sdd:plan

To convert a draft task specification into an implementation-ready plan, you run the draft task file through a multi-agent workflow that orchestrates parallel analysis, architecture synthesis, decomposition, and verification with built-in quality gates.

Multi-agent plan refinement works by coordinating parallel research, codebase, and business analysis agents, then synthesizing their findings into an architecture overview before breaking it down into optimized, parallel execution steps.

Yes, you can use human-in-the-loop checkpoints during task decomposition and planning by enabling the flexible configuration options, allowing manual review and intervention throughout the orchestration workflow stages.

The best way to accelerate architecture synthesis is running research, codebase analysis, and business analysis concurrently in parallel, which combines findings rapidly to form a concrete architecture overview and plan.

Yes, the planning workflow supports refine and continue modes for failed quality gates, enforcing judge thresholds and allowing iterations until the task successfully passes verification and promotes to the ready-to-execute queue.

You should avoid a multi-agent planning workflow for software architecture when dealing with simple, single-step tasks, as the overhead of parallel analysis, judge verifications, and decomposition outweighs the benefits of orchestrated refinement.