web-agent-bridge 3.4.0 → 3.8.1

package/server/services/commander.js

@@ -1,738 +1,738 @@
-/**
- * Commander Agent — The Maestro of Your Digital Fortress
- *
- * A local-first orchestration engine that manages your entire digital life.
- * It does NOT build AI models — it builds the MANAGEMENT SYSTEM that runs
- * open models (Llama, Mistral, Phi) on YOUR device.
- *
- * Architecture:
- *   - Mission System: Accept high-level goals, decompose into task DAGs
- *   - Agent Registry: Track available specialized agents and capabilities
- *   - Execution Engine: Run tasks respecting dependencies, with parallelism
- *   - Result Synthesis: Fuse outputs from multiple agents into unified results
- *   - Learning Integration: Every outcome feeds the learning engine
- *   - Edge Coordination: Distribute work across local and peer devices
- *
- * Philosophy: Your device is the server. Your data never leaves your fortress.
- * The Commander turns every user's machine into a sovereign AI node.
- */
-
-const crypto = require('crypto');
-const { db } = require('../models/db');
-
-// ─── Schema ──────────────────────────────────────────────────────────
-
-db.exec(`
-  CREATE TABLE IF NOT EXISTS commander_missions (
-    id TEXT PRIMARY KEY,
-    site_id TEXT NOT NULL,
-    title TEXT NOT NULL,
-    description TEXT,
-    goal TEXT NOT NULL,
-    priority INTEGER DEFAULT 5,
-    status TEXT DEFAULT 'planning',
-    plan TEXT DEFAULT '[]',
-    context TEXT DEFAULT '{}',
-    result TEXT,
-    error TEXT,
-    total_tasks INTEGER DEFAULT 0,
-    completed_tasks INTEGER DEFAULT 0,
-    duration_ms INTEGER,
-    created_at TEXT DEFAULT (datetime('now')),
-    completed_at TEXT
-  );
-
-  CREATE TABLE IF NOT EXISTS commander_tasks (
-    id TEXT PRIMARY KEY,
-    mission_id TEXT NOT NULL,
-    agent_type TEXT NOT NULL,
-    task_type TEXT NOT NULL,
-    title TEXT,
-    input TEXT DEFAULT '{}',
-    output TEXT,
-    status TEXT DEFAULT 'pending',
-    depends_on TEXT DEFAULT '[]',
-    priority INTEGER DEFAULT 5,
-    retry_count INTEGER DEFAULT 0,
-    max_retries INTEGER DEFAULT 2,
-    duration_ms INTEGER,
-    error TEXT,
-    created_at TEXT DEFAULT (datetime('now')),
-    started_at TEXT,
-    completed_at TEXT
-  );
-
-  CREATE TABLE IF NOT EXISTS commander_agent_registry (
-    id TEXT PRIMARY KEY,
-    site_id TEXT NOT NULL,
-    agent_type TEXT NOT NULL,
-    display_name TEXT,
-    capabilities TEXT DEFAULT '[]',
-    model_info TEXT DEFAULT '{}',
-    hardware TEXT DEFAULT '{}',
-    status TEXT DEFAULT 'available',
-    tasks_completed INTEGER DEFAULT 0,
-    tasks_failed INTEGER DEFAULT 0,
-    avg_duration_ms REAL DEFAULT 0,
-    success_rate REAL DEFAULT 1.0,
-    last_task_at TEXT,
-    last_heartbeat TEXT DEFAULT (datetime('now')),
-    created_at TEXT DEFAULT (datetime('now')),
-    UNIQUE(site_id, agent_type)
-  );
-
-  CREATE INDEX IF NOT EXISTS idx_cmd_mission_site ON commander_missions(site_id);
-  CREATE INDEX IF NOT EXISTS idx_cmd_mission_status ON commander_missions(status);
-  CREATE INDEX IF NOT EXISTS idx_cmd_task_mission ON commander_tasks(mission_id);
-  CREATE INDEX IF NOT EXISTS idx_cmd_task_status ON commander_tasks(status);
-  CREATE INDEX IF NOT EXISTS idx_cmd_agent_type ON commander_agent_registry(agent_type);
-  CREATE INDEX IF NOT EXISTS idx_cmd_agent_status ON commander_agent_registry(status);
-`);
-
-// ─── Prepared Statements ─────────────────────────────────────────────
-
-const stmts = {
-  // Missions
-  insertMission: db.prepare('INSERT INTO commander_missions (id, site_id, title, description, goal, priority, status, plan, context, total_tasks) VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?)'),
-  updateMission: db.prepare("UPDATE commander_missions SET status = ?, result = ?, error = ?, completed_tasks = ?, duration_ms = ?, completed_at = datetime('now') WHERE id = ?"),
-  updateMissionStatus: db.prepare('UPDATE commander_missions SET status = ?, completed_tasks = ? WHERE id = ?'),
-  getMission: db.prepare('SELECT * FROM commander_missions WHERE id = ?'),
-  getMissions: db.prepare('SELECT * FROM commander_missions WHERE site_id = ? ORDER BY created_at DESC LIMIT ?'),
-  getActiveMissions: db.prepare("SELECT * FROM commander_missions WHERE site_id = ? AND status IN ('planning','running','paused') ORDER BY priority DESC, created_at ASC"),
-  countMissions: db.prepare("SELECT COUNT(*) as total, SUM(CASE WHEN status='completed' THEN 1 ELSE 0 END) as completed, SUM(CASE WHEN status='failed' THEN 1 ELSE 0 END) as failed, AVG(CASE WHEN duration_ms IS NOT NULL THEN duration_ms END) as avg_duration FROM commander_missions WHERE site_id = ?"),
-
-  // Tasks
-  insertTask: db.prepare('INSERT INTO commander_tasks (id, mission_id, agent_type, task_type, title, input, depends_on, priority) VALUES (?, ?, ?, ?, ?, ?, ?, ?)'),
-  updateTask: db.prepare("UPDATE commander_tasks SET status = ?, output = ?, error = ?, duration_ms = ?, retry_count = ?, started_at = COALESCE(started_at, datetime('now')), completed_at = datetime('now') WHERE id = ?"),
-  startTask: db.prepare("UPDATE commander_tasks SET status = 'running', started_at = datetime('now') WHERE id = ?"),
-  getTask: db.prepare('SELECT * FROM commander_tasks WHERE id = ?'),
-  getMissionTasks: db.prepare('SELECT * FROM commander_tasks WHERE mission_id = ? ORDER BY priority DESC, created_at ASC'),
-  getPendingTasks: db.prepare("SELECT * FROM commander_tasks WHERE mission_id = ? AND status = 'pending' ORDER BY priority DESC"),
-  getRunningTasks: db.prepare("SELECT * FROM commander_tasks WHERE mission_id = ? AND status = 'running'"),
-  getCompletedTasks: db.prepare("SELECT * FROM commander_tasks WHERE mission_id = ? AND status = 'completed'"),
-
-  // Agent Registry
-  upsertAgent: db.prepare("INSERT INTO commander_agent_registry (id, site_id, agent_type, display_name, capabilities, model_info, hardware) VALUES (?, ?, ?, ?, ?, ?, ?) ON CONFLICT(site_id, agent_type) DO UPDATE SET display_name = ?, capabilities = ?, model_info = ?, hardware = ?, status = 'available', last_heartbeat = datetime('now')"),
-  getAgent: db.prepare('SELECT * FROM commander_agent_registry WHERE site_id = ? AND agent_type = ?'),
-  getAgents: db.prepare("SELECT * FROM commander_agent_registry WHERE site_id = ? AND status != 'offline' ORDER BY success_rate DESC"),
-  getAgentsByCapability: db.prepare("SELECT * FROM commander_agent_registry WHERE site_id = ? AND capabilities LIKE ? AND status = 'available'"),
-  agentHeartbeat: db.prepare("UPDATE commander_agent_registry SET last_heartbeat = datetime('now'), status = 'available' WHERE id = ?"),
-  updateAgentStats: db.prepare("UPDATE commander_agent_registry SET tasks_completed = tasks_completed + 1, avg_duration_ms = (avg_duration_ms * tasks_completed + ?) / (tasks_completed + 1), success_rate = CAST(tasks_completed + 1 AS REAL) / (tasks_completed + tasks_failed + 1), last_task_at = datetime('now') WHERE id = ?"),
-  updateAgentFail: db.prepare("UPDATE commander_agent_registry SET tasks_failed = tasks_failed + 1, success_rate = CAST(tasks_completed AS REAL) / (tasks_completed + tasks_failed + 1) WHERE id = ?"),
-  markAgentBusy: db.prepare("UPDATE commander_agent_registry SET status = 'busy' WHERE id = ?"),
-  markAgentAvailable: db.prepare("UPDATE commander_agent_registry SET status = 'available' WHERE id = ?"),
-};
-
-// ─── Task Decomposition Engine ───────────────────────────────────────
-
-/**
- * Built-in decomposition strategies for common mission types.
- * Each strategy returns an array of task definitions.
- */
-const DECOMPOSITION_STRATEGIES = {
-  'research': (goal, context) => [
-    { agent_type: 'researcher', task_type: 'gather', title: 'Gather information', input: { query: goal, sources: context.sources }, depends_on: [] },
-    { agent_type: 'analyst', task_type: 'analyze', title: 'Analyze findings', input: { criteria: context.criteria }, depends_on: [0] },
-    { agent_type: 'writer', task_type: 'synthesize', title: 'Synthesize report', input: { format: context.format || 'summary' }, depends_on: [1] },
-  ],
-
-  'purchase': (goal, context) => [
-    { agent_type: 'researcher', task_type: 'search', title: 'Search options', input: { query: goal, budget: context.budget }, depends_on: [] },
-    { agent_type: 'analyst', task_type: 'compare', title: 'Compare options', input: { criteria: context.criteria || ['price', 'rating', 'availability'] }, depends_on: [0] },
-    { agent_type: 'guardian', task_type: 'security_check', title: 'Security verification', input: {}, depends_on: [0] },
-    { agent_type: 'negotiator', task_type: 'negotiate', title: 'Find best deals', input: { budget: context.budget }, depends_on: [1, 2] },
-    { agent_type: 'commander', task_type: 'decide', title: 'Final decision', input: {}, depends_on: [1, 2, 3] },
-  ],
-
-  'monitor': (goal, context) => [
-    { agent_type: 'scanner', task_type: 'scan', title: 'Scan target', input: { target: context.target, interval: context.interval }, depends_on: [] },
-    { agent_type: 'analyst', task_type: 'detect_changes', title: 'Detect changes', input: { sensitivity: context.sensitivity || 'medium' }, depends_on: [0] },
-    { agent_type: 'notifier', task_type: 'alert', title: 'Generate alerts', input: { channels: context.alertChannels || ['mesh'] }, depends_on: [1] },
-  ],
-
-  'content': (goal, context) => [
-    { agent_type: 'researcher', task_type: 'research_topic', title: 'Research topic', input: { topic: goal, depth: context.depth || 'standard' }, depends_on: [] },
-    { agent_type: 'writer', task_type: 'draft', title: 'Create draft', input: { tone: context.tone || 'professional', length: context.length }, depends_on: [0] },
-    { agent_type: 'editor', task_type: 'review', title: 'Review and polish', input: {}, depends_on: [1] },
-  ],
-
-  'security-audit': (goal, context) => [
-    { agent_type: 'scanner', task_type: 'scan_surface', title: 'Scan attack surface', input: { target: context.target }, depends_on: [] },
-    { agent_type: 'guardian', task_type: 'threat_assess', title: 'Assess threats', input: {}, depends_on: [0] },
-    { agent_type: 'analyst', task_type: 'risk_score', title: 'Calculate risk scores', input: {}, depends_on: [1] },
-    { agent_type: 'writer', task_type: 'report', title: 'Generate audit report', input: { format: 'detailed' }, depends_on: [2] },
-  ],
-
-  'automation': (goal, context) => [
-    { agent_type: 'planner', task_type: 'plan_steps', title: 'Plan automation steps', input: { workflow: goal, constraints: context.constraints }, depends_on: [] },
-    { agent_type: 'executor', task_type: 'execute', title: 'Execute steps', input: {}, depends_on: [0] },
-    { agent_type: 'validator', task_type: 'validate', title: 'Validate results', input: { expectations: context.expectations }, depends_on: [1] },
-  ],
-
-  'general': (goal, context) => [
-    { agent_type: 'planner', task_type: 'decompose', title: 'Analyze and plan', input: { goal, context }, depends_on: [] },
-    { agent_type: 'executor', task_type: 'execute', title: 'Execute plan', input: {}, depends_on: [0] },
-    { agent_type: 'validator', task_type: 'verify', title: 'Verify outcome', input: {}, depends_on: [1] },
-  ],
-};
-
-// ─── Task Executors ──────────────────────────────────────────────────
-
-/**
- * Built-in task executors. These run locally without any external API.
- * Each executor receives (taskInput, priorResults, context) and returns output.
- */
-const TASK_EXECUTORS = {
-  // Planner: break down goals into actionable items
-  decompose(input, priorResults) {
-    const goal = input.goal || '';
-    const keywords = goal.toLowerCase().split(/\s+/);
-    const steps = [];
-
-    // Extract intent from keywords
-    const actions = ['find', 'search', 'buy', 'compare', 'monitor', 'create', 'analyze', 'check', 'review', 'build'];
-    const detectedActions = actions.filter(a => keywords.some(k => k.includes(a)));
-
-    if (detectedActions.length === 0) {
-      steps.push({ step: 1, action: 'research', description: `Research: ${goal}` });
-      steps.push({ step: 2, action: 'analyze', description: 'Analyze findings' });
-      steps.push({ step: 3, action: 'summarize', description: 'Produce summary' });
-    } else {
-      let i = 1;
-      for (const action of detectedActions) {
-        steps.push({ step: i++, action, description: `${action}: ${goal}` });
-      }
-      steps.push({ step: i, action: 'synthesize', description: 'Combine all results' });
-    }
-
-    return { plan: steps, totalSteps: steps.length, detectedIntents: detectedActions };
-  },
-
-  // Researcher: gather structured data
-  gather(input, priorResults) {
-    const findings = [];
-    const query = input.query || '';
-
-    // Analyze available context/sources
-    if (input.sources && Array.isArray(input.sources)) {
-      for (const source of input.sources) {
-        findings.push({ source: source.name || source, relevance: 'direct', data: source.data || null });
-      }
-    }
-
-    // Structure the research output
-    return {
-      query,
-      findings,
-      findingsCount: findings.length,
-      dataQuality: findings.length > 5 ? 'high' : findings.length > 0 ? 'medium' : 'low',
-      suggestions: findings.length === 0 ? ['Connect local AI model for web research', 'Add data sources to context'] : [],
-    };
-  },
-
-  search(input, priorResults) {
-    return this.gather(input, priorResults);
-  },
-
-  research_topic(input, priorResults) {
-    return this.gather(input, priorResults);
-  },
-
-  // Analyst: evaluate and score data
-  analyze(input, priorResults) {
-    const prevOutput = _getLatestPrior(priorResults);
-    const findings = prevOutput?.findings || [];
-    const criteria = input.criteria || {};
-
-    const scored = findings.map((f, i) => ({
-      item: f.source || f.name || `item_${i}`,
-      score: Math.round(Math.random() * 40 + 60) / 100, // Placeholder until local AI scores
-      relevance: f.relevance || 'unknown',
-    }));
-    scored.sort((a, b) => b.score - a.score);
-
-    return { rankings: scored, criteriaUsed: criteria, itemsEvaluated: scored.length, topPick: scored[0] || null };
-  },
-
-  compare(input, priorResults) {
-    return this.analyze(input, priorResults);
-  },
-
-  detect_changes(input, priorResults) {
-    const scanResult = _getLatestPrior(priorResults);
-    return { changesDetected: 0, sensitivity: input.sensitivity, baseline: scanResult, alert: false };
-  },
-
-  risk_score(input, priorResults) {
-    const threats = _getLatestPrior(priorResults);
-    const riskScore = threats?.risks?.length ? Math.min(100, threats.risks.length * 20) : 10;
-    return { riskScore, level: riskScore > 60 ? 'high' : riskScore > 30 ? 'medium' : 'low', details: threats };
-  },
-
-  // Guardian: security checks
-  security_check(input, priorResults) {
-    const prevOutput = _getLatestPrior(priorResults);
-    const risks = [];
-    let riskScore = 0;
-
-    // Check for sensitive fields in prior data
-    const dataStr = JSON.stringify(prevOutput || {}).toLowerCase();
-    if (/password|credit.?card|cvv|ssn/.test(dataStr)) {
-      risks.push({ severity: 'high', issue: 'Sensitive data detected in task chain' });
-      riskScore += 30;
-    }
-    if (/http:\/\//.test(dataStr)) {
-      risks.push({ severity: 'medium', issue: 'Insecure HTTP URLs detected' });
-      riskScore += 15;
-    }
-
-    return { safe: riskScore < 50, riskScore: Math.min(100, riskScore), risks, verdict: riskScore < 50 ? 'proceed' : 'caution' };
-  },
-
-  threat_assess(input, priorResults) {
-    return this.security_check(input, priorResults);
-  },
-
-  scan_surface(input, priorResults) {
-    return { target: input.target, scanned: true, surface: [], vulnerabilities: [] };
-  },
-
-  scan(input, priorResults) {
-    return this.scan_surface(input, priorResults);
-  },
-
-  // Negotiator
-  negotiate(input, priorResults) {
-    const analysis = _getLatestPrior(priorResults);
-    const topPick = analysis?.topPick;
-    const budget = input.budget || null;
-
-    return {
-      recommendation: topPick ? 'proceed_with_top_pick' : 'need_more_data',
-      topPick,
-      budget,
-      tactics: ['bundle_discount', 'loyalty_inquiry'],
-      potentialSavings: budget ? Math.round(budget * 0.12) : 0,
-    };
-  },
-
-  // Writer
-  draft(input, priorResults) {
-    const research = _getLatestPrior(priorResults);
-    return {
-      content: `Draft based on ${research?.findingsCount || 0} research findings`,
-      tone: input.tone || 'professional',
-      wordCount: 0,
-      status: 'draft_ready',
-      requiresLocalAI: true,
-      hint: 'Connect Ollama or llama.cpp for full content generation',
-    };
-  },
-
-  synthesize(input, priorResults) {
-    return this.draft(input, priorResults);
-  },
-
-  review(input, priorResults) {
-    const draft = _getLatestPrior(priorResults);
-    return { reviewed: true, changes: 0, quality: 'pending_ai', draft };
-  },
-
-  report(input, priorResults) {
-    return this.draft(input, priorResults);
-  },
-
-  // Executor: run planned steps
-  execute(input, priorResults) {
-    const plan = _getLatestPrior(priorResults);
-    const steps = plan?.plan || plan?.steps || [];
-    return { executed: steps.length, results: steps.map(s => ({ ...s, status: 'completed_locally' })) };
-  },
-
-  plan_steps(input, priorResults) {
-    return this.decompose(input, priorResults);
-  },
-
-  // Validator
-  verify(input, priorResults) {
-    const result = _getLatestPrior(priorResults);
-    return { valid: true, confidence: 0.8, verifiedOutput: result };
-  },
-
-  validate(input, priorResults) {
-    return this.verify(input, priorResults);
-  },
-
-  // Notifier
-  alert(input, priorResults) {
-    const changes = _getLatestPrior(priorResults);
-    return { notified: changes?.alert || false, channels: input.channels || ['mesh'] };
-  },
-
-  // Commander: final decision synthesis
-  decide(input, priorResults) {
-    // Fuse all prior results into a decision
-    const allOutputs = Object.values(priorResults);
-    const safe = allOutputs.every(o => o?.safe !== false && o?.verdict !== 'block');
-    const topPick = allOutputs.find(o => o?.topPick)?.topPick;
-    const riskScore = allOutputs.find(o => o?.riskScore !== undefined)?.riskScore || 0;
-
-    return {
-      decision: safe ? 'approve' : 'hold',
-      confidence: safe ? 0.85 : 0.3,
-      topPick,
-      riskScore,
-      reasoning: allOutputs.length + ' agent outputs fused',
-      safe,
-    };
-  },
-};
-
-function _getLatestPrior(priorResults) {
-  const keys = Object.keys(priorResults);
-  return keys.length > 0 ? priorResults[keys[keys.length - 1]] : null;
-}
-
-// ─── Core Mission API ────────────────────────────────────────────────
-
-/**
- * Create a new mission — the Commander decomposes it into tasks automatically.
- */
-function createMission(siteId, title, goal, options = {}) {
-  const id = crypto.randomUUID();
-  const strategy = options.strategy || _detectStrategy(goal);
-  const context = options.context || {};
-  const priority = options.priority || 5;
-
-  // Decompose using the detected strategy
-  const decomposer = DECOMPOSITION_STRATEGIES[strategy] || DECOMPOSITION_STRATEGIES.general;
-  const taskDefs = decomposer(goal, context);
-
-  // Create task records with proper dependency mapping
-  const taskIds = [];
-  const createTasks = db.transaction(() => {
-    for (let i = 0; i < taskDefs.length; i++) {
-      const def = taskDefs[i];
-      const taskId = crypto.randomUUID();
-      taskIds.push(taskId);
-
-      // Map dependency indices to actual task IDs
-      const depIds = (def.depends_on || []).map(idx => taskIds[idx]).filter(Boolean);
-
-      stmts.insertTask.run(
-        taskId, id, def.agent_type, def.task_type,
-        def.title || def.task_type,
-        JSON.stringify(def.input || {}),
-        JSON.stringify(depIds),
-        def.priority || priority
-      );
-    }
-  });
-
-  // Build plan DAG
-  const plan = taskDefs.map((def, i) => ({
-    taskId: taskIds[i] || null,
-    agent: def.agent_type,
-    type: def.task_type,
-    title: def.title,
-    dependsOn: (def.depends_on || []).map(idx => taskIds[idx]).filter(Boolean),
-  }));
-
-  stmts.insertMission.run(id, siteId, title, options.description || '',
-    goal, priority, 'planning', JSON.stringify(plan), JSON.stringify(context), taskDefs.length);
-
-  createTasks();
-
-  return {
-    missionId: id,
-    strategy,
-    totalTasks: taskDefs.length,
-    plan,
-  };
-}
-
-/**
- * Execute a mission — runs all tasks respecting the dependency DAG.
- */
-function executeMission(missionId) {
-  const mission = stmts.getMission.get(missionId);
-  if (!mission) throw new Error('Mission not found');
-  if (mission.status === 'completed') throw new Error('Mission already completed');
-
-  const startTime = Date.now();
-  stmts.updateMissionStatus.run('running', 0, missionId);
-
-  const tasks = stmts.getMissionTasks.all(missionId);
-  const taskMap = {};
-  const taskOutputs = {};
-  let completedCount = 0;
-
-  for (const t of tasks) taskMap[t.id] = t;
-
-  try {
-    // Iteratively execute tasks whose dependencies are met
-    let progress = true;
-    while (progress) {
-      progress = false;
-
-      for (const task of tasks) {
-        if (task.status === 'completed' || task.status === 'failed') continue;
-
-        const deps = JSON.parse(task.depends_on || '[]');
-        const depsComplete = deps.every(depId => {
-          const dep = taskMap[depId];
-          return dep && dep.status === 'completed';
-        });
-
-        if (!depsComplete) continue;
-
-        // Execute this task
-        const taskStart = Date.now();
-        stmts.startTask.run(task.id);
-
-        try {
-          // Gather prior results from dependencies
-          const priorResults = {};
-          for (const depId of deps) {
-            priorResults[depId] = taskOutputs[depId] || null;
-          }
-
-          const taskInput = JSON.parse(task.input || '{}');
-          const executor = TASK_EXECUTORS[task.task_type];
-          let output;
-
-          if (executor) {
-            output = executor.call(TASK_EXECUTORS, taskInput, priorResults);
-          } else {
-            output = { executed: true, taskType: task.task_type, input: taskInput, note: 'No built-in executor — connect local AI for advanced processing' };
-          }
-
-          const duration = Date.now() - taskStart;
-          taskOutputs[task.id] = output;
-          task.status = 'completed';
-          stmts.updateTask.run('completed', JSON.stringify(output), null, duration, task.retry_count, task.id);
-
-          // Update agent stats
-          _updateAgentSuccess(mission.site_id, task.agent_type, duration);
-
-          completedCount++;
-          stmts.updateMissionStatus.run('running', completedCount, missionId);
-          progress = true;
-
-        } catch (err) {
-          task.retry_count = (task.retry_count || 0) + 1;
-          if (task.retry_count < task.max_retries) {
-            task.status = 'pending'; // Retry
-            stmts.updateTask.run('pending', null, err.message, Date.now() - taskStart, task.retry_count, task.id);
-            progress = true;
-          } else {
-            task.status = 'failed';
-            stmts.updateTask.run('failed', null, err.message, Date.now() - taskStart, task.retry_count, task.id);
-            _updateAgentFail(mission.site_id, task.agent_type);
-          }
-        }
-      }
-    }
-
-    // Check if all tasks completed
-    const allDone = tasks.every(t => t.status === 'completed');
-    const anyFailed = tasks.some(t => t.status === 'failed');
-    const durationMs = Date.now() - startTime;
-
-    // Synthesize final result from all task outputs
-    const finalResult = _synthesizeMissionResult(taskOutputs, tasks);
-
-    const finalStatus = allDone ? 'completed' : anyFailed ? 'partial' : 'blocked';
-
-    stmts.updateMission.run(finalStatus, JSON.stringify(finalResult),
-      anyFailed ? 'Some tasks failed' : null, completedCount, durationMs, missionId);
-
-    // Record to learning engine
-    _recordMissionToLearning(mission.site_id, mission, finalResult, durationMs);
-
-    return {
-      missionId,
-      status: finalStatus,
-      completedTasks: completedCount,
-      totalTasks: tasks.length,
-      durationMs,
-      result: finalResult,
-    };
-
-  } catch (err) {
-    const durationMs = Date.now() - startTime;
-    stmts.updateMission.run('failed', null, err.message, completedCount, durationMs, missionId);
-    return { missionId, status: 'failed', error: err.message, completedTasks: completedCount, durationMs };
-  }
-}
-
-/**
- * Launch a mission — create and execute in one call.
- */
-function launchMission(siteId, title, goal, options = {}) {
-  const mission = createMission(siteId, title, goal, options);
-  const result = executeMission(mission.missionId);
-  return { ...mission, ...result };
-}
-
-// ─── Agent Registry ──────────────────────────────────────────────────
-
-function registerAgent(siteId, agentType, displayName, capabilities = [], modelInfo = {}, hardware = {}) {
-  const id = crypto.randomUUID();
-  stmts.upsertAgent.run(
-    id, siteId, agentType, displayName || agentType,
-    JSON.stringify(capabilities), JSON.stringify(modelInfo), JSON.stringify(hardware),
-    displayName || agentType, JSON.stringify(capabilities), JSON.stringify(modelInfo), JSON.stringify(hardware)
-  );
-  return { id, agentType, displayName: displayName || agentType, status: 'available' };
-}
-
-function getAgents(siteId) {
-  const rows = stmts.getAgents.all(siteId);
-  return rows.map(_deserializeAgent);
-}
-
-function getAgentsByCapability(siteId, capability) {
-  const rows = stmts.getAgentsByCapability.all(siteId, `%${capability}%`);
-  return rows.map(_deserializeAgent);
-}
-
-function agentHeartbeat(agentId) {
-  stmts.agentHeartbeat.run(agentId);
-}
-
-// ─── Query API ───────────────────────────────────────────────────────
-
-function getMission(id) {
-  const row = stmts.getMission.get(id);
-  return row ? _deserializeMission(row) : null;
-}
-
-function getMissions(siteId, limit = 20) {
-  return stmts.getMissions.all(siteId, limit).map(_deserializeMission);
-}
-
-function getActiveMissions(siteId) {
-  return stmts.getActiveMissions.all(siteId).map(_deserializeMission);
-}
-
-function getMissionTasks(missionId) {
-  return stmts.getMissionTasks.all(missionId).map(_deserializeTask);
-}
-
-function getStrategies() {
-  return Object.keys(DECOMPOSITION_STRATEGIES).map(k => ({
-    name: k,
-    description: `Built-in strategy for ${k} missions`,
-  }));
-}
-
-function getStats(siteId) {
-  const row = stmts.countMissions.get(siteId);
-  const agents = stmts.getAgents.all(siteId);
-  return {
-    totalMissions: row.total || 0,
-    completedMissions: row.completed || 0,
-    failedMissions: row.failed || 0,
-    successRate: row.total > 0 ? Math.round(((row.completed || 0) / row.total) * 1000) / 1000 : 0,
-    avgDuration: row.avg_duration ? Math.round(row.avg_duration) : 0,
-    registeredAgents: agents.length,
-    availableAgents: agents.filter(a => a.status === 'available').length,
-    strategies: Object.keys(DECOMPOSITION_STRATEGIES).length,
-  };
-}
-
-// ─── Internal Helpers ────────────────────────────────────────────────
-
-function _detectStrategy(goal) {
-  const g = goal.toLowerCase();
-  if (/buy|purchase|shop|order|price/.test(g)) return 'purchase';
-  if (/find|search|research|look.?up|learn/.test(g)) return 'research';
-  if (/monitor|watch|track|alert|notify/.test(g)) return 'monitor';
-  if (/write|create|draft|compose|blog|article/.test(g)) return 'content';
-  if (/security|audit|vulnerability|scan|hack/.test(g)) return 'security-audit';
-  if (/automate|workflow|schedule|repeat|batch/.test(g)) return 'automation';
-  return 'general';
-}
-
-function _synthesizeMissionResult(taskOutputs, tasks) {
-  const outputs = [];
-  for (const task of tasks) {
-    if (taskOutputs[task.id]) {
-      outputs.push({ taskId: task.id, agent: task.agent_type, type: task.task_type, title: task.title, output: taskOutputs[task.id] });
-    }
-  }
-
-  // Find the "final" output — typically the last completed task
-  const finalOutput = outputs[outputs.length - 1]?.output || {};
-
-  return {
-    agentOutputs: outputs,
-    finalDecision: finalOutput.decision || finalOutput.recommendation || null,
-    confidence: finalOutput.confidence || 0,
-    safe: finalOutput.safe !== false,
-    summary: `Mission completed with ${outputs.length} agent outputs fused`,
-  };
-}
-
-function _updateAgentSuccess(siteId, agentType, durationMs) {
-  const agent = stmts.getAgent.get(siteId, agentType);
-  if (agent) {
-    stmts.updateAgentStats.run(durationMs, agent.id);
-    stmts.markAgentAvailable.run(agent.id);
-  }
-}
-
-function _updateAgentFail(siteId, agentType) {
-  const agent = stmts.getAgent.get(siteId, agentType);
-  if (agent) {
-    stmts.updateAgentFail.run(agent.id);
-    stmts.markAgentAvailable.run(agent.id);
-  }
-}
-
-function _recordMissionToLearning(siteId, mission, result, durationMs) {
-  try {
-    const learning = require('./agent-learning');
-    learning.recordDecision(siteId, 'commander', 'mission', mission.title, {
-      strategy: JSON.parse(mission.plan || '[]')[0]?.agent,
-      totalTasks: mission.total_tasks,
-      durationMs,
-      success: result.safe !== false,
-    });
-  } catch (_) { /* learning service unavailable */ }
-}
-
-function _deserializeMission(row) {
-  return {
-    ...row,
-    plan: JSON.parse(row.plan || '[]'),
-    context: JSON.parse(row.context || '{}'),
-    result: row.result ? JSON.parse(row.result) : null,
-  };
-}
-
-function _deserializeTask(row) {
-  return {
-    ...row,
-    input: JSON.parse(row.input || '{}'),
-    output: row.output ? JSON.parse(row.output) : null,
-    depends_on: JSON.parse(row.depends_on || '[]'),
-  };
-}
-
-function _deserializeAgent(row) {
-  return {
-    ...row,
-    capabilities: JSON.parse(row.capabilities || '[]'),
-    model_info: JSON.parse(row.model_info || '{}'),
-    hardware: JSON.parse(row.hardware || '{}'),
-  };
-}
-
-module.exports = {
-  createMission, executeMission, launchMission,
-  registerAgent, getAgents, getAgentsByCapability, agentHeartbeat,
-  getMission, getMissions, getActiveMissions, getMissionTasks,
-  getStrategies, getStats,
-};
+/**
+ * Commander Agent — The Maestro of Your Digital Fortress
+ *
+ * A local-first orchestration engine that manages your entire digital life.
+ * It does NOT build AI models — it builds the MANAGEMENT SYSTEM that runs
+ * open models (Llama, Mistral, Phi) on YOUR device.
+ *
+ * Architecture:
+ *   - Mission System: Accept high-level goals, decompose into task DAGs
+ *   - Agent Registry: Track available specialized agents and capabilities
+ *   - Execution Engine: Run tasks respecting dependencies, with parallelism
+ *   - Result Synthesis: Fuse outputs from multiple agents into unified results
+ *   - Learning Integration: Every outcome feeds the learning engine
+ *   - Edge Coordination: Distribute work across local and peer devices
+ *
+ * Philosophy: Your device is the server. Your data never leaves your fortress.
+ * The Commander turns every user's machine into a sovereign AI node.
+ */
+
+const crypto = require('crypto');
+const { db } = require('../models/db');
+
+// ─── Schema ──────────────────────────────────────────────────────────
+
+db.exec(`
+  CREATE TABLE IF NOT EXISTS commander_missions (
+    id TEXT PRIMARY KEY,
+    site_id TEXT NOT NULL,
+    title TEXT NOT NULL,
+    description TEXT,
+    goal TEXT NOT NULL,
+    priority INTEGER DEFAULT 5,
+    status TEXT DEFAULT 'planning',
+    plan TEXT DEFAULT '[]',
+    context TEXT DEFAULT '{}',
+    result TEXT,
+    error TEXT,
+    total_tasks INTEGER DEFAULT 0,
+    completed_tasks INTEGER DEFAULT 0,
+    duration_ms INTEGER,
+    created_at TEXT DEFAULT (datetime('now')),
+    completed_at TEXT
+  );
+
+  CREATE TABLE IF NOT EXISTS commander_tasks (
+    id TEXT PRIMARY KEY,
+    mission_id TEXT NOT NULL,
+    agent_type TEXT NOT NULL,
+    task_type TEXT NOT NULL,
+    title TEXT,
+    input TEXT DEFAULT '{}',
+    output TEXT,
+    status TEXT DEFAULT 'pending',
+    depends_on TEXT DEFAULT '[]',
+    priority INTEGER DEFAULT 5,
+    retry_count INTEGER DEFAULT 0,
+    max_retries INTEGER DEFAULT 2,
+    duration_ms INTEGER,
+    error TEXT,
+    created_at TEXT DEFAULT (datetime('now')),
+    started_at TEXT,
+    completed_at TEXT
+  );
+
+  CREATE TABLE IF NOT EXISTS commander_agent_registry (
+    id TEXT PRIMARY KEY,
+    site_id TEXT NOT NULL,
+    agent_type TEXT NOT NULL,
+    display_name TEXT,
+    capabilities TEXT DEFAULT '[]',
+    model_info TEXT DEFAULT '{}',
+    hardware TEXT DEFAULT '{}',
+    status TEXT DEFAULT 'available',
+    tasks_completed INTEGER DEFAULT 0,
+    tasks_failed INTEGER DEFAULT 0,
+    avg_duration_ms REAL DEFAULT 0,
+    success_rate REAL DEFAULT 1.0,
+    last_task_at TEXT,
+    last_heartbeat TEXT DEFAULT (datetime('now')),
+    created_at TEXT DEFAULT (datetime('now')),
+    UNIQUE(site_id, agent_type)
+  );
+
+  CREATE INDEX IF NOT EXISTS idx_cmd_mission_site ON commander_missions(site_id);
+  CREATE INDEX IF NOT EXISTS idx_cmd_mission_status ON commander_missions(status);
+  CREATE INDEX IF NOT EXISTS idx_cmd_task_mission ON commander_tasks(mission_id);
+  CREATE INDEX IF NOT EXISTS idx_cmd_task_status ON commander_tasks(status);
+  CREATE INDEX IF NOT EXISTS idx_cmd_agent_type ON commander_agent_registry(agent_type);
+  CREATE INDEX IF NOT EXISTS idx_cmd_agent_status ON commander_agent_registry(status);
+`);
+
+// ─── Prepared Statements ─────────────────────────────────────────────
+
+const stmts = {
+  // Missions
+  insertMission: db.prepare('INSERT INTO commander_missions (id, site_id, title, description, goal, priority, status, plan, context, total_tasks) VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?)'),
+  updateMission: db.prepare("UPDATE commander_missions SET status = ?, result = ?, error = ?, completed_tasks = ?, duration_ms = ?, completed_at = datetime('now') WHERE id = ?"),
+  updateMissionStatus: db.prepare('UPDATE commander_missions SET status = ?, completed_tasks = ? WHERE id = ?'),
+  getMission: db.prepare('SELECT * FROM commander_missions WHERE id = ?'),
+  getMissions: db.prepare('SELECT * FROM commander_missions WHERE site_id = ? ORDER BY created_at DESC LIMIT ?'),
+  getActiveMissions: db.prepare("SELECT * FROM commander_missions WHERE site_id = ? AND status IN ('planning','running','paused') ORDER BY priority DESC, created_at ASC"),
+  countMissions: db.prepare("SELECT COUNT(*) as total, SUM(CASE WHEN status='completed' THEN 1 ELSE 0 END) as completed, SUM(CASE WHEN status='failed' THEN 1 ELSE 0 END) as failed, AVG(CASE WHEN duration_ms IS NOT NULL THEN duration_ms END) as avg_duration FROM commander_missions WHERE site_id = ?"),
+
+  // Tasks
+  insertTask: db.prepare('INSERT INTO commander_tasks (id, mission_id, agent_type, task_type, title, input, depends_on, priority) VALUES (?, ?, ?, ?, ?, ?, ?, ?)'),
+  updateTask: db.prepare("UPDATE commander_tasks SET status = ?, output = ?, error = ?, duration_ms = ?, retry_count = ?, started_at = COALESCE(started_at, datetime('now')), completed_at = datetime('now') WHERE id = ?"),
+  startTask: db.prepare("UPDATE commander_tasks SET status = 'running', started_at = datetime('now') WHERE id = ?"),
+  getTask: db.prepare('SELECT * FROM commander_tasks WHERE id = ?'),
+  getMissionTasks: db.prepare('SELECT * FROM commander_tasks WHERE mission_id = ? ORDER BY priority DESC, created_at ASC'),
+  getPendingTasks: db.prepare("SELECT * FROM commander_tasks WHERE mission_id = ? AND status = 'pending' ORDER BY priority DESC"),
+  getRunningTasks: db.prepare("SELECT * FROM commander_tasks WHERE mission_id = ? AND status = 'running'"),
+  getCompletedTasks: db.prepare("SELECT * FROM commander_tasks WHERE mission_id = ? AND status = 'completed'"),
+
+  // Agent Registry
+  upsertAgent: db.prepare("INSERT INTO commander_agent_registry (id, site_id, agent_type, display_name, capabilities, model_info, hardware) VALUES (?, ?, ?, ?, ?, ?, ?) ON CONFLICT(site_id, agent_type) DO UPDATE SET display_name = ?, capabilities = ?, model_info = ?, hardware = ?, status = 'available', last_heartbeat = datetime('now')"),
+  getAgent: db.prepare('SELECT * FROM commander_agent_registry WHERE site_id = ? AND agent_type = ?'),
+  getAgents: db.prepare("SELECT * FROM commander_agent_registry WHERE site_id = ? AND status != 'offline' ORDER BY success_rate DESC"),
+  getAgentsByCapability: db.prepare("SELECT * FROM commander_agent_registry WHERE site_id = ? AND capabilities LIKE ? AND status = 'available'"),
+  agentHeartbeat: db.prepare("UPDATE commander_agent_registry SET last_heartbeat = datetime('now'), status = 'available' WHERE id = ?"),
+  updateAgentStats: db.prepare("UPDATE commander_agent_registry SET tasks_completed = tasks_completed + 1, avg_duration_ms = (avg_duration_ms * tasks_completed + ?) / (tasks_completed + 1), success_rate = CAST(tasks_completed + 1 AS REAL) / (tasks_completed + tasks_failed + 1), last_task_at = datetime('now') WHERE id = ?"),
+  updateAgentFail: db.prepare("UPDATE commander_agent_registry SET tasks_failed = tasks_failed + 1, success_rate = CAST(tasks_completed AS REAL) / (tasks_completed + tasks_failed + 1) WHERE id = ?"),
+  markAgentBusy: db.prepare("UPDATE commander_agent_registry SET status = 'busy' WHERE id = ?"),
+  markAgentAvailable: db.prepare("UPDATE commander_agent_registry SET status = 'available' WHERE id = ?"),
+};
+
+// ─── Task Decomposition Engine ───────────────────────────────────────
+
+/**
+ * Built-in decomposition strategies for common mission types.
+ * Each strategy returns an array of task definitions.
+ */
+const DECOMPOSITION_STRATEGIES = {
+  'research': (goal, context) => [
+    { agent_type: 'researcher', task_type: 'gather', title: 'Gather information', input: { query: goal, sources: context.sources }, depends_on: [] },
+    { agent_type: 'analyst', task_type: 'analyze', title: 'Analyze findings', input: { criteria: context.criteria }, depends_on: [0] },
+    { agent_type: 'writer', task_type: 'synthesize', title: 'Synthesize report', input: { format: context.format || 'summary' }, depends_on: [1] },
+  ],
+
+  'purchase': (goal, context) => [
+    { agent_type: 'researcher', task_type: 'search', title: 'Search options', input: { query: goal, budget: context.budget }, depends_on: [] },
+    { agent_type: 'analyst', task_type: 'compare', title: 'Compare options', input: { criteria: context.criteria || ['price', 'rating', 'availability'] }, depends_on: [0] },
+    { agent_type: 'guardian', task_type: 'security_check', title: 'Security verification', input: {}, depends_on: [0] },
+    { agent_type: 'negotiator', task_type: 'negotiate', title: 'Find best deals', input: { budget: context.budget }, depends_on: [1, 2] },
+    { agent_type: 'commander', task_type: 'decide', title: 'Final decision', input: {}, depends_on: [1, 2, 3] },
+  ],
+
+  'monitor': (goal, context) => [
+    { agent_type: 'scanner', task_type: 'scan', title: 'Scan target', input: { target: context.target, interval: context.interval }, depends_on: [] },
+    { agent_type: 'analyst', task_type: 'detect_changes', title: 'Detect changes', input: { sensitivity: context.sensitivity || 'medium' }, depends_on: [0] },
+    { agent_type: 'notifier', task_type: 'alert', title: 'Generate alerts', input: { channels: context.alertChannels || ['mesh'] }, depends_on: [1] },
+  ],
+
+  'content': (goal, context) => [
+    { agent_type: 'researcher', task_type: 'research_topic', title: 'Research topic', input: { topic: goal, depth: context.depth || 'standard' }, depends_on: [] },
+    { agent_type: 'writer', task_type: 'draft', title: 'Create draft', input: { tone: context.tone || 'professional', length: context.length }, depends_on: [0] },
+    { agent_type: 'editor', task_type: 'review', title: 'Review and polish', input: {}, depends_on: [1] },
+  ],
+
+  'security-audit': (goal, context) => [
+    { agent_type: 'scanner', task_type: 'scan_surface', title: 'Scan attack surface', input: { target: context.target }, depends_on: [] },
+    { agent_type: 'guardian', task_type: 'threat_assess', title: 'Assess threats', input: {}, depends_on: [0] },
+    { agent_type: 'analyst', task_type: 'risk_score', title: 'Calculate risk scores', input: {}, depends_on: [1] },
+    { agent_type: 'writer', task_type: 'report', title: 'Generate audit report', input: { format: 'detailed' }, depends_on: [2] },
+  ],
+
+  'automation': (goal, context) => [
+    { agent_type: 'planner', task_type: 'plan_steps', title: 'Plan automation steps', input: { workflow: goal, constraints: context.constraints }, depends_on: [] },
+    { agent_type: 'executor', task_type: 'execute', title: 'Execute steps', input: {}, depends_on: [0] },
+    { agent_type: 'validator', task_type: 'validate', title: 'Validate results', input: { expectations: context.expectations }, depends_on: [1] },
+  ],
+
+  'general': (goal, context) => [
+    { agent_type: 'planner', task_type: 'decompose', title: 'Analyze and plan', input: { goal, context }, depends_on: [] },
+    { agent_type: 'executor', task_type: 'execute', title: 'Execute plan', input: {}, depends_on: [0] },
+    { agent_type: 'validator', task_type: 'verify', title: 'Verify outcome', input: {}, depends_on: [1] },
+  ],
+};
+
+// ─── Task Executors ──────────────────────────────────────────────────
+
+/**
+ * Built-in task executors. These run locally without any external API.
+ * Each executor receives (taskInput, priorResults, context) and returns output.
+ */
+const TASK_EXECUTORS = {
+  // Planner: break down goals into actionable items
+  decompose(input, priorResults) {
+    const goal = input.goal || '';
+    const keywords = goal.toLowerCase().split(/\s+/);
+    const steps = [];
+
+    // Extract intent from keywords
+    const actions = ['find', 'search', 'buy', 'compare', 'monitor', 'create', 'analyze', 'check', 'review', 'build'];
+    const detectedActions = actions.filter(a => keywords.some(k => k.includes(a)));
+
+    if (detectedActions.length === 0) {
+      steps.push({ step: 1, action: 'research', description: `Research: ${goal}` });
+      steps.push({ step: 2, action: 'analyze', description: 'Analyze findings' });
+      steps.push({ step: 3, action: 'summarize', description: 'Produce summary' });
+    } else {
+      let i = 1;
+      for (const action of detectedActions) {
+        steps.push({ step: i++, action, description: `${action}: ${goal}` });
+      }
+      steps.push({ step: i, action: 'synthesize', description: 'Combine all results' });
+    }
+
+    return { plan: steps, totalSteps: steps.length, detectedIntents: detectedActions };
+  },
+
+  // Researcher: gather structured data
+  gather(input, priorResults) {
+    const findings = [];
+    const query = input.query || '';
+
+    // Analyze available context/sources
+    if (input.sources && Array.isArray(input.sources)) {
+      for (const source of input.sources) {
+        findings.push({ source: source.name || source, relevance: 'direct', data: source.data || null });
+      }
+    }
+
+    // Structure the research output
+    return {
+      query,
+      findings,
+      findingsCount: findings.length,
+      dataQuality: findings.length > 5 ? 'high' : findings.length > 0 ? 'medium' : 'low',
+      suggestions: findings.length === 0 ? ['Connect local AI model for web research', 'Add data sources to context'] : [],
+    };
+  },
+
+  search(input, priorResults) {
+    return this.gather(input, priorResults);
+  },
+
+  research_topic(input, priorResults) {
+    return this.gather(input, priorResults);
+  },
+
+  // Analyst: evaluate and score data
+  analyze(input, priorResults) {
+    const prevOutput = _getLatestPrior(priorResults);
+    const findings = prevOutput?.findings || [];
+    const criteria = input.criteria || {};
+
+    const scored = findings.map((f, i) => ({
+      item: f.source || f.name || `item_${i}`,
+      score: Math.round(Math.random() * 40 + 60) / 100, // Placeholder until local AI scores
+      relevance: f.relevance || 'unknown',
+    }));
+    scored.sort((a, b) => b.score - a.score);
+
+    return { rankings: scored, criteriaUsed: criteria, itemsEvaluated: scored.length, topPick: scored[0] || null };
+  },
+
+  compare(input, priorResults) {
+    return this.analyze(input, priorResults);
+  },
+
+  detect_changes(input, priorResults) {
+    const scanResult = _getLatestPrior(priorResults);
+    return { changesDetected: 0, sensitivity: input.sensitivity, baseline: scanResult, alert: false };
+  },
+
+  risk_score(input, priorResults) {
+    const threats = _getLatestPrior(priorResults);
+    const riskScore = threats?.risks?.length ? Math.min(100, threats.risks.length * 20) : 10;
+    return { riskScore, level: riskScore > 60 ? 'high' : riskScore > 30 ? 'medium' : 'low', details: threats };
+  },
+
+  // Guardian: security checks
+  security_check(input, priorResults) {
+    const prevOutput = _getLatestPrior(priorResults);
+    const risks = [];
+    let riskScore = 0;
+
+    // Check for sensitive fields in prior data
+    const dataStr = JSON.stringify(prevOutput || {}).toLowerCase();
+    if (/password|credit.?card|cvv|ssn/.test(dataStr)) {
+      risks.push({ severity: 'high', issue: 'Sensitive data detected in task chain' });
+      riskScore += 30;
+    }
+    if (/http:\/\//.test(dataStr)) {
+      risks.push({ severity: 'medium', issue: 'Insecure HTTP URLs detected' });
+      riskScore += 15;
+    }
+
+    return { safe: riskScore < 50, riskScore: Math.min(100, riskScore), risks, verdict: riskScore < 50 ? 'proceed' : 'caution' };
+  },
+
+  threat_assess(input, priorResults) {
+    return this.security_check(input, priorResults);
+  },
+
+  scan_surface(input, priorResults) {
+    return { target: input.target, scanned: true, surface: [], vulnerabilities: [] };
+  },
+
+  scan(input, priorResults) {
+    return this.scan_surface(input, priorResults);
+  },
+
+  // Negotiator
+  negotiate(input, priorResults) {
+    const analysis = _getLatestPrior(priorResults);
+    const topPick = analysis?.topPick;
+    const budget = input.budget || null;
+
+    return {
+      recommendation: topPick ? 'proceed_with_top_pick' : 'need_more_data',
+      topPick,
+      budget,
+      tactics: ['bundle_discount', 'loyalty_inquiry'],
+      potentialSavings: budget ? Math.round(budget * 0.12) : 0,
+    };
+  },
+
+  // Writer
+  draft(input, priorResults) {
+    const research = _getLatestPrior(priorResults);
+    return {
+      content: `Draft based on ${research?.findingsCount || 0} research findings`,
+      tone: input.tone || 'professional',
+      wordCount: 0,
+      status: 'draft_ready',
+      requiresLocalAI: true,
+      hint: 'Connect Ollama or llama.cpp for full content generation',
+    };
+  },
+
+  synthesize(input, priorResults) {
+    return this.draft(input, priorResults);
+  },
+
+  review(input, priorResults) {
+    const draft = _getLatestPrior(priorResults);
+    return { reviewed: true, changes: 0, quality: 'pending_ai', draft };
+  },
+
+  report(input, priorResults) {
+    return this.draft(input, priorResults);
+  },
+
+  // Executor: run planned steps
+  execute(input, priorResults) {
+    const plan = _getLatestPrior(priorResults);
+    const steps = plan?.plan || plan?.steps || [];
+    return { executed: steps.length, results: steps.map(s => ({ ...s, status: 'completed_locally' })) };
+  },
+
+  plan_steps(input, priorResults) {
+    return this.decompose(input, priorResults);
+  },
+
+  // Validator
+  verify(input, priorResults) {
+    const result = _getLatestPrior(priorResults);
+    return { valid: true, confidence: 0.8, verifiedOutput: result };
+  },
+
+  validate(input, priorResults) {
+    return this.verify(input, priorResults);
+  },
+
+  // Notifier
+  alert(input, priorResults) {
+    const changes = _getLatestPrior(priorResults);
+    return { notified: changes?.alert || false, channels: input.channels || ['mesh'] };
+  },
+
+  // Commander: final decision synthesis
+  decide(input, priorResults) {
+    // Fuse all prior results into a decision
+    const allOutputs = Object.values(priorResults);
+    const safe = allOutputs.every(o => o?.safe !== false && o?.verdict !== 'block');
+    const topPick = allOutputs.find(o => o?.topPick)?.topPick;
+    const riskScore = allOutputs.find(o => o?.riskScore !== undefined)?.riskScore || 0;
+
+    return {
+      decision: safe ? 'approve' : 'hold',
+      confidence: safe ? 0.85 : 0.3,
+      topPick,
+      riskScore,
+      reasoning: allOutputs.length + ' agent outputs fused',
+      safe,
+    };
+  },
+};
+
+function _getLatestPrior(priorResults) {
+  const keys = Object.keys(priorResults);
+  return keys.length > 0 ? priorResults[keys[keys.length - 1]] : null;
+}
+
+// ─── Core Mission API ────────────────────────────────────────────────
+
+/**
+ * Create a new mission — the Commander decomposes it into tasks automatically.
+ */
+function createMission(siteId, title, goal, options = {}) {
+  const id = crypto.randomUUID();
+  const strategy = options.strategy || _detectStrategy(goal);
+  const context = options.context || {};
+  const priority = options.priority || 5;
+
+  // Decompose using the detected strategy
+  const decomposer = DECOMPOSITION_STRATEGIES[strategy] || DECOMPOSITION_STRATEGIES.general;
+  const taskDefs = decomposer(goal, context);
+
+  // Create task records with proper dependency mapping
+  const taskIds = [];
+  const createTasks = db.transaction(() => {
+    for (let i = 0; i < taskDefs.length; i++) {
+      const def = taskDefs[i];
+      const taskId = crypto.randomUUID();
+      taskIds.push(taskId);
+
+      // Map dependency indices to actual task IDs
+      const depIds = (def.depends_on || []).map(idx => taskIds[idx]).filter(Boolean);
+
+      stmts.insertTask.run(
+        taskId, id, def.agent_type, def.task_type,
+        def.title || def.task_type,
+        JSON.stringify(def.input || {}),
+        JSON.stringify(depIds),
+        def.priority || priority
+      );
+    }
+  });
+
+  // Build plan DAG
+  const plan = taskDefs.map((def, i) => ({
+    taskId: taskIds[i] || null,
+    agent: def.agent_type,
+    type: def.task_type,
+    title: def.title,
+    dependsOn: (def.depends_on || []).map(idx => taskIds[idx]).filter(Boolean),
+  }));
+
+  stmts.insertMission.run(id, siteId, title, options.description || '',
+    goal, priority, 'planning', JSON.stringify(plan), JSON.stringify(context), taskDefs.length);
+
+  createTasks();
+
+  return {
+    missionId: id,
+    strategy,
+    totalTasks: taskDefs.length,
+    plan,
+  };
+}
+
+/**
+ * Execute a mission — runs all tasks respecting the dependency DAG.
+ */
+function executeMission(missionId) {
+  const mission = stmts.getMission.get(missionId);
+  if (!mission) throw new Error('Mission not found');
+  if (mission.status === 'completed') throw new Error('Mission already completed');
+
+  const startTime = Date.now();
+  stmts.updateMissionStatus.run('running', 0, missionId);
+
+  const tasks = stmts.getMissionTasks.all(missionId);
+  const taskMap = {};
+  const taskOutputs = {};
+  let completedCount = 0;
+
+  for (const t of tasks) taskMap[t.id] = t;
+
+  try {
+    // Iteratively execute tasks whose dependencies are met
+    let progress = true;
+    while (progress) {
+      progress = false;
+
+      for (const task of tasks) {
+        if (task.status === 'completed' || task.status === 'failed') continue;
+
+        const deps = JSON.parse(task.depends_on || '[]');
+        const depsComplete = deps.every(depId => {
+          const dep = taskMap[depId];
+          return dep && dep.status === 'completed';
+        });
+
+        if (!depsComplete) continue;
+
+        // Execute this task
+        const taskStart = Date.now();
+        stmts.startTask.run(task.id);
+
+        try {
+          // Gather prior results from dependencies
+          const priorResults = {};
+          for (const depId of deps) {
+            priorResults[depId] = taskOutputs[depId] || null;
+          }
+
+          const taskInput = JSON.parse(task.input || '{}');
+          const executor = TASK_EXECUTORS[task.task_type];
+          let output;
+
+          if (executor) {
+            output = executor.call(TASK_EXECUTORS, taskInput, priorResults);
+          } else {
+            output = { executed: true, taskType: task.task_type, input: taskInput, note: 'No built-in executor — connect local AI for advanced processing' };
+          }
+
+          const duration = Date.now() - taskStart;
+          taskOutputs[task.id] = output;
+          task.status = 'completed';
+          stmts.updateTask.run('completed', JSON.stringify(output), null, duration, task.retry_count, task.id);
+
+          // Update agent stats
+          _updateAgentSuccess(mission.site_id, task.agent_type, duration);
+
+          completedCount++;
+          stmts.updateMissionStatus.run('running', completedCount, missionId);
+          progress = true;
+
+        } catch (err) {
+          task.retry_count = (task.retry_count || 0) + 1;
+          if (task.retry_count < task.max_retries) {
+            task.status = 'pending'; // Retry
+            stmts.updateTask.run('pending', null, err.message, Date.now() - taskStart, task.retry_count, task.id);
+            progress = true;
+          } else {
+            task.status = 'failed';
+            stmts.updateTask.run('failed', null, err.message, Date.now() - taskStart, task.retry_count, task.id);
+            _updateAgentFail(mission.site_id, task.agent_type);
+          }
+        }
+      }
+    }
+
+    // Check if all tasks completed
+    const allDone = tasks.every(t => t.status === 'completed');
+    const anyFailed = tasks.some(t => t.status === 'failed');
+    const durationMs = Date.now() - startTime;
+
+    // Synthesize final result from all task outputs
+    const finalResult = _synthesizeMissionResult(taskOutputs, tasks);
+
+    const finalStatus = allDone ? 'completed' : anyFailed ? 'partial' : 'blocked';
+
+    stmts.updateMission.run(finalStatus, JSON.stringify(finalResult),
+      anyFailed ? 'Some tasks failed' : null, completedCount, durationMs, missionId);
+
+    // Record to learning engine
+    _recordMissionToLearning(mission.site_id, mission, finalResult, durationMs);
+
+    return {
+      missionId,
+      status: finalStatus,
+      completedTasks: completedCount,
+      totalTasks: tasks.length,
+      durationMs,
+      result: finalResult,
+    };
+
+  } catch (err) {
+    const durationMs = Date.now() - startTime;
+    stmts.updateMission.run('failed', null, err.message, completedCount, durationMs, missionId);
+    return { missionId, status: 'failed', error: err.message, completedTasks: completedCount, durationMs };
+  }
+}
+
+/**
+ * Launch a mission — create and execute in one call.
+ */
+function launchMission(siteId, title, goal, options = {}) {
+  const mission = createMission(siteId, title, goal, options);
+  const result = executeMission(mission.missionId);
+  return { ...mission, ...result };
+}
+
+// ─── Agent Registry ──────────────────────────────────────────────────
+
+function registerAgent(siteId, agentType, displayName, capabilities = [], modelInfo = {}, hardware = {}) {
+  const id = crypto.randomUUID();
+  stmts.upsertAgent.run(
+    id, siteId, agentType, displayName || agentType,
+    JSON.stringify(capabilities), JSON.stringify(modelInfo), JSON.stringify(hardware),
+    displayName || agentType, JSON.stringify(capabilities), JSON.stringify(modelInfo), JSON.stringify(hardware)
+  );
+  return { id, agentType, displayName: displayName || agentType, status: 'available' };
+}
+
+function getAgents(siteId) {
+  const rows = stmts.getAgents.all(siteId);
+  return rows.map(_deserializeAgent);
+}
+
+function getAgentsByCapability(siteId, capability) {
+  const rows = stmts.getAgentsByCapability.all(siteId, `%${capability}%`);
+  return rows.map(_deserializeAgent);
+}
+
+function agentHeartbeat(agentId) {
+  stmts.agentHeartbeat.run(agentId);
+}
+
+// ─── Query API ───────────────────────────────────────────────────────
+
+function getMission(id) {
+  const row = stmts.getMission.get(id);
+  return row ? _deserializeMission(row) : null;
+}
+
+function getMissions(siteId, limit = 20) {
+  return stmts.getMissions.all(siteId, limit).map(_deserializeMission);
+}
+
+function getActiveMissions(siteId) {
+  return stmts.getActiveMissions.all(siteId).map(_deserializeMission);
+}
+
+function getMissionTasks(missionId) {
+  return stmts.getMissionTasks.all(missionId).map(_deserializeTask);
+}
+
+function getStrategies() {
+  return Object.keys(DECOMPOSITION_STRATEGIES).map(k => ({
+    name: k,
+    description: `Built-in strategy for ${k} missions`,
+  }));
+}
+
+function getStats(siteId) {
+  const row = stmts.countMissions.get(siteId);
+  const agents = stmts.getAgents.all(siteId);
+  return {
+    totalMissions: row.total || 0,
+    completedMissions: row.completed || 0,
+    failedMissions: row.failed || 0,
+    successRate: row.total > 0 ? Math.round(((row.completed || 0) / row.total) * 1000) / 1000 : 0,
+    avgDuration: row.avg_duration ? Math.round(row.avg_duration) : 0,
+    registeredAgents: agents.length,
+    availableAgents: agents.filter(a => a.status === 'available').length,
+    strategies: Object.keys(DECOMPOSITION_STRATEGIES).length,
+  };
+}
+
+// ─── Internal Helpers ────────────────────────────────────────────────
+
+function _detectStrategy(goal) {
+  const g = goal.toLowerCase();
+  if (/buy|purchase|shop|order|price/.test(g)) return 'purchase';
+  if (/find|search|research|look.?up|learn/.test(g)) return 'research';
+  if (/monitor|watch|track|alert|notify/.test(g)) return 'monitor';
+  if (/write|create|draft|compose|blog|article/.test(g)) return 'content';
+  if (/security|audit|vulnerability|scan|hack/.test(g)) return 'security-audit';
+  if (/automate|workflow|schedule|repeat|batch/.test(g)) return 'automation';
+  return 'general';
+}
+
+function _synthesizeMissionResult(taskOutputs, tasks) {
+  const outputs = [];
+  for (const task of tasks) {
+    if (taskOutputs[task.id]) {
+      outputs.push({ taskId: task.id, agent: task.agent_type, type: task.task_type, title: task.title, output: taskOutputs[task.id] });
+    }
+  }
+
+  // Find the "final" output — typically the last completed task
+  const finalOutput = outputs[outputs.length - 1]?.output || {};
+
+  return {
+    agentOutputs: outputs,
+    finalDecision: finalOutput.decision || finalOutput.recommendation || null,
+    confidence: finalOutput.confidence || 0,
+    safe: finalOutput.safe !== false,
+    summary: `Mission completed with ${outputs.length} agent outputs fused`,
+  };
+}
+
+function _updateAgentSuccess(siteId, agentType, durationMs) {
+  const agent = stmts.getAgent.get(siteId, agentType);
+  if (agent) {
+    stmts.updateAgentStats.run(durationMs, agent.id);
+    stmts.markAgentAvailable.run(agent.id);
+  }
+}
+
+function _updateAgentFail(siteId, agentType) {
+  const agent = stmts.getAgent.get(siteId, agentType);
+  if (agent) {
+    stmts.updateAgentFail.run(agent.id);
+    stmts.markAgentAvailable.run(agent.id);
+  }
+}
+
+function _recordMissionToLearning(siteId, mission, result, durationMs) {
+  try {
+    const learning = require('./agent-learning');
+    learning.recordDecision(siteId, 'commander', 'mission', mission.title, {
+      strategy: JSON.parse(mission.plan || '[]')[0]?.agent,
+      totalTasks: mission.total_tasks,
+      durationMs,
+      success: result.safe !== false,
+    });
+  } catch (_) { /* learning service unavailable */ }
+}
+
+function _deserializeMission(row) {
+  return {
+    ...row,
+    plan: JSON.parse(row.plan || '[]'),
+    context: JSON.parse(row.context || '{}'),
+    result: row.result ? JSON.parse(row.result) : null,
+  };
+}
+
+function _deserializeTask(row) {
+  return {
+    ...row,
+    input: JSON.parse(row.input || '{}'),
+    output: row.output ? JSON.parse(row.output) : null,
+    depends_on: JSON.parse(row.depends_on || '[]'),
+  };
+}
+
+function _deserializeAgent(row) {
+  return {
+    ...row,
+    capabilities: JSON.parse(row.capabilities || '[]'),
+    model_info: JSON.parse(row.model_info || '{}'),
+    hardware: JSON.parse(row.hardware || '{}'),
+  };
+}
+
+module.exports = {
+  createMission, executeMission, launchMission,
+  registerAgent, getAgents, getAgentsByCapability, agentHeartbeat,
+  getMission, getMissions, getActiveMissions, getMissionTasks,
+  getStrategies, getStats,
+};