subagent-driven-development

Task decomposition with fresh subagents isolates context per task, preventing leakage across work streams. Load a plan with distinct tasks, dispatch a new subagent for each one, and enforce mandatory reviews between tasks to maintain quality control and catch errors early.

Fresh subagents per task eliminate accumulated context bias and reasoning drift. Each subagent starts clean, focused only on its assigned work, reducing mistakes and enabling strict inter-task reviews before proceeding to dependent work.

Mandatory reviews enforce a plan-driven sequence: complete one task, review its output against the plan, apply feedback if needed, mark completion, then dispatch the next subagent. This gating prevents cascading errors and keeps work aligned with the original plan.

Tasks execute sequentially with no parallel dispatch. This enforces quality control and ensures each task can depend on verified outputs from prior tasks, making the workflow predictable and reviewable.

Use this approach for complex projects with multiple independent tasks, diverse required skills, and strict quality gates. It's ideal when context leakage between tasks poses risk or when you need auditable reviews at each stage.