session:start

Start a multi-agent orchestration session by running the harness session start command with a clear context flag describing the goal. Wait for worker agents to join before creating and assigning discrete, file-scoped tasks to coordinate test environments effectively.

No, a leader agent cannot execute tasks directly during multi-agent session coordination. The leader must delegate all work to worker agents, avoiding conflicting edits and uncoordinated effort while breaking goals into independent tasks with acceptance criteria.

The best way to delegate tasks in a harness session is to break the high-level goal into independent, file-scoped tasks with clear acceptance criteria, wait for capable agents to join, and then assign tasks using the task assignment and role transfer APIs.

Monitor agent progress and send signals during a multi-agent workflow by polling session status in machine-readable form. Observe progress and send signals to safely redirect or pause worker agents when coordination adjustments are needed.

Transfer leadership in a multi-agent orchestration session when the current leader needs to step down or when another agent is better suited to coordinate remaining tasks. Use the role transfer API to safely hand off session control without losing state.

Yes, you need to wait for agents to join before creating tasks in a harness session. Starting a session with a clear context and waiting for joins ensures capable worker agents are available to receive assignments before task creation begins.