v3.6.19: Learning Pipeline Fixes & 25 New Tests

What's New

Fixes 6 learning pipeline issues discovered via brutal honesty audit, with event-driven trajectory lifecycle, correct object return type detection across all test generators, and real learning metrics in fleet_status.

Fixed

• Event-driven trajectory lifecycle replaces fragile polling loop — no more race conditions or memory leaks
• Object return type ({ valid: boolean }) no longer misclassified as boolean in all 5 test generators
• memory_usage now returns real vector and namespace counts instead of zeros
• fleet_status includes learning metrics (patterns, experiences, trajectories, vectors, dream cycles)
• memory_store persistence check simplified — no more unnecessary duck-typing

Added

• 25 new unit tests covering every fix
• queryCount() on UnifiedMemoryManager with SQL-injection-safe allowlist
• ReasoningBank learning system documentation guide

Changed

• Embedding dimensions consolidated to 384-dim (all-MiniLM-L6-v2) for ReasoningBank/PatternStore

Getting Started

npx agentic-qe init --auto

See CHANGELOG for full details.

v3.6.18: Agent Teams MCP Integration

What's New

6 new MCP tools for Agent Teams management (ADR-064), giving you full visibility and control over domain-scoped agent teams:

• team_list — List all domain teams with membership and task progress
• team_health — Get real-time health metrics for any domain team
• team_message — Send typed messages between agents with validation
• team_broadcast — Broadcast alerts or commands to entire domain teams
• team_scale — Scale teams up or down dynamically
• team_rebalance — Optimize agent distribution across teams

Enriched existing tools:

• agent_list now shows team membership (domain, role, team size) for every agent
• fleet_status now includes a teams summary (active teams, agents in teams, healthy count)

Reliability improvements:

• Duplicate agent guard prevents the same agent from joining a team twice
• Race-safe auto-wiring when spawning concurrent agents into the same domain
• Runtime message type validation rejects invalid types before delivery
• Best-effort team wiring never blocks agent spawns

Getting Started

npx agentic-qe init --auto

See CHANGELOG for full details.

v3.6.17: QE Queen Coordinator & HNSW Semantic Search

What's New

The QE Queen Coordinator now runs its full 10-phase protocol end-to-end. Semantic search via HNSW actually works — store a value, query with natural language, get scored results back. Hierarchical topology assigns domain leads and workers. MCP connections survive parallel load.

Highlights

• HNSW semantic search — memory_query with semantic: true for natural language queries with cosine similarity scoring
• Vector storage on write — memory_store indexes entries so semantic search finds them immediately
• Hierarchical agent topology — First agent per domain becomes lead, subsequent become workers
• MCP reconnection — Exponential backoff with request buffering and replay
• Real experience capture — Tool executions record patterns via ExperienceCaptureService
• Stdio transport hardening — 120s timeout, retry with backoff, backpressure handling
• Bidirectional cloud sync — Phase 5 cloud-to-local pull support

Getting Started

npx agentic-qe init --auto

See CHANGELOG for full details.

v3.6.16: Node.js node:test Support & Smart Assertions

What's New

• Node.js node:test runner — 5th supported test framework, zero-dependency testing on Node.js 18+
• Smart assertions — automatically infers test expectations from function name prefixes (is* → boolean, get* → not undefined, create* → truthy) and return types
• Destructured parameter handling — correctly handles ({ a, b }) and ([x, y]) patterns
• 12 bug fixes from comprehensive devil's advocate review, including critical regex ordering and double-lambda fixes

Getting Started

npx agentic-qe init --auto

Generate tests with the new node:test framework:

npx aqe test generate src/my-file.ts --framework node-test

See CHANGELOG for full details.

v3.6.15: Fix Test Generation & Cloud Sync

What's New

Test Generation Fixed (#295)

aqe test generate now produces runnable, correct test code across all 4 frameworks (Jest, Vitest, Mocha, Pytest). Fixed 11 bugs including undefined variables, tests for non-exported internals, wrong mock paths, and files not being written to disk.

Cloud Sync Working

Fixed PostgreSQL module loading in esbuild bundles and improved error handling. Cloud sync now properly connects and writes data instead of silently falling back to mock mode.

Data Protection

Added comprehensive rules to prevent accidental database operations by AI agents.

Getting Started

npx agentic-qe init --auto

See CHANGELOG for full details.

v3.6.14: Fix test OOM crashes and CI failures

What's New

Resolves DevPod OOM crashes during test execution and fixes CI artifact upload failures. Test suite now runs reliably in memory-constrained environments.

Fixed

• MCP unit test timeouts and DevPod OOM crashes (#294, #251) — Mocked task executor at module level to prevent real fleet initialization (~200-400MB per init). Reduced fleet init from ~31x to ~5x per test file.
• Vitest parallel execution causing memory exhaustion — Single-fork execution and disabled file parallelism prevent OOM in constrained environments.
• CI artifact upload 403 Forbidden (#294) — Added actions: write permission to workflow files.
• Flaky topology optimizer test (#251) — Widened convergence tolerance and added retry.
• Test assertions using 128-dim instead of 768-dim — Aligned with v3.6.12's HNSW dimension fix.

Getting Started

npx agentic-qe init --auto

See CHANGELOG for full details.

v3.6.13

v3.6.13 — Fix 5 critical issues from #292

Bug Fixes

• test_generate_enhanced: Fixed Agent Booster intercepting test generation tasks + restored real test code pass-through. MCP now returns actual vitest assertions instead of V2 placeholder stubs.
• Vector dimensions (768 vs 128): Fixed 6 locations where HNSW embedding dimension defaulted to 128, causing vector index crashes.
• coverage_analyze_sublinear: Added per-file coverageByFile data to MCP results + automatic test runner detection and coverage collection.
• Security scanning: Extended secret detection to all file types, added generic SECRET_KEY pattern, added known-CVE dependency checking (21 vulns detected in test fixture).
• Init hooks: Added post-write verification logging for .claude/settings.json hook entries.

Additions

• CLAUDE.md: Added mandatory Bug Fix Verification rules (reproduction-first, MCP-CLI parity, per-issue evidence)

Files Changed

18 files across coordination, domains, governance, integrations, MCP handlers, and init phases.

Closes #292

Full Changelog: v3.6.12...v3.6.13

v3.6.12

v3.6.12

Bug Fixes

• fix: prevent SQLite corruption from concurrent writers — Created centralized safe-db.ts opener with WAL + busy_timeout enforced on every database open. Replaced all 28 direct new Database() calls across v3/src, hooks, helpers, and scripts. Root cause of repeated memory.db corruption was 40+ unprotected concurrent writers from hook processes.

• fix: HNSW dimension mismatch — Changed hardcoded dimension from 128 to 768 to match transformer embedding output

• fix: real security scanning — Replaced stub security scan with actual file-system analysis (dependency audit, secret detection, permission checks)

• fix: hooks directory creation — Phase 07 now creates .claude/hooks/ directory before writing hook files

• fix: coverage embedder dimension — Updated HNSW dimension constant to 768

• fix: ghost coverage analyzer — Updated stale HNSW references

• fix: sublinear analyzer — Fixed HNSW index initialization parameters

• fix: HNSW index cosine metric — Corrected similarity metric configuration

• fix: coordinator constants — Aligned embedding dimensions across modules

• fix: database data recovery — Rebuilt corrupted memory.db from backup + cloud sync (ruvector-postgres), recovering 5,152 patterns, 1,871 GOAP actions, and historical learning data

Full Changelog

v3.6.11...v3.6.12

v3.6.11: MCP stability, multi-language support, smart init

What's New

This release resolves 9 GitHub issues (#274-#282) with improvements to MCP stability, multi-language support, init workflow, and metric accuracy.

Highlights

• Smart aqe init --auto — Detects existing AQE hooks and updates them in-place. No more duplicates on re-init.
• Multi-language file discovery — All CLI commands now support 12+ languages (Python, Go, Rust, Java, Kotlin, Ruby, C#, PHP, Swift, C/C++, Scala)
• MCP crash protection — Malformed tool calls return structured errors instead of killing the connection
• HNSW auto-resize — Vector dimension mismatches are handled gracefully instead of panicking
• LOC counter accuracy — Node-native file counting with expanded directory exclusions

Bug Fixes

• MCP server crash on malformed requests (#274)
• code_index limited to JS/TS only (#275)
• Coverage handler returned synthetic file paths (#276)
• HNSW vector dimension mismatch panic (#277)
• Model router negative complexity scores (#278)
• LOC counter inaccuracy for non-JS projects (#281)

Getting Started

npx agentic-qe init --auto

See CHANGELOG for full details.

v3.6.10: QCSD Production Telemetry, Eval-Driven Workflow, KG Test Generation

What's New

Completes the QCSD lifecycle with the Production Telemetry Swarm, adds eval-driven development tooling, and integrates Knowledge Graphs into test generation for smarter, faster test creation.

Added

• QCSD Production Telemetry Swarm (#271) — 5th QCSD lifecycle phase collecting DORA metrics via GitHub API
• Eval-driven development workflow — Skill evaluations as the primary development loop
• Skill quality scorer & conflict detector — Deterministic quality scores and overlap detection
• Eval runner P1 grading — Negative control tests, finding count enforcement, weighted rubric scoring, adaptive rubric
• KG-assisted test generation — Knowledge Graph integration with 276x faster HNSW vector loading

Fixed

• Self-learning hooks now record data from all CLI hook interactions
• CI workflow tolerates Vitest worker fork crash when all tests pass

Security

• Bumped tar 7.5.9 and ajv 8.18.0 (CVE-2025-69873 ReDoS fix)

Getting Started

npx agentic-qe@3.6.10 init --auto

See CHANGELOG for full details.