workflow-execute

Autonomous workflow execution is orchestrated by discovering active sessions, planning tasks, and coordinating agents. The system supports lazy task loading, parallel execution, and real-time progress tracking to keep workflows moving without manual input.

Session discovery identifies the right active workflow automatically. This allows the orchestration system to select the correct context before planning tasks and coordinating agents, reducing the labor and error associated with manual session selection.

Yes, execution strategy is parsed directly from plan files such as plan.json or IMPL_PLAN.md. The system extracts the strategy from these files to determine the optimal run model, supporting both sequential and parallel task processing.

The orchestrator supports auto-resume to continue interrupted workflows and optional auto-commit after task completion. Auto-commit captures changes with concise messages, ensuring progress is saved without requiring manual git operations.

Lazy loading of task definitions minimizes upfront I/O by only loading tasks when needed. This maximizes throughput and optimizes performance during dynamic task orchestration, especially in complex workflows with many planned tasks.

Progress tracking is handled using TodoWrite-based updates during task orchestration. This provides real-time status monitoring from start to completion, allowing you to see which tasks are pending, in progress, or completed across the workflow.