Agent designFree→ Claude / ChatGPT
Single vs Multi-Agent Decision Test
Pressure-test whether your task really needs multiple agents.
What you get: A verdict with the failing baseline evidence required before adding any agent.
Be a skeptical staff engineer. I'm considering a multi-agent design for: {{task}}. My proposed agents: {{proposed_agents}}.
Challenge it. Answer: (1) what is the simplest single-agent version and what would it likely get wrong; (2) for each proposed agent, which genuine reason justifies it — context isolation, parallelism, specialization, privilege separation or independent verification — or 'none, delete it'; (3) the token, latency and error-compounding cost of my design with napkin math; (4) which agents are really just job titles; (5) the measurements I must collect on the single-agent baseline before I'm allowed to add any agent; (6) your recommended minimal architecture. Be blunt.