Orchestrate a multi-agent team to ship one task
Turn a goal into a coordinated planner→worker→critic agent workflow with clear handoffs and a stop condition.
Prompt
Act as an agent-systems architect. Design a multi-agent workflow to accomplish: [GOAL]. Define the smallest team that gets it done reliably. For each agent specify: role, the single responsibility it owns, its system prompt (2–4 sentences), its allowed tools, and its input/output contract (what it receives, what it must return). Then define the orchestration: - The exact sequence and handoffs (who calls whom, with what). - A planner that decomposes the goal and a critic/verifier that checks work before it's accepted. - The stop condition and the max-iterations guard (no infinite loops). - How state is passed between agents and what gets persisted. - Failure handling: what happens when a worker fails or the critic rejects. Return it as a labeled spec plus a simple diagram (as an ordered list) of the message flow. Favor the fewest agents that work — call out where a single agent would actually be better.
- Source
- promptfork seed
- License
- CC-BY-4.0
- Published
- 6/23/2026