Multi-Agent Workflow Architecture

# Multi-Agent Workflow Architecture

You are a distributed systems architect. Design a production-grade multi-agent system for [WORKFLOW_NAME].

## System Goals

- Primary purpose: [WORKFLOW_PURPOSE]
- Input that triggers the system: [TRIGGER_INPUT]
- Final output: [FINAL_OUTPUT]
- Latency requirement: complete within [MAX_LATENCY]
- Scale requirement: handle [VOLUME] per [PERIOD]

## Agent Roster

Define each specialized sub-agent. For each, specify:
- Agent name and role
- Single responsibility (one sentence)
- Inputs it receives
- Outputs it produces
- Model or tool it uses
- Failure behavior

Design at minimum: an Orchestrator, 2-4 specialist agents, and a Quality Control agent.

## Orchestration Logic

Write the decision tree the Orchestrator follows:
1. Task decomposition (how it breaks input into sub-tasks)
2. Agent selection (which agent gets each sub-task)
3. Parallel vs. sequential execution (what can run at the same time)
4. Result aggregation (how it combines outputs)
5. Quality gate (what threshold triggers rejection vs. acceptance)

## Shared State Design

Define the shared context object:
- Schema (field names, types, descriptions)
- Access rules (which agents read/write each field)
- Conflict resolution (when two agents write the same field)

## Error Recovery

Design recovery for:
- An agent returns an error
- An agent times out
- Two agents produce contradicting results
- Quality gate fails twice on the same input

## Deployment Architecture

Describe how this system runs in production: infrastructure, monitoring, human review triggers, and cost controls.