Squad Station 2

One big agent can do a lot. But try to make the same agent research a topic, write a draft, fact-check itself, and then format the output cleanly, and somewhere in the middle it starts mixing the roles. The researcher brain and the editor brain step on each other. You end up with a draft that's half-cited and the fact-checker quietly vanished three turns ago.

Multi-agent systems exist because specialization works. squad-station-2 is an npm package for building a small team of cooperating agents — a squad — where each agent has a role and the framework routes work between them. Researcher hands to writer. Writer hands to editor. Editor hands to formatter. Each one stays in its lane.

It's for developers who've hit the limits of a single agent and want the clearest possible way to break the work into roles. Not a sprawling "society of mind" with fifty agents — a small, readable squad with a handful of well-defined parts. The kind of structure you can draw on a whiteboard and hold in your head.

The package gives you the orchestration primitives: define the agents, define how work moves between them, let the framework run the loop. You still bring the models, the prompts, and the tools. What you get back is a pattern for keeping them organized.

Pair it with a shared task log — Questlog is a natural fit — so every agent in the squad can see what the others have done. Pair it with AI Agent Types if the squad spans services, and with UAP (Universal Agent Protocol) when they need to talk across process boundaries.

Start with two agents and one handoff. Add a third when you miss the structure.