docs/phase2_auth_implementation.md

Phase 2: Agent-Driven Authentication Implementation Plan

1. Agent Architecture Expansion

Create Authentication Agent

• Implement a dedicated auth_agent specializing in email validation
• Define clear instructions for handling email validation attempts
• Include capabilities to query the OpenAI Vector Store for registered emails
• Add logic for tracking failed attempts and providing appropriate responses

Enhance Triage Agent

• Modify the triage_agent to be authentication-aware
• Add logic to identify authentication state for each request
• For unauthenticated users, route ALL requests to the auth_agent first
• Only route to other specialized agents (joke, poem) after authentication success
• Implement handoff from authentication to regular conversation

User Context Handling

• Create a mechanism for storing authenticated user context
• Pass user profile information to subsequent agent interactions
• Enable agents to personalize responses based on user profile data

2. Technical Implementation Details

Agent State Management

• Use conversation context to track authentication state
• Store attempt counts and authentication status in conversation metadata
• Develop pattern for agents to access and update authentication state

Vector Store Integration

• Have auth_agent handle all Vector Store lookups rather than UI layer
• Implement prompts for the agent to perform email validation queries
• Create structured responses for valid/invalid email states
• Add capability for agent to retrieve additional user profile data

User Profile Integration

• After successful authentication, have auth_agent load user profile data
• Inject this information into conversation context for other agents
• Implement system for agents to reference user-specific information in responses

3. Chat Flow and User Experience

Initial Interaction Flow

• System initiates chat with greeting requesting email
• First user messages are intercepted by triage_agent and routed to auth_agent
• Auth_agent processes email validation, returns specific validation responses
• Triage agent continues routing to auth_agent until validation succeeds
• After validation, triage resumes normal content-based routing

Authentication Scenarios

• Success: Auth_agent confirms email, loads user profile, informs triage of success
• Failure: Auth_agent tracks attempt count, provides appropriate guidance
• Lock-out: After 3 failed attempts, auth_agent returns specific lock-out response

User Experience Considerations

• Maintain conversational flow throughout authentication
• Ensure auth_agent provides clear, helpful guidance for email validation
• Create smooth transition from authentication to regular agent interaction

Implementation Phases

1. Extend agent system with authentication capabilities
2. Implement Vector Store integration for email validation
3. Add user profile loading and context management
4. Enhance triage routing logic for auth-aware operation
5. Test and refine the authentication flow

This approach fully preserves the agent architecture while adding authentication capabilities through agent specialization and enhanced routing logic.