subagent-driven-development

Automating plan execution with subagents requires a structured plan input, which the skill reads once to extract tasks, dispatches each to a fresh subagent via task delegation, and finalizes after sequential reviews.

Fresh subagents prevent state leakage by isolating each task in its own context. This ensures independent execution and maintains clean boundaries between tasks during automated plan orchestration.

The two-stage review process validates delegated tasks sequentially, first checking spec compliance and then performing code-quality validation, ensuring tasks are marked completed only after passing both reviews.

Subagent-driven delegation applies to multi-task plans where tasks are independent. Dependent tasks that do not benefit from isolated execution and automatic quality checks are not suited for this orchestration approach.

Subagent orchestration requires no external dependencies or components. It operates using a structured plan input and built-in task delegation to dispatch fresh subagents and perform two-stage reviews.

Task dispatch reads the structured plan once, extracts all tasks, and drives parallel or sequential work as appropriate, delegating each task to fresh subagents and performing two-stage reviews before marking completion.