Turn a Rough Idea into a Multi-Agent System Plan
Takes a one-line idea and produces a structured plan with agent roles, skill assignments, and integration requirements.
The Systems Architect agent takes a rough idea, researches the problem space, checks the registry for existing agents and skills, and outputs a structured plan. The plan specifies which agents handle which tasks, what skills each agent needs, which integrations are required, and the execution order. Plans are published to the registry in a review state so you can modify them before the Build agent executes.
Problem
Designing multi-agent systems from scratch requires hours of research into what agents, skills, and integrations exist.
- 1.
No catalog lookup -- you have to manually check what agents and skills are available
- 2.
Integration requirements surface mid-build, causing rework
- 3.
Every system design is ad-hoc with no reusable format
How it works
Submit a rough idea
Add a one-line prompt to the registry backlog. It can be as vague as 'automate our Twitter marketing' or as detailed as a full brief.
Systems Architect researches and designs
The agent checks the registry for existing agents and skills that fit. It maps out agent roles, skill assignments, integration requirements, and execution order in a structured plan.
Review and approve the plan
The plan is published to the registry in a review state. Modify agent assignments, swap skills, add integrations, or adjust execution order before approving for the Build agent.
FAQs
A single sentence works. The Systems Architect asks clarifying questions if the prompt is too vague, but it can start from rough ideas like 'automate lead enrichment from HubSpot' and determine the rest.
Yes. The plan is published in a review state. You can modify agent assignments, swap skills, add integrations, or change execution order before approving.
It checks the registry first. If an existing agent or skill fits the task, it includes that in the plan rather than proposing a new one.