subagent-driven-development

Subagent orchestration executes an implementation plan by dispatching a fresh delegate_task subagent for each task. It enforces a two-stage review process, checking spec compliance and code quality before final integration testing.

Spec compliance is enforced through a mandatory two-stage review loop. A dedicated reviewer first checks the implementation against the spec, then a separate code-quality review runs before the task is marked complete.

TDD with pytest is implemented by instructing per-task subagents to write failing tests first. The subagent verifies pass/fail behavior and runs broader test checks to prevent regressions before submitting the task for review.

Yes, subagent-driven development applies to plan-driven software changes where tasks are mostly independent. It extracts all tasks up front from the plan, providing full task context to each implementer subagent to avoid context pollution.

A fresh subagent is dispatched per task to avoid context pollution and keep work tightly scoped. This per-task delegation ensures each implementer only sees the context relevant to its specific piece of the implementation plan.

A final integration review is needed once all individual tasks are completed and approved. It catches cross-task inconsistencies, confirms readiness to merge, and runs a full test suite to verify the entire implementation plan.