subagent-driven-development

To avoid context pollution during implementation planning, delegate each independent task to a fresh subagent. This isolates execution details, keeping the main session focused on orchestration and maintaining fast iteration.

Subagent task orchestration involves dispatching one fresh subagent per independent task and managing staged reviews. The main conversation coordinates the workflow while subagents handle the narrow implementation details and revisions.

Spec compliance and code quality review work through a two-stage process applied after each subagent task. Every task is reviewed first for spec compliance and then for code quality before proceeding to the next step.

You need subagents for implementation planning when executing plan-driven software work that must stay in the same session but risks overloading a single conversation. This approach prevents context pollution across multiple fast iterations.

When a subagent task is blocked or needs more context, the system uses clear escalation to manage the issue. This ensures the main orchestrating session receives the necessary information to unblock the implementation plan.

Subagent-driven development is designed for plan-driven coding where a task plan already exists. It focuses on executing and reviewing independent tasks rather than generating the initial implementation plan.