subagent-driven-development

Execute multi-step implementation plans by dispatching each task to a fresh subagent, ensuring work stays isolated without shared context. This approach prevents mixing tasks together and maintains focus throughout the plan execution.

A two-stage review process verifies every completed task first for spec compliance, ensuring it meets requirements, and then for code quality. This dual verification happens before marking each task complete and moving to the next one.

Use isolated subagents for task orchestration when executing same-session plans with mostly independent tasks, such as feature delivery, refactors, or structured bug fixes. Tasks must be separable and completable without shared context between them.

Verify spec compliance during implementation by running a dedicated review stage that checks each completed task against the original plan requirements before assessing code quality. This ensures the implementation matches the defined specifications.

Subagent-driven development does not work well for dependent tasks requiring shared context. It is designed for implementation plans where tasks can be separated and completed independently without needing shared state between execution units.

The best way to orchestrate independent implementation tasks is assigning each task to a fresh subagent with controlled handoff to finishing workflows. This includes implementer dispatch, spec compliance verification, code quality review, and task tracking.