subagent-driven-development

Subagent-driven development dispatches a fresh isolated subagent for each decomposed task unit, executing implementation plans continuously within the current session without human-in-the-loop delays between tasks.

The two-stage review runs spec compliance verification first, then evaluates code quality after each dispatched task completes, ensuring implementations meet requirements before assessing code structure within the current session.

Preventing context leakage requires dispatching a fresh subagent per task with isolated context boundaries, ensuring each implementation unit executes independently without inheriting prior task state within the session.

Yes, continuous execution dispatches fresh subagents sequentially without human-in-the-loop checkpoints between tasks, triggering the two-stage spec compliance and code quality reviews automatically to enable fast iteration across the full implementation plan.

Telemetry hooks coordinate the planning, execution, and finishing workflows during subagent task dispatch, but the core two-stage review loop operates independently of telemetry to ensure spec compliance and code quality.

Avoid subagent-driven development when tasks cannot be decomposed into independent units or when strict human-in-the-loop approval is required between tasks, as it relies on continuous execution with automated two-stage reviews for isolated task dispatch.