subagent-driven-development

Task delegation to subagents works by dispatching a fresh subagent per task to isolate context and prevent state leakage. This orchestration pattern executes complex plans by aggregating independent subtasks while maintaining fresh context for each implementation step.

A two-stage review process for code quality assurance checks spec compliance first, followed by code quality validation. This mechanism dispatches reviewer subagents after implementation to catch issues early and ensure plan-driven software development produces auditable results.

You can parallelize independent subtasks in a plan-driven workflow by dispatching subagents per task and aggregating the results. This scalable workflow requires clear task extraction to ensure subtasks remain independent enough for parallel execution without state leakage.

Preventing state leakage when executing multi-step plans requires using a fresh subagent per task to isolate context. This strict per-task workflow ensures each subagent operates independently, preventing previous task state from corrupting subsequent implementations.

You need a predefined plan with clearly extracted tasks to use subagent-driven task execution. This orchestration pattern applies to plan-driven software development where independent subtasks require strict spec conformance and code quality checks through implementation and review steps.

You should not use a subagent orchestration pattern for tasks that lack clear independence or require shared mutable state across steps. This pattern requires clearly extractable tasks to maintain strict per-task workflow isolation and ensure auditable implementation and review results.