Agentic-Compaction-and-Pipleline-by-GLM-5/delegation-system/core/request-classifier.ts

/**
 * Request Classifier
 * 
 * Fast request analysis for delegation decisions.
 * Classifies requests by complexity and determines optimal routing.
 */

import {
  RequestClassification,
  RequestComplexity,
  ClassifiableRequest,
  AgentType
} from './types';

// ============================================================================
// Classification Rules
// ============================================================================

interface ClassificationRule {
  pattern: RegExp | string;
  complexity: RequestComplexity;
  agentType: AgentType;
  weight: number;
}

const CLASSIFICATION_RULES: ClassificationRule[] = [
  // Quick patterns - instant response
  { pattern: /^(what is|explain|show me|list|get|find)\s/i, complexity: 'quick', agentType: 'fast-responder', weight: 0.9 },
  { pattern: /status|help|version|info$/i, complexity: 'quick', agentType: 'fast-responder', weight: 0.95 },
  { pattern: /^(yes|no|ok|thanks|done)$/i, complexity: 'quick', agentType: 'fast-responder', weight: 0.99 },
  
  // Code patterns - moderate complexity
  { pattern: /(review|check|analyze|inspect)\s+(this|the|my)?\s*(code|file|function)/i, complexity: 'moderate', agentType: 'reviewer', weight: 0.85 },
  { pattern: /(fix|solve|debug|resolve)\s+/i, complexity: 'moderate', agentType: 'analyzer', weight: 0.8 },
  { pattern: /(add|update|modify|change)\s+(a|the|this)?\s*(function|method|class)/i, complexity: 'moderate', agentType: 'coder', weight: 0.85 },
  
  // Research patterns - streaming
  { pattern: /(research|investigate|explore|search)\s+/i, complexity: 'streaming', agentType: 'researcher', weight: 0.8 },
  { pattern: /(analyze|audit|examine)\s+(the|entire|whole)?\s*(codebase|project|repo)/i, complexity: 'streaming', agentType: 'explorer', weight: 0.85 },
  
  // Complex patterns - need main agent or parallel delegation
  { pattern: /(refactor|rewrite|restructure|redesign)\s+/i, complexity: 'complex', agentType: 'planner', weight: 0.9 },
  { pattern: /(implement|build|create|develop)\s+(a|an|the)?\s*(new|feature|system)/i, complexity: 'complex', agentType: 'planner', weight: 0.85 },
  { pattern: /(architect|design|plan)\s+/i, complexity: 'complex', agentType: 'planner', weight: 0.9 },
  { pattern: /migrate|upgrade|port\s+/i, complexity: 'complex', agentType: 'planner', weight: 0.85 },
];

// Keywords that indicate complexity
const COMPLEXITY_INDICATORS = {
  high: ['architecture', 'system', 'multiple', 'integrate', 'refactor', 'migrate', 'redesign'],
  medium: ['implement', 'feature', 'function', 'module', 'component', 'service', 'api'],
  low: ['fix', 'update', 'change', 'add', 'remove', 'rename', 'comment']
};

// Time estimates by complexity (ms)
const TIME_ESTIMATES: Record<RequestComplexity, { min: number; max: number }> = {
  quick: { min: 100, max: 2000 },
  moderate: { min: 2000, max: 15000 },
  complex: { min: 15000, max: 120000 },
  streaming: { min: 5000, max: 60000 }
};

// ============================================================================
// Request Classifier Class
// ============================================================================

export class RequestClassifier {
  private cachedClassifications: Map<string, RequestClassification> = new Map();
  private readonly cacheMaxSize: number = 1000;
  
  // Capability mapping by request type
  private readonly typeCapabilities: Record<ClassifiableRequest['type'], string[]> = {
    code: ['syntax', 'semantics', 'patterns', 'best-practices'],
    question: ['knowledge', 'explanation', 'examples'],
    task: ['execution', 'planning', 'coordination'],
    analysis: ['inspection', 'metrics', 'patterns', 'security'],
    review: ['quality', 'standards', 'best-practices', 'security'],
    refactor: ['patterns', 'optimization', 'clean-code'],
    debug: ['tracing', 'analysis', 'diagnosis']
  };

  /**
   * Classify a request for delegation decisions
   */
  async classify(request: ClassifiableRequest): Promise<RequestClassification> {
    // Check cache first
    const cacheKey = this.getCacheKey(request);
    const cached = this.cachedClassifications.get(cacheKey);
    if (cached) {
      return { ...cached, confidence: Math.max(0, cached.confidence - 0.1) }; // Slightly lower confidence for cached
    }

    // Analyze request
    const analysis = await this.analyzeRequest(request);
    
    // Determine complexity
    const complexity = this.determineComplexity(request, analysis);
    
    // Get recommended agent
    const recommendedAgent = this.getRecommendedAgent(request, complexity, analysis);
    
    // Calculate estimated time
    const estimatedTime = this.estimateTime(complexity, analysis);
    
    // Determine capabilities needed
    const requiredCapabilities = this.getRequiredCapabilities(request, analysis);
    
    // Build classification
    const classification: RequestClassification = {
      complexity,
      score: analysis.complexityScore,
      confidence: analysis.confidence,
      recommendedAgent,
      estimatedTime,
      canDelegate: this.canDelegate(complexity, analysis),
      delegationPriority: this.getPriority(request, analysis),
      requiredCapabilities,
      contextRequirements: {
        files: analysis.fileCount,
        depth: analysis.contextDepth,
        history: analysis.needsHistory
      }
    };

    // Cache the result
    this.cacheClassification(cacheKey, classification);

    return classification;
  }

  /**
   * Quick classification for fast-path decisions (< 50ms target)
   */
  quickClassify(content: string): { complexity: RequestComplexity; agent: AgentType } {
    // Fast pattern matching
    for (const rule of CLASSIFICATION_RULES) {
      if (typeof rule.pattern === 'string') {
        if (content.toLowerCase().includes(rule.pattern.toLowerCase())) {
          return { complexity: rule.complexity, agent: rule.agentType };
        }
      } else {
        if (rule.pattern.test(content)) {
          return { complexity: rule.complexity, agent: rule.agentType };
        }
      }
    }

    // Default based on length
    const length = content.length;
    if (length < 100) return { complexity: 'quick', agent: 'fast-responder' };
    if (length < 500) return { complexity: 'moderate', agent: 'coder' };
    return { complexity: 'complex', agent: 'planner' };
  }

  // ============================================================================
  // Private Methods
  // ============================================================================

  private async analyzeRequest(request: ClassifiableRequest): Promise<{
    complexityScore: number;
    confidence: number;
    fileCount: number;
    contextDepth: 'shallow' | 'medium' | 'deep';
    needsHistory: boolean;
    matchedRules: ClassificationRule[];
    keywordDensity: Record<string, number>;
  }> {
    const content = request.content.toLowerCase();
    
    // Match rules
    const matchedRules: ClassificationRule[] = [];
    for (const rule of CLASSIFICATION_RULES) {
      const pattern = typeof rule.pattern === 'string' 
        ? new RegExp(rule.pattern, 'i')
        : rule.pattern;
      if (pattern.test(content)) {
        matchedRules.push(rule);
      }
    }

    // Calculate keyword density
    const keywordDensity: Record<string, number> = {};
    for (const [level, keywords] of Object.entries(COMPLEXITY_INDICATORS)) {
      keywordDensity[level] = keywords.reduce((count, kw) => {
        return count + (content.includes(kw) ? 1 : 0);
      }, 0) / keywords.length;
    }

    // Calculate complexity score (0-1)
    let complexityScore = 0;
    
    // Length factor
    complexityScore += Math.min(request.content.length / 2000, 0.3);
    
    // Rule matching factor
    if (matchedRules.length > 0) {
      const avgWeight = matchedRules.reduce((sum, r) => sum + r.weight, 0) / matchedRules.length;
      complexityScore += avgWeight * 0.4;
    }
    
    // Keyword density factor
    complexityScore += keywordDensity.high * 0.2;
    complexityScore += keywordDensity.medium * 0.1;

    // File count factor
    const fileCount = request.files?.length || 0;
    complexityScore += Math.min(fileCount / 10, 0.1);

    // Normalize
    complexityScore = Math.min(complexityScore, 1);

    // Determine context depth
    let contextDepth: 'shallow' | 'medium' | 'deep' = 'shallow';
    if (complexityScore > 0.6 || fileCount > 3) contextDepth = 'deep';
    else if (complexityScore > 0.3 || fileCount > 1) contextDepth = 'medium';

    // Check if history is needed
    const needsHistory = /context|previous|earlier|before|last|history/i.test(content);

    // Calculate confidence
    let confidence = 0.5;
    if (matchedRules.length > 0) confidence += 0.3;
    if (keywordDensity.high > 0 || keywordDensity.medium > 0) confidence += 0.1;
    confidence = Math.min(confidence, 0.95);

    return {
      complexityScore,
      confidence,
      fileCount,
      contextDepth,
      needsHistory,
      matchedRules,
      keywordDensity
    };
  }

  private determineComplexity(
    request: ClassifiableRequest, 
    analysis: { complexityScore: number; matchedRules: ClassificationRule[] }
  ): RequestComplexity {
    // Check matched rules first
    if (analysis.matchedRules.length > 0) {
      // Get the highest weight rule's complexity
      const topRule = analysis.matchedRules.reduce((best, rule) => 
        rule.weight > best.weight ? rule : best
      );
      return topRule.complexity;
    }

    // Fall back to score-based classification
    if (analysis.complexityScore < 0.25) return 'quick';
    if (analysis.complexityScore < 0.5) return 'moderate';
    if (analysis.complexityScore < 0.75) return 'streaming';
    return 'complex';
  }

  private getRecommendedAgent(
    request: ClassifiableRequest,
    complexity: RequestComplexity,
    analysis: { matchedRules: ClassificationRule[] }
  ): AgentType {
    // Check matched rules
    if (analysis.matchedRules.length > 0) {
      const topRule = analysis.matchedRules.reduce((best, rule) => 
        rule.weight > best.weight ? rule : best
      );
      return topRule.agentType;
    }

    // Map request type to agent
    const typeToAgent: Record<ClassifiableRequest['type'], AgentType> = {
      code: 'coder',
      question: 'fast-responder',
      task: 'executor',
      analysis: 'analyzer',
      review: 'reviewer',
      refactor: 'coder',
      debug: 'analyzer'
    };

    return typeToAgent[request.type] || 'fast-responder';
  }

  private estimateTime(
    complexity: RequestComplexity,
    analysis: { complexityScore: number; fileCount: number }
  ): number {
    const base = TIME_ESTIMATES[complexity];
    let estimate = base.min + (base.max - base.min) * analysis.complexityScore;
    
    // Add time for files
    estimate += analysis.fileCount * 500;
    
    return Math.round(estimate);
  }

  private canDelegate(complexity: RequestComplexity, analysis: any): boolean {
    // Quick and moderate can always be delegated
    if (complexity === 'quick' || complexity === 'moderate') return true;
    
    // Streaming can be delegated with progress
    if (complexity === 'streaming') return true;
    
    // Complex depends on confidence
    return analysis.confidence > 0.7;
  }

  private getPriority(
    request: ClassifiableRequest,
    analysis: any
  ): 'low' | 'medium' | 'high' | 'critical' {
    // Check metadata for explicit priority
    if (request.metadata?.priority) {
      return request.metadata.priority;
    }

    // Determine from analysis
    if (analysis.keywordDensity?.high > 0.5) return 'high';
    if (analysis.complexityScore > 0.7) return 'high';
    if (analysis.complexityScore > 0.4) return 'medium';
    return 'low';
  }

  private getRequiredCapabilities(
    request: ClassifiableRequest,
    analysis: any
  ): string[] {
    const capabilities = new Set<string>();
    
    // Add type-based capabilities
    const typeCaps = this.typeCapabilities[request.type] || [];
    typeCaps.forEach(c => capabilities.add(c));
    
    // Add based on content analysis
    if (/security|auth|encrypt/i.test(request.content)) capabilities.add('security');
    if (/test|spec|coverage/i.test(request.content)) capabilities.add('testing');
    if (/performance|optim|speed/i.test(request.content)) capabilities.add('performance');
    if (/doc|comment|readme/i.test(request.content)) capabilities.add('documentation');
    
    return Array.from(capabilities);
  }

  private getCacheKey(request: ClassifiableRequest): string {
    return `${request.type}:${request.content.slice(0, 100)}:${request.files?.length || 0}`;
  }

  private cacheClassification(key: string, classification: RequestClassification): void {
    // Enforce max size
    if (this.cachedClassifications.size >= this.cacheMaxSize) {
      // Remove oldest entry
      const firstKey = this.cachedClassifications.keys().next().value;
      if (firstKey) {
        this.cachedClassifications.delete(firstKey);
      }
    }
    this.cachedClassifications.set(key, classification);
  }

  /**
   * Clear classification cache
   */
  clearCache(): void {
    this.cachedClassifications.clear();
  }
}

// ============================================================================
// Factory Function
// ============================================================================

export function createRequestClassifier(): RequestClassifier {
  return new RequestClassifier();
}

// ============================================================================
// Export
// ============================================================================

export default RequestClassifier;