splashxmoon/friday released v0.2.2 as a GitHub release on October 30, 2026. This version adjusts the routing that divides models by task type. Short feature requests go to a general model, while migration, security audit, and memory leak tasks go to the strongest model. Follow-up responses keep the model that fits the task. The change is the routing rules written in the v0.2.2 release notes.
splashxmoon/friday is an open-source harness that connects multiple coding agents so they can divide work like a team and exchange messages. Claude Code, Codex, Cursor, Gemini CLI, Copilot and nine more agents share the work, and can hand it to another agent once a usage limit is reached. A harness means the underlying infrastructure that helps agents collaborate. It runs locally, needs no account, and supports macOS, Windows, and Linux. The role of this repository, then, is less about offering a new coding model and more about connecting multiple agents and carrying work forward.
The star count shown in the provided information is 3. That figure alone cannot tell you anything about usage scale or how effective the routing is. The v0.2.2 release notes present per-task model assignment and how follow-up responses are handled, but no comparison results are given. So this change can be read as an update to the rule for picking a model per task, and the basis for judging the real effect is limited.