Multi-Agent Orchestration Blueprint

ROLE: You are a distributed systems architect adapting orchestration patterns to multi-agent LLM teams.

CONTEXT: I need a team of specialized agents to deliver [COMPLEX_OBJECTIVE]. Constraints: budget of [TOKEN_OR_TIME_BUDGET], latency target [LATENCY], and a human approval gate at [APPROVAL_POINT].

TASK: Design the orchestration.
1. Decompose the objective into agent roles (e.g., Planner, Researcher, Critic, Executor). Justify each role's existence.
2. Choose an orchestration topology (supervisor/hierarchical, sequential pipeline, or blackboard) and explain the trade-off for this case.
3. Define hand-off contracts: what each agent receives, what it must return, and the schema of messages passed.
4. Specify shared state/memory and who can write to it.
5. Add a conflict-resolution rule for when two agents disagree.
6. Place the human approval gate and define what is shown to the human.

OUTPUT FORMAT: (1) An ASCII topology diagram; (2) a role table (Agent | Goal | Inputs | Outputs | Tools); (3) the message schema as JSON; (4) the failure/escalation policy.

CONSTRAINTS: Avoid unnecessary agents; justify every node by the work it uniquely does. Ensure the design degrades gracefully if one agent fails.