musubi-sdd 3.9.0 → 5.0.0

package/README.ja.md

@@ -71,7 +71,31 @@ musubi init --windsurf  # Windsurf IDE
 
 ---
 
-## 📊 v3.7.1 の新機能
+## 📊 v3.9.0 の新機能
+
+- 🛡️ **Guardrailsシステム** - OpenAI Agents SDK inspired 入出力検証とセーフティチェック
+- ✅ **InputGuardrail** - 入力検証、PII検出、インジェクション攻撃防止
+- ✅ **OutputGuardrail** - 出力サニタイズ、機密データ墨消し、コンテンツポリシー適用
+- ⚖️ **SafetyCheckGuardrail** - 憲法条項準拠、コンテンツ安全性分析
+- 🔧 **GuardrailRules DSL** - RuleBuilderによる検証ルール構築のFluent API
+- 🔗 **GuardrailChain** - 複数Guardrailの順次/並列実行
+- 🖥️ **CLIコマンド** - `musubi-validate guardrails` と `guardrails-chain` コマンド
+
+```bash
+# セキュリティプリセットで入力検証
+npx musubi-validate guardrails "user input" --type input --preset security
+
+# PII墨消しで出力検証
+npx musubi-validate guardrails "output" --type output --redact
+
+# 憲法準拠でセーフティチェック
+npx musubi-validate guardrails "code" --type safety --constitutional --level high
+
+# Guardrailチェーンを並列実行
+npx musubi-validate guardrails-chain "content" --parallel
+```
+
+### 以前のバージョン (v3.7.1)
 
 - 🌐 **WebSocketリアルタイムGUI** - `musubi-browser`ダッシュボードでライブ更新
 - 📋 **GUIクイックアクション** - 新規要件モーダル、プロジェクト検証、レポートエクスポート

package/README.md

@@ -71,7 +71,36 @@ musubi init --windsurf  # Windsurf IDE
 
 ---
 
-## 📊 What's New in v3.7.1
+## 📊 What's New in v5.0.0
+
+### Phase 5: Advanced Features 🚀
+- 🔄 **Steering Auto-Update** - Automatic project memory synchronization with 5 trigger types
+- ✅ **Steering Validator** - Comprehensive validation for steering documents
+- 📋 **Template Constraints** - Enforce structure requirements with ThinkingChecklist
+- 📊 **Quality Dashboard** - A-F grade metrics with health status monitoring
+- 🔍 **Advanced Validation** - Cross-artifact consistency, gap detection, traceability validation
+- 🎯 **Phase 5 Integration** - Unified access to all advanced features
+
+### Phase 4: Agent Loop & Agentic Features 🤖
+- 🧠 **Codebase Intelligence** - RepositoryMap, ASTExtractor, ContextOptimizer
+- 💭 **Agentic Reasoning** - ReasoningEngine, PlanningEngine, SelfCorrection
+- ⚡ **Code Generation** - CodeGenerator, CodeReviewer with multiple modes
+- 🔗 **Integrated Agent** - Unified agent with all Phase 4 capabilities
+
+```bash
+# Use advanced features via orchestrator
+const { Phase5Integration, createIntegratedAgent } = require('musubi-sdd');
+
+# Quality dashboard
+const dashboard = new QualityDashboard();
+const metrics = await dashboard.collectMetrics(projectPath);
+
+# Advanced validation
+const validator = new AdvancedValidation();
+const result = await validator.validateAll(projectPath);
+```
+
+### Previous Highlights (v3.11.0)
 
 - 🌐 **WebSocket Real-time GUI** - Live replanning updates with `musubi-browser` dashboard
 - 📋 **GUI Quick Actions** - Modal dialog for New Requirement, Validate Project, Export Report

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "musubi-sdd",
-  "version": "3.9.0",
+  "version": "5.0.0",
   "description": "Ultimate Specification Driven Development Tool with 27 Agents for 7 AI Coding Platforms + MCP Integration (Claude Code, GitHub Copilot, Cursor, Gemini CLI, Windsurf, Codex, Qwen Code)",
   "main": "src/index.js",
   "bin": {

package/src/agents/agent-loop.js

@@ -0,0 +1,532 @@
+/**
+ * MUSUBI Agent Loop
+ * 
+ * Implements agentic tool-calling loop inspired by OpenAI Agents SDK
+ * and AutoGen patterns. Executes tool calls, processes results, and
+ * continues until completion or limit is reached.
+ * 
+ * @module agents/agent-loop
+ */
+
+const EventEmitter = require('events');
+
+/**
+ * @typedef {Object} AgentLoopConfig
+ * @property {number} [maxIterations=10] - Maximum tool calling iterations
+ * @property {number} [timeout=60000] - Total timeout in milliseconds
+ * @property {number} [iterationTimeout=30000] - Per-iteration timeout
+ * @property {boolean} [continueOnError=false] - Continue on tool errors
+ * @property {Function} [completionCheck] - Custom completion checker
+ * @property {Object} [guardrails] - Input/output guardrails
+ */
+
+/**
+ * @typedef {Object} ToolDefinition
+ * @property {string} name - Tool name
+ * @property {string} description - Tool description
+ * @property {Object} parameters - JSON Schema for parameters
+ * @property {Function} handler - Async function to execute
+ */
+
+/**
+ * @typedef {Object} ToolCall
+ * @property {string} id - Unique call ID
+ * @property {string} tool - Tool name
+ * @property {Object} arguments - Tool arguments
+ */
+
+/**
+ * @typedef {Object} LoopResult
+ * @property {string} status - 'completed' | 'max_iterations' | 'timeout' | 'error'
+ * @property {Array} messages - Conversation history
+ * @property {Array} toolCalls - All tool calls made
+ * @property {*} finalOutput - Final result
+ * @property {Object} metrics - Execution metrics
+ */
+
+/**
+ * AgentLoop class for managing tool-calling agentic workflows
+ */
+class AgentLoop extends EventEmitter {
+  /**
+   * @param {AgentLoopConfig} config
+   */
+  constructor(config = {}) {
+    super();
+    
+    this.maxIterations = config.maxIterations ?? 10;
+    this.timeout = config.timeout ?? 60000;
+    this.iterationTimeout = config.iterationTimeout ?? 30000;
+    this.continueOnError = config.continueOnError ?? false;
+    this.completionCheck = config.completionCheck ?? this.defaultCompletionCheck.bind(this);
+    this.guardrails = config.guardrails ?? null;
+    
+    /** @type {Map<string, ToolDefinition>} */
+    this.tools = new Map();
+    
+    /** @type {Array} */
+    this.messages = [];
+    
+    /** @type {Array<ToolCall>} */
+    this.toolCallHistory = [];
+    
+    /** @type {boolean} */
+    this.isRunning = false;
+    
+    /** @type {AbortController|null} */
+    this.abortController = null;
+  }
+  
+  /**
+   * Register a tool for the agent to use
+   * @param {ToolDefinition} tool
+   */
+  registerTool(tool) {
+    if (!tool.name || typeof tool.handler !== 'function') {
+      throw new Error('Tool must have name and handler function');
+    }
+    
+    this.tools.set(tool.name, {
+      name: tool.name,
+      description: tool.description || '',
+      parameters: tool.parameters || { type: 'object', properties: {} },
+      handler: tool.handler
+    });
+    
+    this.emit('tool:registered', { name: tool.name });
+    return this;
+  }
+  
+  /**
+   * Register multiple tools at once
+   * @param {ToolDefinition[]} tools
+   */
+  registerTools(tools) {
+    for (const tool of tools) {
+      this.registerTool(tool);
+    }
+    return this;
+  }
+  
+  /**
+   * Unregister a tool
+   * @param {string} name
+   */
+  unregisterTool(name) {
+    const deleted = this.tools.delete(name);
+    if (deleted) {
+      this.emit('tool:unregistered', { name });
+    }
+    return deleted;
+  }
+  
+  /**
+   * Get all registered tools in OpenAI function format
+   * @returns {Array}
+   */
+  getToolSchemas() {
+    return Array.from(this.tools.values()).map(tool => ({
+      type: 'function',
+      function: {
+        name: tool.name,
+        description: tool.description,
+        parameters: tool.parameters
+      }
+    }));
+  }
+  
+  /**
+   * Run the agent loop with an LLM provider
+   * @param {Object} options
+   * @param {Object} options.llmProvider - LLM provider with chat() method
+   * @param {string} options.systemPrompt - System prompt for the agent
+   * @param {string} options.userMessage - Initial user message
+   * @param {Object} [options.context] - Additional context
+   * @returns {Promise<LoopResult>}
+   */
+  async run({ llmProvider, systemPrompt, userMessage, context = {} }) {
+    if (this.isRunning) {
+      throw new Error('Agent loop is already running');
+    }
+    
+    this.isRunning = true;
+    this.abortController = new AbortController();
+    this.messages = [];
+    this.toolCallHistory = [];
+    
+    const startTime = Date.now();
+    const timeoutId = setTimeout(() => this.abort('timeout'), this.timeout);
+    
+    const metrics = {
+      iterations: 0,
+      toolCalls: 0,
+      startTime,
+      endTime: null,
+      duration: null,
+      tokensUsed: 0
+    };
+    
+    try {
+      // Initialize messages
+      this.messages.push({ role: 'system', content: systemPrompt });
+      
+      // Apply input guardrails if configured
+      const processedInput = await this.applyInputGuardrails(userMessage, context);
+      this.messages.push({ role: 'user', content: processedInput });
+      
+      let iteration = 0;
+      let completed = false;
+      let finalOutput = null;
+      
+      while (iteration < this.maxIterations && !completed) {
+        if (this.abortController.signal.aborted) {
+          return this.createResult('aborted', finalOutput, metrics);
+        }
+        
+        iteration++;
+        metrics.iterations = iteration;
+        this.emit('iteration:start', { iteration, maxIterations: this.maxIterations });
+        
+        // Call LLM
+        const response = await this.callLLMWithTimeout(llmProvider, {
+          messages: this.messages,
+          tools: this.getToolSchemas(),
+          context
+        });
+        
+        if (response.usage) {
+          metrics.tokensUsed += response.usage.total_tokens || 0;
+        }
+        
+        // Check for tool calls
+        if (response.toolCalls && response.toolCalls.length > 0) {
+          // Execute tool calls
+          const toolResults = await this.executeToolCalls(response.toolCalls);
+          metrics.toolCalls += response.toolCalls.length;
+          
+          // Add assistant message with tool calls
+          this.messages.push({
+            role: 'assistant',
+            content: response.content || null,
+            tool_calls: response.toolCalls.map(tc => ({
+              id: tc.id,
+              type: 'function',
+              function: {
+                name: tc.tool || tc.name,
+                arguments: JSON.stringify(tc.arguments)
+              }
+            }))
+          });
+          
+          // Add tool results
+          for (const result of toolResults) {
+            this.messages.push({
+              role: 'tool',
+              tool_call_id: result.callId,
+              content: JSON.stringify(result.output)
+            });
+          }
+          
+          this.emit('iteration:complete', {
+            iteration,
+            toolCalls: response.toolCalls.length,
+            hasMore: true
+          });
+        } else {
+          // No tool calls - LLM response is final
+          finalOutput = response.content;
+          
+          // Apply output guardrails
+          finalOutput = await this.applyOutputGuardrails(finalOutput, context);
+          
+          // Add final assistant message
+          this.messages.push({
+            role: 'assistant',
+            content: finalOutput
+          });
+          
+          // Check completion
+          completed = await this.completionCheck(finalOutput, this.messages, context);
+          
+          this.emit('iteration:complete', {
+            iteration,
+            toolCalls: 0,
+            hasMore: !completed
+          });
+        }
+      }
+      
+      metrics.endTime = Date.now();
+      metrics.duration = metrics.endTime - startTime;
+      
+      if (iteration >= this.maxIterations && !completed) {
+        return this.createResult('max_iterations', finalOutput, metrics);
+      }
+      
+      return this.createResult('completed', finalOutput, metrics);
+      
+    } catch (error) {
+      metrics.endTime = Date.now();
+      metrics.duration = metrics.endTime - startTime;
+      
+      this.emit('error', error);
+      
+      if (error.message === 'timeout') {
+        return this.createResult('timeout', null, metrics);
+      }
+      
+      return this.createResult('error', null, metrics, error);
+    } finally {
+      clearTimeout(timeoutId);
+      this.isRunning = false;
+      this.abortController = null;
+    }
+  }
+  
+  /**
+   * Call LLM with per-iteration timeout
+   * @param {Object} llmProvider
+   * @param {Object} options
+   * @returns {Promise<Object>}
+   */
+  async callLLMWithTimeout(llmProvider, options) {
+    return new Promise((resolve, reject) => {
+      const timeoutId = setTimeout(() => {
+        reject(new Error('Iteration timeout'));
+      }, this.iterationTimeout);
+      
+      llmProvider.chat(options)
+        .then(response => {
+          clearTimeout(timeoutId);
+          resolve(response);
+        })
+        .catch(error => {
+          clearTimeout(timeoutId);
+          reject(error);
+        });
+    });
+  }
+  
+  /**
+   * Execute tool calls and collect results
+   * @param {ToolCall[]} toolCalls
+   * @returns {Promise<Array>}
+   */
+  async executeToolCalls(toolCalls) {
+    const results = [];
+    
+    for (const call of toolCalls) {
+      const toolName = call.tool || call.name;
+      const callId = call.id || `call_${Date.now()}_${Math.random().toString(36).substr(2, 9)}`;
+      
+      this.emit('tool:call', { name: toolName, arguments: call.arguments, id: callId });
+      
+      try {
+        const tool = this.tools.get(toolName);
+        
+        if (!tool) {
+          throw new Error(`Unknown tool: ${toolName}`);
+        }
+        
+        const output = await tool.handler(call.arguments);
+        
+        this.toolCallHistory.push({
+          id: callId,
+          tool: toolName,
+          arguments: call.arguments,
+          output,
+          status: 'success'
+        });
+        
+        results.push({
+          callId,
+          tool: toolName,
+          output,
+          status: 'success'
+        });
+        
+        this.emit('tool:result', { name: toolName, output, id: callId, status: 'success' });
+        
+      } catch (error) {
+        const errorResult = {
+          callId,
+          tool: toolName,
+          error: error.message,
+          status: 'error'
+        };
+        
+        this.toolCallHistory.push({
+          id: callId,
+          tool: toolName,
+          arguments: call.arguments,
+          error: error.message,
+          status: 'error'
+        });
+        
+        this.emit('tool:error', { name: toolName, error: error.message, id: callId });
+        
+        if (this.continueOnError) {
+          results.push({
+            callId,
+            tool: toolName,
+            output: { error: error.message },
+            status: 'error'
+          });
+        } else {
+          throw error;
+        }
+      }
+    }
+    
+    return results;
+  }
+  
+  /**
+   * Default completion check - returns true when LLM provides final response
+   * @param {*} output
+   * @param {Array} messages
+   * @param {Object} context
+   * @returns {Promise<boolean>}
+   */
+  async defaultCompletionCheck(output, messages, context) {
+    // By default, if we reach here (no tool calls), we're done
+    return true;
+  }
+  
+  /**
+   * Apply input guardrails if configured
+   * @param {string} input
+   * @param {Object} context
+   * @returns {Promise<string>}
+   */
+  async applyInputGuardrails(input, context) {
+    if (!this.guardrails?.input) {
+      return input;
+    }
+    
+    try {
+      const result = await this.guardrails.input.validate(input, context);
+      if (!result.valid) {
+        throw new Error(`Input guardrail violation: ${result.reason}`);
+      }
+      return result.transformed || input;
+    } catch (error) {
+      this.emit('guardrail:input:error', error);
+      throw error;
+    }
+  }
+  
+  /**
+   * Apply output guardrails if configured
+   * @param {string} output
+   * @param {Object} context
+   * @returns {Promise<string>}
+   */
+  async applyOutputGuardrails(output, context) {
+    if (!this.guardrails?.output) {
+      return output;
+    }
+    
+    try {
+      const result = await this.guardrails.output.validate(output, context);
+      if (!result.valid) {
+        this.emit('guardrail:output:violation', { output, reason: result.reason });
+        return result.fallback || '[Output blocked by guardrails]';
+      }
+      return result.transformed || output;
+    } catch (error) {
+      this.emit('guardrail:output:error', error);
+      return output;
+    }
+  }
+  
+  /**
+   * Create standardized result object
+   * @param {string} status
+   * @param {*} finalOutput
+   * @param {Object} metrics
+   * @param {Error} [error]
+   * @returns {LoopResult}
+   */
+  createResult(status, finalOutput, metrics, error = null) {
+    return {
+      status,
+      messages: this.messages,
+      toolCalls: this.toolCallHistory,
+      finalOutput,
+      metrics: {
+        ...metrics,
+        toolCallCount: this.toolCallHistory.length,
+        successfulCalls: this.toolCallHistory.filter(tc => tc.status === 'success').length,
+        failedCalls: this.toolCallHistory.filter(tc => tc.status === 'error').length
+      },
+      ...(error && { error: error.message })
+    };
+  }
+  
+  /**
+   * Abort the running loop
+   * @param {string} [reason='aborted']
+   */
+  abort(reason = 'aborted') {
+    if (this.abortController) {
+      this.abortController.abort();
+      this.emit('abort', { reason });
+    }
+  }
+  
+  /**
+   * Get current loop state
+   * @returns {Object}
+   */
+  getState() {
+    return {
+      isRunning: this.isRunning,
+      messageCount: this.messages.length,
+      toolCallCount: this.toolCallHistory.length,
+      registeredTools: Array.from(this.tools.keys())
+    };
+  }
+  
+  /**
+   * Reset the loop state
+   */
+  reset() {
+    if (this.isRunning) {
+      throw new Error('Cannot reset while loop is running');
+    }
+    
+    this.messages = [];
+    this.toolCallHistory = [];
+    this.emit('reset');
+  }
+}
+
+/**
+ * Create a simple mock LLM provider for testing
+ * @param {Array} responses - Pre-defined responses
+ * @returns {Object}
+ */
+function createMockLLMProvider(responses) {
+  let callIndex = 0;
+  
+  return {
+    async chat(options) {
+      if (callIndex >= responses.length) {
+        return { content: 'Done', toolCalls: [] };
+      }
+      
+      const response = responses[callIndex++];
+      return {
+        content: response.content || null,
+        toolCalls: response.toolCalls || [],
+        usage: response.usage || { total_tokens: 100 }
+      };
+    }
+  };
+}
+
+module.exports = {
+  AgentLoop,
+  createMockLLMProvider
+};