ai-dev-workflow

Multi-agent coordination uses a structured workflow system that manages sub-agent communication and state across sequential phases: Task, Explore, Research, Spec, Build, Review, and Refactor. Each phase creates persistent files in `.tasks/{task-name}/` so agents can hand off work with full context, enabling cross-agent collaboration without losing state between sessions.

Yes. The workflow creates a per-task directory structure with dedicated files for each phase (task.md, explore.md, research.md, spec.md, review.md, refactor.md) and a status.md log. This persists all decisions, findings, and progress across sessions so you can pause and resume work without losing context.

Start with `/task` to initialize `.tasks/{task-name}/`, then progress sequentially through `/explore`, `/research`, `/spec`, `/build`, `/review`, and `/refactor`. Each command writes outputs to its phase file and updates status.md, creating clear checkpoints and enabling plan-mode decision points between phases.

Status updates and activity logging write to status.md within the task directory, creating an audit trail of decisions and phase transitions. Each phase generates its own output file, so you can review what happened at each stage and trace dependencies across the entire workflow.

The workflow manages sub-agent communication and state automatically. Phase-specific files and shared task context enable agents to pass information forward without manual coordination, reducing handoff friction and ensuring each stage has the context it needs.