sdd

Spec compliance is enforced before code review by executing each plan task in an isolated subagent and running a two-stage review gate. The controller routes reviewer findings back to the subagent and repeats fix-and-recheck loops until the spec compliance gate is fully satisfied.

The best way to execute tightly coupled tasks without parallel agents is serial execution via controller-orchestrated subagents. The controller reads the plan once, extracts tasks, and dispatches them sequentially to isolated subagents to prevent context pollution and maintain strict task focus.

You orchestrate subagents for iterative review by dispatching a fresh subagent per task and routing reviewer findings back to the controller. The controller re-dispatches tasks on NEEDS_CONTEXT, BLOCKED, or DONE_WITH_CONCERNS statuses, repeating the review loop until all approvals pass.

Yes, subagent-driven task execution is specifically designed for plans with a low task count or tightly coupled work. It suits sequential development scenarios where parallelization is impractical, ensuring each isolated subagent maintains focus without context pollution.

The controller re-dispatches tasks on a DONE_WITH_CONCERNS status to resolve lingering issues flagged during the two-stage review. This iterative re-dispatch ensures both spec compliance and code quality gates are fully approved before the final implementation review completes.

You should avoid serial execution for workflow orchestration when your plan contains many independent tasks and parallel agent teams are available. This approach is strictly optimized for tightly coupled work and low-count task plans where parallelization is impractical.