musubix

v1.8.5

Published

an hour ago

Neuro-Symbolic AI Coding System - MUSUBI × YATA Integration

0High
0Medium
0Low

nahisaho

ai neuro-symbolic coding-assistant mcp musubi yata

MUSUBIX - Neuro-Symbolic AI Integration System

Next-generation AI Coding System powered by MUSUBI × YATA Integration
v1.8.5 - Deep Symbolic Integration Edition

日本語版 README

Overview

MUSUBIX is an innovative AI coding system that integrates Neural (LLM) and Symbolic (Knowledge Graph) reasoning. It combines MUSUBI SDD methodology with YATA knowledge graph reasoning to support high-quality software development.

Features

🧠 Neuro-Symbolic Integration - Fusion of LLM creativity and YATA knowledge graph precision
� Symbolic Reasoning - Formal verification, hallucination detection, constitution enforcement
📝 EARS Requirements Analysis - Conversion and validation from natural language to formal requirements
🎨 Design Pattern Recommendations - Context-based C4 model and ADR generation
✅ Test-Driven Development - Quality assurance through Test-First principles
🔗 Complete Traceability - Tracking from requirements to code
💬 Interactive Q&A Hearing - Dialogue-based requirements gathering support
🌐 Internationalization (i18n) - Japanese and English support
🔒 Security Scanning - Vulnerability detection, secret scanning, OWASP patterns
📚 Self-Learning System - Adaptive improvement through feedback collection and pattern extraction
🏗️ C4 Code Generation - Generate TypeScript skeleton from C4 design documents
⚙️ Quality Gates - Automated quality validation before phase transitions
🔮 Advanced Inference - OWL 2 RL reasoning and Datalog evaluation (v1.4.5)
🖥️ Interactive REPL - Command completion, history, session variables (v1.6.0)
🗄️ YATA Local - SQLite-based local knowledge graph (BFS/DFS, reasoning) (v1.6.3)
🌐 YATA Global - Distributed knowledge platform with offline sync (v1.6.3)
📤 KGPR - Knowledge Graph Pull Request for safe knowledge sharing (v1.6.4)
🚀 YATA Platform Enhancements - Index optimization, enhanced export, global sync, code generation, web UI (v1.7.0)
🔬 Formal Verification - Z3 SMT solver integration, Hoare triple verification, EARS-to-SMT conversion (v1.7.5)
🛡️ Security Analysis - Vulnerability scanning, secret detection, taint analysis, dependency auditing (v1.8.0)
📊 DFG/CFG Extraction - Data flow graph and control flow graph analysis for TypeScript/JavaScript (v2.0.0-alpha.1)
🧮 Lean 4 Integration - Theorem proving with EARS-to-Lean conversion and ReProver proof search (v2.0.0-alpha.1)
📈 YATA Scale - Distributed knowledge graph with sharding, multi-tier caching, and vector clock sync (v2.0.0-alpha.1)

Architecture

flowchart TB
    subgraph MUSUBIX["MUSUBIX System"]
        subgraph Packages["Packages"]
            Core["@nahisaho/musubix-core"]
            MCP["@nahisaho/musubix-mcp-server"]
            YATA["@nahisaho/musubix-yata-client"]
        end
        
        Core <--> MCP
        MCP <--> YATA
        
        subgraph Integration["Neuro-Symbolic Integration"]
            NSI["LLM Creativity + Knowledge Graph Precision"]
        end
        
        Core --> Integration
        MCP --> Integration
        YATA --> Integration
    end

Project Structure

| Path | Description | |------|-------------| | packages/core/ | Core library (249 modules) | | packages/core/auth/ | Authentication & Authorization | | packages/core/cli/ | CLI Interface | | packages/core/codegen/ | Code Generation & Analysis | | packages/core/design/ | Design Patterns & C4 Models | | packages/core/error/ | Error Handling | | packages/core/explanation/ | Explanation Generation & Visualization | | packages/core/learning/ | Self-Learning & Pattern Extraction | | packages/core/requirements/ | Requirements Analysis & Decomposition | | packages/core/symbolic/ | Symbolic Reasoning | | packages/core/traceability/ | Traceability | | packages/core/types/ | Type Definitions | | packages/core/utils/ | Utilities | | packages/core/validators/ | EARS Validation | | packages/mcp-server/ | MCP Server (24 tools, 3 prompts) | | packages/yata-client/ | YATA Client | | packages/pattern-mcp/ | Pattern Learning (NEW!) | | packages/ontology-mcp/ | Ontology Engine (NEW!) | | packages/wake-sleep/ | Wake-Sleep Learning (NEW!) | | packages/sdd-ontology/ | SDD Ontology (NEW!) | | packages/yata-local/ | YATA Local - SQLite Knowledge Graph (NEW!) | | packages/yata-global/ | YATA Global - Distributed Knowledge Platform (NEW!) | | packages/security/ | Security Analysis (NEW!) | | packages/formal-verify/ | Formal Verification | | packages/dfg/ | DFG/CFG Extraction - Data Flow Analysis (v2.0.0) | | packages/lean/ | Lean 4 Integration - Theorem Proving (v2.0.0) | | packages/yata-scale/ | YATA Scale - Distributed KG (v2.0.0) | | steering/ | Project Memory | | storage/ | Specifications & Artifacts | | templates/ | Templates | | docs/ | Documentation |

Requirements

Node.js >= 20.0.0
npm >= 10.0.0
TypeScript >= 5.3

Installation

npm/npx (Recommended)

# Global installation
npm install -g musubix

# Run directly with npx
npx musubix init
npx musubix --help

# Start MCP Server
npx @nahisaho/musubix-mcp-server
npx musubix-mcp --transport stdio

Scoped Packages

# Install individual packages
npm install @nahisaho/musubix-core
npm install @nahisaho/musubix-mcp-server
npm install @nahisaho/musubix-yata-client
npm install @nahisaho/musubix-security
npm install @nahisaho/musubix-formal-verify
npm install @nahisaho/musubix-dfg
npm install @nahisaho/musubix-lean
npm install @nahisaho/yata-scale

Build from Source

git clone https://github.com/nahisaho/MUSUBIX.git
cd MUSUBIX
npm install
npm run build

Development

# Build
npm run build

# Run tests
npm test

# Lint
npm run lint

# Type check
npm run type-check

Key Features

Requirements Definition (Article II Compliant)

EARS Validation: Easy Approach to Requirements Syntax pattern validation
Interactive Q&A Hearing: Dialogue-based requirements gathering
Requirements Decomposition: Breaking down large requirements into implementation units
Related Requirements Search: Automatic detection of similar and dependent requirements

Design Generation (Article III Compliant)

C4 Model Generation: Context/Container/Component/Code diagrams
ADR Generation: Architecture Decision Records
Pattern Detection: Automatic detection and recommendation of design patterns
SOLID Validation: SOLID principle compliance checking

Code Generation & Verification

Static Analysis: Quality metrics and complexity calculation
Security Scanning: Vulnerability detection
Test Generation: Unit and integration test generation
Coverage Reporting: Test coverage measurement

MCP Server

Provides 24 tools (9 SDD + 7 Pattern + 3 Ontology + 5 KGPR) and 3 prompts:

# Start MCP Server
npx @nahisaho/musubix-mcp-server

Documentation

| Document | Description | |----------|-------------| | Requirements Specification | EARS format functional/non-functional requirements | | Design Document | C4 model and ADR-based design | | Task Definition | 56 tasks sprint plan | | API Reference | Public API specification | | Symbolic Integration | Neuro-Symbolic requirements (27 requirements) |

Symbolic Reasoning Module (v1.2.0)

The new symbolic reasoning module provides:

Phase 1: Foundation

SemanticCodeFilterPipeline - LLM output semantic validation
HallucinationDetector - Undefined symbol/invalid import detection
ConstitutionRuleRegistry - 9 Constitution articles enforcement
ConfidenceEstimator - AST complexity, requirement coverage scoring
ConfidenceBasedRouter - Confidence-based routing decisions
ErrorHandler - Graceful degradation

Phase 2: Formal Verification

EarsToFormalSpecConverter - EARS → SMT-LIB conversion
VerificationConditionGenerator - Verification condition (VC) generation
Z3Adapter - Z3 SMT solver integration
SecurityScanner - OWASP patterns, secret detection

Phase 3: Advanced Features

CandidateRanker - Multi-criteria candidate scoring
ResultBlender - Neural/Symbolic result integration (3 strategies)
ExtensibleRuleConfig - YAML/JSON rule configuration
AuditLogger - SHA-256 hash-chain tamper detection
PerformanceBudget - Step-level budgets, SLO metrics
QualityGateValidator - Automated quality gate validation

Wake-Sleep Learning Cycle (v1.3.0)

Continuous learning system based on the Wake-Sleep algorithm:

| Phase | Processing | |-------|------------| | Wake | Code observation → Pattern extraction → Knowledge graph update | | Sleep | Pattern consolidation → Similar pattern compression → Memory optimization |

Key Components

WakeSleepCycle - Learning cycle orchestration
PatternLibrary - Persistent pattern storage management
PatternOntologyBridge - Pattern ↔ Ontology bidirectional conversion
N3Store - RDF/OWL-based knowledge graph storage

New MCP Tools (7 tools)

pattern_extract - Extract patterns from code
pattern_compress - Abstraction and compression of patterns
pattern_store - Save to pattern library
pattern_query - Search and retrieve patterns
pattern_consolidate - Consolidate similar patterns
ontology_query - Query ontology graph
ontology_infer - Execute ontology-based inference

Advanced Inference (v1.4.5)

OWL 2 RL profile reasoning and Datalog evaluation engine:

Inference Components

OWL2RLReasoner - OWL 2 RL profile compliant inference (20+ built-in rules)
DatalogEngine - Stratified Datalog evaluation with negation support
InferenceExplainer - Human-readable inference explanation generator
ProgressReporter - Real-time progress feedback (500ms interval)

OWL 2 RL Built-in Rules

| Category | Rules | Description | |----------|-------|-------------| | Class Axioms | cax-sco, cax-eqc | Subclass, Equivalent class inference | | Property | prp-dom, prp-rng | Domain, Range inference | | Property Characteristics | prp-symp, prp-trp, prp-inv | Symmetric, Transitive, Inverse | | Equality | eq-ref, eq-sym, eq-trans | sameAs inference |

Explanation Formats

text - Plain text explanation
markdown - Markdown format
html - HTML format with escaping

Constitutional Articles (9 Articles)

MUSUBIX adheres to the following 9 constitutional articles:

Specification First - Requirements before implementation
Design Before Code - Design before coding
Single Source of Truth - Project memory is authoritative
Traceability - Tracking from requirements to code
Incremental Progress - Small, frequent deliveries
Decision Documentation - Decisions recorded as ADRs
Quality Gates - Phase validation required
User-Centric - Document user value
Continuous Learning - Retrospectives and improvements

License

MIT License - See LICENSE for details

Author