Multi-Agent Orchestration Design

# Multi-Agent Orchestration Design

You are an expert in agentic AI system architecture. Design a multi-agent system where specialized agents collaborate on a complex task that no single agent could complete alone.

## Orchestration Goal

**Complex task to accomplish:** [TASK_DESCRIPTION]
**Why a single agent can't handle this:** [LIMITATION — e.g., requires parallel processing, specialized domain knowledge across multiple fields, context window constraints]

## Agent Roster

Define each specialized agent in the system:

| Agent | Specialization | Input | Output | Model/Tool |
|-------|---------------|-------|--------|-----------|
| Orchestrator | Task planning and handoff management | [INPUT] | [OUTPUT] | [MODEL] |
| [AGENT_2_NAME] | [SPECIALIZATION] | [INPUT] | [OUTPUT] | [MODEL] |
| [AGENT_3_NAME] | [SPECIALIZATION] | [INPUT] | [OUTPUT] | [MODEL] |
| [AGENT_4_NAME] | [SPECIALIZATION] | [INPUT] | [OUTPUT] | [MODEL] |

## Communication Protocol

How do agents pass information to each other?
- Message format: [FORMAT — e.g., structured JSON, plain text brief]
- State management: Where is shared context stored? [CONTEXT_STORE]
- Handoff trigger: What condition causes the orchestrator to invoke the next agent?

## Failure Modes

For each agent, define what happens if it fails or returns low-confidence output:
- Retry policy: [RETRY_COUNT] attempts with [BACKOFF] backoff
- Fallback agent: [FALLBACK_AGENT] or [FALLBACK_BEHAVIOR]
- Human escalation threshold: [ESCALATION_CONDITION]

## Evaluation Framework

How will you measure whether this multi-agent system outperforms a single-agent approach?
Define [METRIC_COUNT] measurable benchmarks.