Background Agents: Triggers, Isolation, and Reporting Back

An agent can start because something happened rather than because a person asked it a question. It can then work with nobody watching, in a separate copy of the project, and finish by leaving a result for a person to review. Such a design is called a background agent.

Forge, our coding agent, is a natural example. When a test on the main branch fails, Forge starts on its own, tries to repair the failure on a new branch, and opens a change for an engineer to read. With nobody waiting or watching, the design must settle how a run starts, stays harmless, checks itself, and reports.