v1.4.0: Multilingual ASTs, 90-Query Benchmark, Architecture Presets, and MCP Polish

Changelog

Format: Keep a Changelog, versioning: SemVer.

[1.4.0] - 2026-10-17

Added

• 4 New Languages: Full AST extraction for Go, Rust, Java, and C# via Tree-sitter WASM (packages/parser/src/extractors/).
• Expanded Multilingual Benchmark: 90 hand-labeled queries across TS, PHP, Python, Go, Rust, Java; verified Full Graph Pipeline superiority (Recall@1 0.843, Recall@5 0.956, MRR 0.917, Wilcoxon $p = 3.51 \times 10^{-3} < 0.05$).
• 5 Extended Architecture Presets: Clean Architecture, Hexagonal (Ports & Adapters), Onion, Feature-Sliced Design (FSD), and Modular Monolith via --preset <name> in ruas lint and ruas check.
• Node.js >= 22 Preflight Wrapper: packages/cli/bin/ruas.js with clear diagnostics for native node:sqlite.
• Real Stdio MCP Integration Suite: Comprehensive end-to-end testing of all 8 MCP tools over live JSON-RPC stdio subprocess transport.
• Client Configuration Snippets: Ready-to-use MCP configs in examples/mcp-configs/ for Cursor, Claude Code, Claude Desktop, Cline, and Windsurf.
• V8 Coverage & Blocking CI Gates: 86.9% line coverage with strict blocking thresholds and 100/100 architecture purity gate in CI.
• Visuals & VHS Demo: Official 4:1 banner (assets/banner.png), UI screenshot (assets/ui-screenshot.png), and terminal recording script (docs/demo.tape).
• Repo Hygiene & .ruasignore: Relocated sample playgrounds and graph exports to examples/; .ruasignore support for clean dogfooding.

[1.3.0]

• Fuzzy Architectural Boundary Linter (ruas lint / ruas check), MCP lint_architecture.

[1.2.0]

• Systematic fuzzy engine: Mamdani FIS, fuzzy matcher, fuzzy Dijkstra, smell detection, visual fuzzy search.

[1.0.0]

• Dijkstra / Yen top-K over Bayesian-calibrated call graph.

Ruas v1.3.0 - Systematic Fuzzy Logic, Architectural Boundary Linter, and MCP lint_architecture

🌿 Ruas v1.3.0: Systematic Fuzzy Logic & Architectural Boundary Linter

Ruas v1.3.0 introduces systematic approximate reasoning (Zadeh 1965 / Mamdani FIS), the first continuous Fuzzy Architectural Boundary Linter (ruas lint / ruas check), and the 8th native MCP tool lint_architecture for AI agents.

🌟 Key Highlights

1. Systematic Fuzzy Inference Engine (FuzzyInferenceEngine):

   • Continuous triangular (trimf), trapezoidal (trapmf), sigmoidal (sigmf), and Gaussian (gaussmf) membership functions.
   • Mamdani fuzzy rule evaluation and centroid defuzzification for structural code fragility, coupling metrics, and code smells.
   • Linguistic membership profiles (Clean, Fair, Fragile, Critical).

2. Fuzzy Architectural Boundary Linter (ruas lint / ruas check):

   • Zero-dependency glob matcher resolving file paths into architectural layers (domain, application, infrastructure, presentation).
   • Calculates continuous fuzzy violation degrees $\mu_{\text{violation}}(e) \in [0, 1]$ blending edge coupling intensity (calls, extends, imports, references), Bayesian edge confidence, and destination afferent blast radius.
   • Continuous Architectural Purity Score (zsh-100%$) and letter grade (A, B, C, D, F).
   • Automated Clean Architecture generator (ruas lint --init).
   • Strict CI/CD quality gatekeeper mode (ruas check --strict --json).

3. Fuzzy-Weighted Dijkstra Traversal (ruas path --fuzzy):

   • Computes minimal-friction dependency trajectories through Mamdani FIS edge resistance multipliers (e) \in [1.0, 3.5]$.
   • Seamlessly finds optimal routes bypassing congested hubs and fragile boundary crossings.

4. Fuzzy Code Smell Detection (ruas smell):

   • Multi-signal defuzzified certainty for dead code candidates and God Components.

5. AI Agent Tool #8: lint_architecture (MCP):

   • Exposes automated architectural compliance checks directly to Claude Desktop, Cursor, Antigravity, Cline, and other MCP-enabled AI coding agents.

6. Quality & Test Coverage:

   • 20/20 test suites passing (100/100 tests).
   • 100% clean TypeScript strict typechecking.