overview.md

Agent Overview

Agent Ecosystem

ISC-CodeConnect employs a sophisticated multi-agent system, each specialized for specific aspects of Salesforce development assistance. This document provides a comprehensive overview of all agents and their interactions.

Agent Hierarchy

Tier 1: Orchestration Agents

These agents manage the overall system workflow and coordination.

1. Supervisor Agent

• Primary Role: Query decomposition and stage orchestration
• Responsibilities:
  • Break complex queries into executable stages
  • Manage stage execution loop
  • Coordinate inter-agent communication
  • Handle workflow exceptions
• Input: User queries, chat history
• Output: Stage definitions, execution plans

2. Router Agent

• Primary Role: Query analysis and agent selection
• Responsibilities:
  • Analyze query intent and code type
  • Route to appropriate specialized agents
  • Handle classification edge cases
  • Maintain routing decision logs
• Input: Stage queries from Supervisor
• Output: Agent selection decisions

Tier 2: Context & Retrieval Agents

These agents prepare and execute knowledge retrieval operations.

3. Context Analyzer Agent

• Primary Role: Retrieval context preparation
• Responsibilities:
  • Analyze query requirements
  • Prepare optimal search queries
  • Context enhancement and refinement
  • Query transformation for vector search
• Input: Routed queries
• Output: Enhanced retrieval contexts

4. Retriever Agent

• Primary Role: Vector database retrieval
• Responsibilities:
  • Execute parallel context queries
  • Retrieve relevant code chunks
  • Score and rank results
  • Filter and deduplicate content
• Input: Search contexts from Context Analyzer
• Output: Ranked relevant code chunks

5. Research Agent

• Primary Role: Detailed analysis and planning
• Responsibilities:
  • Perform comprehensive query research
  • Ground facts and validate context
  • Create detailed execution roadmaps
  • Reduce hallucination through verification
• Input: Stage query, chat history, retrieved context
• Output: Detailed research reports and roadmaps

Tier 3: Specialized Code Agents

These agents handle specific Salesforce development tasks.

6-8. Apex Agents

Specialized for Apex code operations:

• Apex Creation Agent: New class/trigger generation
• Apex Modification Agent: Existing code updates
• Apex Explanation Agent: Code analysis and documentation

9-11. LWC Agents

Specialized for Lightning Web Component operations:

• LWC Creation Agent: New component generation
• LWC Modification Agent: Component updates
• LWC Explanation Agent: Component analysis

Tier 4: Generation Pipeline Agents

Multi-step generation workflows for complex tasks.

Apex Test Generation Pipeline (5 Agents)

1. Test Planner Agent

   • Plans comprehensive test strategies
   • Analyzes code coverage requirements
   • Defines test scenarios and cases

2. Setup Generator Agent

   • Creates test data setup code
   • Handles test environment configuration
   • Manages test data relationships

3. Stub Generator Agent

   • Generates mocking and stubbing code
   • Handles external service mocks
   • Creates test doubles and fakes

4. Test Method Generator Agent

   • Generates test methods in batches
   • Implements test assertions
   • Handles test method organization

5. Apex Test Generator Agent

   • Combines all generation steps
   • Provides complete test class
   • Validates test completeness

LWC Generation Pipeline (4 Agents)

1. LWC HTML Generator Agent

   • Creates component HTML templates
   • Implements responsive design patterns
   • Handles accessibility requirements

2. LWC JS Generator Agent

   • Develops JavaScript controller logic
   • Implements event handling
   • Manages component lifecycle

3. LWC Apex Controller Agent

   • Creates backend Apex controllers
   • Implements data access patterns
   • Handles security and permissions

4. LWC Generator Agent

   • Summarizes complete generation
   • Validates component integration
   • Provides usage instructions

Tier 5: Support & Utility Agents

12. Follow-up Handler Agent

• Primary Role: Multi-turn conversation enhancement
• Responsibilities:
  • Analyze chat history for context
  • Prepare enhanced follow-up queries
  • Maintain conversation continuity
  • Handle clarification requests

13. General Query Processor Agent

• Primary Role: Fallback query processing
• Responsibilities:
  • Handle unclassified queries
  • Perform multi-collection retrieval
  • Provide general Salesforce assistance
  • Error recovery and graceful degradation

14. Chat Handler Agent

• Primary Role: Basic interaction management
• Responsibilities:
  • Handle greetings and confirmations
  • Manage session initialization
  • Process simple acknowledgments
  • Route to appropriate specialized agents

Additional System Agents

Error Handler Agent

• Primary Role: Error management and user experience
• Responsibilities:
  • Provide user-friendly error messages
  • Log error details for debugging
  • Suggest resolution steps
  • Maintain system stability

Chat Title Agent

• Primary Role: Conversation organization
• Responsibilities:
  • Generate contextual chat titles
  • Analyze query content for titles
  • Maintain title consistency
  • Support conversation management

Agent Communication Patterns

Synchronous Communication

• Supervisor ↔ Router: Direct query routing
• Router ↔ Specialized Agents: Task assignment
• Context Analyzer ↔ Retriever: Context-aware retrieval

Asynchronous Communication

• Parallel Retrieval: Multiple context queries
• Pipeline Processing: Sequential agent execution
• Background Processing: Title generation, logging

Memory Sharing

• Session Memory: Shared across all agents in conversation
• Context Memory: Retrieved content available to downstream agents
• Persistent Memory: Historical context for follow-up processing

Agent Performance Metrics

Individual Agent Metrics

• Response Time: Average processing duration
• Success Rate: Successful task completion percentage
• Error Rate: Failed operations percentage
• Context Utilization: Effectiveness of retrieved context

Pipeline Metrics

• End-to-End Latency: Complete workflow duration
• Parallel Efficiency: Concurrent processing effectiveness
• Memory Usage: Session and persistent memory consumption
• Throughput: Queries processed per unit time

Agent Development Guidelines

Single Responsibility Principle

Each agent should have a clearly defined, focused responsibility to ensure maintainability and scalability.

Error Handling

Implement robust error handling with graceful fallback to support agents when primary agents fail.

Memory Efficiency

Optimize context usage and memory consumption to maintain system performance at scale.

Prompt Engineering

Use clear, specific prompts tailored to each agent's domain expertise and expected outputs.

Next Steps

Core Agent Documentation

• Supervisor Agent Details - Query decomposition and stage orchestration
• Router Agent Implementation - Query analysis and agent selection
• Follow-up Handler Agent - Multi-turn conversation enhancement
• Query Analysis Agent - Query intent analysis and classification
• Research Agent - Detailed analysis and planning

Pipeline Documentation

• Test Generation Pipeline - 5-stage Apex test generation workflow
• LWC Generation Pipeline - 4-stage Lightning Web Component generation

Specialized Agent Documentation

• Specialized Agents Deep Dive - Code analysis, error handling, and system support agents

Related System Documentation

• Vector Database Setup - Milvus configuration and collection management
• Milvus Connection Manager - Connection pooling and health monitoring
• Retriever Factory - Vector database abstraction and specialized retrievers