conductor-orchestrator

You orchestrate parallel execution across multiple agents by defining a DAG-based plan that dispatches tasks via primary and fallback paths. This coordinates planning, execution, and evaluation to automate complex multi-agent workflows.

The Evaluate-Loop workflow for agent coordination is an end-to-end process that drives goal-driven task execution. It handles DAG planning, parallel execution, governance by a Board of Directors, and crash-safe resumption using a metadata state store.

You start a goal-driven orchestration task by providing a clear high-level goal, such as 'Add Stripe payment integration'. The system then plans and coordinates the execution across available agents automatically.

Yes, you can resume multi-agent task execution after interruptions. The system uses atomic metadata writes to track state, enabling exact crash-safe resumption of the Evaluate-Loop without losing progress.

Yes, the orchestration workflow supports dual dispatch paths. It defaults to superpowers for primary dispatch and uses legacy loop agents as a fallback to maintain continuity during parallel execution.

Using a DAG-based plan for agent coordination requires a metadata state store and relies on a file-based message bus for inter-agent communication. It is designed for goal-driven tracks and may not suit simple single-agent tasks.