subagent-driven-development

Task orchestration for complex plans is automated by dispatching independent subagents to execute individual tasks with a two-stage review process. This approach isolates context, validates spec compliance and code quality, and tracks progress until all tasks are approved.

A two-stage review process evaluates plan execution by first running a spec compliance review, followed by a code quality review for every task. This mechanism ensures implementation meets requirements before checking code standards, maintaining quality during rapid iteration.

Cross-task context contamination is prevented by dispatching fresh subagents for each individual task. Isolating subagent context ensures that independent tasks execute without interfering with each other, enabling clean incremental validation across the same session.

Subagent-driven task execution supports rapid feature iteration by splitting multi-step features into independent tasks, running them sequentially or in parallel, and validating results through automated reviews before proceeding to the next task.

Progress across multiple independent tasks is tracked end-to-end by deriving tasks from a defined plan and monitoring the two-stage review status. The system re-reviews failed tasks until approved, ensuring complete plan execution visibility.

A predefined implementation plan and a list of tasks are required before starting subagent orchestration. The skill uses these inputs to dispatch implementer, spec-compliance reviewer, and code-quality reviewer subagents for each task.