attocode 0.2.3 → 0.2.5

package/dist/src/agent/feature-initializer.js

@@ -0,0 +1,677 @@
+/**
+ * Feature initialization logic extracted from ProductionAgent.
+ * Initializes all integration managers based on configuration.
+ */
+import { isFeatureEnabled, getEnabledFeatures, } from '../defaults.js';
+import { HookManager, MemoryManager, PlanningManager, ObservabilityManager, SafetyManager, RoutingManager, MultiAgentManager, ReActManager, ExecutionPolicyManager, ThreadManager, RulesManager, DEFAULT_RULE_SOURCES, ExecutionEconomicsManager, STANDARD_BUDGET, AgentRegistry, createCancellationManager, createResourceManager, createLSPManager, createSemanticCacheManager, createSkillManager, createContextEngineering, createCodebaseContext, createSwarmOrchestrator, createThrottledProvider, FREE_TIER_THROTTLE, PAID_TIER_THROTTLE, createWorkLog, createVerificationGate, createTypeCheckerState, createInjectionBudgetManager, createSelfImprovementProtocol, createSubagentOutputStore, createAutoCheckpointManager, createToolRecommendationEngine, createInteractivePlanner, createRecursiveContext, createLearningStore, createCompactor, createAutoCompactionManager, createTaskManager, createSerperSearchTool, } from '../integrations/index.js';
+import { resolvePolicyProfile, } from '../integrations/safety/policy-engine.js';
+import { createTraceCollector } from '../tracing/trace-collector.js';
+import { modelRegistry } from '../costs/index.js';
+import { getModelContextLength } from '../integrations/utilities/openrouter-pricing.js';
+import { createComponentLogger } from '../integrations/utilities/logger.js';
+import { createBoundSpawnAgentTool, createBoundSpawnAgentsParallelTool, } from '../tools/agent.js';
+import { createTaskTools, } from '../tools/tasks.js';
+import { createAgentStateMachine } from '../core/agent-state-machine.js';
+const log = createComponentLogger('FeatureInitializer');
+// =============================================================================
+// FEATURE INITIALIZATION
+// =============================================================================
+/**
+ * Initialize all enabled features on an agent.
+ * This is the extracted body of ProductionAgent.initializeFeatures().
+ * All `this.` references are replaced with `agent.`.
+ */
+export function initializeFeatures(agent) {
+    // Debug output only when DEBUG env var is set
+    if (process.env.DEBUG) {
+        const features = getEnabledFeatures(agent.config);
+        log.debug('Initializing with features', { features: features.join(', ') });
+    }
+    // Hooks & Plugins
+    if (isFeatureEnabled(agent.config.hooks) && isFeatureEnabled(agent.config.plugins)) {
+        agent.hooks = new HookManager(agent.config.hooks, agent.config.plugins);
+    }
+    // Memory
+    if (isFeatureEnabled(agent.config.memory)) {
+        agent.memory = new MemoryManager(agent.config.memory);
+    }
+    // Planning & Reflection
+    if (isFeatureEnabled(agent.config.planning) && isFeatureEnabled(agent.config.reflection)) {
+        agent.planning = new PlanningManager(agent.config.planning, agent.config.reflection);
+    }
+    // Observability
+    if (isFeatureEnabled(agent.config.observability)) {
+        agent.observability = new ObservabilityManager(agent.config.observability);
+        // Lesson 26: Full trace capture
+        const traceCaptureConfig = agent.config.observability.traceCapture;
+        if (traceCaptureConfig?.enabled) {
+            agent.traceCollector = createTraceCollector({
+                enabled: true,
+                outputDir: traceCaptureConfig.outputDir ?? '.traces',
+                captureMessageContent: traceCaptureConfig.captureMessageContent ?? true,
+                captureToolResults: traceCaptureConfig.captureToolResults ?? true,
+                analyzeCacheBoundaries: traceCaptureConfig.analyzeCacheBoundaries ?? true,
+                filePattern: traceCaptureConfig.filePattern ?? 'trace-{sessionId}-{timestamp}.jsonl',
+                enableConsoleOutput: false,
+            });
+        }
+    }
+    // Safety (Sandbox + Human-in-Loop)
+    if (isFeatureEnabled(agent.config.sandbox) || isFeatureEnabled(agent.config.humanInLoop)) {
+        agent.safety = new SafetyManager(isFeatureEnabled(agent.config.sandbox) ? agent.config.sandbox : false, isFeatureEnabled(agent.config.humanInLoop) ? agent.config.humanInLoop : false, isFeatureEnabled(agent.config.policyEngine) ? agent.config.policyEngine : false);
+    }
+    if (isFeatureEnabled(agent.config.policyEngine)) {
+        const rootPolicy = resolvePolicyProfile({
+            policyEngine: agent.config.policyEngine,
+            sandboxConfig: isFeatureEnabled(agent.config.sandbox) ? agent.config.sandbox : undefined,
+        });
+        agent.emit({
+            type: 'policy.profile.resolved',
+            profile: rootPolicy.profileName,
+            context: 'root',
+            selectionSource: rootPolicy.metadata.selectionSource,
+            usedLegacyMappings: rootPolicy.metadata.usedLegacyMappings,
+            legacySources: rootPolicy.metadata.legacyMappingSources,
+        });
+        if (rootPolicy.metadata.usedLegacyMappings) {
+            agent.emit({
+                type: 'policy.legacy.fallback.used',
+                profile: rootPolicy.profileName,
+                sources: rootPolicy.metadata.legacyMappingSources,
+                warnings: rootPolicy.metadata.warnings,
+            });
+        }
+    }
+    // Routing
+    if (isFeatureEnabled(agent.config.routing)) {
+        agent.routing = new RoutingManager(agent.config.routing);
+    }
+    // Multi-Agent (Lesson 17)
+    if (isFeatureEnabled(agent.config.multiAgent)) {
+        const roles = (agent.config.multiAgent.roles || []).map((r) => ({
+            name: r.name,
+            description: r.description,
+            systemPrompt: r.systemPrompt,
+            capabilities: r.capabilities,
+            authority: r.authority,
+            model: r.model,
+        }));
+        agent.multiAgent = new MultiAgentManager(agent.provider, Array.from(agent.tools.values()), roles);
+    }
+    // ReAct (Lesson 18)
+    if (isFeatureEnabled(agent.config.react)) {
+        agent.react = new ReActManager(agent.provider, Array.from(agent.tools.values()), {
+            maxSteps: agent.config.react.maxSteps,
+            stopOnAnswer: agent.config.react.stopOnAnswer,
+            includeReasoning: agent.config.react.includeReasoning,
+        });
+    }
+    // Execution Policies (Lesson 23)
+    if (isFeatureEnabled(agent.config.executionPolicy)) {
+        agent.executionPolicy = new ExecutionPolicyManager({
+            defaultPolicy: agent.config.executionPolicy.defaultPolicy,
+            toolPolicies: agent.config.executionPolicy.toolPolicies,
+            intentAware: agent.config.executionPolicy.intentAware,
+            intentConfidenceThreshold: agent.config.executionPolicy.intentConfidenceThreshold,
+        });
+    }
+    // Thread Management (Lesson 24)
+    if (isFeatureEnabled(agent.config.threads)) {
+        agent.threadManager = new ThreadManager();
+    }
+    // Rules System (Lesson 12)
+    if (isFeatureEnabled(agent.config.rules)) {
+        const ruleSources = agent.config.rules.sources || DEFAULT_RULE_SOURCES;
+        agent.rules = new RulesManager({
+            enabled: true,
+            sources: ruleSources,
+            watch: agent.config.rules.watch,
+        });
+        // Load rules asynchronously - tracked for ensureReady()
+        agent.initPromises.push(agent.rules.loadRules().catch(err => {
+            log.warn('Failed to load rules', { error: String(err) });
+        }));
+    }
+    // Economics System (Token Budget) - always enabled
+    // Use custom budget if provided (subagents use SUBAGENT_BUDGET), otherwise STANDARD_BUDGET
+    const baseBudget = agent.config.budget ?? STANDARD_BUDGET;
+    agent.economics = new ExecutionEconomicsManager({
+        ...baseBudget,
+        // Use maxIterations from config as absolute safety cap
+        maxIterations: agent.config.maxIterations,
+        targetIterations: Math.min(baseBudget.targetIterations ?? 20, agent.config.maxIterations),
+    }, agent._sharedEconomicsState ?? undefined, agent.agentId);
+    // Enable incremental token accounting for the root agent.
+    // Estimate baseline from system prompt size (~4 chars/token).
+    // This activates incremental mode in recordLLMUsage() so only marginal
+    // tokens per call are counted toward the budget, preventing quadratic growth.
+    const systemPromptLength = agent.config.systemPrompt?.length ?? 0;
+    const estimatedBaselineTokens = Math.ceil(systemPromptLength / 4);
+    if (estimatedBaselineTokens > 0) {
+        agent.economics.setBaseline(estimatedBaselineTokens);
+    }
+    // Phase 2.2: Agent State Machine - formalizes phase tracking
+    // Always enabled - provides structured phase transitions with metrics
+    agent.stateMachine = createAgentStateMachine();
+    // Forward state machine phase transitions as subagent.phase events
+    const phaseMap = {
+        exploring: 'exploring', planning: 'planning', acting: 'executing', verifying: 'completing',
+    };
+    const unsubStateMachine = agent.stateMachine.subscribe(event => {
+        if (event.type === 'phase.changed') {
+            agent.emit({
+                type: 'subagent.phase',
+                agentId: agent.agentId,
+                phase: phaseMap[event.transition.to] ?? 'exploring',
+            });
+        }
+    });
+    agent.unsubscribers.push(unsubStateMachine);
+    // Work Log - compaction-resilient summary of agent work
+    // Always enabled - minimal overhead and critical for long-running tasks
+    agent.workLog = createWorkLog();
+    // Verification Gate - opt-in completion verification
+    if (agent.config.verificationCriteria) {
+        agent.verificationGate = createVerificationGate(agent.config.verificationCriteria);
+    }
+    // TypeScript compilation checking — auto-detect and enable
+    {
+        const cwd = agent.config.workingDirectory || process.cwd();
+        const tcState = createTypeCheckerState(cwd);
+        agent.typeCheckerState = tcState;
+        if (tcState.tsconfigDir && !agent.verificationGate) {
+            agent.verificationGate = createVerificationGate({
+                requireCompilation: true,
+                compilationMaxAttempts: 8,
+            });
+        }
+    }
+    // Phase 2-4: Orchestration & Advanced modules (always enabled, lightweight)
+    agent.injectionBudget = createInjectionBudgetManager();
+    agent.selfImprovement = createSelfImprovementProtocol(undefined, agent.learningStore ?? undefined);
+    agent.subagentOutputStore = createSubagentOutputStore({ persistToFile: false });
+    agent.autoCheckpointManager = createAutoCheckpointManager({ enabled: true });
+    agent.toolRecommendation = createToolRecommendationEngine();
+    // Agent Registry - always enabled for subagent support
+    agent.agentRegistry = new AgentRegistry();
+    // Load user agents asynchronously - tracked for ensureReady()
+    agent.initPromises.push(agent.agentRegistry.loadUserAgents().catch(err => {
+        log.warn('Failed to load user agents', { error: String(err) });
+    }));
+    // Register spawn_agent tool so LLM can delegate to subagents
+    const boundSpawnTool = createBoundSpawnAgentTool((name, task, constraints) => agent.spawnAgent(name, task, constraints));
+    agent.tools.set(boundSpawnTool.name, boundSpawnTool);
+    // Register spawn_agents_parallel tool for parallel subagent execution
+    const boundParallelSpawnTool = createBoundSpawnAgentsParallelTool((tasks) => agent.spawnAgentsParallel(tasks));
+    agent.tools.set(boundParallelSpawnTool.name, boundParallelSpawnTool);
+    // Task Manager - Claude Code-style task system for coordination
+    agent.taskManager = createTaskManager();
+    // Forward task events (with cleanup tracking for EventEmitter-based managers)
+    const taskCreatedHandler = (data) => {
+        agent.emit({ type: 'task.created', task: data.task });
+    };
+    agent.taskManager.on('task.created', taskCreatedHandler);
+    agent.unsubscribers.push(() => agent.taskManager?.off('task.created', taskCreatedHandler));
+    const taskUpdatedHandler = (data) => {
+        agent.emit({ type: 'task.updated', task: data.task });
+    };
+    agent.taskManager.on('task.updated', taskUpdatedHandler);
+    agent.unsubscribers.push(() => agent.taskManager?.off('task.updated', taskUpdatedHandler));
+    // Register task tools
+    const taskTools = createTaskTools(agent.taskManager);
+    for (const tool of taskTools) {
+        agent.tools.set(tool.name, tool);
+    }
+    // Built-in web search (Serper API) — gracefully handles missing API key
+    const serperCustomTool = createSerperSearchTool();
+    agent.tools.set('web_search', {
+        name: serperCustomTool.name,
+        description: serperCustomTool.description,
+        parameters: serperCustomTool.inputSchema,
+        execute: serperCustomTool.execute,
+        dangerLevel: 'safe',
+    });
+    // Swarm Mode (experimental)
+    if (agent.config.swarm) {
+        const swarmConfig = agent.config.swarm;
+        // Wrap provider with request throttle to prevent 429 rate limiting.
+        // All subagents share agent.provider by reference (line 4398),
+        // so wrapping here throttles ALL downstream LLM calls.
+        if (swarmConfig.throttle !== false) {
+            const throttleConfig = swarmConfig.throttle === 'paid'
+                ? PAID_TIER_THROTTLE
+                : swarmConfig.throttle === 'free' || swarmConfig.throttle === undefined
+                    ? FREE_TIER_THROTTLE
+                    : swarmConfig.throttle;
+            agent.provider = createThrottledProvider(agent.provider, throttleConfig);
+        }
+        // Pass codebaseContext so the decomposer can ground tasks in actual project files
+        swarmConfig.codebaseContext = agent.codebaseContext ?? undefined;
+        agent.swarmOrchestrator = createSwarmOrchestrator(swarmConfig, agent.provider, agent.agentRegistry, (name, task) => agent.spawnAgent(name, task), agent.blackboard ?? undefined);
+        // Override parent budget pool with swarm's much larger pool so spawnAgent()
+        // allocates from the swarm budget (e.g. 10M tokens) instead of the parent's
+        // generic pool (200K tokens). Without this, workers get 5K emergency budget.
+        agent.budgetPool = agent.swarmOrchestrator.getBudgetPool().pool;
+        // Phase 3.1+3.2: Set shared state so workers inherit it via buildContext()
+        agent._sharedContextState = agent.swarmOrchestrator.getSharedContextState();
+        agent._sharedEconomicsState = agent.swarmOrchestrator.getSharedEconomicsState();
+    }
+    // Cancellation Support
+    if (isFeatureEnabled(agent.config.cancellation)) {
+        agent.cancellation = createCancellationManager();
+        // Forward cancellation events (with cleanup tracking)
+        const unsubCancellation = agent.cancellation.subscribe(event => {
+            if (event.type === 'cancellation.requested') {
+                agent.emit({ type: 'cancellation.requested', reason: event.reason });
+            }
+        });
+        agent.unsubscribers.push(unsubCancellation);
+    }
+    // Resource Monitoring
+    if (isFeatureEnabled(agent.config.resources)) {
+        agent.resourceManager = createResourceManager({
+            enabled: agent.config.resources.enabled,
+            maxMemoryMB: agent.config.resources.maxMemoryMB,
+            maxCpuTimeSec: agent.config.resources.maxCpuTimeSec,
+            maxConcurrentOps: agent.config.resources.maxConcurrentOps,
+            warnThreshold: agent.config.resources.warnThreshold,
+            criticalThreshold: agent.config.resources.criticalThreshold,
+        });
+    }
+    // LSP (Language Server Protocol) Support
+    if (isFeatureEnabled(agent.config.lsp)) {
+        agent.lspManager = createLSPManager({
+            enabled: agent.config.lsp.enabled,
+            autoDetect: agent.config.lsp.autoDetect,
+            servers: agent.config.lsp.servers,
+            timeout: agent.config.lsp.timeout,
+        });
+        // Auto-start is done lazily on first use to avoid startup delays
+    }
+    // Semantic Cache Support
+    if (isFeatureEnabled(agent.config.semanticCache)) {
+        agent.semanticCache = createSemanticCacheManager({
+            enabled: agent.config.semanticCache.enabled,
+            threshold: agent.config.semanticCache.threshold,
+            maxSize: agent.config.semanticCache.maxSize,
+            ttl: agent.config.semanticCache.ttl,
+        });
+        // Forward cache events (with cleanup tracking)
+        const unsubSemanticCache = agent.semanticCache.subscribe(event => {
+            if (event.type === 'cache.hit') {
+                agent.emit({ type: 'cache.hit', query: event.query, similarity: event.similarity });
+            }
+            else if (event.type === 'cache.miss') {
+                agent.emit({ type: 'cache.miss', query: event.query });
+            }
+            else if (event.type === 'cache.set') {
+                agent.emit({ type: 'cache.set', query: event.query });
+            }
+        });
+        agent.unsubscribers.push(unsubSemanticCache);
+    }
+    // Skills Support
+    if (isFeatureEnabled(agent.config.skills)) {
+        agent.skillManager = createSkillManager({
+            enabled: agent.config.skills.enabled,
+            directories: agent.config.skills.directories,
+            loadBuiltIn: agent.config.skills.loadBuiltIn,
+            autoActivate: agent.config.skills.autoActivate,
+        });
+        // Load skills asynchronously - tracked for ensureReady()
+        agent.initPromises.push(agent.skillManager.loadSkills()
+            .then(() => { }) // Convert to void
+            .catch(err => {
+            log.warn('Failed to load skills', { error: String(err) });
+        }));
+    }
+    // Context Engineering (Manus-inspired tricks P, Q, R, S, T)
+    // Always enabled - these are performance optimizations
+    agent.contextEngineering = createContextEngineering({
+        enableCacheOptimization: true,
+        enableRecitation: true,
+        enableReversibleCompaction: true,
+        enableFailureTracking: true,
+        enableDiversity: false, // Off by default - can cause unexpected behavior
+        staticPrefix: agent.config.systemPrompt,
+        recitationFrequency: 5,
+        maxFailures: 30,
+        maxReferences: 50,
+    });
+    // Bind shared context state for cross-worker failure learning (swarm workers only)
+    if (agent._sharedContextState) {
+        agent.contextEngineering.setSharedState(agent._sharedContextState);
+    }
+    // Codebase Context - intelligent code selection for context management
+    // Analyzes repo structure and selects relevant code within token budgets
+    if (agent.config.codebaseContext !== false) {
+        const codebaseConfig = typeof agent.config.codebaseContext === 'object'
+            ? agent.config.codebaseContext
+            : {};
+        agent.codebaseContext = createCodebaseContext({
+            root: codebaseConfig.root ?? process.cwd(),
+            includePatterns: codebaseConfig.includePatterns,
+            excludePatterns: codebaseConfig.excludePatterns,
+            maxFileSize: codebaseConfig.maxFileSize ?? 100 * 1024, // 100KB
+            tokensPerChar: 0.25,
+            analyzeDependencies: true,
+            cacheResults: true,
+            cacheTTL: 5 * 60 * 1000, // 5 minutes
+        });
+        // Forward trace collector so codebase analysis can emit codebase.map entries.
+        if (agent.traceCollector) {
+            agent.codebaseContext.traceCollector = agent.traceCollector;
+        }
+        // Connect LSP manager to codebase context for enhanced code selection
+        // This enables LSP-based relevance boosting (Phase 4.1)
+        if (agent.lspManager) {
+            agent.codebaseContext.setLSPManager(agent.lspManager);
+        }
+    }
+    // Forward context engineering events (with cleanup tracking)
+    const unsubContextEngineering = agent.contextEngineering.on(event => {
+        switch (event.type) {
+            case 'failure.recorded':
+                agent.observability?.logger?.warn('Failure recorded', {
+                    action: event.failure.action,
+                    category: event.failure.category,
+                });
+                break;
+            case 'failure.pattern':
+                agent.observability?.logger?.warn('Failure pattern detected', {
+                    type: event.pattern.type,
+                    description: event.pattern.description,
+                });
+                agent.emit({ type: 'error', error: `Pattern: ${event.pattern.description}` });
+                break;
+            case 'recitation.injected':
+                agent.observability?.logger?.debug('Recitation injected', {
+                    iteration: event.iteration,
+                });
+                break;
+        }
+    });
+    agent.unsubscribers.push(unsubContextEngineering);
+    // Interactive Planning (conversational + editable planning)
+    if (isFeatureEnabled(agent.config.interactivePlanning)) {
+        const interactiveConfig = typeof agent.config.interactivePlanning === 'object'
+            ? agent.config.interactivePlanning
+            : {};
+        agent.interactivePlanner = createInteractivePlanner({
+            autoCheckpoint: interactiveConfig.enableCheckpoints ?? true,
+            confirmBeforeExecute: interactiveConfig.requireApproval ?? true,
+            maxCheckpoints: 20,
+            autoPauseAtDecisions: true,
+        });
+        // Forward planner events to observability (with cleanup tracking)
+        const unsubInteractivePlanner = agent.interactivePlanner.on(event => {
+            switch (event.type) {
+                case 'plan.created':
+                    agent.observability?.logger?.info('Interactive plan created', {
+                        planId: event.plan.id,
+                        stepCount: event.plan.steps.length,
+                    });
+                    break;
+                case 'step.completed':
+                    agent.observability?.logger?.debug('Plan step completed', {
+                        stepId: event.step.id,
+                        status: event.step.status,
+                    });
+                    break;
+                case 'plan.cancelled':
+                    agent.observability?.logger?.info('Plan cancelled', { reason: event.reason });
+                    break;
+                case 'checkpoint.created':
+                    agent.observability?.logger?.debug('Plan checkpoint created', {
+                        checkpointId: event.checkpoint.id,
+                    });
+                    break;
+            }
+        });
+        agent.unsubscribers.push(unsubInteractivePlanner);
+    }
+    // Recursive Context (RLM - Recursive Language Models)
+    // Enables on-demand context exploration for large codebases
+    if (isFeatureEnabled(agent.config.recursiveContext)) {
+        const recursiveConfig = typeof agent.config.recursiveContext === 'object'
+            ? agent.config.recursiveContext
+            : {};
+        agent.recursiveContext = createRecursiveContext({
+            maxDepth: recursiveConfig.maxRecursionDepth ?? 5,
+            snippetTokens: recursiveConfig.maxSnippetTokens ?? 2000,
+            synthesisTokens: 1000,
+            totalBudget: 50000,
+            cacheResults: recursiveConfig.cacheNavigationResults ?? true,
+        });
+        // Note: File system source should be registered when needed with proper glob/readFile functions
+        // This is deferred to allow flexible configuration
+        // Forward RLM events (with cleanup tracking)
+        const unsubRecursiveContext = agent.recursiveContext.on(event => {
+            switch (event.type) {
+                case 'process.started':
+                    agent.observability?.logger?.debug('RLM process started', {
+                        query: event.query,
+                        depth: event.depth,
+                    });
+                    break;
+                case 'navigation.command':
+                    agent.observability?.logger?.debug('RLM navigation command', {
+                        command: event.command,
+                        depth: event.depth,
+                    });
+                    break;
+                case 'process.completed':
+                    agent.observability?.logger?.debug('RLM process completed', {
+                        stats: event.stats,
+                    });
+                    break;
+                case 'budget.warning':
+                    agent.observability?.logger?.warn('RLM budget warning', {
+                        remaining: event.remaining,
+                        total: event.total,
+                    });
+                    break;
+            }
+        });
+        agent.unsubscribers.push(unsubRecursiveContext);
+    }
+    // Learning Store (cross-session learning from failures)
+    // Connects to the failure tracker in contextEngineering for automatic learning extraction
+    if (isFeatureEnabled(agent.config.learningStore)) {
+        const learningConfig = typeof agent.config.learningStore === 'object'
+            ? agent.config.learningStore
+            : {};
+        agent.learningStore = createLearningStore({
+            dbPath: learningConfig.dbPath ?? '.agent/learnings.db',
+            requireValidation: learningConfig.requireValidation ?? true,
+            autoValidateThreshold: learningConfig.autoValidateThreshold ?? 0.9,
+            maxLearnings: learningConfig.maxLearnings ?? 500,
+        });
+        // Connect to the failure tracker if available
+        if (agent.contextEngineering) {
+            const failureTracker = agent.contextEngineering.getFailureTracker();
+            if (failureTracker) {
+                agent.learningStore.connectFailureTracker(failureTracker);
+            }
+        }
+        // Forward learning events to observability (with cleanup tracking)
+        const unsubLearningStore = agent.learningStore.on(event => {
+            switch (event.type) {
+                case 'learning.proposed':
+                    agent.observability?.logger?.info('Learning proposed', {
+                        learningId: event.learning.id,
+                        description: event.learning.description,
+                    });
+                    agent.emit({
+                        type: 'learning.proposed',
+                        learningId: event.learning.id,
+                        description: event.learning.description,
+                    });
+                    break;
+                case 'learning.validated':
+                    agent.observability?.logger?.info('Learning validated', {
+                        learningId: event.learningId,
+                    });
+                    agent.emit({ type: 'learning.validated', learningId: event.learningId });
+                    break;
+                case 'learning.applied':
+                    agent.observability?.logger?.debug('Learning applied', {
+                        learningId: event.learningId,
+                        context: event.context,
+                    });
+                    agent.emit({
+                        type: 'learning.applied',
+                        learningId: event.learningId,
+                        context: event.context,
+                    });
+                    break;
+                case 'pattern.extracted':
+                    agent.observability?.logger?.info('Pattern extracted as learning', {
+                        pattern: event.pattern.description,
+                        learningId: event.learning.id,
+                    });
+                    break;
+            }
+        });
+        agent.unsubscribers.push(unsubLearningStore);
+    }
+    // Auto-Compaction Manager (sophisticated context compaction)
+    // Uses the Compactor for LLM-based summarization with threshold monitoring
+    if (isFeatureEnabled(agent.config.compaction)) {
+        const compactionConfig = typeof agent.config.compaction === 'object'
+            ? agent.config.compaction
+            : {};
+        // Create the compactor (requires provider for LLM summarization)
+        agent.compactor = createCompactor(agent.provider, {
+            enabled: true,
+            tokenThreshold: compactionConfig.tokenThreshold ?? 80000,
+            preserveRecentCount: compactionConfig.preserveRecentCount ?? 10,
+            preserveToolResults: compactionConfig.preserveToolResults ?? true,
+            summaryMaxTokens: compactionConfig.summaryMaxTokens ?? 2000,
+            summaryModel: compactionConfig.summaryModel,
+        });
+        // Create the auto-compaction manager with threshold monitoring
+        // Wire reversible compaction through contextEngineering when available
+        const compactHandler = agent.contextEngineering
+            ? async (messages) => {
+                // Use contextEngineering's reversible compaction to preserve references
+                const summarize = async (msgs) => {
+                    // Use the basic compactor's summarization capability
+                    const result = await agent.compactor.compact(msgs);
+                    return result.summary;
+                };
+                const contextMsgs = messages.map(m => ({
+                    role: m.role,
+                    content: typeof m.content === 'string' ? m.content : JSON.stringify(m.content),
+                }));
+                const result = await agent.contextEngineering.compact(contextMsgs, summarize);
+                const tokensBefore = agent.compactor.estimateTokens(messages);
+                const tokensAfter = agent.compactor.estimateTokens([{ role: 'assistant', content: result.summary }]);
+                return {
+                    summary: result.summary + (result.reconstructionPrompt ? `\n\n${result.reconstructionPrompt}` : ''),
+                    tokensBefore,
+                    tokensAfter,
+                    preservedMessages: [{ role: 'assistant', content: result.summary }],
+                    references: result.references,
+                };
+            }
+            : undefined;
+        // Get model's actual context window - try OpenRouter first (real API data),
+        // then fall back to hardcoded ModelRegistry, then config, then default
+        const openRouterContext = getModelContextLength(agent.config.model || '');
+        const registryInfo = modelRegistry.getModel(agent.config.model || '');
+        const registryContext = registryInfo?.capabilities?.maxContextTokens;
+        const maxContextTokens = agent.config.maxContextTokens
+            ?? openRouterContext // From OpenRouter API (e.g., GLM-4.7 = 202752)
+            ?? registryContext // From hardcoded registry (Claude, GPT-4o, etc.)
+            ?? 200000; // Fallback to 200K
+        agent.autoCompactionManager = createAutoCompactionManager(agent.compactor, {
+            mode: compactionConfig.mode ?? 'auto',
+            warningThreshold: 0.70, // Warn at 70% of model's context
+            autoCompactThreshold: 0.80, // Compact at 80% (changed from 0.90)
+            hardLimitThreshold: 0.95, // Hard limit at 95%
+            preserveRecentUserMessages: Math.ceil((compactionConfig.preserveRecentCount ?? 10) / 2),
+            preserveRecentAssistantMessages: Math.ceil((compactionConfig.preserveRecentCount ?? 10) / 2),
+            cooldownMs: 60000, // 1 minute cooldown
+            maxContextTokens, // Dynamic from model registry or config
+            compactHandler, // Use reversible compaction when contextEngineering is available
+        });
+        // Forward compactor events to observability (with cleanup tracking)
+        const unsubCompactor = agent.compactor.on(event => {
+            switch (event.type) {
+                case 'compaction.start':
+                    agent.observability?.logger?.info('Compaction started', {
+                        messageCount: event.messageCount,
+                    });
+                    break;
+                case 'compaction.complete':
+                    agent.observability?.logger?.info('Compaction complete', {
+                        tokensBefore: event.result.tokensBefore,
+                        tokensAfter: event.result.tokensAfter,
+                        compactedCount: event.result.compactedCount,
+                    });
+                    break;
+                case 'compaction.error':
+                    agent.observability?.logger?.error('Compaction error', {
+                        error: event.error,
+                    });
+                    break;
+            }
+        });
+        agent.unsubscribers.push(unsubCompactor);
+        // Forward auto-compaction events (with cleanup tracking)
+        const unsubAutoCompaction = agent.autoCompactionManager.on((event) => {
+            switch (event.type) {
+                case 'autocompaction.warning':
+                    agent.observability?.logger?.warn('Context approaching limit', {
+                        currentTokens: event.currentTokens,
+                        ratio: event.ratio,
+                    });
+                    agent.emit({
+                        type: 'compaction.warning',
+                        currentTokens: event.currentTokens,
+                        threshold: Math.round(event.ratio * (agent.config.maxContextTokens ?? 200000)),
+                    });
+                    break;
+                case 'autocompaction.triggered':
+                    agent.observability?.logger?.info('Auto-compaction triggered', {
+                        mode: event.mode,
+                        currentTokens: event.currentTokens,
+                    });
+                    break;
+                case 'autocompaction.completed':
+                    agent.observability?.logger?.info('Auto-compaction completed', {
+                        tokensBefore: event.tokensBefore,
+                        tokensAfter: event.tokensAfter,
+                        reduction: event.reduction,
+                    });
+                    agent.emit({
+                        type: 'compaction.auto',
+                        tokensBefore: event.tokensBefore,
+                        tokensAfter: event.tokensAfter,
+                        messagesCompacted: event.tokensBefore - event.tokensAfter,
+                    });
+                    break;
+                case 'autocompaction.hard_limit':
+                    agent.observability?.logger?.error('Context hard limit reached', {
+                        currentTokens: event.currentTokens,
+                        ratio: event.ratio,
+                    });
+                    break;
+                case 'autocompaction.emergency_truncate':
+                    agent.observability?.logger?.warn('Emergency truncation performed', {
+                        reason: event.reason,
+                        messagesBefore: event.messagesBefore,
+                        messagesAfter: event.messagesAfter,
+                    });
+                    break;
+            }
+        });
+        agent.unsubscribers.push(unsubAutoCompaction);
+    }
+    // Note: FileChangeTracker requires a database instance which is not
+    // available at this point. Use initFileChangeTracker() to enable it
+    // after the agent is constructed with a database reference.
+    // This allows the feature to be optional and not require SQLite at all times.
+}
+//# sourceMappingURL=feature-initializer.js.map