subagent-driven-development

Subagent-driven development is a workflow that orchestrates implementation plans by dispatching fresh subagents for individual tasks within a single session. It prevents context bleed and accelerates delivery by keeping each task's context isolated.

You execute tasks by dispatching fresh subagents for individual tasks within your current session. This approach provides controlled parallel-like execution while maintaining isolated context for each task to reduce cross-task leakage.

Yes, you can use a single combined reviewer for non-trivial tasks. This ensures consistent quality checks across all tasks dispatched to subagents, providing a consolidated final review of the implementation plan results.

Subagent orchestration uses deterministic, sequential execution with concurrency set to one. This provides predictable coordination and easier debugging rather than true parallel processing, while still enabling controlled parallel-like execution.

You should avoid subagent-driven development for tasks that are not largely independent. The workflow is designed for largely independent tasks that require isolated context, so tightly coupled tasks may not benefit from this orchestration approach.

The best way to handle issues is through the clear escalation path provided by the subagent-driven development workflow. When issues arise during task implementation, you can escalate them while maintaining the combined spec and quality reviewer workflow.