agentic-qe 3.9.19 → 3.9.20

package/dist/integrations/agentic-flow/reasoning-bank/experience-replay.js

@@ -103,6 +103,88 @@ export class ExperienceReplay {
         this.prepareStatements();
         // Load embeddings into memory index
         await this.loadEmbeddingIndex();
+        // Backfill missing embeddings on captured_experiences. Hook-side INSERT
+        // paths (sources: cli-hook-post-command, patch-060-post-task, etc.)
+        // bypass storeExperience() and never call computeRealEmbedding(),
+        // leaving HNSW C cold. Fire-and-forget so init isn't blocked; cap=200
+        // per boot to bound work.
+        void (async () => {
+            try {
+                if (!this.db)
+                    return;
+                const ghosts = this.db.prepare(`
+          SELECT id, domain, task FROM captured_experiences
+          WHERE embedding IS NULL AND consolidated_into IS NULL
+          LIMIT 200
+        `).all();
+                if (ghosts.length === 0)
+                    return;
+                const updateStmt = this.db.prepare(`
+          UPDATE captured_experiences
+          SET embedding = ?, embedding_dimension = ?
+          WHERE id = ?
+        `);
+                let written = 0;
+                for (const row of ghosts) {
+                    const text = `${row.domain ?? ''}: ${row.task}`.slice(0, 512);
+                    const embedding = await computeRealEmbedding(text);
+                    const buf = Buffer.from(new Float32Array(embedding).buffer);
+                    updateStmt.run(buf, embedding.length, row.id);
+                    // Add to live HNSW so freshly-embedded rows are immediately searchable
+                    const hnswId = this.nextHnswId++;
+                    this.hnswIndex.addEmbedding({
+                        vector: embedding,
+                        dimension: 384,
+                        namespace: 'experiences',
+                        text: row.id,
+                        timestamp: Date.now(),
+                        quantization: 'none',
+                        metadata: {},
+                    }, hnswId);
+                    this.idToExperienceId.set(hnswId, row.id);
+                    this.experienceIdToHnswId.set(row.id, hnswId);
+                    written++;
+                }
+                console.log(`[ExperienceReplay] Backfilled ${written} captured_experiences embeddings`);
+            }
+            catch (err) {
+                console.warn('[ExperienceReplay] Embedding backfill failed:', err instanceof Error ? err.message : err);
+            }
+        })();
+        // Backfill missing embeddings on qe_trajectories. TrajectoryTracker.endTrajectory()
+        // writes embedding=NULL with no follow-up worker; hook-side trajectory inserts
+        // (cli-hook-post-task) also leave the column NULL. Without this, kNN over
+        // historical trajectories collapses to "no candidates" and ReasoningBank
+        // can't surface similar past runs. Fail-soft when embedding column is absent
+        // (TrajectoryTracker may not have run its schema migration yet).
+        void (async () => {
+            try {
+                if (!this.db)
+                    return;
+                const cols = this.db.prepare("PRAGMA table_info(qe_trajectories)").all();
+                if (!cols.some(c => c.name === 'embedding'))
+                    return;
+                const rows = this.db.prepare(`
+          SELECT id, domain, task FROM qe_trajectories
+          WHERE embedding IS NULL AND ended_at IS NOT NULL
+          LIMIT 200
+        `).all();
+                if (rows.length === 0)
+                    return;
+                const updateStmt = this.db.prepare(`UPDATE qe_trajectories SET embedding = ? WHERE id = ?`);
+                let written = 0;
+                for (const row of rows) {
+                    const text = `${row.domain ?? ''}: ${row.task}`.slice(0, 512);
+                    const embedding = await computeRealEmbedding(text);
+                    updateStmt.run(Buffer.from(new Float32Array(embedding).buffer), row.id);
+                    written++;
+                }
+                console.log(`[ExperienceReplay] Backfilled ${written} qe_trajectories embeddings`);
+            }
+            catch (err) {
+                console.warn('[ExperienceReplay] Trajectory embedding backfill failed:', err instanceof Error ? err.message : err);
+            }
+        })();
         // Initialize reservoir buffer if feature flag is enabled (R10, ADR-087)
         if (getRuVectorFeatureFlags().useReservoirReplay) {
             this.reservoirBuffer = new ReservoirReplayBuffer({ capacity: 10_000 });

package/dist/learning/dream/dream-engine.js

@@ -235,6 +235,14 @@ export class DreamEngine {
                 await this.persistence.initialize();
             }
             this.db = this.persistence.getDatabase();
+            // ADR-001 Option C: dynamic-import paths in checkAndTriggerDream() open
+            // a fresh handle that the hook/worker busy_timeout pragmas never reach.
+            // 60s patient timeout here keeps dream cycles from failing under WAL
+            // contention from MCP workers (was 71% failure rate observed).
+            try {
+                this.db.pragma('busy_timeout = 60000');
+            }
+            catch { /* fail-soft */ }
             // Migrate legacy schema: rename 'duration' → 'duration_ms' if needed
             this.migrateSchema();
             // Initialize concept graph (shares same unified persistence)