subagent-driven-development

To coordinate independent tasks with subagents without context leakage, dispatch a fresh subagent for each task in your plan. This isolates execution context and enables rapid iteration across session-long plans.

A two-stage review workflow routes completed tasks from implementers to reviewers sequentially. The first stage checks spec compliance, and the second stage evaluates code quality to maintain structured discipline.

Use subagent-driven development for session-long plans containing mostly independent tasks. It is designed to accelerate delivery by maintaining isolated contexts and structured status tracking across multiple dispatched subagents.

Escalation rules for blocked subagent tasks require structured status tracking using BLOCKED, NEEDS_CONTEXT, and DONE states. These states enforce discipline and maintain quality throughout the task dispatch workflow.

Yes, you need structured prompts for subagent task dispatch. Enforcing structured prompts ensures task extraction, plan walking, and status tracking maintain discipline and quality during isolated subagent execution.

The best way to handle context isolation when automating multi-task workflows is dispatching fresh subagents per task. This approach ensures isolated context and rapid iteration for complex plans.