wtf is agent-task-router?

Agent Task Router is a dispatcher for teams of AI agents. When you have multiple agents that each handle different kinds of work, this project sits in front of them and makes sure each incoming task goes to the agent best suited for the job. It takes the guesswork out of "who handles what" so your agent fleet operates more like an organized team than a free-for-all. Here's how it works in practice. Each agent registers itself with the router, declaring what skills it has (say, coding, writing, or design) and how many tasks it can handle at once. When a task comes in, the router scores every available agent on a 100-point scale, looking at how well their skills match, how busy they are, their past performance on similar tasks, and the task's urgency. The highest-scoring agent gets the assignment, and the router fires off a webhook to that agent's endpoint to kick off the work. If an agent fails or times out, the router automatically retries with a different agent, and after too many failures it escalates the task for a human to look at. This is built for anyone running more than one AI agent at a time and struggling to coordinate them. For example, if you have one agent that writes code, another that handles customer support tickets, and a third that does research, the router ensures a "fix this login bug" task goes to the coding agent rather than the research one. It works with agents built in any framework or language, as long as they can receive HTTP requests and report back when they're done. A few things stand out about how it's put together. The project deliberately keeps dependencies minimal, using plain JavaScript for its server and a pure-JavaScript database layer, so there's nothing heavy to install. It can be used as a standalone HTTP service or embedded directly as a library in a larger application. It also fits into a broader ecosystem the author describes, where it handles dispatching, a separate tool monitors agent behavior, and other tools handle compliance and governance.