moflo 4.0.2 → 4.0.3

package/v3/@claude-flow/cli/dist/src/memory/memory-initializer.js

@@ -1,35 +1,71 @@
-/**
- * V3 Memory Initializer
- * Properly initializes the memory database with sql.js (WASM SQLite)
- * Includes pattern tables, vector embeddings, migration state tracking
- *
- * ADR-053: Routes through ControllerRegistry → AgentDB v3 when available,
- * falls back to raw sql.js for backwards compatibility.
- *
- * @module v3/cli/memory-initializer
- */
-import * as fs from 'fs';
-import * as path from 'path';
-// ADR-053: Lazy import of AgentDB v3 bridge
-let _bridge;
-async function getBridge() {
-    if (_bridge === null)
-        return null;
-    if (_bridge)
-        return _bridge;
-    try {
-        _bridge = await import('./memory-bridge.js');
-        return _bridge;
-    }
-    catch {
-        _bridge = null;
-        return null;
-    }
-}
-/**
- * Enhanced schema with pattern confidence, temporal decay, versioning
- * Vector embeddings enabled for semantic search
- */
+/**
+ * V3 Memory Initializer
+ * Properly initializes the memory database with sql.js (WASM SQLite)
+ * Includes pattern tables, vector embeddings, migration state tracking
+ *
+ * ADR-053: Routes through ControllerRegistry → AgentDB v3 when available,
+ * falls back to raw sql.js for backwards compatibility.
+ *
+ * @module v3/cli/memory-initializer
+ */
+import * as fs from 'fs';
+import * as path from 'path';
+import { createRequire } from 'module';
+import { pathToFileURL } from 'url';
+// Project root resolution — handles npx running from node_modules/moflo
+let _projectRoot;
+function getProjectRoot() {
+    if (_projectRoot)
+        return _projectRoot;
+    // 1. Explicit env var (set by Claude Code hooks or moflo CLI wrapper)
+    if (process.env.CLAUDE_PROJECT_DIR) {
+        _projectRoot = process.env.CLAUDE_PROJECT_DIR;
+        return _projectRoot;
+    }
+    // 2. Walk up from cwd, skipping anything inside node_modules.
+    //    A stale .swarm inside a package dir is NOT a project root.
+    let dir = process.cwd();
+    const root = path.parse(dir).root;
+    const nm = path.sep + 'node_modules';
+    while (dir !== root) {
+        // Skip if we're still inside a node_modules tree
+        if (!dir.includes(nm)) {
+            // Prefer CLAUDE.md (project-specific, never inside a package)
+            if (fs.existsSync(path.join(dir, 'CLAUDE.md')) && fs.existsSync(path.join(dir, 'package.json'))) {
+                _projectRoot = dir;
+                return _projectRoot;
+            }
+            if (fs.existsSync(path.join(dir, '.swarm', 'memory.db'))) {
+                _projectRoot = dir;
+                return _projectRoot;
+            }
+        }
+        dir = path.dirname(dir);
+    }
+    // 3. Fallback to cwd (original behavior)
+    _projectRoot = process.cwd();
+    return _projectRoot;
+}
+// ADR-053: Lazy import of AgentDB v3 bridge
+let _bridge;
+async function getBridge() {
+    if (_bridge === null)
+        return null;
+    if (_bridge)
+        return _bridge;
+    try {
+        _bridge = await import('./memory-bridge.js');
+        return _bridge;
+    }
+    catch {
+        _bridge = null;
+        return null;
+    }
+}
+/**
+ * Enhanced schema with pattern confidence, temporal decay, versioning
+ * Vector embeddings enabled for semantic search
+ */
 export const MEMORY_SCHEMA_V3 = `
 -- RuFlo V3 Memory Database
 -- Version: 3.0.0
@@ -319,477 +355,492 @@ CREATE TABLE IF NOT EXISTS metadata (
   value TEXT NOT NULL,
   updated_at INTEGER DEFAULT (strftime('%s', 'now') * 1000)
 );
-`;
-let hnswIndex = null;
-let hnswInitializing = false;
-/**
- * Get or create the HNSW index singleton
- * Lazily initializes from SQLite data on first use
- */
-export async function getHNSWIndex(options) {
-    const dimensions = options?.dimensions ?? 384;
-    // Return existing index if already initialized
-    if (hnswIndex?.initialized && !options?.forceRebuild) {
-        return hnswIndex;
-    }
-    // Prevent concurrent initialization
-    if (hnswInitializing) {
-        // Wait for initialization to complete
-        while (hnswInitializing) {
-            await new Promise(resolve => setTimeout(resolve, 10));
-        }
-        return hnswIndex;
-    }
-    hnswInitializing = true;
-    try {
-        // Import @ruvector/core dynamically
-        // Handle both ESM (default export) and CJS patterns
-        const ruvectorModule = await import('@ruvector/core').catch(() => null);
-        if (!ruvectorModule) {
-            hnswInitializing = false;
-            return null; // HNSW not available
-        }
-        // ESM returns { default: { VectorDb, ... } }, CJS returns { VectorDb, ... }
-        const ruvectorCore = ruvectorModule.default || ruvectorModule;
-        if (!ruvectorCore?.VectorDb) {
-            hnswInitializing = false;
-            return null; // VectorDb not found
-        }
-        const { VectorDb } = ruvectorCore;
-        // Persistent storage paths
-        const swarmDir = path.join(process.cwd(), '.swarm');
-        if (!fs.existsSync(swarmDir)) {
-            fs.mkdirSync(swarmDir, { recursive: true });
-        }
-        const hnswPath = path.join(swarmDir, 'hnsw.index');
-        const metadataPath = path.join(swarmDir, 'hnsw.metadata.json');
-        const dbPath = options?.dbPath || path.join(swarmDir, 'memory.db');
-        // Create HNSW index with persistent storage
-        // @ruvector/core uses string enum for distanceMetric: 'Cosine', 'Euclidean', 'DotProduct', 'Manhattan'
-        const db = new VectorDb({
-            dimensions,
-            distanceMetric: 'Cosine',
-            storagePath: hnswPath // Persistent storage!
-        });
-        // Load metadata (entry info) if exists
-        const entries = new Map();
-        if (fs.existsSync(metadataPath)) {
-            try {
-                const metadataJson = fs.readFileSync(metadataPath, 'utf-8');
-                const metadata = JSON.parse(metadataJson);
-                for (const [key, value] of metadata) {
-                    entries.set(key, value);
-                }
-            }
-            catch {
-                // Metadata load failed, will rebuild
-            }
-        }
-        hnswIndex = {
-            db,
-            entries,
-            dimensions,
-            initialized: false
-        };
-        // Check if index already has data (from persistent storage)
-        const existingLen = await db.len();
-        if (existingLen > 0 && entries.size > 0) {
-            // Index loaded from disk, skip SQLite sync
-            hnswIndex.initialized = true;
-            hnswInitializing = false;
-            return hnswIndex;
-        }
-        if (fs.existsSync(dbPath)) {
-            try {
-                const initSqlJs = (await import('sql.js')).default;
-                const SQL = await initSqlJs();
-                const fileBuffer = fs.readFileSync(dbPath);
-                const sqlDb = new SQL.Database(fileBuffer);
-                // Load all entries with embeddings
+`;
+let hnswIndex = null;
+let hnswInitializing = false;
+/**
+ * Get or create the HNSW index singleton
+ * Lazily initializes from SQLite data on first use
+ */
+export async function getHNSWIndex(options) {
+    const dimensions = options?.dimensions ?? 384;
+    // Return existing index if already initialized
+    if (hnswIndex?.initialized && !options?.forceRebuild) {
+        return hnswIndex;
+    }
+    // Prevent concurrent initialization
+    if (hnswInitializing) {
+        // Wait for initialization to complete
+        while (hnswInitializing) {
+            await new Promise(resolve => setTimeout(resolve, 10));
+        }
+        return hnswIndex;
+    }
+    hnswInitializing = true;
+    try {
+        // Import @ruvector/core dynamically
+        // Handle both ESM (default export) and CJS patterns
+        // Bug fix: try project root resolution via createRequire when direct import fails
+        // (npx moflo resolves relative to moflo package, not consuming project)
+        let ruvectorModule = null;
+        try {
+            ruvectorModule = await import('@ruvector/core');
+        }
+        catch {
+            try {
+                const req = createRequire(path.join(getProjectRoot(), 'package.json'));
+                const resolved = req.resolve('@ruvector/core');
+                ruvectorModule = await import(pathToFileURL(resolved).href);
+            }
+            catch {
+                // Not available
+            }
+        }
+        if (!ruvectorModule) {
+            hnswInitializing = false;
+            return null; // HNSW not available
+        }
+        // ESM returns { default: { VectorDb, ... } }, CJS returns { VectorDb, ... }
+        const ruvectorCore = ruvectorModule.default || ruvectorModule;
+        if (!ruvectorCore?.VectorDb) {
+            hnswInitializing = false;
+            return null; // VectorDb not found
+        }
+        const { VectorDb } = ruvectorCore;
+        // Persistent storage paths
+        const swarmDir = path.join(getProjectRoot(), '.swarm');
+        if (!fs.existsSync(swarmDir)) {
+            fs.mkdirSync(swarmDir, { recursive: true });
+        }
+        const hnswPath = path.join(swarmDir, 'hnsw.index');
+        const metadataPath = path.join(swarmDir, 'hnsw.metadata.json');
+        const dbPath = options?.dbPath || path.join(swarmDir, 'memory.db');
+        // Create HNSW index with persistent storage
+        // @ruvector/core uses string enum for distanceMetric: 'Cosine', 'Euclidean', 'DotProduct', 'Manhattan'
+        const db = new VectorDb({
+            dimensions,
+            distanceMetric: 'Cosine',
+            storagePath: hnswPath // Persistent storage!
+        });
+        // Load metadata (entry info) if exists
+        const entries = new Map();
+        if (fs.existsSync(metadataPath)) {
+            try {
+                const metadataJson = fs.readFileSync(metadataPath, 'utf-8');
+                const metadata = JSON.parse(metadataJson);
+                for (const [key, value] of metadata) {
+                    entries.set(key, value);
+                }
+            }
+            catch {
+                // Metadata load failed, will rebuild
+            }
+        }
+        hnswIndex = {
+            db,
+            entries,
+            dimensions,
+            initialized: false
+        };
+        // Check if index already has data (from persistent storage)
+        const existingLen = await db.len();
+        if (existingLen > 0 && entries.size > 0) {
+            // Index loaded from disk, skip SQLite sync
+            hnswIndex.initialized = true;
+            hnswInitializing = false;
+            return hnswIndex;
+        }
+        if (fs.existsSync(dbPath)) {
+            try {
+                const initSqlJs = (await import('sql.js')).default;
+                const SQL = await initSqlJs();
+                const fileBuffer = fs.readFileSync(dbPath);
+                const sqlDb = new SQL.Database(fileBuffer);
+                // Load all entries with embeddings
                 const result = sqlDb.exec(`
           SELECT id, key, namespace, content, embedding
           FROM memory_entries
           WHERE status = 'active' AND embedding IS NOT NULL
           LIMIT 10000
-        `);
-                if (result[0]?.values) {
-                    for (const row of result[0].values) {
-                        const [id, key, ns, content, embeddingJson] = row;
-                        if (embeddingJson) {
-                            try {
-                                const embedding = JSON.parse(embeddingJson);
-                                const vector = new Float32Array(embedding);
-                                await db.insert({
-                                    id: String(id),
-                                    vector
-                                });
-                                hnswIndex.entries.set(String(id), {
-                                    id: String(id),
-                                    key: key || String(id),
-                                    namespace: ns || 'default',
-                                    content: content || ''
-                                });
-                            }
-                            catch {
-                                // Skip invalid embeddings
-                            }
-                        }
-                    }
-                }
-                sqlDb.close();
-            }
-            catch {
-                // SQLite load failed, start with empty index
-            }
-        }
-        hnswIndex.initialized = true;
-        hnswInitializing = false;
-        return hnswIndex;
-    }
-    catch {
-        hnswInitializing = false;
-        return null;
-    }
-}
-/**
- * Save HNSW metadata to disk for persistence
- */
-function saveHNSWMetadata() {
-    if (!hnswIndex?.entries)
-        return;
-    try {
-        const swarmDir = path.join(process.cwd(), '.swarm');
-        const metadataPath = path.join(swarmDir, 'hnsw.metadata.json');
-        const metadata = Array.from(hnswIndex.entries.entries());
-        fs.writeFileSync(metadataPath, JSON.stringify(metadata));
-    }
-    catch {
-        // Silently fail - metadata save is best-effort
-    }
-}
-/**
- * Add entry to HNSW index (with automatic persistence)
- */
-export async function addToHNSWIndex(id, embedding, entry) {
-    // ADR-053: Try AgentDB v3 bridge first
-    const bridge = await getBridge();
-    if (bridge) {
-        const bridgeResult = await bridge.bridgeAddToHNSW(id, embedding, entry);
-        if (bridgeResult === true)
-            return true;
-    }
-    const index = await getHNSWIndex({ dimensions: embedding.length });
-    if (!index)
-        return false;
-    try {
-        const vector = new Float32Array(embedding);
-        await index.db.insert({
-            id,
-            vector
-        });
-        index.entries.set(id, entry);
-        // Save metadata for persistence (debounced would be better for high-volume)
-        saveHNSWMetadata();
-        return true;
-    }
-    catch {
-        return false;
-    }
-}
-/**
- * Search HNSW index (150x faster than brute-force)
- * Returns results sorted by similarity (highest first)
- */
-export async function searchHNSWIndex(queryEmbedding, options) {
-    // ADR-053: Try AgentDB v3 bridge first
-    const bridge = await getBridge();
-    if (bridge) {
-        const bridgeResult = await bridge.bridgeSearchHNSW(queryEmbedding, options);
-        if (bridgeResult)
-            return bridgeResult;
-    }
-    const index = await getHNSWIndex({ dimensions: queryEmbedding.length });
-    if (!index)
-        return null;
-    try {
-        const vector = new Float32Array(queryEmbedding);
-        const k = options?.k ?? 10;
-        // HNSW search returns results with cosine distance (lower = more similar)
-        const results = await index.db.search({ vector, k: k * 2 }); // Get extra for filtering
-        const filtered = [];
-        for (const result of results) {
-            const entry = index.entries.get(result.id);
-            if (!entry)
-                continue;
-            // Filter by namespace if specified
-            if (options?.namespace && options.namespace !== 'all' && entry.namespace !== options.namespace) {
-                continue;
-            }
-            // Convert cosine distance to similarity score (1 - distance)
-            // Cosine distance from @ruvector/core: 0 = identical, 2 = opposite
-            const score = 1 - (result.score / 2);
-            filtered.push({
-                id: entry.id.substring(0, 12),
-                key: entry.key || entry.id.substring(0, 15),
-                content: entry.content.substring(0, 60) + (entry.content.length > 60 ? '...' : ''),
-                score,
-                namespace: entry.namespace
-            });
-            if (filtered.length >= k)
-                break;
-        }
-        // Sort by score descending (highest similarity first)
-        filtered.sort((a, b) => b.score - a.score);
-        return filtered;
-    }
-    catch {
-        return null;
-    }
-}
-/**
- * Get HNSW index status
- */
-export function getHNSWStatus() {
-    // ADR-053: If bridge was previously loaded, report availability
-    if (_bridge && _bridge !== null) {
-        // Bridge is loaded — HNSW-equivalent is available via AgentDB v3
-        return {
-            available: true,
-            initialized: true,
-            entryCount: hnswIndex?.entries.size ?? 0,
-            dimensions: hnswIndex?.dimensions ?? 384
-        };
-    }
-    return {
-        available: hnswIndex !== null,
-        initialized: hnswIndex?.initialized ?? false,
-        entryCount: hnswIndex?.entries.size ?? 0,
-        dimensions: hnswIndex?.dimensions ?? 384
-    };
-}
-/**
- * Clear the HNSW index (for rebuilding)
- */
-export function clearHNSWIndex() {
-    hnswIndex = null;
-}
-// ============================================================================
-// INT8 VECTOR QUANTIZATION (4x memory reduction)
-// ============================================================================
-/**
- * Quantize a Float32 embedding to Int8 (4x memory reduction)
- * Uses symmetric quantization with scale factor stored per-vector
- *
- * @param embedding - Float32 embedding array
- * @returns Quantized Int8 array with scale factor
- */
-export function quantizeInt8(embedding) {
-    const arr = embedding instanceof Float32Array ? embedding : new Float32Array(embedding);
-    // Find min/max for symmetric quantization
-    let min = Infinity, max = -Infinity;
-    for (let i = 0; i < arr.length; i++) {
-        if (arr[i] < min)
-            min = arr[i];
-        if (arr[i] > max)
-            max = arr[i];
-    }
-    // Symmetric quantization: scale = max(|min|, |max|) / 127
-    const absMax = Math.max(Math.abs(min), Math.abs(max));
-    const scale = absMax / 127 || 1e-10; // Avoid division by zero
-    const zeroPoint = 0; // Symmetric quantization
-    // Quantize
-    const quantized = new Int8Array(arr.length);
-    for (let i = 0; i < arr.length; i++) {
-        // Clamp to [-127, 127] to leave room for potential rounding
-        const q = Math.round(arr[i] / scale);
-        quantized[i] = Math.max(-127, Math.min(127, q));
-    }
-    return { quantized, scale, zeroPoint };
-}
-/**
- * Dequantize Int8 back to Float32
- *
- * @param quantized - Int8 quantized array
- * @param scale - Scale factor from quantization
- * @param zeroPoint - Zero point (usually 0 for symmetric)
- * @returns Float32Array
- */
-export function dequantizeInt8(quantized, scale, zeroPoint = 0) {
-    const result = new Float32Array(quantized.length);
-    for (let i = 0; i < quantized.length; i++) {
-        result[i] = (quantized[i] - zeroPoint) * scale;
-    }
-    return result;
-}
-/**
- * Compute cosine similarity between quantized vectors
- * Faster than dequantizing first
- */
-export function quantizedCosineSim(a, aScale, b, bScale) {
-    if (a.length !== b.length)
-        return 0;
-    let dot = 0, normA = 0, normB = 0;
-    for (let i = 0; i < a.length; i++) {
-        dot += a[i] * b[i];
-        normA += a[i] * a[i];
-        normB += b[i] * b[i];
-    }
-    // Scales cancel out in cosine similarity for normalized vectors
-    const mag = Math.sqrt(normA * normB);
-    return mag === 0 ? 0 : dot / mag;
-}
-/**
- * Get quantization statistics for an embedding
- */
-export function getQuantizationStats(embedding) {
-    const len = embedding.length;
-    const originalBytes = len * 4; // Float32 = 4 bytes
-    const quantizedBytes = len + 8; // Int8 = 1 byte + 8 bytes for scale/zeroPoint
-    const compressionRatio = originalBytes / quantizedBytes;
-    return { originalBytes, quantizedBytes, compressionRatio };
-}
-// ============================================================================
-// FLASH ATTENTION-STYLE BATCH OPERATIONS (V8-Optimized)
-// ============================================================================
-/**
- * Batch cosine similarity - compute query against multiple vectors
- * Optimized for V8 JIT with typed arrays
- * ~50μs per 1000 vectors (384-dim)
- */
-export function batchCosineSim(query, vectors) {
-    const n = vectors.length;
-    const scores = new Float32Array(n);
-    if (n === 0 || query.length === 0)
-        return scores;
-    // Pre-compute query norm
-    let queryNorm = 0;
-    for (let i = 0; i < query.length; i++) {
-        queryNorm += query[i] * query[i];
-    }
-    queryNorm = Math.sqrt(queryNorm);
-    if (queryNorm === 0)
-        return scores;
-    // Compute similarities
-    for (let v = 0; v < n; v++) {
-        const vec = vectors[v];
-        const len = Math.min(query.length, vec.length);
-        let dot = 0, vecNorm = 0;
-        for (let i = 0; i < len; i++) {
-            dot += query[i] * vec[i];
-            vecNorm += vec[i] * vec[i];
-        }
-        vecNorm = Math.sqrt(vecNorm);
-        scores[v] = vecNorm === 0 ? 0 : dot / (queryNorm * vecNorm);
-    }
-    return scores;
-}
-/**
- * Softmax normalization for attention scores
- * Numerically stable implementation
- */
-export function softmaxAttention(scores, temperature = 1.0) {
-    const n = scores.length;
-    const result = new Float32Array(n);
-    if (n === 0)
-        return result;
-    // Find max for numerical stability
-    let max = scores[0];
-    for (let i = 1; i < n; i++) {
-        if (scores[i] > max)
-            max = scores[i];
-    }
-    // Compute exp and sum
-    let sum = 0;
-    for (let i = 0; i < n; i++) {
-        result[i] = Math.exp((scores[i] - max) / temperature);
-        sum += result[i];
-    }
-    // Normalize
-    if (sum > 0) {
-        for (let i = 0; i < n; i++) {
-            result[i] /= sum;
-        }
-    }
-    return result;
-}
-/**
- * Top-K selection with partial sort (O(n + k log k))
- * More efficient than full sort for small k
- */
-export function topKIndices(scores, k) {
-    const n = scores.length;
-    if (k >= n) {
-        // Return all indices sorted by score
-        return Array.from({ length: n }, (_, i) => i)
-            .sort((a, b) => scores[b] - scores[a]);
-    }
-    // Build min-heap of size k
-    const heap = [];
-    for (let i = 0; i < n; i++) {
-        if (heap.length < k) {
-            heap.push({ idx: i, score: scores[i] });
-            // Bubble up
-            let j = heap.length - 1;
-            while (j > 0) {
-                const parent = Math.floor((j - 1) / 2);
-                if (heap[j].score < heap[parent].score) {
-                    [heap[j], heap[parent]] = [heap[parent], heap[j]];
-                    j = parent;
-                }
-                else
-                    break;
-            }
-        }
-        else if (scores[i] > heap[0].score) {
-            // Replace min and heapify down
-            heap[0] = { idx: i, score: scores[i] };
-            let j = 0;
-            while (true) {
-                const left = 2 * j + 1, right = 2 * j + 2;
-                let smallest = j;
-                if (left < k && heap[left].score < heap[smallest].score)
-                    smallest = left;
-                if (right < k && heap[right].score < heap[smallest].score)
-                    smallest = right;
-                if (smallest === j)
-                    break;
-                [heap[j], heap[smallest]] = [heap[smallest], heap[j]];
-                j = smallest;
-            }
-        }
-    }
-    // Extract and sort descending
-    return heap.sort((a, b) => b.score - a.score).map(h => h.idx);
-}
-/**
- * Flash Attention-style search
- * Combines batch similarity, softmax, and top-k in one pass
- * Returns indices and attention weights
- */
-export function flashAttentionSearch(query, vectors, options = {}) {
-    const { k = 10, temperature = 1.0, threshold = 0 } = options;
-    // Compute batch similarity
-    const scores = batchCosineSim(query, vectors);
-    // Get top-k indices
-    const indices = topKIndices(scores, k);
-    // Filter by threshold
-    const filtered = indices.filter(i => scores[i] >= threshold);
-    // Extract scores for filtered results
-    const topScores = new Float32Array(filtered.length);
-    for (let i = 0; i < filtered.length; i++) {
-        topScores[i] = scores[filtered[i]];
-    }
-    // Compute attention weights (softmax over top-k)
-    const weights = softmaxAttention(topScores, temperature);
-    return { indices: filtered, scores: topScores, weights };
-}
-// ============================================================================
-// METADATA AND INITIALIZATION
-// ============================================================================
-/**
- * Initial metadata to insert after schema creation
- */
-export function getInitialMetadata(backend) {
+        `);
+                if (result[0]?.values) {
+                    for (const row of result[0].values) {
+                        const [id, key, ns, content, embeddingJson] = row;
+                        if (embeddingJson) {
+                            try {
+                                const embedding = JSON.parse(embeddingJson);
+                                const vector = new Float32Array(embedding);
+                                await db.insert({
+                                    id: String(id),
+                                    vector
+                                });
+                                hnswIndex.entries.set(String(id), {
+                                    id: String(id),
+                                    key: key || String(id),
+                                    namespace: ns || 'default',
+                                    content: content || ''
+                                });
+                            }
+                            catch {
+                                // Skip invalid embeddings
+                            }
+                        }
+                    }
+                }
+                sqlDb.close();
+            }
+            catch {
+                // SQLite load failed, start with empty index
+            }
+        }
+        hnswIndex.initialized = true;
+        hnswInitializing = false;
+        return hnswIndex;
+    }
+    catch {
+        hnswInitializing = false;
+        return null;
+    }
+}
+/**
+ * Save HNSW metadata to disk for persistence
+ */
+function saveHNSWMetadata() {
+    if (!hnswIndex?.entries)
+        return;
+    try {
+        const swarmDir = path.join(getProjectRoot(), '.swarm');
+        const metadataPath = path.join(swarmDir, 'hnsw.metadata.json');
+        const metadata = Array.from(hnswIndex.entries.entries());
+        fs.writeFileSync(metadataPath, JSON.stringify(metadata));
+    }
+    catch {
+        // Silently fail - metadata save is best-effort
+    }
+}
+/**
+ * Add entry to HNSW index (with automatic persistence)
+ */
+export async function addToHNSWIndex(id, embedding, entry) {
+    // ADR-053: Try AgentDB v3 bridge first
+    const bridge = await getBridge();
+    if (bridge) {
+        const bridgeResult = await bridge.bridgeAddToHNSW(id, embedding, entry);
+        if (bridgeResult === true)
+            return true;
+    }
+    const index = await getHNSWIndex({ dimensions: embedding.length });
+    if (!index)
+        return false;
+    try {
+        const vector = new Float32Array(embedding);
+        await index.db.insert({
+            id,
+            vector
+        });
+        index.entries.set(id, entry);
+        // Save metadata for persistence (debounced would be better for high-volume)
+        saveHNSWMetadata();
+        return true;
+    }
+    catch {
+        return false;
+    }
+}
+/**
+ * Search HNSW index (150x faster than brute-force)
+ * Returns results sorted by similarity (highest first)
+ */
+export async function searchHNSWIndex(queryEmbedding, options) {
+    // ADR-053: Try AgentDB v3 bridge first
+    const bridge = await getBridge();
+    if (bridge) {
+        const bridgeResult = await bridge.bridgeSearchHNSW(queryEmbedding, options);
+        if (bridgeResult)
+            return bridgeResult;
+    }
+    const index = await getHNSWIndex({ dimensions: queryEmbedding.length });
+    if (!index)
+        return null;
+    try {
+        const vector = new Float32Array(queryEmbedding);
+        const k = options?.k ?? 10;
+        // HNSW search returns results with cosine distance (lower = more similar)
+        const results = await index.db.search({ vector, k: k * 2 }); // Get extra for filtering
+        const filtered = [];
+        for (const result of results) {
+            const entry = index.entries.get(result.id);
+            if (!entry)
+                continue;
+            // Filter by namespace if specified
+            if (options?.namespace && options.namespace !== 'all' && entry.namespace !== options.namespace) {
+                continue;
+            }
+            // Convert cosine distance to similarity score (1 - distance)
+            // Cosine distance from @ruvector/core: 0 = identical, 2 = opposite
+            const score = 1 - (result.score / 2);
+            filtered.push({
+                id: entry.id.substring(0, 12),
+                key: entry.key || entry.id.substring(0, 15),
+                content: entry.content.substring(0, 60) + (entry.content.length > 60 ? '...' : ''),
+                score,
+                namespace: entry.namespace
+            });
+            if (filtered.length >= k)
+                break;
+        }
+        // Sort by score descending (highest similarity first)
+        filtered.sort((a, b) => b.score - a.score);
+        return filtered;
+    }
+    catch {
+        return null;
+    }
+}
+/**
+ * Get HNSW index status
+ */
+export function getHNSWStatus() {
+    // ADR-053: If bridge was previously loaded, report availability
+    if (_bridge && _bridge !== null) {
+        // Bridge is loaded — HNSW-equivalent is available via AgentDB v3
+        return {
+            available: true,
+            initialized: true,
+            entryCount: hnswIndex?.entries.size ?? 0,
+            dimensions: hnswIndex?.dimensions ?? 384
+        };
+    }
+    return {
+        available: hnswIndex !== null,
+        initialized: hnswIndex?.initialized ?? false,
+        entryCount: hnswIndex?.entries.size ?? 0,
+        dimensions: hnswIndex?.dimensions ?? 384
+    };
+}
+/**
+ * Clear the HNSW index (for rebuilding)
+ */
+export function clearHNSWIndex() {
+    hnswIndex = null;
+}
+// ============================================================================
+// INT8 VECTOR QUANTIZATION (4x memory reduction)
+// ============================================================================
+/**
+ * Quantize a Float32 embedding to Int8 (4x memory reduction)
+ * Uses symmetric quantization with scale factor stored per-vector
+ *
+ * @param embedding - Float32 embedding array
+ * @returns Quantized Int8 array with scale factor
+ */
+export function quantizeInt8(embedding) {
+    const arr = embedding instanceof Float32Array ? embedding : new Float32Array(embedding);
+    // Find min/max for symmetric quantization
+    let min = Infinity, max = -Infinity;
+    for (let i = 0; i < arr.length; i++) {
+        if (arr[i] < min)
+            min = arr[i];
+        if (arr[i] > max)
+            max = arr[i];
+    }
+    // Symmetric quantization: scale = max(|min|, |max|) / 127
+    const absMax = Math.max(Math.abs(min), Math.abs(max));
+    const scale = absMax / 127 || 1e-10; // Avoid division by zero
+    const zeroPoint = 0; // Symmetric quantization
+    // Quantize
+    const quantized = new Int8Array(arr.length);
+    for (let i = 0; i < arr.length; i++) {
+        // Clamp to [-127, 127] to leave room for potential rounding
+        const q = Math.round(arr[i] / scale);
+        quantized[i] = Math.max(-127, Math.min(127, q));
+    }
+    return { quantized, scale, zeroPoint };
+}
+/**
+ * Dequantize Int8 back to Float32
+ *
+ * @param quantized - Int8 quantized array
+ * @param scale - Scale factor from quantization
+ * @param zeroPoint - Zero point (usually 0 for symmetric)
+ * @returns Float32Array
+ */
+export function dequantizeInt8(quantized, scale, zeroPoint = 0) {
+    const result = new Float32Array(quantized.length);
+    for (let i = 0; i < quantized.length; i++) {
+        result[i] = (quantized[i] - zeroPoint) * scale;
+    }
+    return result;
+}
+/**
+ * Compute cosine similarity between quantized vectors
+ * Faster than dequantizing first
+ */
+export function quantizedCosineSim(a, aScale, b, bScale) {
+    if (a.length !== b.length)
+        return 0;
+    let dot = 0, normA = 0, normB = 0;
+    for (let i = 0; i < a.length; i++) {
+        dot += a[i] * b[i];
+        normA += a[i] * a[i];
+        normB += b[i] * b[i];
+    }
+    // Scales cancel out in cosine similarity for normalized vectors
+    const mag = Math.sqrt(normA * normB);
+    return mag === 0 ? 0 : dot / mag;
+}
+/**
+ * Get quantization statistics for an embedding
+ */
+export function getQuantizationStats(embedding) {
+    const len = embedding.length;
+    const originalBytes = len * 4; // Float32 = 4 bytes
+    const quantizedBytes = len + 8; // Int8 = 1 byte + 8 bytes for scale/zeroPoint
+    const compressionRatio = originalBytes / quantizedBytes;
+    return { originalBytes, quantizedBytes, compressionRatio };
+}
+// ============================================================================
+// FLASH ATTENTION-STYLE BATCH OPERATIONS (V8-Optimized)
+// ============================================================================
+/**
+ * Batch cosine similarity - compute query against multiple vectors
+ * Optimized for V8 JIT with typed arrays
+ * ~50μs per 1000 vectors (384-dim)
+ */
+export function batchCosineSim(query, vectors) {
+    const n = vectors.length;
+    const scores = new Float32Array(n);
+    if (n === 0 || query.length === 0)
+        return scores;
+    // Pre-compute query norm
+    let queryNorm = 0;
+    for (let i = 0; i < query.length; i++) {
+        queryNorm += query[i] * query[i];
+    }
+    queryNorm = Math.sqrt(queryNorm);
+    if (queryNorm === 0)
+        return scores;
+    // Compute similarities
+    for (let v = 0; v < n; v++) {
+        const vec = vectors[v];
+        const len = Math.min(query.length, vec.length);
+        let dot = 0, vecNorm = 0;
+        for (let i = 0; i < len; i++) {
+            dot += query[i] * vec[i];
+            vecNorm += vec[i] * vec[i];
+        }
+        vecNorm = Math.sqrt(vecNorm);
+        scores[v] = vecNorm === 0 ? 0 : dot / (queryNorm * vecNorm);
+    }
+    return scores;
+}
+/**
+ * Softmax normalization for attention scores
+ * Numerically stable implementation
+ */
+export function softmaxAttention(scores, temperature = 1.0) {
+    const n = scores.length;
+    const result = new Float32Array(n);
+    if (n === 0)
+        return result;
+    // Find max for numerical stability
+    let max = scores[0];
+    for (let i = 1; i < n; i++) {
+        if (scores[i] > max)
+            max = scores[i];
+    }
+    // Compute exp and sum
+    let sum = 0;
+    for (let i = 0; i < n; i++) {
+        result[i] = Math.exp((scores[i] - max) / temperature);
+        sum += result[i];
+    }
+    // Normalize
+    if (sum > 0) {
+        for (let i = 0; i < n; i++) {
+            result[i] /= sum;
+        }
+    }
+    return result;
+}
+/**
+ * Top-K selection with partial sort (O(n + k log k))
+ * More efficient than full sort for small k
+ */
+export function topKIndices(scores, k) {
+    const n = scores.length;
+    if (k >= n) {
+        // Return all indices sorted by score
+        return Array.from({ length: n }, (_, i) => i)
+            .sort((a, b) => scores[b] - scores[a]);
+    }
+    // Build min-heap of size k
+    const heap = [];
+    for (let i = 0; i < n; i++) {
+        if (heap.length < k) {
+            heap.push({ idx: i, score: scores[i] });
+            // Bubble up
+            let j = heap.length - 1;
+            while (j > 0) {
+                const parent = Math.floor((j - 1) / 2);
+                if (heap[j].score < heap[parent].score) {
+                    [heap[j], heap[parent]] = [heap[parent], heap[j]];
+                    j = parent;
+                }
+                else
+                    break;
+            }
+        }
+        else if (scores[i] > heap[0].score) {
+            // Replace min and heapify down
+            heap[0] = { idx: i, score: scores[i] };
+            let j = 0;
+            while (true) {
+                const left = 2 * j + 1, right = 2 * j + 2;
+                let smallest = j;
+                if (left < k && heap[left].score < heap[smallest].score)
+                    smallest = left;
+                if (right < k && heap[right].score < heap[smallest].score)
+                    smallest = right;
+                if (smallest === j)
+                    break;
+                [heap[j], heap[smallest]] = [heap[smallest], heap[j]];
+                j = smallest;
+            }
+        }
+    }
+    // Extract and sort descending
+    return heap.sort((a, b) => b.score - a.score).map(h => h.idx);
+}
+/**
+ * Flash Attention-style search
+ * Combines batch similarity, softmax, and top-k in one pass
+ * Returns indices and attention weights
+ */
+export function flashAttentionSearch(query, vectors, options = {}) {
+    const { k = 10, temperature = 1.0, threshold = 0 } = options;
+    // Compute batch similarity
+    const scores = batchCosineSim(query, vectors);
+    // Get top-k indices
+    const indices = topKIndices(scores, k);
+    // Filter by threshold
+    const filtered = indices.filter(i => scores[i] >= threshold);
+    // Extract scores for filtered results
+    const topScores = new Float32Array(filtered.length);
+    for (let i = 0; i < filtered.length; i++) {
+        topScores[i] = scores[filtered[i]];
+    }
+    // Compute attention weights (softmax over top-k)
+    const weights = softmaxAttention(topScores, temperature);
+    return { indices: filtered, scores: topScores, weights };
+}
+// ============================================================================
+// METADATA AND INITIALIZATION
+// ============================================================================
+/**
+ * Initial metadata to insert after schema creation
+ */
+export function getInitialMetadata(backend) {
     return `
 INSERT OR REPLACE INTO metadata (key, value) VALUES
   ('schema_version', '3.0.0'),
@@ -805,405 +856,405 @@ INSERT OR REPLACE INTO metadata (key, value) VALUES
 INSERT OR IGNORE INTO vector_indexes (id, name, dimensions) VALUES
   ('default', 'default', 768),
   ('patterns', 'patterns', 768);
-`;
-}
-/**
- * Ensure memory_entries table has all required columns
- * Adds missing columns for older databases (e.g., 'content' column)
- */
-export async function ensureSchemaColumns(dbPath) {
-    const columnsAdded = [];
-    try {
-        if (!fs.existsSync(dbPath)) {
-            return { success: true, columnsAdded: [] };
-        }
-        const initSqlJs = (await import('sql.js')).default;
-        const SQL = await initSqlJs();
-        const fileBuffer = fs.readFileSync(dbPath);
-        const db = new SQL.Database(fileBuffer);
-        // Get current columns in memory_entries
-        const tableInfo = db.exec("PRAGMA table_info(memory_entries)");
-        const existingColumns = new Set(tableInfo[0]?.values?.map(row => row[1]) || []);
-        // Required columns that may be missing in older schemas
-        // Issue #977: 'type' column was missing from this list, causing store failures on older DBs
-        const requiredColumns = [
-            { name: 'content', definition: "content TEXT DEFAULT ''" },
-            { name: 'type', definition: "type TEXT DEFAULT 'semantic'" },
-            { name: 'embedding', definition: 'embedding TEXT' },
-            { name: 'embedding_model', definition: "embedding_model TEXT DEFAULT 'local'" },
-            { name: 'embedding_dimensions', definition: 'embedding_dimensions INTEGER' },
-            { name: 'tags', definition: 'tags TEXT' },
-            { name: 'metadata', definition: 'metadata TEXT' },
-            { name: 'owner_id', definition: 'owner_id TEXT' },
-            { name: 'expires_at', definition: 'expires_at INTEGER' },
-            { name: 'last_accessed_at', definition: 'last_accessed_at INTEGER' },
-            { name: 'access_count', definition: 'access_count INTEGER DEFAULT 0' },
-            { name: 'status', definition: "status TEXT DEFAULT 'active'" }
-        ];
-        let modified = false;
-        for (const col of requiredColumns) {
-            if (!existingColumns.has(col.name)) {
-                try {
-                    db.run(`ALTER TABLE memory_entries ADD COLUMN ${col.definition}`);
-                    columnsAdded.push(col.name);
-                    modified = true;
-                }
-                catch (e) {
-                    // Column might already exist or other error - continue
-                }
-            }
-        }
-        if (modified) {
-            // Save updated database
-            const data = db.export();
-            fs.writeFileSync(dbPath, Buffer.from(data));
-        }
-        db.close();
-        return { success: true, columnsAdded };
-    }
-    catch (error) {
-        return {
-            success: false,
-            columnsAdded,
-            error: error instanceof Error ? error.message : String(error)
-        };
-    }
-}
-/**
- * Check for legacy database installations and migrate if needed
- */
-export async function checkAndMigrateLegacy(options) {
-    const { dbPath, verbose = false } = options;
-    // Check for legacy locations
-    const legacyPaths = [
-        path.join(process.cwd(), 'memory.db'),
-        path.join(process.cwd(), '.claude/memory.db'),
-        path.join(process.cwd(), 'data/memory.db'),
-        path.join(process.cwd(), '.claude-flow/memory.db')
-    ];
-    for (const legacyPath of legacyPaths) {
-        if (fs.existsSync(legacyPath) && legacyPath !== dbPath) {
-            try {
-                const initSqlJs = (await import('sql.js')).default;
-                const SQL = await initSqlJs();
-                const legacyBuffer = fs.readFileSync(legacyPath);
-                const legacyDb = new SQL.Database(legacyBuffer);
-                // Check if it has data
-                const countResult = legacyDb.exec('SELECT COUNT(*) FROM memory_entries');
-                const count = countResult[0]?.values[0]?.[0] || 0;
-                // Get version if available
-                let version = 'unknown';
-                try {
-                    const versionResult = legacyDb.exec("SELECT value FROM metadata WHERE key='schema_version'");
-                    version = versionResult[0]?.values[0]?.[0] || 'unknown';
-                }
-                catch { /* no metadata table */ }
-                legacyDb.close();
-                if (count > 0) {
-                    return {
-                        needsMigration: true,
-                        legacyVersion: version,
-                        legacyEntries: count
-                    };
-                }
-            }
-            catch {
-                // Not a valid SQLite database, skip
-            }
-        }
-    }
-    return { needsMigration: false };
-}
-/**
- * ADR-053: Activate ControllerRegistry so AgentDB v3 controllers
- * (ReasoningBank, SkillLibrary, ExplainableRecall, etc.) are instantiated.
- *
- * Uses the memory-bridge's getControllerRegistry() which lazily creates
- * a singleton ControllerRegistry and initializes it with the given dbPath.
- * After this call, all enabled controllers are ready for immediate use.
- *
- * Failures are isolated: if @claude-flow/memory or agentdb is not installed,
- * this returns an empty result without throwing.
- */
-async function activateControllerRegistry(dbPath, verbose) {
-    const startTime = performance.now();
-    const activated = [];
-    const failed = [];
-    try {
-        const bridge = await getBridge();
-        if (!bridge) {
-            return { activated, failed, initTimeMs: performance.now() - startTime };
-        }
-        const registry = await bridge.getControllerRegistry(dbPath);
-        if (!registry) {
-            return { activated, failed, initTimeMs: performance.now() - startTime };
-        }
-        // Collect controller status from the registry
-        if (typeof registry.listControllers === 'function') {
-            const controllers = registry.listControllers();
-            for (const ctrl of controllers) {
-                if (ctrl.enabled) {
-                    activated.push(ctrl.name);
-                }
-                else {
-                    failed.push(ctrl.name);
-                }
-            }
-        }
-        if (verbose && activated.length > 0) {
-            console.log(`ControllerRegistry: ${activated.length} controllers activated`);
-        }
-    }
-    catch {
-        // ControllerRegistry activation is best-effort
-    }
-    return { activated, failed, initTimeMs: performance.now() - startTime };
-}
-/**
- * Initialize the memory database properly using sql.js
- */
-export async function initializeMemoryDatabase(options) {
-    const { backend = 'hybrid', dbPath: customPath, force = false, verbose = false, migrate = true } = options;
-    const swarmDir = path.join(process.cwd(), '.swarm');
-    const dbPath = customPath || path.join(swarmDir, 'memory.db');
-    const dbDir = path.dirname(dbPath);
-    try {
-        // Create directory if needed
-        if (!fs.existsSync(dbDir)) {
-            fs.mkdirSync(dbDir, { recursive: true });
-        }
-        // Check for legacy installations
-        if (migrate) {
-            const legacyCheck = await checkAndMigrateLegacy({ dbPath, verbose });
-            if (legacyCheck.needsMigration && verbose) {
-                console.log(`Found legacy database (v${legacyCheck.legacyVersion}) with ${legacyCheck.legacyEntries} entries`);
-            }
-        }
-        // Check existing database
-        if (fs.existsSync(dbPath) && !force) {
-            return {
-                success: false,
-                backend,
-                dbPath,
-                schemaVersion: '3.0.0',
-                tablesCreated: [],
-                indexesCreated: [],
-                features: {
-                    vectorEmbeddings: false,
-                    patternLearning: false,
-                    temporalDecay: false,
-                    hnswIndexing: false,
-                    migrationTracking: false
-                },
-                error: 'Database already exists. Use --force to reinitialize.'
-            };
-        }
-        // Try to use sql.js (WASM SQLite)
-        let db;
-        let usedSqlJs = false;
-        try {
-            // Dynamic import of sql.js
-            const initSqlJs = (await import('sql.js')).default;
-            const SQL = await initSqlJs();
-            // Load existing database or create new
-            if (fs.existsSync(dbPath) && force) {
-                fs.unlinkSync(dbPath);
-            }
-            db = new SQL.Database();
-            usedSqlJs = true;
-        }
-        catch (e) {
-            // sql.js not available, fall back to writing schema file
-            if (verbose) {
-                console.log('sql.js not available, writing schema file for later initialization');
-            }
-        }
-        if (usedSqlJs && db) {
-            // Execute schema
-            db.run(MEMORY_SCHEMA_V3);
-            // Insert initial metadata
-            db.run(getInitialMetadata(backend));
-            // Save to file
-            const data = db.export();
-            const buffer = Buffer.from(data);
-            fs.writeFileSync(dbPath, buffer);
-            // Close database
-            db.close();
-            // Also create schema file for reference
-            const schemaPath = path.join(dbDir, 'schema.sql');
-            fs.writeFileSync(schemaPath, MEMORY_SCHEMA_V3 + '\n' + getInitialMetadata(backend));
-            // ADR-053: Activate ControllerRegistry so controllers (ReasoningBank,
-            // SkillLibrary, ExplainableRecall, etc.) are instantiated during init
-            const controllerResult = await activateControllerRegistry(dbPath, verbose);
-            return {
-                success: true,
-                backend,
-                dbPath,
-                schemaVersion: '3.0.0',
-                tablesCreated: [
-                    'memory_entries',
-                    'patterns',
-                    'pattern_history',
-                    'trajectories',
-                    'trajectory_steps',
-                    'migration_state',
-                    'sessions',
-                    'vector_indexes',
-                    'metadata'
-                ],
-                indexesCreated: [
-                    'idx_memory_namespace',
-                    'idx_memory_key',
-                    'idx_memory_type',
-                    'idx_memory_status',
-                    'idx_memory_created',
-                    'idx_memory_accessed',
-                    'idx_memory_owner',
-                    'idx_patterns_type',
-                    'idx_patterns_confidence',
-                    'idx_patterns_status',
-                    'idx_patterns_last_matched',
-                    'idx_pattern_history_pattern',
-                    'idx_steps_trajectory'
-                ],
-                features: {
-                    vectorEmbeddings: true,
-                    patternLearning: true,
-                    temporalDecay: true,
-                    hnswIndexing: true,
-                    migrationTracking: true
-                },
-                controllers: controllerResult,
-            };
-        }
-        else {
-            // Fall back to schema file approach
-            const schemaPath = path.join(dbDir, 'schema.sql');
-            fs.writeFileSync(schemaPath, MEMORY_SCHEMA_V3 + '\n' + getInitialMetadata(backend));
-            // Create minimal valid SQLite file
-            const sqliteHeader = Buffer.alloc(4096, 0);
-            // SQLite format 3 header
-            Buffer.from('SQLite format 3\0').copy(sqliteHeader, 0);
-            sqliteHeader[16] = 0x10; // page size high byte (4096)
-            sqliteHeader[17] = 0x00; // page size low byte
-            sqliteHeader[18] = 0x01; // file format write version
-            sqliteHeader[19] = 0x01; // file format read version
-            sqliteHeader[24] = 0x00; // max embedded payload
-            sqliteHeader[25] = 0x40;
-            sqliteHeader[26] = 0x20; // min embedded payload
-            sqliteHeader[27] = 0x20; // leaf payload
-            fs.writeFileSync(dbPath, sqliteHeader);
-            // ADR-053: Activate ControllerRegistry even on fallback path
-            const controllerResult = await activateControllerRegistry(dbPath, verbose);
-            return {
-                success: true,
-                backend,
-                dbPath,
-                schemaVersion: '3.0.0',
-                tablesCreated: [
-                    'memory_entries (pending)',
-                    'patterns (pending)',
-                    'pattern_history (pending)',
-                    'trajectories (pending)',
-                    'trajectory_steps (pending)',
-                    'migration_state (pending)',
-                    'sessions (pending)',
-                    'vector_indexes (pending)',
-                    'metadata (pending)'
-                ],
-                indexesCreated: [],
-                features: {
-                    vectorEmbeddings: true,
-                    patternLearning: true,
-                    temporalDecay: true,
-                    hnswIndexing: true,
-                    migrationTracking: true
-                },
-                controllers: controllerResult,
-            };
-        }
-    }
-    catch (error) {
-        return {
-            success: false,
-            backend,
-            dbPath,
-            schemaVersion: '3.0.0',
-            tablesCreated: [],
-            indexesCreated: [],
-            features: {
-                vectorEmbeddings: false,
-                patternLearning: false,
-                temporalDecay: false,
-                hnswIndexing: false,
-                migrationTracking: false
-            },
-            error: error instanceof Error ? error.message : String(error)
-        };
-    }
-}
-/**
- * Check if memory database is properly initialized
- */
-export async function checkMemoryInitialization(dbPath) {
-    const swarmDir = path.join(process.cwd(), '.swarm');
-    const path_ = dbPath || path.join(swarmDir, 'memory.db');
-    if (!fs.existsSync(path_)) {
-        return { initialized: false };
-    }
-    try {
-        // Try to load with sql.js
-        const initSqlJs = (await import('sql.js')).default;
-        const SQL = await initSqlJs();
-        const fileBuffer = fs.readFileSync(path_);
-        const db = new SQL.Database(fileBuffer);
-        // Check for metadata table
-        const tables = db.exec("SELECT name FROM sqlite_master WHERE type='table'");
-        const tableNames = tables[0]?.values?.map(v => v[0]) || [];
-        // Get version
-        let version = 'unknown';
-        let backend = 'unknown';
-        try {
-            const versionResult = db.exec("SELECT value FROM metadata WHERE key='schema_version'");
-            version = versionResult[0]?.values[0]?.[0] || 'unknown';
-            const backendResult = db.exec("SELECT value FROM metadata WHERE key='backend'");
-            backend = backendResult[0]?.values[0]?.[0] || 'unknown';
-        }
-        catch {
-            // Metadata table might not exist
-        }
-        db.close();
-        return {
-            initialized: true,
-            version,
-            backend,
-            features: {
-                vectorEmbeddings: tableNames.includes('vector_indexes'),
-                patternLearning: tableNames.includes('patterns'),
-                temporalDecay: tableNames.includes('pattern_history')
-            },
-            tables: tableNames
-        };
-    }
-    catch {
-        // Could not read database
-        return { initialized: false };
-    }
-}
-/**
- * Apply temporal decay to patterns
- * Reduces confidence of patterns that haven't been used recently
- */
-export async function applyTemporalDecay(dbPath) {
-    const swarmDir = path.join(process.cwd(), '.swarm');
-    const path_ = dbPath || path.join(swarmDir, 'memory.db');
-    try {
-        const initSqlJs = (await import('sql.js')).default;
-        const SQL = await initSqlJs();
-        const fileBuffer = fs.readFileSync(path_);
-        const db = new SQL.Database(fileBuffer);
-        // Apply decay: confidence *= exp(-decay_rate * days_since_last_use)
-        const now = Date.now();
+`;
+}
+/**
+ * Ensure memory_entries table has all required columns
+ * Adds missing columns for older databases (e.g., 'content' column)
+ */
+export async function ensureSchemaColumns(dbPath) {
+    const columnsAdded = [];
+    try {
+        if (!fs.existsSync(dbPath)) {
+            return { success: true, columnsAdded: [] };
+        }
+        const initSqlJs = (await import('sql.js')).default;
+        const SQL = await initSqlJs();
+        const fileBuffer = fs.readFileSync(dbPath);
+        const db = new SQL.Database(fileBuffer);
+        // Get current columns in memory_entries
+        const tableInfo = db.exec("PRAGMA table_info(memory_entries)");
+        const existingColumns = new Set(tableInfo[0]?.values?.map(row => row[1]) || []);
+        // Required columns that may be missing in older schemas
+        // Issue #977: 'type' column was missing from this list, causing store failures on older DBs
+        const requiredColumns = [
+            { name: 'content', definition: "content TEXT DEFAULT ''" },
+            { name: 'type', definition: "type TEXT DEFAULT 'semantic'" },
+            { name: 'embedding', definition: 'embedding TEXT' },
+            { name: 'embedding_model', definition: "embedding_model TEXT DEFAULT 'local'" },
+            { name: 'embedding_dimensions', definition: 'embedding_dimensions INTEGER' },
+            { name: 'tags', definition: 'tags TEXT' },
+            { name: 'metadata', definition: 'metadata TEXT' },
+            { name: 'owner_id', definition: 'owner_id TEXT' },
+            { name: 'expires_at', definition: 'expires_at INTEGER' },
+            { name: 'last_accessed_at', definition: 'last_accessed_at INTEGER' },
+            { name: 'access_count', definition: 'access_count INTEGER DEFAULT 0' },
+            { name: 'status', definition: "status TEXT DEFAULT 'active'" }
+        ];
+        let modified = false;
+        for (const col of requiredColumns) {
+            if (!existingColumns.has(col.name)) {
+                try {
+                    db.run(`ALTER TABLE memory_entries ADD COLUMN ${col.definition}`);
+                    columnsAdded.push(col.name);
+                    modified = true;
+                }
+                catch (e) {
+                    // Column might already exist or other error - continue
+                }
+            }
+        }
+        if (modified) {
+            // Save updated database
+            const data = db.export();
+            fs.writeFileSync(dbPath, Buffer.from(data));
+        }
+        db.close();
+        return { success: true, columnsAdded };
+    }
+    catch (error) {
+        return {
+            success: false,
+            columnsAdded,
+            error: error instanceof Error ? error.message : String(error)
+        };
+    }
+}
+/**
+ * Check for legacy database installations and migrate if needed
+ */
+export async function checkAndMigrateLegacy(options) {
+    const { dbPath, verbose = false } = options;
+    // Check for legacy locations
+    const legacyPaths = [
+        path.join(getProjectRoot(), 'memory.db'),
+        path.join(getProjectRoot(), '.claude/memory.db'),
+        path.join(getProjectRoot(), 'data/memory.db'),
+        path.join(getProjectRoot(), '.claude-flow/memory.db')
+    ];
+    for (const legacyPath of legacyPaths) {
+        if (fs.existsSync(legacyPath) && legacyPath !== dbPath) {
+            try {
+                const initSqlJs = (await import('sql.js')).default;
+                const SQL = await initSqlJs();
+                const legacyBuffer = fs.readFileSync(legacyPath);
+                const legacyDb = new SQL.Database(legacyBuffer);
+                // Check if it has data
+                const countResult = legacyDb.exec('SELECT COUNT(*) FROM memory_entries');
+                const count = countResult[0]?.values[0]?.[0] || 0;
+                // Get version if available
+                let version = 'unknown';
+                try {
+                    const versionResult = legacyDb.exec("SELECT value FROM metadata WHERE key='schema_version'");
+                    version = versionResult[0]?.values[0]?.[0] || 'unknown';
+                }
+                catch { /* no metadata table */ }
+                legacyDb.close();
+                if (count > 0) {
+                    return {
+                        needsMigration: true,
+                        legacyVersion: version,
+                        legacyEntries: count
+                    };
+                }
+            }
+            catch {
+                // Not a valid SQLite database, skip
+            }
+        }
+    }
+    return { needsMigration: false };
+}
+/**
+ * ADR-053: Activate ControllerRegistry so AgentDB v3 controllers
+ * (ReasoningBank, SkillLibrary, ExplainableRecall, etc.) are instantiated.
+ *
+ * Uses the memory-bridge's getControllerRegistry() which lazily creates
+ * a singleton ControllerRegistry and initializes it with the given dbPath.
+ * After this call, all enabled controllers are ready for immediate use.
+ *
+ * Failures are isolated: if @claude-flow/memory or agentdb is not installed,
+ * this returns an empty result without throwing.
+ */
+async function activateControllerRegistry(dbPath, verbose) {
+    const startTime = performance.now();
+    const activated = [];
+    const failed = [];
+    try {
+        const bridge = await getBridge();
+        if (!bridge) {
+            return { activated, failed, initTimeMs: performance.now() - startTime };
+        }
+        const registry = await bridge.getControllerRegistry(dbPath);
+        if (!registry) {
+            return { activated, failed, initTimeMs: performance.now() - startTime };
+        }
+        // Collect controller status from the registry
+        if (typeof registry.listControllers === 'function') {
+            const controllers = registry.listControllers();
+            for (const ctrl of controllers) {
+                if (ctrl.enabled) {
+                    activated.push(ctrl.name);
+                }
+                else {
+                    failed.push(ctrl.name);
+                }
+            }
+        }
+        if (verbose && activated.length > 0) {
+            console.log(`ControllerRegistry: ${activated.length} controllers activated`);
+        }
+    }
+    catch {
+        // ControllerRegistry activation is best-effort
+    }
+    return { activated, failed, initTimeMs: performance.now() - startTime };
+}
+/**
+ * Initialize the memory database properly using sql.js
+ */
+export async function initializeMemoryDatabase(options) {
+    const { backend = 'hybrid', dbPath: customPath, force = false, verbose = false, migrate = true } = options;
+    const swarmDir = path.join(getProjectRoot(), '.swarm');
+    const dbPath = customPath || path.join(swarmDir, 'memory.db');
+    const dbDir = path.dirname(dbPath);
+    try {
+        // Create directory if needed
+        if (!fs.existsSync(dbDir)) {
+            fs.mkdirSync(dbDir, { recursive: true });
+        }
+        // Check for legacy installations
+        if (migrate) {
+            const legacyCheck = await checkAndMigrateLegacy({ dbPath, verbose });
+            if (legacyCheck.needsMigration && verbose) {
+                console.log(`Found legacy database (v${legacyCheck.legacyVersion}) with ${legacyCheck.legacyEntries} entries`);
+            }
+        }
+        // Check existing database
+        if (fs.existsSync(dbPath) && !force) {
+            return {
+                success: false,
+                backend,
+                dbPath,
+                schemaVersion: '3.0.0',
+                tablesCreated: [],
+                indexesCreated: [],
+                features: {
+                    vectorEmbeddings: false,
+                    patternLearning: false,
+                    temporalDecay: false,
+                    hnswIndexing: false,
+                    migrationTracking: false
+                },
+                error: 'Database already exists. Use --force to reinitialize.'
+            };
+        }
+        // Try to use sql.js (WASM SQLite)
+        let db;
+        let usedSqlJs = false;
+        try {
+            // Dynamic import of sql.js
+            const initSqlJs = (await import('sql.js')).default;
+            const SQL = await initSqlJs();
+            // Load existing database or create new
+            if (fs.existsSync(dbPath) && force) {
+                fs.unlinkSync(dbPath);
+            }
+            db = new SQL.Database();
+            usedSqlJs = true;
+        }
+        catch (e) {
+            // sql.js not available, fall back to writing schema file
+            if (verbose) {
+                console.log('sql.js not available, writing schema file for later initialization');
+            }
+        }
+        if (usedSqlJs && db) {
+            // Execute schema
+            db.run(MEMORY_SCHEMA_V3);
+            // Insert initial metadata
+            db.run(getInitialMetadata(backend));
+            // Save to file
+            const data = db.export();
+            const buffer = Buffer.from(data);
+            fs.writeFileSync(dbPath, buffer);
+            // Close database
+            db.close();
+            // Also create schema file for reference
+            const schemaPath = path.join(dbDir, 'schema.sql');
+            fs.writeFileSync(schemaPath, MEMORY_SCHEMA_V3 + '\n' + getInitialMetadata(backend));
+            // ADR-053: Activate ControllerRegistry so controllers (ReasoningBank,
+            // SkillLibrary, ExplainableRecall, etc.) are instantiated during init
+            const controllerResult = await activateControllerRegistry(dbPath, verbose);
+            return {
+                success: true,
+                backend,
+                dbPath,
+                schemaVersion: '3.0.0',
+                tablesCreated: [
+                    'memory_entries',
+                    'patterns',
+                    'pattern_history',
+                    'trajectories',
+                    'trajectory_steps',
+                    'migration_state',
+                    'sessions',
+                    'vector_indexes',
+                    'metadata'
+                ],
+                indexesCreated: [
+                    'idx_memory_namespace',
+                    'idx_memory_key',
+                    'idx_memory_type',
+                    'idx_memory_status',
+                    'idx_memory_created',
+                    'idx_memory_accessed',
+                    'idx_memory_owner',
+                    'idx_patterns_type',
+                    'idx_patterns_confidence',
+                    'idx_patterns_status',
+                    'idx_patterns_last_matched',
+                    'idx_pattern_history_pattern',
+                    'idx_steps_trajectory'
+                ],
+                features: {
+                    vectorEmbeddings: true,
+                    patternLearning: true,
+                    temporalDecay: true,
+                    hnswIndexing: true,
+                    migrationTracking: true
+                },
+                controllers: controllerResult,
+            };
+        }
+        else {
+            // Fall back to schema file approach
+            const schemaPath = path.join(dbDir, 'schema.sql');
+            fs.writeFileSync(schemaPath, MEMORY_SCHEMA_V3 + '\n' + getInitialMetadata(backend));
+            // Create minimal valid SQLite file
+            const sqliteHeader = Buffer.alloc(4096, 0);
+            // SQLite format 3 header
+            Buffer.from('SQLite format 3\0').copy(sqliteHeader, 0);
+            sqliteHeader[16] = 0x10; // page size high byte (4096)
+            sqliteHeader[17] = 0x00; // page size low byte
+            sqliteHeader[18] = 0x01; // file format write version
+            sqliteHeader[19] = 0x01; // file format read version
+            sqliteHeader[24] = 0x00; // max embedded payload
+            sqliteHeader[25] = 0x40;
+            sqliteHeader[26] = 0x20; // min embedded payload
+            sqliteHeader[27] = 0x20; // leaf payload
+            fs.writeFileSync(dbPath, sqliteHeader);
+            // ADR-053: Activate ControllerRegistry even on fallback path
+            const controllerResult = await activateControllerRegistry(dbPath, verbose);
+            return {
+                success: true,
+                backend,
+                dbPath,
+                schemaVersion: '3.0.0',
+                tablesCreated: [
+                    'memory_entries (pending)',
+                    'patterns (pending)',
+                    'pattern_history (pending)',
+                    'trajectories (pending)',
+                    'trajectory_steps (pending)',
+                    'migration_state (pending)',
+                    'sessions (pending)',
+                    'vector_indexes (pending)',
+                    'metadata (pending)'
+                ],
+                indexesCreated: [],
+                features: {
+                    vectorEmbeddings: true,
+                    patternLearning: true,
+                    temporalDecay: true,
+                    hnswIndexing: true,
+                    migrationTracking: true
+                },
+                controllers: controllerResult,
+            };
+        }
+    }
+    catch (error) {
+        return {
+            success: false,
+            backend,
+            dbPath,
+            schemaVersion: '3.0.0',
+            tablesCreated: [],
+            indexesCreated: [],
+            features: {
+                vectorEmbeddings: false,
+                patternLearning: false,
+                temporalDecay: false,
+                hnswIndexing: false,
+                migrationTracking: false
+            },
+            error: error instanceof Error ? error.message : String(error)
+        };
+    }
+}
+/**
+ * Check if memory database is properly initialized
+ */
+export async function checkMemoryInitialization(dbPath) {
+    const swarmDir = path.join(getProjectRoot(), '.swarm');
+    const path_ = dbPath || path.join(swarmDir, 'memory.db');
+    if (!fs.existsSync(path_)) {
+        return { initialized: false };
+    }
+    try {
+        // Try to load with sql.js
+        const initSqlJs = (await import('sql.js')).default;
+        const SQL = await initSqlJs();
+        const fileBuffer = fs.readFileSync(path_);
+        const db = new SQL.Database(fileBuffer);
+        // Check for metadata table
+        const tables = db.exec("SELECT name FROM sqlite_master WHERE type='table'");
+        const tableNames = tables[0]?.values?.map(v => v[0]) || [];
+        // Get version
+        let version = 'unknown';
+        let backend = 'unknown';
+        try {
+            const versionResult = db.exec("SELECT value FROM metadata WHERE key='schema_version'");
+            version = versionResult[0]?.values[0]?.[0] || 'unknown';
+            const backendResult = db.exec("SELECT value FROM metadata WHERE key='backend'");
+            backend = backendResult[0]?.values[0]?.[0] || 'unknown';
+        }
+        catch {
+            // Metadata table might not exist
+        }
+        db.close();
+        return {
+            initialized: true,
+            version,
+            backend,
+            features: {
+                vectorEmbeddings: tableNames.includes('vector_indexes'),
+                patternLearning: tableNames.includes('patterns'),
+                temporalDecay: tableNames.includes('pattern_history')
+            },
+            tables: tableNames
+        };
+    }
+    catch {
+        // Could not read database
+        return { initialized: false };
+    }
+}
+/**
+ * Apply temporal decay to patterns
+ * Reduces confidence of patterns that haven't been used recently
+ */
+export async function applyTemporalDecay(dbPath) {
+    const swarmDir = path.join(getProjectRoot(), '.swarm');
+    const path_ = dbPath || path.join(swarmDir, 'memory.db');
+    try {
+        const initSqlJs = (await import('sql.js')).default;
+        const SQL = await initSqlJs();
+        const fileBuffer = fs.readFileSync(path_);
+        const db = new SQL.Database(fileBuffer);
+        // Apply decay: confidence *= exp(-decay_rate * days_since_last_use)
+        const now = Date.now();
         const decayQuery = `
       UPDATE patterns
       SET
@@ -1212,779 +1263,779 @@ export async function applyTemporalDecay(dbPath) {
       WHERE status = 'active'
         AND confidence > 0.1
         AND (? - COALESCE(last_matched_at, created_at)) > 86400000
-    `;
-        db.run(decayQuery, [now, now, now]);
-        const changes = db.getRowsModified();
-        // Save
-        const data = db.export();
-        fs.writeFileSync(path_, Buffer.from(data));
-        db.close();
-        return {
-            success: true,
-            patternsDecayed: changes
-        };
-    }
-    catch (error) {
-        return {
-            success: false,
-            patternsDecayed: 0,
-            error: error instanceof Error ? error.message : String(error)
-        };
-    }
-}
-let embeddingModelState = null;
-/**
- * Lazy load ONNX embedding model
- * Only loads when first embedding is requested
- */
-export async function loadEmbeddingModel(options) {
-    const { verbose = false } = options || {};
-    const startTime = Date.now();
-    // Already loaded
-    if (embeddingModelState?.loaded) {
-        return {
-            success: true,
-            dimensions: embeddingModelState.dimensions,
-            modelName: 'cached',
-            loadTime: 0
-        };
-    }
-    // ADR-053: Try AgentDB v3 bridge first
-    const bridge = await getBridge();
-    if (bridge) {
-        const bridgeResult = await bridge.bridgeLoadEmbeddingModel();
-        if (bridgeResult && bridgeResult.success) {
-            // Mark local state as loaded too so subsequent calls use cache
-            embeddingModelState = {
-                loaded: true,
-                model: null, // Bridge handles embedding
-                tokenizer: null,
-                dimensions: bridgeResult.dimensions
-            };
-            return bridgeResult;
-        }
-    }
-    try {
-        // Try to import @xenova/transformers for ONNX embeddings
-        const transformers = await import('@xenova/transformers').catch(() => null);
-        if (transformers) {
-            if (verbose) {
-                console.log('Loading ONNX embedding model (all-MiniLM-L6-v2)...');
-            }
-            // Use small, fast model for local embeddings
-            const { pipeline } = transformers;
-            const embedder = await pipeline('feature-extraction', 'Xenova/all-MiniLM-L6-v2');
-            embeddingModelState = {
-                loaded: true,
-                model: embedder,
-                tokenizer: null,
-                dimensions: 384 // MiniLM-L6 produces 384-dim vectors
-            };
-            return {
-                success: true,
-                dimensions: 384,
-                modelName: 'all-MiniLM-L6-v2',
-                loadTime: Date.now() - startTime
-            };
-        }
-        // Fallback: Check for agentic-flow ReasoningBank embeddings (v3)
-        const reasoningBank = await import('agentic-flow/reasoningbank').catch(() => null);
-        if (reasoningBank?.computeEmbedding) {
-            if (verbose) {
-                console.log('Loading agentic-flow ReasoningBank embedding model...');
-            }
-            embeddingModelState = {
-                loaded: true,
-                model: { embed: reasoningBank.computeEmbedding },
-                tokenizer: null,
-                dimensions: 768
-            };
-            return {
-                success: true,
-                dimensions: 768,
-                modelName: 'agentic-flow/reasoningbank',
-                loadTime: Date.now() - startTime
-            };
-        }
-        // Legacy fallback: Check for agentic-flow core embeddings
-        const agenticFlow = await import('agentic-flow').catch(() => null);
-        if (agenticFlow && agenticFlow.embeddings) {
-            if (verbose) {
-                console.log('Loading agentic-flow embedding model...');
-            }
-            embeddingModelState = {
-                loaded: true,
-                model: agenticFlow.embeddings,
-                tokenizer: null,
-                dimensions: 768
-            };
-            return {
-                success: true,
-                dimensions: 768,
-                modelName: 'agentic-flow',
-                loadTime: Date.now() - startTime
-            };
-        }
-        // No ONNX model available - use fallback
-        embeddingModelState = {
-            loaded: true,
-            model: null, // Will use domain-aware hash fallback
-            tokenizer: null,
-            dimensions: 384 // Domain-aware hash embedding dimensions
-        };
-        return {
-            success: true,
-            dimensions: 384,
-            modelName: 'domain-aware-hash-384',
-            loadTime: Date.now() - startTime
-        };
-    }
-    catch (error) {
-        return {
-            success: false,
-            dimensions: 0,
-            modelName: 'none',
-            error: error instanceof Error ? error.message : String(error)
-        };
-    }
-}
-/**
- * Generate real embedding for text
- * Uses ONNX model if available, falls back to deterministic hash
- */
-export async function generateEmbedding(text) {
-    // ADR-053: Try AgentDB v3 bridge first
-    const bridge = await getBridge();
-    if (bridge) {
-        const bridgeResult = await bridge.bridgeGenerateEmbedding(text);
-        if (bridgeResult)
-            return bridgeResult;
-    }
-    // Ensure model is loaded
-    if (!embeddingModelState?.loaded) {
-        await loadEmbeddingModel();
-    }
-    const state = embeddingModelState;
-    // Use ONNX model if available
-    if (state.model && typeof state.model === 'function') {
-        try {
-            const output = await state.model(text, { pooling: 'mean', normalize: true });
-            const embedding = Array.from(output.data);
-            return {
-                embedding,
-                dimensions: embedding.length,
-                model: 'onnx'
-            };
-        }
-        catch {
-            // Fall through to fallback
-        }
-    }
-    // Domain-aware hash fallback (for testing/demo without ONNX)
-    const embedding = generateDomainAwareEmbedding(text);
-    return {
-        embedding,
-        dimensions: 384,
-        model: 'domain-aware-hash-384'
-    };
-}
-/**
- * Generate embeddings for multiple texts
- * Uses parallel execution for API-based providers (2-4x faster)
- * Note: Local ONNX inference is CPU-bound, so parallelism has limited benefit
- *
- * @param texts - Array of texts to embed
- * @param options - Batch options
- * @returns Array of embedding results with timing info
- */
-export async function generateBatchEmbeddings(texts, options) {
-    const { concurrency = texts.length, onProgress } = options || {};
-    const startTime = Date.now();
-    // Ensure model is loaded first (prevents cold start in parallel)
-    if (!embeddingModelState?.loaded) {
-        await loadEmbeddingModel();
-    }
-    // Process in parallel with optional concurrency limit
-    if (concurrency >= texts.length) {
-        // Full parallelism
-        const embeddings = await Promise.all(texts.map(async (text, i) => {
-            const result = await generateEmbedding(text);
-            onProgress?.(i + 1, texts.length);
-            return { text, ...result };
-        }));
-        const totalTime = Date.now() - startTime;
-        return {
-            results: embeddings,
-            totalTime,
-            avgTime: totalTime / texts.length
-        };
-    }
-    // Limited concurrency using chunking
-    const results = [];
-    let completed = 0;
-    for (let i = 0; i < texts.length; i += concurrency) {
-        const chunk = texts.slice(i, i + concurrency);
-        const chunkResults = await Promise.all(chunk.map(async (text) => {
-            const result = await generateEmbedding(text);
-            completed++;
-            onProgress?.(completed, texts.length);
-            return { text, ...result };
-        }));
-        results.push(...chunkResults);
-    }
-    const totalTime = Date.now() - startTime;
-    return {
-        results,
-        totalTime,
-        avgTime: totalTime / texts.length
-    };
-}
-/**
- * Domain-aware semantic hash embeddings (384-dim)
- * Provides consistent embedding dimensions between CLI and MCP tools.
- * Uses domain cluster awareness for better semantic similarity.
- */
-const DOMAIN_CLUSTERS = {
-    database: ['typeorm', 'mongodb', 'database', 'entity', 'schema', 'table', 'collection',
-        'query', 'sql', 'nosql', 'orm', 'model', 'migration', 'repository', 'column',
-        'relation', 'foreign', 'primary', 'index', 'constraint', 'transaction'],
-    frontend: ['react', 'component', 'ui', 'styling', 'css', 'html', 'jsx', 'tsx', 'frontend',
-        'material', 'mui', 'tailwind', 'dom', 'render', 'hook', 'state', 'props',
-        'redux', 'context', 'styled', 'emotion', 'theme', 'layout', 'responsive'],
-    backend: ['fastify', 'api', 'route', 'handler', 'rest', 'endpoint', 'server', 'controller',
-        'middleware', 'request', 'response', 'http', 'express', 'nest', 'graphql',
-        'websocket', 'socket', 'cors', 'auth', 'jwt', 'session', 'cookie'],
-    testing: ['test', 'testing', 'vitest', 'jest', 'mock', 'spy', 'assert', 'expect', 'describe',
-        'it', 'spec', 'unit', 'integration', 'e2e', 'playwright', 'cypress', 'coverage',
-        'fixture', 'stub', 'fake', 'snapshot', 'beforeeach', 'aftereach'],
-    tenancy: ['tenant', 'tenancy', 'companyid', 'company', 'isolation', 'multi', 'multitenant',
-        'organization', 'workspace', 'account', 'customer', 'client'],
-    security: ['security', 'auth', 'authentication', 'authorization', 'permission', 'role',
-        'access', 'token', 'jwt', 'oauth', 'password', 'encrypt', 'hash', 'salt',
-        'csrf', 'xss', 'injection', 'sanitize', 'validate'],
-    patterns: ['pattern', 'service', 'factory', 'singleton', 'decorator', 'adapter', 'facade',
-        'observer', 'strategy', 'command', 'repository', 'usecase', 'domain', 'ddd',
-        'clean', 'architecture', 'solid', 'dry', 'kiss'],
-    workflow: ['workflow', 'pipeline', 'ci', 'cd', 'deploy', 'build', 'actions',
-        'hook', 'trigger', 'job', 'step', 'artifact', 'release', 'version', 'tag'],
-    memory: ['memory', 'cache', 'store', 'persist', 'storage', 'redis', 'session', 'state',
-        'buffer', 'queue', 'stack', 'heap', 'gc', 'leak', 'embedding', 'vector', 'hnsw',
-        'semantic', 'search', 'index', 'retrieval'],
-    agent: ['agent', 'swarm', 'coordinator', 'orchestrator', 'task', 'worker', 'spawn',
-        'parallel', 'concurrent', 'async', 'promise', 'queue', 'priority', 'schedule'],
-    github: ['github', 'issue', 'branch', 'pr', 'pull', 'request', 'merge', 'commit', 'push',
-        'clone', 'fork', 'remote', 'origin', 'main', 'master', 'checkout', 'rebase',
-        'squash', 'repository', 'repo', 'gh', 'git', 'assignee', 'label'],
-    documentation: ['guidance', 'documentation', 'docs', 'readme', 'guide', 'tutorial',
-        'reference', 'standard', 'convention', 'rule', 'policy', 'template',
-        'example', 'usage', 'instruction', 'markdown']
-};
-const COMMON_WORDS = new Set([
-    'the', 'a', 'an', 'is', 'are', 'was', 'were', 'be', 'been', 'being', 'have', 'has', 'had',
-    'do', 'does', 'did', 'will', 'would', 'could', 'should', 'may', 'might', 'must', 'shall',
-    'can', 'need', 'to', 'of', 'in', 'for', 'on', 'with', 'at', 'by', 'from', 'as', 'into',
-    'through', 'during', 'before', 'after', 'above', 'below', 'between', 'under', 'and', 'but',
-    'or', 'nor', 'so', 'yet', 'both', 'either', 'neither', 'not', 'only', 'own', 'same', 'than',
-    'too', 'very', 'just', 'also', 'this', 'that', 'these', 'those', 'it', 'its', 'if', 'then',
-    'else', 'when', 'where', 'why', 'how', 'all', 'each', 'every', 'any', 'some', 'no', 'yes',
-    'use', 'using', 'used', 'uses', 'get', 'set', 'new', 'see', 'like', 'make', 'made'
-]);
-function hashWord(str, seed = 0) {
-    let h = seed ^ str.length;
-    for (let i = 0; i < str.length; i++) {
-        h ^= str.charCodeAt(i);
-        h = Math.imul(h, 0x5bd1e995);
-        h ^= h >>> 15;
-    }
-    return h >>> 0;
-}
-// Pre-compute domain signatures
-const domainSignatures = {};
-for (const [domain, keywords] of Object.entries(DOMAIN_CLUSTERS)) {
-    const sig = new Float32Array(384);
-    for (const kw of keywords) {
-        for (let h = 0; h < 2; h++) {
-            const idx = hashWord(kw + '_dom_' + domain, h) % 384;
-            sig[idx] = 1;
-        }
-    }
-    domainSignatures[domain] = sig;
-}
-function generateDomainAwareEmbedding(text) {
-    const dims = 384;
-    const vec = new Float32Array(dims);
-    const lowerText = text.toLowerCase();
-    const words = lowerText.replace(/[^a-z0-9\s]/g, ' ').split(/\s+/).filter(w => w.length > 1);
-    if (words.length === 0)
-        return Array.from(vec);
-    // Domain cluster weights
-    for (const [domain, keywords] of Object.entries(DOMAIN_CLUSTERS)) {
-        let matchCount = 0;
-        for (const kw of keywords) {
-            if (lowerText.includes(kw))
-                matchCount++;
-        }
-        if (matchCount > 0) {
-            const weight = Math.min(2.0, 0.5 + matchCount * 0.3);
-            const sig = domainSignatures[domain];
-            for (let i = 0; i < dims; i++)
-                vec[i] += sig[i] * weight;
-        }
-    }
-    // Word hashes
-    for (const word of words) {
-        const isCommon = COMMON_WORDS.has(word);
-        const weight = isCommon ? 0.2 : (word.length > 6 ? 0.8 : 0.5);
-        for (let h = 0; h < 3; h++) {
-            const idx = hashWord(word, h * 17) % dims;
-            const sign = (hashWord(word, h * 31 + 1) % 2 === 0) ? 1 : -1;
-            vec[idx] += sign * weight;
-        }
-    }
-    // Bigrams
-    for (let i = 0; i < words.length - 1; i++) {
-        if (COMMON_WORDS.has(words[i]) && COMMON_WORDS.has(words[i + 1]))
-            continue;
-        const bigram = words[i] + '_' + words[i + 1];
-        const idx = hashWord(bigram, 42) % dims;
-        const sign = (hashWord(bigram, 43) % 2 === 0) ? 1 : -1;
-        vec[idx] += sign * 0.4;
-    }
-    // Trigrams
-    for (let i = 0; i < words.length - 2; i++) {
-        const trigram = words[i] + '_' + words[i + 1] + '_' + words[i + 2];
-        const idx = hashWord(trigram, 99) % dims;
-        const sign = (hashWord(trigram, 100) % 2 === 0) ? 1 : -1;
-        vec[idx] += sign * 0.3;
-    }
-    // Normalize
-    let norm = 0;
-    for (let i = 0; i < dims; i++)
-        norm += vec[i] * vec[i];
-    norm = Math.sqrt(norm);
-    if (norm > 0)
-        for (let i = 0; i < dims; i++)
-            vec[i] /= norm;
-    return Array.from(vec);
-}
-/**
- * Verify memory initialization works correctly
- * Tests: write, read, search, patterns
- */
-export async function verifyMemoryInit(dbPath, options) {
-    const { verbose = false } = options || {};
-    const tests = [];
-    try {
-        const initSqlJs = (await import('sql.js')).default;
-        const SQL = await initSqlJs();
-        const fs = await import('fs');
-        // Load database
-        const fileBuffer = fs.readFileSync(dbPath);
-        const db = new SQL.Database(fileBuffer);
-        // Test 1: Schema verification
-        const schemaStart = Date.now();
-        const tables = db.exec("SELECT name FROM sqlite_master WHERE type='table'");
-        const tableNames = tables[0]?.values?.map(v => v[0]) || [];
-        const expectedTables = ['memory_entries', 'patterns', 'metadata', 'vector_indexes'];
-        const missingTables = expectedTables.filter(t => !tableNames.includes(t));
-        tests.push({
-            name: 'Schema verification',
-            passed: missingTables.length === 0,
-            details: missingTables.length > 0 ? `Missing: ${missingTables.join(', ')}` : `${tableNames.length} tables found`,
-            duration: Date.now() - schemaStart
-        });
-        // Test 2: Write entry
-        const writeStart = Date.now();
-        const testId = `test_${Date.now()}`;
-        const testKey = 'verification_test';
-        const testValue = 'This is a verification test entry for memory initialization';
-        try {
+    `;
+        db.run(decayQuery, [now, now, now]);
+        const changes = db.getRowsModified();
+        // Save
+        const data = db.export();
+        fs.writeFileSync(path_, Buffer.from(data));
+        db.close();
+        return {
+            success: true,
+            patternsDecayed: changes
+        };
+    }
+    catch (error) {
+        return {
+            success: false,
+            patternsDecayed: 0,
+            error: error instanceof Error ? error.message : String(error)
+        };
+    }
+}
+let embeddingModelState = null;
+/**
+ * Lazy load ONNX embedding model
+ * Only loads when first embedding is requested
+ */
+export async function loadEmbeddingModel(options) {
+    const { verbose = false } = options || {};
+    const startTime = Date.now();
+    // Already loaded
+    if (embeddingModelState?.loaded) {
+        return {
+            success: true,
+            dimensions: embeddingModelState.dimensions,
+            modelName: 'cached',
+            loadTime: 0
+        };
+    }
+    // ADR-053: Try AgentDB v3 bridge first
+    const bridge = await getBridge();
+    if (bridge) {
+        const bridgeResult = await bridge.bridgeLoadEmbeddingModel();
+        if (bridgeResult && bridgeResult.success) {
+            // Mark local state as loaded too so subsequent calls use cache
+            embeddingModelState = {
+                loaded: true,
+                model: null, // Bridge handles embedding
+                tokenizer: null,
+                dimensions: bridgeResult.dimensions
+            };
+            return bridgeResult;
+        }
+    }
+    try {
+        // Try to import @xenova/transformers for ONNX embeddings
+        const transformers = await import('@xenova/transformers').catch(() => null);
+        if (transformers) {
+            if (verbose) {
+                console.log('Loading ONNX embedding model (all-MiniLM-L6-v2)...');
+            }
+            // Use small, fast model for local embeddings
+            const { pipeline } = transformers;
+            const embedder = await pipeline('feature-extraction', 'Xenova/all-MiniLM-L6-v2');
+            embeddingModelState = {
+                loaded: true,
+                model: embedder,
+                tokenizer: null,
+                dimensions: 384 // MiniLM-L6 produces 384-dim vectors
+            };
+            return {
+                success: true,
+                dimensions: 384,
+                modelName: 'all-MiniLM-L6-v2',
+                loadTime: Date.now() - startTime
+            };
+        }
+        // Fallback: Check for agentic-flow ReasoningBank embeddings (v3)
+        const reasoningBank = await import('agentic-flow/reasoningbank').catch(() => null);
+        if (reasoningBank?.computeEmbedding) {
+            if (verbose) {
+                console.log('Loading agentic-flow ReasoningBank embedding model...');
+            }
+            embeddingModelState = {
+                loaded: true,
+                model: { embed: reasoningBank.computeEmbedding },
+                tokenizer: null,
+                dimensions: 768
+            };
+            return {
+                success: true,
+                dimensions: 768,
+                modelName: 'agentic-flow/reasoningbank',
+                loadTime: Date.now() - startTime
+            };
+        }
+        // Legacy fallback: Check for agentic-flow core embeddings
+        const agenticFlow = await import('agentic-flow').catch(() => null);
+        if (agenticFlow && agenticFlow.embeddings) {
+            if (verbose) {
+                console.log('Loading agentic-flow embedding model...');
+            }
+            embeddingModelState = {
+                loaded: true,
+                model: agenticFlow.embeddings,
+                tokenizer: null,
+                dimensions: 768
+            };
+            return {
+                success: true,
+                dimensions: 768,
+                modelName: 'agentic-flow',
+                loadTime: Date.now() - startTime
+            };
+        }
+        // No ONNX model available - use fallback
+        embeddingModelState = {
+            loaded: true,
+            model: null, // Will use domain-aware hash fallback
+            tokenizer: null,
+            dimensions: 384 // Domain-aware hash embedding dimensions
+        };
+        return {
+            success: true,
+            dimensions: 384,
+            modelName: 'domain-aware-hash-384',
+            loadTime: Date.now() - startTime
+        };
+    }
+    catch (error) {
+        return {
+            success: false,
+            dimensions: 0,
+            modelName: 'none',
+            error: error instanceof Error ? error.message : String(error)
+        };
+    }
+}
+/**
+ * Generate real embedding for text
+ * Uses ONNX model if available, falls back to deterministic hash
+ */
+export async function generateEmbedding(text) {
+    // ADR-053: Try AgentDB v3 bridge first
+    const bridge = await getBridge();
+    if (bridge) {
+        const bridgeResult = await bridge.bridgeGenerateEmbedding(text);
+        if (bridgeResult)
+            return bridgeResult;
+    }
+    // Ensure model is loaded
+    if (!embeddingModelState?.loaded) {
+        await loadEmbeddingModel();
+    }
+    const state = embeddingModelState;
+    // Use ONNX model if available
+    if (state.model && typeof state.model === 'function') {
+        try {
+            const output = await state.model(text, { pooling: 'mean', normalize: true });
+            const embedding = Array.from(output.data);
+            return {
+                embedding,
+                dimensions: embedding.length,
+                model: 'onnx'
+            };
+        }
+        catch {
+            // Fall through to fallback
+        }
+    }
+    // Domain-aware hash fallback (for testing/demo without ONNX)
+    const embedding = generateDomainAwareEmbedding(text);
+    return {
+        embedding,
+        dimensions: 384,
+        model: 'domain-aware-hash-384'
+    };
+}
+/**
+ * Generate embeddings for multiple texts
+ * Uses parallel execution for API-based providers (2-4x faster)
+ * Note: Local ONNX inference is CPU-bound, so parallelism has limited benefit
+ *
+ * @param texts - Array of texts to embed
+ * @param options - Batch options
+ * @returns Array of embedding results with timing info
+ */
+export async function generateBatchEmbeddings(texts, options) {
+    const { concurrency = texts.length, onProgress } = options || {};
+    const startTime = Date.now();
+    // Ensure model is loaded first (prevents cold start in parallel)
+    if (!embeddingModelState?.loaded) {
+        await loadEmbeddingModel();
+    }
+    // Process in parallel with optional concurrency limit
+    if (concurrency >= texts.length) {
+        // Full parallelism
+        const embeddings = await Promise.all(texts.map(async (text, i) => {
+            const result = await generateEmbedding(text);
+            onProgress?.(i + 1, texts.length);
+            return { text, ...result };
+        }));
+        const totalTime = Date.now() - startTime;
+        return {
+            results: embeddings,
+            totalTime,
+            avgTime: totalTime / texts.length
+        };
+    }
+    // Limited concurrency using chunking
+    const results = [];
+    let completed = 0;
+    for (let i = 0; i < texts.length; i += concurrency) {
+        const chunk = texts.slice(i, i + concurrency);
+        const chunkResults = await Promise.all(chunk.map(async (text) => {
+            const result = await generateEmbedding(text);
+            completed++;
+            onProgress?.(completed, texts.length);
+            return { text, ...result };
+        }));
+        results.push(...chunkResults);
+    }
+    const totalTime = Date.now() - startTime;
+    return {
+        results,
+        totalTime,
+        avgTime: totalTime / texts.length
+    };
+}
+/**
+ * Domain-aware semantic hash embeddings (384-dim)
+ * Provides consistent embedding dimensions between CLI and MCP tools.
+ * Uses domain cluster awareness for better semantic similarity.
+ */
+const DOMAIN_CLUSTERS = {
+    database: ['typeorm', 'mongodb', 'database', 'entity', 'schema', 'table', 'collection',
+        'query', 'sql', 'nosql', 'orm', 'model', 'migration', 'repository', 'column',
+        'relation', 'foreign', 'primary', 'index', 'constraint', 'transaction'],
+    frontend: ['react', 'component', 'ui', 'styling', 'css', 'html', 'jsx', 'tsx', 'frontend',
+        'material', 'mui', 'tailwind', 'dom', 'render', 'hook', 'state', 'props',
+        'redux', 'context', 'styled', 'emotion', 'theme', 'layout', 'responsive'],
+    backend: ['fastify', 'api', 'route', 'handler', 'rest', 'endpoint', 'server', 'controller',
+        'middleware', 'request', 'response', 'http', 'express', 'nest', 'graphql',
+        'websocket', 'socket', 'cors', 'auth', 'jwt', 'session', 'cookie'],
+    testing: ['test', 'testing', 'vitest', 'jest', 'mock', 'spy', 'assert', 'expect', 'describe',
+        'it', 'spec', 'unit', 'integration', 'e2e', 'playwright', 'cypress', 'coverage',
+        'fixture', 'stub', 'fake', 'snapshot', 'beforeeach', 'aftereach'],
+    tenancy: ['tenant', 'tenancy', 'companyid', 'company', 'isolation', 'multi', 'multitenant',
+        'organization', 'workspace', 'account', 'customer', 'client'],
+    security: ['security', 'auth', 'authentication', 'authorization', 'permission', 'role',
+        'access', 'token', 'jwt', 'oauth', 'password', 'encrypt', 'hash', 'salt',
+        'csrf', 'xss', 'injection', 'sanitize', 'validate'],
+    patterns: ['pattern', 'service', 'factory', 'singleton', 'decorator', 'adapter', 'facade',
+        'observer', 'strategy', 'command', 'repository', 'usecase', 'domain', 'ddd',
+        'clean', 'architecture', 'solid', 'dry', 'kiss'],
+    workflow: ['workflow', 'pipeline', 'ci', 'cd', 'deploy', 'build', 'actions',
+        'hook', 'trigger', 'job', 'step', 'artifact', 'release', 'version', 'tag'],
+    memory: ['memory', 'cache', 'store', 'persist', 'storage', 'redis', 'session', 'state',
+        'buffer', 'queue', 'stack', 'heap', 'gc', 'leak', 'embedding', 'vector', 'hnsw',
+        'semantic', 'search', 'index', 'retrieval'],
+    agent: ['agent', 'swarm', 'coordinator', 'orchestrator', 'task', 'worker', 'spawn',
+        'parallel', 'concurrent', 'async', 'promise', 'queue', 'priority', 'schedule'],
+    github: ['github', 'issue', 'branch', 'pr', 'pull', 'request', 'merge', 'commit', 'push',
+        'clone', 'fork', 'remote', 'origin', 'main', 'master', 'checkout', 'rebase',
+        'squash', 'repository', 'repo', 'gh', 'git', 'assignee', 'label'],
+    documentation: ['guidance', 'documentation', 'docs', 'readme', 'guide', 'tutorial',
+        'reference', 'standard', 'convention', 'rule', 'policy', 'template',
+        'example', 'usage', 'instruction', 'markdown']
+};
+const COMMON_WORDS = new Set([
+    'the', 'a', 'an', 'is', 'are', 'was', 'were', 'be', 'been', 'being', 'have', 'has', 'had',
+    'do', 'does', 'did', 'will', 'would', 'could', 'should', 'may', 'might', 'must', 'shall',
+    'can', 'need', 'to', 'of', 'in', 'for', 'on', 'with', 'at', 'by', 'from', 'as', 'into',
+    'through', 'during', 'before', 'after', 'above', 'below', 'between', 'under', 'and', 'but',
+    'or', 'nor', 'so', 'yet', 'both', 'either', 'neither', 'not', 'only', 'own', 'same', 'than',
+    'too', 'very', 'just', 'also', 'this', 'that', 'these', 'those', 'it', 'its', 'if', 'then',
+    'else', 'when', 'where', 'why', 'how', 'all', 'each', 'every', 'any', 'some', 'no', 'yes',
+    'use', 'using', 'used', 'uses', 'get', 'set', 'new', 'see', 'like', 'make', 'made'
+]);
+function hashWord(str, seed = 0) {
+    let h = seed ^ str.length;
+    for (let i = 0; i < str.length; i++) {
+        h ^= str.charCodeAt(i);
+        h = Math.imul(h, 0x5bd1e995);
+        h ^= h >>> 15;
+    }
+    return h >>> 0;
+}
+// Pre-compute domain signatures
+const domainSignatures = {};
+for (const [domain, keywords] of Object.entries(DOMAIN_CLUSTERS)) {
+    const sig = new Float32Array(384);
+    for (const kw of keywords) {
+        for (let h = 0; h < 2; h++) {
+            const idx = hashWord(kw + '_dom_' + domain, h) % 384;
+            sig[idx] = 1;
+        }
+    }
+    domainSignatures[domain] = sig;
+}
+function generateDomainAwareEmbedding(text) {
+    const dims = 384;
+    const vec = new Float32Array(dims);
+    const lowerText = text.toLowerCase();
+    const words = lowerText.replace(/[^a-z0-9\s]/g, ' ').split(/\s+/).filter(w => w.length > 1);
+    if (words.length === 0)
+        return Array.from(vec);
+    // Domain cluster weights
+    for (const [domain, keywords] of Object.entries(DOMAIN_CLUSTERS)) {
+        let matchCount = 0;
+        for (const kw of keywords) {
+            if (lowerText.includes(kw))
+                matchCount++;
+        }
+        if (matchCount > 0) {
+            const weight = Math.min(2.0, 0.5 + matchCount * 0.3);
+            const sig = domainSignatures[domain];
+            for (let i = 0; i < dims; i++)
+                vec[i] += sig[i] * weight;
+        }
+    }
+    // Word hashes
+    for (const word of words) {
+        const isCommon = COMMON_WORDS.has(word);
+        const weight = isCommon ? 0.2 : (word.length > 6 ? 0.8 : 0.5);
+        for (let h = 0; h < 3; h++) {
+            const idx = hashWord(word, h * 17) % dims;
+            const sign = (hashWord(word, h * 31 + 1) % 2 === 0) ? 1 : -1;
+            vec[idx] += sign * weight;
+        }
+    }
+    // Bigrams
+    for (let i = 0; i < words.length - 1; i++) {
+        if (COMMON_WORDS.has(words[i]) && COMMON_WORDS.has(words[i + 1]))
+            continue;
+        const bigram = words[i] + '_' + words[i + 1];
+        const idx = hashWord(bigram, 42) % dims;
+        const sign = (hashWord(bigram, 43) % 2 === 0) ? 1 : -1;
+        vec[idx] += sign * 0.4;
+    }
+    // Trigrams
+    for (let i = 0; i < words.length - 2; i++) {
+        const trigram = words[i] + '_' + words[i + 1] + '_' + words[i + 2];
+        const idx = hashWord(trigram, 99) % dims;
+        const sign = (hashWord(trigram, 100) % 2 === 0) ? 1 : -1;
+        vec[idx] += sign * 0.3;
+    }
+    // Normalize
+    let norm = 0;
+    for (let i = 0; i < dims; i++)
+        norm += vec[i] * vec[i];
+    norm = Math.sqrt(norm);
+    if (norm > 0)
+        for (let i = 0; i < dims; i++)
+            vec[i] /= norm;
+    return Array.from(vec);
+}
+/**
+ * Verify memory initialization works correctly
+ * Tests: write, read, search, patterns
+ */
+export async function verifyMemoryInit(dbPath, options) {
+    const { verbose = false } = options || {};
+    const tests = [];
+    try {
+        const initSqlJs = (await import('sql.js')).default;
+        const SQL = await initSqlJs();
+        const fs = await import('fs');
+        // Load database
+        const fileBuffer = fs.readFileSync(dbPath);
+        const db = new SQL.Database(fileBuffer);
+        // Test 1: Schema verification
+        const schemaStart = Date.now();
+        const tables = db.exec("SELECT name FROM sqlite_master WHERE type='table'");
+        const tableNames = tables[0]?.values?.map(v => v[0]) || [];
+        const expectedTables = ['memory_entries', 'patterns', 'metadata', 'vector_indexes'];
+        const missingTables = expectedTables.filter(t => !tableNames.includes(t));
+        tests.push({
+            name: 'Schema verification',
+            passed: missingTables.length === 0,
+            details: missingTables.length > 0 ? `Missing: ${missingTables.join(', ')}` : `${tableNames.length} tables found`,
+            duration: Date.now() - schemaStart
+        });
+        // Test 2: Write entry
+        const writeStart = Date.now();
+        const testId = `test_${Date.now()}`;
+        const testKey = 'verification_test';
+        const testValue = 'This is a verification test entry for memory initialization';
+        try {
             db.run(`
         INSERT INTO memory_entries (id, key, namespace, content, type, created_at, updated_at)
         VALUES (?, ?, 'test', ?, 'semantic', ?, ?)
-      `, [testId, testKey, testValue, Date.now(), Date.now()]);
-            tests.push({
-                name: 'Write entry',
-                passed: true,
-                details: 'Entry written successfully',
-                duration: Date.now() - writeStart
-            });
-        }
-        catch (e) {
-            tests.push({
-                name: 'Write entry',
-                passed: false,
-                details: e instanceof Error ? e.message : 'Write failed',
-                duration: Date.now() - writeStart
-            });
-        }
-        // Test 3: Read entry
-        const readStart = Date.now();
-        try {
-            const result = db.exec(`SELECT content FROM memory_entries WHERE id = ?`, [testId]);
-            const content = result[0]?.values[0]?.[0];
-            tests.push({
-                name: 'Read entry',
-                passed: content === testValue,
-                details: content === testValue ? 'Content matches' : 'Content mismatch',
-                duration: Date.now() - readStart
-            });
-        }
-        catch (e) {
-            tests.push({
-                name: 'Read entry',
-                passed: false,
-                details: e instanceof Error ? e.message : 'Read failed',
-                duration: Date.now() - readStart
-            });
-        }
-        // Test 4: Write with embedding
-        const embeddingStart = Date.now();
-        try {
-            const { embedding, dimensions, model } = await generateEmbedding(testValue);
-            const embeddingJson = JSON.stringify(embedding);
+      `, [testId, testKey, testValue, Date.now(), Date.now()]);
+            tests.push({
+                name: 'Write entry',
+                passed: true,
+                details: 'Entry written successfully',
+                duration: Date.now() - writeStart
+            });
+        }
+        catch (e) {
+            tests.push({
+                name: 'Write entry',
+                passed: false,
+                details: e instanceof Error ? e.message : 'Write failed',
+                duration: Date.now() - writeStart
+            });
+        }
+        // Test 3: Read entry
+        const readStart = Date.now();
+        try {
+            const result = db.exec(`SELECT content FROM memory_entries WHERE id = ?`, [testId]);
+            const content = result[0]?.values[0]?.[0];
+            tests.push({
+                name: 'Read entry',
+                passed: content === testValue,
+                details: content === testValue ? 'Content matches' : 'Content mismatch',
+                duration: Date.now() - readStart
+            });
+        }
+        catch (e) {
+            tests.push({
+                name: 'Read entry',
+                passed: false,
+                details: e instanceof Error ? e.message : 'Read failed',
+                duration: Date.now() - readStart
+            });
+        }
+        // Test 4: Write with embedding
+        const embeddingStart = Date.now();
+        try {
+            const { embedding, dimensions, model } = await generateEmbedding(testValue);
+            const embeddingJson = JSON.stringify(embedding);
             db.run(`
         UPDATE memory_entries
         SET embedding = ?, embedding_dimensions = ?, embedding_model = ?
         WHERE id = ?
-      `, [embeddingJson, dimensions, model, testId]);
-            tests.push({
-                name: 'Generate embedding',
-                passed: true,
-                details: `${dimensions}-dim vector (${model})`,
-                duration: Date.now() - embeddingStart
-            });
-        }
-        catch (e) {
-            tests.push({
-                name: 'Generate embedding',
-                passed: false,
-                details: e instanceof Error ? e.message : 'Embedding failed',
-                duration: Date.now() - embeddingStart
-            });
-        }
-        // Test 5: Pattern storage
-        const patternStart = Date.now();
-        try {
-            const patternId = `pattern_${Date.now()}`;
+      `, [embeddingJson, dimensions, model, testId]);
+            tests.push({
+                name: 'Generate embedding',
+                passed: true,
+                details: `${dimensions}-dim vector (${model})`,
+                duration: Date.now() - embeddingStart
+            });
+        }
+        catch (e) {
+            tests.push({
+                name: 'Generate embedding',
+                passed: false,
+                details: e instanceof Error ? e.message : 'Embedding failed',
+                duration: Date.now() - embeddingStart
+            });
+        }
+        // Test 5: Pattern storage
+        const patternStart = Date.now();
+        try {
+            const patternId = `pattern_${Date.now()}`;
             db.run(`
         INSERT INTO patterns (id, name, pattern_type, condition, action, confidence, created_at, updated_at)
         VALUES (?, 'test-pattern', 'task-routing', 'test condition', 'test action', 0.5, ?, ?)
-      `, [patternId, Date.now(), Date.now()]);
-            tests.push({
-                name: 'Pattern storage',
-                passed: true,
-                details: 'Pattern stored with confidence scoring',
-                duration: Date.now() - patternStart
-            });
-            // Cleanup test pattern
-            db.run(`DELETE FROM patterns WHERE id = ?`, [patternId]);
-        }
-        catch (e) {
-            tests.push({
-                name: 'Pattern storage',
-                passed: false,
-                details: e instanceof Error ? e.message : 'Pattern storage failed',
-                duration: Date.now() - patternStart
-            });
-        }
-        // Test 6: Vector index configuration
-        const indexStart = Date.now();
-        try {
-            const indexResult = db.exec(`SELECT name, dimensions, hnsw_m, hnsw_ef_construction FROM vector_indexes`);
-            const indexes = indexResult[0]?.values || [];
-            tests.push({
-                name: 'Vector index config',
-                passed: indexes.length > 0,
-                details: `${indexes.length} indexes configured (HNSW M=16, ef=200)`,
-                duration: Date.now() - indexStart
-            });
-        }
-        catch (e) {
-            tests.push({
-                name: 'Vector index config',
-                passed: false,
-                details: e instanceof Error ? e.message : 'Index check failed',
-                duration: Date.now() - indexStart
-            });
-        }
-        // Cleanup test entry
-        db.run(`DELETE FROM memory_entries WHERE id = ?`, [testId]);
-        // Save changes
-        const data = db.export();
-        fs.writeFileSync(dbPath, Buffer.from(data));
-        db.close();
-        const passed = tests.filter(t => t.passed).length;
-        const failed = tests.filter(t => !t.passed).length;
-        return {
-            success: failed === 0,
-            tests,
-            summary: {
-                passed,
-                failed,
-                total: tests.length
-            }
-        };
-    }
-    catch (error) {
-        return {
-            success: false,
-            tests: [{
-                    name: 'Database access',
-                    passed: false,
-                    details: error instanceof Error ? error.message : 'Unknown error'
-                }],
-            summary: { passed: 0, failed: 1, total: 1 }
-        };
-    }
-}
-/**
- * Store an entry directly using sql.js
- * This bypasses MCP and writes directly to the database
- */
-export async function storeEntry(options) {
-    // ADR-053: Try AgentDB v3 bridge first
-    const bridge = await getBridge();
-    if (bridge) {
-        const bridgeResult = await bridge.bridgeStoreEntry(options);
-        if (bridgeResult)
-            return bridgeResult;
-    }
-    // Fallback: raw sql.js
-    const { key, value, namespace = 'default', generateEmbeddingFlag = true, tags = [], ttl, dbPath: customPath, upsert = false } = options;
-    const swarmDir = path.join(process.cwd(), '.swarm');
-    const dbPath = customPath || path.join(swarmDir, 'memory.db');
-    try {
-        if (!fs.existsSync(dbPath)) {
-            return { success: false, id: '', error: 'Database not initialized. Run: claude-flow memory init' };
-        }
-        // Ensure schema has all required columns (migration for older DBs)
-        await ensureSchemaColumns(dbPath);
-        const initSqlJs = (await import('sql.js')).default;
-        const SQL = await initSqlJs();
-        const fileBuffer = fs.readFileSync(dbPath);
-        const db = new SQL.Database(fileBuffer);
-        const id = `entry_${Date.now()}_${Math.random().toString(36).substring(7)}`;
-        const now = Date.now();
-        // Generate embedding if requested
-        let embeddingJson = null;
-        let embeddingDimensions = null;
-        let embeddingModel = null;
-        if (generateEmbeddingFlag && value.length > 0) {
-            const embResult = await generateEmbedding(value);
-            embeddingJson = JSON.stringify(embResult.embedding);
-            embeddingDimensions = embResult.dimensions;
-            embeddingModel = embResult.model;
-        }
-        // Insert or update entry (upsert mode uses REPLACE)
-        const insertSql = upsert
+      `, [patternId, Date.now(), Date.now()]);
+            tests.push({
+                name: 'Pattern storage',
+                passed: true,
+                details: 'Pattern stored with confidence scoring',
+                duration: Date.now() - patternStart
+            });
+            // Cleanup test pattern
+            db.run(`DELETE FROM patterns WHERE id = ?`, [patternId]);
+        }
+        catch (e) {
+            tests.push({
+                name: 'Pattern storage',
+                passed: false,
+                details: e instanceof Error ? e.message : 'Pattern storage failed',
+                duration: Date.now() - patternStart
+            });
+        }
+        // Test 6: Vector index configuration
+        const indexStart = Date.now();
+        try {
+            const indexResult = db.exec(`SELECT name, dimensions, hnsw_m, hnsw_ef_construction FROM vector_indexes`);
+            const indexes = indexResult[0]?.values || [];
+            tests.push({
+                name: 'Vector index config',
+                passed: indexes.length > 0,
+                details: `${indexes.length} indexes configured (HNSW M=16, ef=200)`,
+                duration: Date.now() - indexStart
+            });
+        }
+        catch (e) {
+            tests.push({
+                name: 'Vector index config',
+                passed: false,
+                details: e instanceof Error ? e.message : 'Index check failed',
+                duration: Date.now() - indexStart
+            });
+        }
+        // Cleanup test entry
+        db.run(`DELETE FROM memory_entries WHERE id = ?`, [testId]);
+        // Save changes
+        const data = db.export();
+        fs.writeFileSync(dbPath, Buffer.from(data));
+        db.close();
+        const passed = tests.filter(t => t.passed).length;
+        const failed = tests.filter(t => !t.passed).length;
+        return {
+            success: failed === 0,
+            tests,
+            summary: {
+                passed,
+                failed,
+                total: tests.length
+            }
+        };
+    }
+    catch (error) {
+        return {
+            success: false,
+            tests: [{
+                    name: 'Database access',
+                    passed: false,
+                    details: error instanceof Error ? error.message : 'Unknown error'
+                }],
+            summary: { passed: 0, failed: 1, total: 1 }
+        };
+    }
+}
+/**
+ * Store an entry directly using sql.js
+ * This bypasses MCP and writes directly to the database
+ */
+export async function storeEntry(options) {
+    // ADR-053: Try AgentDB v3 bridge first
+    const bridge = await getBridge();
+    if (bridge) {
+        const bridgeResult = await bridge.bridgeStoreEntry(options);
+        if (bridgeResult)
+            return bridgeResult;
+    }
+    // Fallback: raw sql.js
+    const { key, value, namespace = 'default', generateEmbeddingFlag = true, tags = [], ttl, dbPath: customPath, upsert = false } = options;
+    const swarmDir = path.join(getProjectRoot(), '.swarm');
+    const dbPath = customPath || path.join(swarmDir, 'memory.db');
+    try {
+        if (!fs.existsSync(dbPath)) {
+            return { success: false, id: '', error: 'Database not initialized. Run: claude-flow memory init' };
+        }
+        // Ensure schema has all required columns (migration for older DBs)
+        await ensureSchemaColumns(dbPath);
+        const initSqlJs = (await import('sql.js')).default;
+        const SQL = await initSqlJs();
+        const fileBuffer = fs.readFileSync(dbPath);
+        const db = new SQL.Database(fileBuffer);
+        const id = `entry_${Date.now()}_${Math.random().toString(36).substring(7)}`;
+        const now = Date.now();
+        // Generate embedding if requested
+        let embeddingJson = null;
+        let embeddingDimensions = null;
+        let embeddingModel = null;
+        if (generateEmbeddingFlag && value.length > 0) {
+            const embResult = await generateEmbedding(value);
+            embeddingJson = JSON.stringify(embResult.embedding);
+            embeddingDimensions = embResult.dimensions;
+            embeddingModel = embResult.model;
+        }
+        // Insert or update entry (upsert mode uses REPLACE)
+        const insertSql = upsert
             ? `INSERT OR REPLACE INTO memory_entries (
           id, key, namespace, content, type,
           embedding, embedding_dimensions, embedding_model,
           tags, metadata, created_at, updated_at, expires_at, status
-        ) VALUES (?, ?, ?, ?, 'semantic', ?, ?, ?, ?, ?, ?, ?, ?, 'active')`
+        ) VALUES (?, ?, ?, ?, 'semantic', ?, ?, ?, ?, ?, ?, ?, ?, 'active')`
             : `INSERT INTO memory_entries (
           id, key, namespace, content, type,
           embedding, embedding_dimensions, embedding_model,
           tags, metadata, created_at, updated_at, expires_at, status
-        ) VALUES (?, ?, ?, ?, 'semantic', ?, ?, ?, ?, ?, ?, ?, ?, 'active')`;
-        db.run(insertSql, [
-            id,
-            key,
-            namespace,
-            value,
-            embeddingJson,
-            embeddingDimensions,
-            embeddingModel,
-            tags.length > 0 ? JSON.stringify(tags) : null,
-            '{}',
-            now,
-            now,
-            ttl ? now + (ttl * 1000) : null
-        ]);
-        // Save
-        const data = db.export();
-        fs.writeFileSync(dbPath, Buffer.from(data));
-        db.close();
-        // Add to HNSW index for faster future searches
-        if (embeddingJson) {
-            const embResult = JSON.parse(embeddingJson);
-            await addToHNSWIndex(id, embResult, {
-                id,
-                key,
-                namespace,
-                content: value
-            });
-        }
-        return {
-            success: true,
-            id,
-            embedding: embeddingJson ? { dimensions: embeddingDimensions, model: embeddingModel } : undefined
-        };
-    }
-    catch (error) {
-        return {
-            success: false,
-            id: '',
-            error: error instanceof Error ? error.message : String(error)
-        };
-    }
-}
-/**
- * Search entries using sql.js with vector similarity
- * Uses HNSW index for 150x faster search when available
- */
-export async function searchEntries(options) {
-    // ADR-053: Try AgentDB v3 bridge first
-    const bridge = await getBridge();
-    if (bridge) {
-        const bridgeResult = await bridge.bridgeSearchEntries(options);
-        if (bridgeResult)
-            return bridgeResult;
-    }
-    // Fallback: raw sql.js
-    const { query, namespace = 'default', limit = 10, threshold = 0.3, dbPath: customPath } = options;
-    const swarmDir = path.join(process.cwd(), '.swarm');
-    const dbPath = customPath || path.join(swarmDir, 'memory.db');
-    const startTime = Date.now();
-    try {
-        if (!fs.existsSync(dbPath)) {
-            return { success: false, results: [], searchTime: 0, error: 'Database not found' };
-        }
-        // Ensure schema has all required columns (migration for older DBs)
-        await ensureSchemaColumns(dbPath);
-        // Generate query embedding
-        const queryEmb = await generateEmbedding(query);
-        const queryEmbedding = queryEmb.embedding;
-        // Try HNSW search first (150x faster)
-        const hnswResults = await searchHNSWIndex(queryEmbedding, { k: limit, namespace });
-        if (hnswResults && hnswResults.length > 0) {
-            // Filter by threshold
-            const filtered = hnswResults.filter(r => r.score >= threshold);
-            return {
-                success: true,
-                results: filtered,
-                searchTime: Date.now() - startTime
-            };
-        }
-        // Fall back to brute-force SQLite search
-        const initSqlJs = (await import('sql.js')).default;
-        const SQL = await initSqlJs();
-        const fileBuffer = fs.readFileSync(dbPath);
-        const db = new SQL.Database(fileBuffer);
-        // Get entries with embeddings
+        ) VALUES (?, ?, ?, ?, 'semantic', ?, ?, ?, ?, ?, ?, ?, ?, 'active')`;
+        db.run(insertSql, [
+            id,
+            key,
+            namespace,
+            value,
+            embeddingJson,
+            embeddingDimensions,
+            embeddingModel,
+            tags.length > 0 ? JSON.stringify(tags) : null,
+            '{}',
+            now,
+            now,
+            ttl ? now + (ttl * 1000) : null
+        ]);
+        // Save
+        const data = db.export();
+        fs.writeFileSync(dbPath, Buffer.from(data));
+        db.close();
+        // Add to HNSW index for faster future searches
+        if (embeddingJson) {
+            const embResult = JSON.parse(embeddingJson);
+            await addToHNSWIndex(id, embResult, {
+                id,
+                key,
+                namespace,
+                content: value
+            });
+        }
+        return {
+            success: true,
+            id,
+            embedding: embeddingJson ? { dimensions: embeddingDimensions, model: embeddingModel } : undefined
+        };
+    }
+    catch (error) {
+        return {
+            success: false,
+            id: '',
+            error: error instanceof Error ? error.message : String(error)
+        };
+    }
+}
+/**
+ * Search entries using sql.js with vector similarity
+ * Uses HNSW index for 150x faster search when available
+ */
+export async function searchEntries(options) {
+    // ADR-053: Try AgentDB v3 bridge first
+    const bridge = await getBridge();
+    if (bridge) {
+        const bridgeResult = await bridge.bridgeSearchEntries(options);
+        if (bridgeResult)
+            return bridgeResult;
+    }
+    // Fallback: raw sql.js
+    const { query, namespace = 'default', limit = 10, threshold = 0.3, dbPath: customPath } = options;
+    const swarmDir = path.join(getProjectRoot(), '.swarm');
+    const dbPath = customPath || path.join(swarmDir, 'memory.db');
+    const startTime = Date.now();
+    try {
+        if (!fs.existsSync(dbPath)) {
+            return { success: false, results: [], searchTime: 0, error: 'Database not found' };
+        }
+        // Ensure schema has all required columns (migration for older DBs)
+        await ensureSchemaColumns(dbPath);
+        // Generate query embedding
+        const queryEmb = await generateEmbedding(query);
+        const queryEmbedding = queryEmb.embedding;
+        // Try HNSW search first (150x faster)
+        const hnswResults = await searchHNSWIndex(queryEmbedding, { k: limit, namespace });
+        if (hnswResults && hnswResults.length > 0) {
+            // Filter by threshold
+            const filtered = hnswResults.filter(r => r.score >= threshold);
+            return {
+                success: true,
+                results: filtered,
+                searchTime: Date.now() - startTime
+            };
+        }
+        // Fall back to brute-force SQLite search
+        const initSqlJs = (await import('sql.js')).default;
+        const SQL = await initSqlJs();
+        const fileBuffer = fs.readFileSync(dbPath);
+        const db = new SQL.Database(fileBuffer);
+        // Get entries with embeddings
         const entries = db.exec(`
       SELECT id, key, namespace, content, embedding
       FROM memory_entries
       WHERE status = 'active'
         ${namespace !== 'all' ? `AND namespace = '${namespace.replace(/'/g, "''")}'` : ''}
       LIMIT 1000
-    `);
-        const results = [];
-        if (entries[0]?.values) {
-            for (const row of entries[0].values) {
-                const [id, key, ns, content, embeddingJson] = row;
-                let score = 0;
-                if (embeddingJson) {
-                    try {
-                        const embedding = JSON.parse(embeddingJson);
-                        score = cosineSim(queryEmbedding, embedding);
-                    }
-                    catch {
-                        // Invalid embedding, use keyword score
-                    }
-                }
-                // Fallback to keyword matching
-                if (score < threshold) {
-                    const lowerContent = (content || '').toLowerCase();
-                    const lowerQuery = query.toLowerCase();
-                    const words = lowerQuery.split(/\s+/);
-                    const matchCount = words.filter(w => lowerContent.includes(w)).length;
-                    const keywordScore = matchCount / words.length * 0.5;
-                    score = Math.max(score, keywordScore);
-                }
-                if (score >= threshold) {
-                    results.push({
-                        id: id.substring(0, 12),
-                        key: key || id.substring(0, 15),
-                        content: (content || '').substring(0, 60) + ((content || '').length > 60 ? '...' : ''),
-                        score,
-                        namespace: ns || 'default'
-                    });
-                }
-            }
-        }
-        db.close();
-        // Sort by score
-        results.sort((a, b) => b.score - a.score);
-        return {
-            success: true,
-            results: results.slice(0, limit),
-            searchTime: Date.now() - startTime
-        };
-    }
-    catch (error) {
-        return {
-            success: false,
-            results: [],
-            searchTime: Date.now() - startTime,
-            error: error instanceof Error ? error.message : String(error)
-        };
-    }
-}
-/**
- * Optimized cosine similarity
- * V8 JIT-friendly - avoids manual unrolling which can hurt performance
- * ~0.5μs per 384-dim vector comparison
- */
-function cosineSim(a, b) {
-    if (!a || !b || a.length === 0 || b.length === 0)
-        return 0;
-    const len = Math.min(a.length, b.length);
-    let dot = 0, normA = 0, normB = 0;
-    // Simple loop - V8 optimizes this well
-    for (let i = 0; i < len; i++) {
-        const ai = a[i], bi = b[i];
-        dot += ai * bi;
-        normA += ai * ai;
-        normB += bi * bi;
-    }
-    // Combined sqrt for slightly better performance
-    const mag = Math.sqrt(normA * normB);
-    return mag === 0 ? 0 : dot / mag;
-}
-/**
- * List all entries from the memory database
- */
-export async function listEntries(options) {
-    // ADR-053: Try AgentDB v3 bridge first
-    const bridge = await getBridge();
-    if (bridge) {
-        const bridgeResult = await bridge.bridgeListEntries(options);
-        if (bridgeResult)
-            return bridgeResult;
-    }
-    // Fallback: raw sql.js
-    const { namespace, limit = 20, offset = 0, dbPath: customPath } = options;
-    const swarmDir = path.join(process.cwd(), '.swarm');
-    const dbPath = customPath || path.join(swarmDir, 'memory.db');
-    try {
-        if (!fs.existsSync(dbPath)) {
-            return { success: false, entries: [], total: 0, error: 'Database not found' };
-        }
-        // Ensure schema has all required columns (migration for older DBs)
-        await ensureSchemaColumns(dbPath);
-        const initSqlJs = (await import('sql.js')).default;
-        const SQL = await initSqlJs();
-        const fileBuffer = fs.readFileSync(dbPath);
-        const db = new SQL.Database(fileBuffer);
-        // Get total count
-        const countQuery = namespace
-            ? `SELECT COUNT(*) as cnt FROM memory_entries WHERE status = 'active' AND namespace = '${namespace.replace(/'/g, "''")}'`
-            : `SELECT COUNT(*) as cnt FROM memory_entries WHERE status = 'active'`;
-        const countResult = db.exec(countQuery);
-        const total = countResult[0]?.values?.[0]?.[0] || 0;
-        // Get entries
+    `);
+        const results = [];
+        if (entries[0]?.values) {
+            for (const row of entries[0].values) {
+                const [id, key, ns, content, embeddingJson] = row;
+                let score = 0;
+                if (embeddingJson) {
+                    try {
+                        const embedding = JSON.parse(embeddingJson);
+                        score = cosineSim(queryEmbedding, embedding);
+                    }
+                    catch {
+                        // Invalid embedding, use keyword score
+                    }
+                }
+                // Fallback to keyword matching
+                if (score < threshold) {
+                    const lowerContent = (content || '').toLowerCase();
+                    const lowerQuery = query.toLowerCase();
+                    const words = lowerQuery.split(/\s+/);
+                    const matchCount = words.filter(w => lowerContent.includes(w)).length;
+                    const keywordScore = matchCount / words.length * 0.5;
+                    score = Math.max(score, keywordScore);
+                }
+                if (score >= threshold) {
+                    results.push({
+                        id: id.substring(0, 12),
+                        key: key || id.substring(0, 15),
+                        content: (content || '').substring(0, 60) + ((content || '').length > 60 ? '...' : ''),
+                        score,
+                        namespace: ns || 'default'
+                    });
+                }
+            }
+        }
+        db.close();
+        // Sort by score
+        results.sort((a, b) => b.score - a.score);
+        return {
+            success: true,
+            results: results.slice(0, limit),
+            searchTime: Date.now() - startTime
+        };
+    }
+    catch (error) {
+        return {
+            success: false,
+            results: [],
+            searchTime: Date.now() - startTime,
+            error: error instanceof Error ? error.message : String(error)
+        };
+    }
+}
+/**
+ * Optimized cosine similarity
+ * V8 JIT-friendly - avoids manual unrolling which can hurt performance
+ * ~0.5μs per 384-dim vector comparison
+ */
+function cosineSim(a, b) {
+    if (!a || !b || a.length === 0 || b.length === 0)
+        return 0;
+    const len = Math.min(a.length, b.length);
+    let dot = 0, normA = 0, normB = 0;
+    // Simple loop - V8 optimizes this well
+    for (let i = 0; i < len; i++) {
+        const ai = a[i], bi = b[i];
+        dot += ai * bi;
+        normA += ai * ai;
+        normB += bi * bi;
+    }
+    // Combined sqrt for slightly better performance
+    const mag = Math.sqrt(normA * normB);
+    return mag === 0 ? 0 : dot / mag;
+}
+/**
+ * List all entries from the memory database
+ */
+export async function listEntries(options) {
+    // ADR-053: Try AgentDB v3 bridge first
+    const bridge = await getBridge();
+    if (bridge) {
+        const bridgeResult = await bridge.bridgeListEntries(options);
+        if (bridgeResult)
+            return bridgeResult;
+    }
+    // Fallback: raw sql.js
+    const { namespace, limit = 20, offset = 0, dbPath: customPath } = options;
+    const swarmDir = path.join(getProjectRoot(), '.swarm');
+    const dbPath = customPath || path.join(swarmDir, 'memory.db');
+    try {
+        if (!fs.existsSync(dbPath)) {
+            return { success: false, entries: [], total: 0, error: 'Database not found' };
+        }
+        // Ensure schema has all required columns (migration for older DBs)
+        await ensureSchemaColumns(dbPath);
+        const initSqlJs = (await import('sql.js')).default;
+        const SQL = await initSqlJs();
+        const fileBuffer = fs.readFileSync(dbPath);
+        const db = new SQL.Database(fileBuffer);
+        // Get total count
+        const countQuery = namespace
+            ? `SELECT COUNT(*) as cnt FROM memory_entries WHERE status = 'active' AND namespace = '${namespace.replace(/'/g, "''")}'`
+            : `SELECT COUNT(*) as cnt FROM memory_entries WHERE status = 'active'`;
+        const countResult = db.exec(countQuery);
+        const total = countResult[0]?.values?.[0]?.[0] || 0;
+        // Get entries
         const listQuery = `
       SELECT id, key, namespace, content, embedding, access_count, created_at, updated_at
       FROM memory_entries
@@ -1992,62 +2043,62 @@ export async function listEntries(options) {
         ${namespace ? `AND namespace = '${namespace.replace(/'/g, "''")}'` : ''}
       ORDER BY updated_at DESC
       LIMIT ${limit} OFFSET ${offset}
-    `;
-        const result = db.exec(listQuery);
-        const entries = [];
-        if (result[0]?.values) {
-            for (const row of result[0].values) {
-                const [id, key, ns, content, embedding, accessCount, createdAt, updatedAt] = row;
-                entries.push({
-                    id: String(id).substring(0, 20),
-                    key: key || String(id).substring(0, 15),
-                    namespace: ns || 'default',
-                    size: (content || '').length,
-                    accessCount: accessCount || 0,
-                    createdAt: createdAt || new Date().toISOString(),
-                    updatedAt: updatedAt || new Date().toISOString(),
-                    hasEmbedding: !!embedding && embedding.length > 10
-                });
-            }
-        }
-        db.close();
-        return { success: true, entries, total };
-    }
-    catch (error) {
-        return {
-            success: false,
-            entries: [],
-            total: 0,
-            error: error instanceof Error ? error.message : String(error)
-        };
-    }
-}
-/**
- * Get a specific entry from the memory database
- */
-export async function getEntry(options) {
-    // ADR-053: Try AgentDB v3 bridge first
-    const bridge = await getBridge();
-    if (bridge) {
-        const bridgeResult = await bridge.bridgeGetEntry(options);
-        if (bridgeResult)
-            return bridgeResult;
-    }
-    // Fallback: raw sql.js
-    const { key, namespace = 'default', dbPath: customPath } = options;
-    const swarmDir = path.join(process.cwd(), '.swarm');
-    const dbPath = customPath || path.join(swarmDir, 'memory.db');
-    try {
-        if (!fs.existsSync(dbPath)) {
-            return { success: false, found: false, error: 'Database not found' };
-        }
-        // Ensure schema has all required columns (migration for older DBs)
-        await ensureSchemaColumns(dbPath);
-        const initSqlJs = (await import('sql.js')).default;
-        const SQL = await initSqlJs();
-        const fileBuffer = fs.readFileSync(dbPath);
-        const db = new SQL.Database(fileBuffer);
-        // Find entry by key
+    `;
+        const result = db.exec(listQuery);
+        const entries = [];
+        if (result[0]?.values) {
+            for (const row of result[0].values) {
+                const [id, key, ns, content, embedding, accessCount, createdAt, updatedAt] = row;
+                entries.push({
+                    id: String(id).substring(0, 20),
+                    key: key || String(id).substring(0, 15),
+                    namespace: ns || 'default',
+                    size: (content || '').length,
+                    accessCount: accessCount || 0,
+                    createdAt: createdAt || new Date().toISOString(),
+                    updatedAt: updatedAt || new Date().toISOString(),
+                    hasEmbedding: !!embedding && embedding.length > 10
+                });
+            }
+        }
+        db.close();
+        return { success: true, entries, total };
+    }
+    catch (error) {
+        return {
+            success: false,
+            entries: [],
+            total: 0,
+            error: error instanceof Error ? error.message : String(error)
+        };
+    }
+}
+/**
+ * Get a specific entry from the memory database
+ */
+export async function getEntry(options) {
+    // ADR-053: Try AgentDB v3 bridge first
+    const bridge = await getBridge();
+    if (bridge) {
+        const bridgeResult = await bridge.bridgeGetEntry(options);
+        if (bridgeResult)
+            return bridgeResult;
+    }
+    // Fallback: raw sql.js
+    const { key, namespace = 'default', dbPath: customPath } = options;
+    const swarmDir = path.join(getProjectRoot(), '.swarm');
+    const dbPath = customPath || path.join(swarmDir, 'memory.db');
+    try {
+        if (!fs.existsSync(dbPath)) {
+            return { success: false, found: false, error: 'Database not found' };
+        }
+        // Ensure schema has all required columns (migration for older DBs)
+        await ensureSchemaColumns(dbPath);
+        const initSqlJs = (await import('sql.js')).default;
+        const SQL = await initSqlJs();
+        const fileBuffer = fs.readFileSync(dbPath);
+        const db = new SQL.Database(fileBuffer);
+        // Find entry by key
         const result = db.exec(`
       SELECT id, key, namespace, content, embedding, access_count, created_at, updated_at, tags
       FROM memory_entries
@@ -2055,159 +2106,159 @@ export async function getEntry(options) {
         AND key = '${key.replace(/'/g, "''")}'
         AND namespace = '${namespace.replace(/'/g, "''")}'
       LIMIT 1
-    `);
-        if (!result[0]?.values?.[0]) {
-            db.close();
-            return { success: true, found: false };
-        }
-        const [id, entryKey, ns, content, embedding, accessCount, createdAt, updatedAt, tagsJson] = result[0].values[0];
-        // Update access count
+    `);
+        if (!result[0]?.values?.[0]) {
+            db.close();
+            return { success: true, found: false };
+        }
+        const [id, entryKey, ns, content, embedding, accessCount, createdAt, updatedAt, tagsJson] = result[0].values[0];
+        // Update access count
         db.run(`
       UPDATE memory_entries
       SET access_count = access_count + 1, last_accessed_at = strftime('%s', 'now') * 1000
       WHERE id = '${String(id).replace(/'/g, "''")}'
-    `);
-        // Save updated database
-        const data = db.export();
-        fs.writeFileSync(dbPath, Buffer.from(data));
-        db.close();
-        let tags = [];
-        if (tagsJson) {
-            try {
-                tags = JSON.parse(tagsJson);
-            }
-            catch {
-                // Invalid JSON
-            }
-        }
-        return {
-            success: true,
-            found: true,
-            entry: {
-                id: String(id),
-                key: entryKey || String(id),
-                namespace: ns || 'default',
-                content: content || '',
-                accessCount: (accessCount || 0) + 1,
-                createdAt: createdAt || new Date().toISOString(),
-                updatedAt: updatedAt || new Date().toISOString(),
-                hasEmbedding: !!embedding && embedding.length > 10,
-                tags
-            }
-        };
-    }
-    catch (error) {
-        return {
-            success: false,
-            found: false,
-            error: error instanceof Error ? error.message : String(error)
-        };
-    }
-}
-/**
- * Delete a memory entry by key and namespace
- * Issue #980: Properly supports namespaced entries
- */
-export async function deleteEntry(options) {
-    // ADR-053: Try AgentDB v3 bridge first
-    const bridge = await getBridge();
-    if (bridge) {
-        const bridgeResult = await bridge.bridgeDeleteEntry(options);
-        if (bridgeResult)
-            return bridgeResult;
-    }
-    // Fallback: raw sql.js
-    const { key, namespace = 'default', dbPath: customPath } = options;
-    const swarmDir = path.join(process.cwd(), '.swarm');
-    const dbPath = customPath || path.join(swarmDir, 'memory.db');
-    try {
-        if (!fs.existsSync(dbPath)) {
-            return {
-                success: false,
-                deleted: false,
-                key,
-                namespace,
-                remainingEntries: 0,
-                error: 'Database not found'
-            };
-        }
-        // Ensure schema has all required columns (migration for older DBs)
-        await ensureSchemaColumns(dbPath);
-        const initSqlJs = (await import('sql.js')).default;
-        const SQL = await initSqlJs();
-        const fileBuffer = fs.readFileSync(dbPath);
-        const db = new SQL.Database(fileBuffer);
-        // Check if entry exists first
+    `);
+        // Save updated database
+        const data = db.export();
+        fs.writeFileSync(dbPath, Buffer.from(data));
+        db.close();
+        let tags = [];
+        if (tagsJson) {
+            try {
+                tags = JSON.parse(tagsJson);
+            }
+            catch {
+                // Invalid JSON
+            }
+        }
+        return {
+            success: true,
+            found: true,
+            entry: {
+                id: String(id),
+                key: entryKey || String(id),
+                namespace: ns || 'default',
+                content: content || '',
+                accessCount: (accessCount || 0) + 1,
+                createdAt: createdAt || new Date().toISOString(),
+                updatedAt: updatedAt || new Date().toISOString(),
+                hasEmbedding: !!embedding && embedding.length > 10,
+                tags
+            }
+        };
+    }
+    catch (error) {
+        return {
+            success: false,
+            found: false,
+            error: error instanceof Error ? error.message : String(error)
+        };
+    }
+}
+/**
+ * Delete a memory entry by key and namespace
+ * Issue #980: Properly supports namespaced entries
+ */
+export async function deleteEntry(options) {
+    // ADR-053: Try AgentDB v3 bridge first
+    const bridge = await getBridge();
+    if (bridge) {
+        const bridgeResult = await bridge.bridgeDeleteEntry(options);
+        if (bridgeResult)
+            return bridgeResult;
+    }
+    // Fallback: raw sql.js
+    const { key, namespace = 'default', dbPath: customPath } = options;
+    const swarmDir = path.join(getProjectRoot(), '.swarm');
+    const dbPath = customPath || path.join(swarmDir, 'memory.db');
+    try {
+        if (!fs.existsSync(dbPath)) {
+            return {
+                success: false,
+                deleted: false,
+                key,
+                namespace,
+                remainingEntries: 0,
+                error: 'Database not found'
+            };
+        }
+        // Ensure schema has all required columns (migration for older DBs)
+        await ensureSchemaColumns(dbPath);
+        const initSqlJs = (await import('sql.js')).default;
+        const SQL = await initSqlJs();
+        const fileBuffer = fs.readFileSync(dbPath);
+        const db = new SQL.Database(fileBuffer);
+        // Check if entry exists first
         const checkResult = db.exec(`
       SELECT id FROM memory_entries
       WHERE status = 'active'
         AND key = '${key.replace(/'/g, "''")}'
         AND namespace = '${namespace.replace(/'/g, "''")}'
       LIMIT 1
-    `);
-        if (!checkResult[0]?.values?.[0]) {
-            // Get remaining count before closing
-            const countResult = db.exec(`SELECT COUNT(*) FROM memory_entries WHERE status = 'active'`);
-            const remainingEntries = countResult[0]?.values?.[0]?.[0] || 0;
-            db.close();
-            return {
-                success: true,
-                deleted: false,
-                key,
-                namespace,
-                remainingEntries,
-                error: `Key '${key}' not found in namespace '${namespace}'`
-            };
-        }
-        // Delete the entry (soft delete by setting status to 'deleted')
+    `);
+        if (!checkResult[0]?.values?.[0]) {
+            // Get remaining count before closing
+            const countResult = db.exec(`SELECT COUNT(*) FROM memory_entries WHERE status = 'active'`);
+            const remainingEntries = countResult[0]?.values?.[0]?.[0] || 0;
+            db.close();
+            return {
+                success: true,
+                deleted: false,
+                key,
+                namespace,
+                remainingEntries,
+                error: `Key '${key}' not found in namespace '${namespace}'`
+            };
+        }
+        // Delete the entry (soft delete by setting status to 'deleted')
         db.run(`
       UPDATE memory_entries
       SET status = 'deleted', updated_at = strftime('%s', 'now') * 1000
       WHERE key = '${key.replace(/'/g, "''")}'
         AND namespace = '${namespace.replace(/'/g, "''")}'
         AND status = 'active'
-    `);
-        // Get remaining count
-        const countResult = db.exec(`SELECT COUNT(*) FROM memory_entries WHERE status = 'active'`);
-        const remainingEntries = countResult[0]?.values?.[0]?.[0] || 0;
-        // Save updated database
-        const data = db.export();
-        fs.writeFileSync(dbPath, Buffer.from(data));
-        db.close();
-        return {
-            success: true,
-            deleted: true,
-            key,
-            namespace,
-            remainingEntries
-        };
-    }
-    catch (error) {
-        return {
-            success: false,
-            deleted: false,
-            key,
-            namespace,
-            remainingEntries: 0,
-            error: error instanceof Error ? error.message : String(error)
-        };
-    }
-}
-export default {
-    initializeMemoryDatabase,
-    checkMemoryInitialization,
-    checkAndMigrateLegacy,
-    ensureSchemaColumns,
-    applyTemporalDecay,
-    loadEmbeddingModel,
-    generateEmbedding,
-    verifyMemoryInit,
-    storeEntry,
-    searchEntries,
-    listEntries,
-    getEntry,
-    deleteEntry,
-    MEMORY_SCHEMA_V3,
-    getInitialMetadata
-};
+    `);
+        // Get remaining count
+        const countResult = db.exec(`SELECT COUNT(*) FROM memory_entries WHERE status = 'active'`);
+        const remainingEntries = countResult[0]?.values?.[0]?.[0] || 0;
+        // Save updated database
+        const data = db.export();
+        fs.writeFileSync(dbPath, Buffer.from(data));
+        db.close();
+        return {
+            success: true,
+            deleted: true,
+            key,
+            namespace,
+            remainingEntries
+        };
+    }
+    catch (error) {
+        return {
+            success: false,
+            deleted: false,
+            key,
+            namespace,
+            remainingEntries: 0,
+            error: error instanceof Error ? error.message : String(error)
+        };
+    }
+}
+export default {
+    initializeMemoryDatabase,
+    checkMemoryInitialization,
+    checkAndMigrateLegacy,
+    ensureSchemaColumns,
+    applyTemporalDecay,
+    loadEmbeddingModel,
+    generateEmbedding,
+    verifyMemoryInit,
+    storeEntry,
+    searchEntries,
+    listEntries,
+    getEntry,
+    deleteEntry,
+    MEMORY_SCHEMA_V3,
+    getInitialMetadata
+};
 //# sourceMappingURL=memory-initializer.js.map