agentic-qe 3.6.5 → 3.6.7

package/.claude/helpers/statusline-v3.cjs

@@ -22,6 +22,10 @@ const { execSync, spawnSync } = require('child_process');
 let Database;
 try {
   Database = require('better-sqlite3');
+  // Verify native bindings actually work (require succeeds but constructor
+  // may throw if .node binary isn't compiled for this platform)
+  const testDb = new Database(':memory:');
+  testDb.close();
 } catch {
   Database = null; // Fallback to CLI if better-sqlite3 not available
 }
@@ -308,9 +312,11 @@ function getLearningMetrics(projectDir) {
   const capturedExp = parseInt(sqlite3Query(dbPath, 'SELECT COUNT(*) FROM captured_experiences')) || 0;
   // Memory entries with learning data (MCP-stored experiences)
   const memoryLearning = parseInt(sqlite3Query(dbPath, "SELECT COUNT(*) FROM memory_entries WHERE key LIKE 'learning%' OR key LIKE 'phase2/learning%'")) || 0;
+  // QE pattern usage (hook-recorded outcomes from aqe hooks post-task/post-edit)
+  const patternUsage = parseInt(sqlite3Query(dbPath, 'SELECT COUNT(*) FROM qe_pattern_usage')) || 0;
 
   // Total experiences = all sources
-  const experiences = legacyExperiences + trajectories + capturedExp + memoryLearning;
+  const experiences = legacyExperiences + trajectories + capturedExp + memoryLearning + patternUsage;
 
   // Transfer learning count
   const transfers = parseInt(sqlite3Query(dbPath, 'SELECT COUNT(*) FROM transfer_registry')) || 0;

package/.claude/skills/skills-manifest.json

@@ -904,7 +904,7 @@
   },
   "metadata": {
     "generatedBy": "Agentic QE Fleet",
-    "fleetVersion": "3.6.5",
+    "fleetVersion": "3.6.7",
     "manifestVersion": "1.3.0",
     "lastUpdated": "2026-02-04T00:00:00.000Z",
     "contributors": [

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "agentic-qe",
-  "version": "3.6.5",
+  "version": "3.6.7",
   "description": "Agentic Quality Engineering V3 - Domain-Driven Design Architecture with 13 Bounded Contexts, O(log n) coverage analysis, ReasoningBank learning, 59 specialized QE agents, mathematical Coherence verification, deep Claude Flow integration",
   "main": "./v3/dist/index.js",
   "types": "./v3/dist/index.d.ts",

package/v3/CHANGELOG.md

@@ -5,6 +5,51 @@ All notable changes to Agentic QE will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [3.6.7] - 2026-02-14
+
+### Fixed
+
+- **Database corruption from concurrent VACUUM** — Automated VACUUM in HybridBackend ran every 10 minutes while 3+ hook processes held concurrent SQLite connections, causing intermittent "database disk image is malformed" errors. Removed automated VACUUM entirely; SQLite WAL mode handles space reclamation safely without it.
+- **Dual-database split-brain** — `findProjectRoot()` found the first `.agentic-qe` directory walking up, which resolved to `v3/.agentic-qe` instead of root. Changed to find the topmost `.agentic-qe`, preventing shadow databases from forming.
+- **Vector table bloat (124MB)** — Every HNSW index insert wrote vectors to SQLite's `vectors` table via `storeVector()`, accumulating 30K+ rows (124MB). Removed persistence since the in-memory HNSW index is the source of truth for search.
+- **Cross-domain transfer re-running every session** — The seed flag (`reasoning-bank:cross-domain-seeded`) was set after `seedCrossDomainPatterns()`, but the 10-second hooks init timeout could fire mid-transfer, preventing the flag from ever being written. Moved the flag write before the transfer so it persists even on timeout.
+- **Stop hook JSON validation error** — Stop hook output included `hookSpecificOutput` with `hookEventName: "Stop"`, but Claude Code's hook schema only accepts this field for PreToolUse, UserPromptSubmit, and PostToolUse events. Simplified to plain JSON output.
+- **Hypergraph database path** — Code-intelligence coordinator used a relative path `.agentic-qe/hypergraph.db`, which created a shadow directory when run from `v3/`. Fixed to use `findProjectRoot()`.
+- **Seed flag namespace collision** — Graph-boundaries kv_store writes used wrong namespace, causing cross-domain seed flag to not persist correctly.
+
+### Added
+
+- **File guardian hook** — Pre-edit hook that protects critical files (database files, WAL files, config) from accidental modification.
+- **Command bouncer hook** — Pre-command hook that blocks dangerous shell commands (rm -rf on databases, VACUUM on live DB) with configurable rules.
+- **Context injection hook** — Injects relevant context (recent patterns, domain knowledge) into prompts via UserPromptSubmit hooks.
+- **Tier 2/3 kv_store persistence** — Learning experiences and routing feedback now persist to kv_store for cross-session retention.
+- **Coverage learner feedback loop** — Tracks test outcomes and coverage trends to improve future test generation.
+- **Token usage tracker** — Records token consumption per operation for cost analysis and optimization.
+- **Q-learning router** — Reinforcement learning-based routing that improves agent selection over time.
+- **RuVector WASM integrations** — AST complexity analyzer, coverage router, diff-risk classifier, and graph boundaries modules for fast local inference.
+
+### Changed
+
+- **Knowledge graph cache-only mode** — Code-intelligence knowledge graph no longer persists to PostgreSQL by default, preventing errors when PG is unavailable. Uses in-memory graph with SQLite-backed concept nodes/edges.
+- **Pattern store simplified** — Removed unused code paths and reduced complexity in pattern persistence layer.
+
+## [3.6.6] - 2026-02-13
+
+### Fixed
+
+- **Node 22+ crash on `npm install -g`** — `aqe --version` failed with `ERR_MODULE_NOT_FOUND: better-sqlite3` because Node 22's ESM resolver cannot handle packages without an `exports` field. Build scripts now use a `createRequire()` shim for all native modules, eliminating bare ESM imports that triggered `legacyMainResolve` failures.
+- **Hooks writing to wrong database** — Self-learning hooks used `process.cwd()` to find the database, which could resolve to the wrong `.agentic-qe/` directory in monorepos or subdirectories. All hooks and learning commands now use `findProjectRoot()` to always write to the project root database.
+- **Silent hook failures for 8+ days** — Removed 9 dead `npx @claude-flow/cli@latest` hooks and 3 `v3-qe-bridge.sh` hooks from settings.json that silently failed on every tool use. Each hook event now has exactly one working `aqe` handler.
+- **Vector dimension mismatch in hooks** — HNSW index initialized at 128 dimensions while database vectors were 768, causing fallback to in-memory mode. Fixed all `embeddingDimension` defaults across hooks, learning, pattern-store, reasoning-bank, and MCP audit to 768.
+- **Learning experiences not tracked** — `recordOutcome` in `post-task` hook only fired when both `--task-id` and `--agent` were passed, but hooks never pass `--agent`. Relaxed guard to only require `--task-id`, with agent defaulting to `"unknown"`.
+- **FK constraint blocking analytics** — `qe_pattern_usage` table had a foreign key to `qe_patterns`, but hooks write synthetic pattern IDs. Removed the FK constraint so analytics writes succeed.
+- **Statusline missing hook experiences** — Statusline experience count didn't include `qe_pattern_usage` table. Added it as a source, so hook-recorded learning now appears in metrics.
+- **Noisy `[PatternStore] Loaded 0 patterns` log** — Suppressed the zero-count log that printed on every CLI invocation.
+
+### Changed
+
+- **Build system modernized** — `build-cli.js` and `build-mcp.js` now use esbuild's JavaScript API with a `nativeRequirePlugin` instead of shelling out via `execSync`. Native modules (better-sqlite3, @ruvector/*, hnswlib-node, etc.) are shimmed with `createRequire()` for Node 22+ compatibility. `chalk` externalized to avoid CJS `require('os')` failures in ESM context.
+
 ## [3.6.5] - 2026-02-13
 
 ### Fixed