+    } catch (error) {
+      console.warn(`[SkillInvoker] Failed to load metadata for skill ${skillId}:`, error);
+    }
+  }
+
+  /**
+   * Get list of all available skills (with metadata)
+   */
+  async listAvailableSkills(): Promise<SkillInfo[]> {
+    await this.initialize();
+
+    // Load metadata for all skills if not already loaded
+    for (const skillId of this.skillCache.keys()) {
+      await this.loadSkillMetadata(skillId);
+    }
+
+    return Array.from(this.skillCache.values());
+  }
+
+  /**
+   * Reload all skills (e.g., after adding/removing a skill via CLI)
+   * Clears the cache and re-discovers all skills
+   */
+  async reload(): Promise<void> {
+    // Reset initialization state to allow re-discovery
+    this.initialized = false;
+    this.skillCache.clear();
+    
+    // Re-initialize
+    await this.initialize();
+  }
+
+  /**
+   * Match the most relevant skill based on user input
+   * NOTE: SKILL_TRIGGERS disabled. Let LLM decide based on system prompt.
+   * Returns null to indicate no explicit match - LLM should use its own judgment.
+   */
+  async matchSkill(userInput: string): Promise<SkillMatcherResult | null> {
+    // SKILL_TRIGGERS is disabled for experiment purposes.
+    // The LLM should decide which skill to use based on system prompt information.
+    return null;
+  }
+
+  /**
+   * Get skill details
+   */
+  async getSkillDetails(skillId: string): Promise<SkillInfo | null> {
+    await this.initialize();
+    return this.skillLoader.getSkill(skillId) || null;
+  }
+
+  /**
+   * Execute skill
+   */
+  async executeSkill(params: SkillExecutionParams): Promise<SkillExecutionResult> {
+    // Ensure initialized
+    await this.initialize();
+
+    // Load skill metadata if not already loaded
+    await this.loadSkillMetadata(params.skillId);
+
+    const skill = this.skillCache.get(params.skillId);
+
+    if (!skill) {
+      return {
+        success: false,
+        error: `Skill not found: ${params.skillId}`
+      };
+    }
+
+    // Generate task ID (if not provided)
+    const taskId = params.taskId || `${params.skillId}-${Date.now()}`;
+
+    try {
+      // Execute based on skillId
+      const executor = this.getSkillExecutor(skill.id);
+      const result = await executor.execute(skill, { ...params, taskId });
+
+      // Add skillPath and workspaceDir to result (delay creation until actually needed)
+      if (result.success) {
+        // Get skillPath directly from skillDirectories (more reliable)
+        result.skillPath = this.skillLoader.getSkillDirectory?.(params.skillId) || skill.skillsPath || '';
+        if (result.nextSteps && result.nextSteps.length > 0) {
+          result.workspaceDir = getWorkspaceDir(taskId);
+          // Don't pre-create workspace - only create when actually used by LLM
+        }
+      }
+
+      return result;
+    } catch (error: any) {
+      return {
+        success: false,
+        error: error.message
+      };
+    }
+  }
+
+  /**
+   * Clean up workspace based on execution result
+   * @param result Skill execution result
+   * @param executionMode Execution mode
+   * @returns Whether cleanup was performed
+   */
+  async cleanupAfterExecution(result: SkillExecutionResult, executionMode: ExecutionMode): Promise<boolean> {
+    if (!result.workspaceDir) {
+      return false;
+    }
+
+    // YOLO mode: auto cleanup
+    if (executionMode === ExecutionMode.YOLO) {
+      await cleanupWorkspace(result.workspaceDir, result.preserveFiles || []);
+      return true;
+    }
+
+    // Other modes: don't auto cleanup, let user decide
+    return false;
+  }
+
+  /**
+   * Get cleanup prompt (for asking user)
+   */
+  async getCleanupPrompt(result: SkillExecutionResult): Promise<string | null> {
+    if (!result.workspaceDir) {
+      return null;
+    }
+
+    const info = await getCleanupInfo(result.workspaceDir);
+    if (info.files.length === 0) {
+      return null;
+    }
+
+    return `Task completed! Workspace directory contains the following files:\n` +
+      `📁 ${result.workspaceDir}\n` +
+      `Files: ${info.files.join(', ')}\n` +
+      `Size: ${info.totalSize}\n\n` +
+      `Do you want to clean up these temporary files?`;
+  }
+
+  /**
+   * Get executor for skill
+   * Unified dynamic approach - all skills use GenericSkillExecutor
+   */
+  private getSkillExecutor(skillId: string): SkillExecutor {
+    return new GenericSkillExecutor();
+  }
+
+  // ============================================================================
+  // Remote Mode Tool Support Methods
+  // ============================================================================
+
+  /**
+   * Check if it's a Skill tool
+   * Used for remote mode tool execution
+   */
+  isSkillTool(toolName: string): boolean {
+    // Check if it's a skill ID in cache
+    return this.skillCache.has(toolName);
+  }
+
+  /**
+   * Get all Skill definitions (for syncing to remote server)
+   * NOTE: triggers field is empty since SKILL_TRIGGERS is disabled
+   */
+  getAllSkillDefinitions(): Array<{
+    id: string;
+    name: string;
+    description: string;
+    category: string;
+    triggers: string[];
+  }> {
+    const definitions: Array<{
+      id: string;
+      name: string;
+      description: string;
+      category: string;
+      triggers: string[];
+    }> = [];
+
+    for (const skill of this.skillCache.values()) {
+      definitions.push({
+        id: skill.id,
+        name: skill.name,
+        description: skill.description,
+        category: skill.category,
+        triggers: [] // SKILL_TRIGGERS disabled - LLM decides based on description
+      });
+    }
+
+    return definitions;
+  }
+
+  /**
+   * Execute Skill tool (for remote mode tool execution)
+   * @param toolName - Tool name (skillId)
+   * @param params - Tool parameters
+   * @returns Execution result
+   */
+  async executeSkillTool(
+    toolName: string,
+    params: Record<string, any>
+  ): Promise<{ success: boolean; result?: any; error?: string }> {
+    // Check if skill exists in cache
+    if (!this.skillCache.has(toolName)) {
+      return { success: false, error: `Skill not found: ${toolName}` };
+    }
+
+    try {
+      const result = await this.executeSkill({
+        skillId: toolName,
+        taskDescription: params.taskDescription || params.description || '',
+        inputFile: params.inputFile,
+        outputFile: params.outputFile,
+        options: params.options || {}
+      });
+
+      return { success: result.success, result };
+    } catch (error: any) {
+      return { success: false, error: error.message };
+    }
+  }
+
+  /**
+   * Get all available Skill ID list
+   * NOTE: SKILL_TRIGGERS disabled - return all skill IDs from cache
+   */
+  getAvailableSkillIds(): string[] {
+    return Array.from(this.skillCache.keys());
+  }
+}
+
+// ============================================================
+// Skill Executor Interface and Implementation
+// ============================================================
+
+interface SkillExecutor {
+  execute(skill: SkillInfo, params: SkillExecutionParams): Promise<SkillExecutionResult>;
+}
+
+/**
+ * Generic Skill Executor - Unified dynamic approach for all skills
+ */
+class GenericSkillExecutor implements SkillExecutor {
+  async execute(skill: SkillInfo, params: SkillExecutionParams): Promise<SkillExecutionResult> {
+    const outputMessages: string[] = [];
+    const files: string[] = [];
+    const nextSteps: ExecutionStep[] = [];
+
+    outputMessages.push(`## ${skill.name} Skill - Execution Guide\n`);
+    outputMessages.push(`**Task**: ${params.taskDescription}\n`);
+
+    try {
+      // Generate task ID
+      const taskId = params.taskId || `${skill.id}-${Date.now()}`;
+
+      // Read complete skill documentation
+      const skillPath = skill.skillsPath;
+      const skillMdPath = path.join(skillPath, 'SKILL.md');
+      files.push(skillMdPath);
+
+      // Read SKILL.md content
+      const skillContent = await fs.readFile(skillMdPath, 'utf-8');
+
+      // Extract relevant content based on task type and generate execution steps
+      const taskContent = await this.extractRelevantContent(skill, params, skillContent, nextSteps, taskId);
+      outputMessages.push(taskContent);
+
+      // Add input/output files to list if they exist
+      if (params.inputFile) files.push(params.inputFile);
+      if (params.outputFile) files.push(params.outputFile);
+
+      return {
+        success: true,
+        output: outputMessages.join('\n'),
+        files: files,
+        nextSteps: nextSteps,
+        requiresManualExecution: true
+      };
+    } catch (error: any) {
+      return {
+        success: false,
+        error: error.message
+      };
+    }
+  }
+
+  /**
+   * Extract relevant skill content dynamically
+   */
+  private async extractRelevantContent(
+    skill: SkillInfo,
+    params: SkillExecutionParams,
+    fullContent: string,
+    nextSteps: ExecutionStep[],
+    taskId: string
+  ): Promise<string> {
+    const workspaceBase = getWorkspaceDescription();
+    const taskWorkspace = `${workspaceBase}/${taskId}`;
+
+    // Dynamically discover files in skill directory
+    const skillPath = skill.skillsPath;
+    let allFiles: string[] = [];
+
+    try {
+      const entries = await fs.readdir(skillPath, { withFileTypes: true });
+      for (const entry of entries) {
+        const fullPath = path.join(skillPath, entry.name);
+        if (entry.isFile()) {
+          allFiles.push(fullPath);
+        } else if (entry.isDirectory()) {
+          // Recursively list files in subdirectories (limited depth)
+          const subEntries = await fs.readdir(fullPath, { withFileTypes: true });
+          for (const subEntry of subEntries) {
+            if (subEntry.isFile()) {
+              allFiles.push(path.join(fullPath, subEntry.name));
+            }
+          }
+        }
+      }
+    } catch {
+      // Fallback to just SKILL.md if directory can't be read
+      allFiles = [path.join(skillPath, 'SKILL.md')];
+    }
+
+    // Step 1: Read SKILL.md first
+    const skillMdPathOnly = path.join(skillPath, 'SKILL.md');
+    nextSteps.push({
+      step: 1,
+      action: 'Read SKILL.md to understand the skill workflow',
+      description: `Read: ${skillMdPathOnly}`,
+      reason: 'Understand the skill workflow and best practices from the main documentation'
+    });
+
+    // Step 2: Explore skill directory and read reference files if needed (optional)
+    nextSteps.push({
+      step: 2,
+      action: 'Explore skill directory and read reference files (if needed)',
+      description: `Explore: ${skillPath}`,
+      reason: 'Discover available reference files and read them based on SKILL.md guidance'
+    });
+
+    nextSteps.push({
+      step: 3,
+      action: 'Analyze documentation, verify data/content completeness, and design approach',
+      description: `For content creation: ensure all info/materials collected. For info retrieval: ensure all required data retrieved. Then design execution plan for: ${taskWorkspace}`,
+      reason: 'Review requirements, verify data/content completeness, fill gaps if needed, then plan execution based on the documentation'
+    });
+
+    nextSteps.push({
+      step: 4,
+      action: 'Execute your plan',
+      description: 'Create workspace, write code, run scripts, verify output',
+      reason: 'Execute the task using your own understanding'
+    });
+
+    return `### Skill Execution\n\n` +
+           `**Your task**: ${params.taskDescription}\n\n` +
+           `**Step 1**: Read SKILL.md to understand the skill workflow\n` +
+           `  - read_file: ${skillMdPathOnly}\n\n` +
+           `**Step 2**: Explore skill directory and read reference files (if needed)\n` +
+           `  - ListDirectory(path="${skillPath}")\n` +
+           `  - read_file relevant .md and script files as needed\n\n` +
+           `Then analyze the documentation and create your own execution plan.\n\n` +
+           `**Workspace**: \`${taskWorkspace}\`\n\n` +
+           `**⚠️ Windows Path Execution**: Use absolute paths, NOT \`cd && command\`:\n` +
+           `  - ✅ Correct: \`node "${taskWorkspace}/script.js"\`\n` +
+           `  - ❌ Wrong: \`cd "${taskWorkspace}" && node script.js\` (fails in PowerShell 5.1)\n` +
+           `  - ✅ Correct: \`python "${taskWorkspace}/script.py"\`\n\n` +
+           `**📦 Dependency Management**:\n` +
+           `  - Use \`skillPath\` parameter to install dependencies to skill's node_modules (persists across invocations)\n` +
+           `  - ✅ Correct: Bash(command="npm install <package>", skillPath="${skillPath}")\n` +
+           `  - ✅ Correct: Bash(command="npm install", skillPath="${skillPath}")\n` +
+           `  - Dependencies are saved to: <userSkillsPath>/<skillName>/node_modules\n` +
+           `  - 💡 Tip: Install once, reuse forever - no need to reinstall on each call!\n` +
+           `  - After install, scripts can use require() directly with NODE_PATH auto-set\n` +
+           `  - NODE_PATH precedence: skill's node_modules > xAgent's node_modules\n` +
+           `  - Manual execution (Windows): set "NODE_PATH=${skillPath}/node_modules;<xAgentPath>/node_modules" && node script.js\n` +
+           `  - Manual execution (Linux/Mac): NODE_PATH=${skillPath}/node_modules:${process.cwd()}/node_modules node script.js\n` +
+           `  - ⚠️ If skillPath approach fails (module not found errors): Install directly in workspace\n` +
+           `    Priority: skill's node_modules > workspace's node_modules > xAgent's node_modules\n\n` +
+           `**🧹 Cleanup**: Delete all intermediate/temporary files when task completes:\n` +
+           `  - Remove: all files generated during the task\n` +
+           `  - Keep: Only the final output file (output.pptx/docx/xlsx/pdf or other file format required by user)\n` +
+           `  - ⚠️ If user needs to check results or make adjustments: RETAIN intermediate/temporary files for debugging\n\n` +
+           `**Instructions**: read_file the documentation, understand the API, and create your own execution plan.\n` +
+           `**If you encounter issues**: Explain what went wrong and suggest a different approach.\n`;
+  }
+}
+// ============================================================
+// ============================================================
+
+/**
+ * Execute skill - LLM analyzes SKILL.md and generates its own steps
+ * @param skill Skill to execute
+ * @param params Execution parameters
+ * @returns Execution result with guidance
+ */
+export async function executeSkill(
+  skill: SkillInfo,
+  params: SkillExecutionParams
+): Promise<SkillExecutionResult> {
+  const executor = new GenericSkillExecutor();
+  return executor.execute(skill, params);
+}
+
+// ============================================================
+// Singleton Instance and Exports
+// ============================================================
+
+let skillInvokerInstance: SkillInvoker | null = null;
+
+export function getSkillInvoker(): SkillInvoker {
+  if (!skillInvokerInstance) {
+    skillInvokerInstance = new SkillInvoker();
+  }
+  return skillInvokerInstance;
+}