ADR-033: Multi-Agent Scripts Deployment

Status

Accepted

Context

During the DDD refactoring (September 9-12), we created scripts for multi-agent coordination that have not been deployed. These scripts (deploy_multi_agent_coordinator.sh, validate_multi_agent_operation.sh) implement the May 27 vision of specialized agents working in coordination. Recent success with parallel agent deployment (Bug #166) validates this approach.

Decision

We will deploy and operationalize the existing multi-agent coordination scripts, making multi-agent deployment the default approach for all development work that can benefit from specialization.

Agent Specialization Model

• Claude Code: Backend, services, investigations, broad analysis
• Cursor Agent: UI, templates, user experience, specific file edits
• Chief Architect: Gameplans, strategy, architectural decisions
• Lead Developer: Agent coordination, prompt creation, validation

Deployment Protocol

1. Default to multi-agent unless explicitly justified otherwise
2. Parallel execution when tasks are independent
3. Sequential handoffs when dependencies exist
4. Cross-validation at defined checkpoints

Consequences

Positive

• Specialized expertise: Each agent optimized for their strengths
• Parallel throughput: Multiple tasks advance simultaneously
• Built-in validation: Agents check each other’s work
• Reduced context switching: Agents maintain focus on their domain
• Natural team dynamics: Mimics human development teams

Negative

• Coordination overhead: Managing multiple agents requires effort
• Potential conflicts: Agents may disagree on approaches
• Resource consumption: Multiple agents use more compute
• Complexity increase: More moving parts to track

Neutral

• GitHub tracking requirement: Must maintain clear issue updates
• Session log multiplication: Each agent creates logs
• Prompt engineering needs: Each agent needs specific instructions

Implementation

Immediate Actions (Today)

# Deploy the coordinator
./scripts/deploy_multi_agent_coordinator.sh

# Validate operation
./scripts/validate_multi_agent_operation.sh

Week 1: Formalize Protocols

• Document handoff procedures
• Create cross-validation checkpoints
• Establish conflict resolution process

Week 2: Optimize Patterns

• Analyze coordination successes/failures
• Refine agent specializations
• Create reusable prompt templates

Success Metrics

• Parallel task completion rate
• Cross-validation catch rate
• Time to completion vs single agent
• Conflict frequency and resolution time

References

• Original vision: May 27, 2025
• Pattern-029: Multi-Agent Coordination
• Issue #118: INFR-AGENT Multi-Agent Coordinator
• Current example: Bug #166 parallel execution