subagent-driven-development

To orchestrate parallel tasks with fresh subagents, you dispatch a dedicated subagent per task within a single session, ensuring isolated context. This approach reduces context switching and improves iteration speed for plans with mostly independent tasks.

The two-stage review workflow runs a spec-compliance review first, followed by a code-quality review after each subagent completes its task. It then iterates with feedback until the task meets approval criteria before moving forward.

To execute implementation plans by distributing work to subagents, provide your plan and trigger the workflow. Per-task subagents handle each item with automatic two-stage reviews, followed by a final finish step to merge all results.

Subagent orchestration is designed for plans with mostly independent tasks. Plans containing heavily dependent tasks may not suit this approach, as it dispatches fresh subagents per task and relies on a final merge step rather than handling complex inter-task dependencies.

Yes, you can run sequential and parallel task delegation within the same session. This keeps context cohesive for sequential execution and supports parallel task delegation when plans allow independent operations.

Using fresh subagents per task ensures isolated context and focused execution for each item. This prevents context bloat and interference between tasks, reducing context switching and improving overall iteration speed.