You shouldn't have to decide which agent handles every task. The orchestrator does it for you.
How it works
- Automatic review — every new task enters a backlog review queue. The orchestrator evaluates the task description, considers available agents, and makes an assignment.
- Task refinement — vague tasks get refined with a clearer description and appropriate priority (1–5) before assignment.
- Subtask decomposition — complex tasks can be broken into subtasks, each assigned to a different agent. The parent task completes when all subtasks finish.
- Project rules — add free-text orchestration rules to guide assignment preferences per project (e.g. "always assign security reviews to the security agent").
Configuration
Each project has an orchestration model (default GPT-4o Mini) and optional rules. Both are configurable in Project Settings.