sdd:plan

To refine a draft task specification, this Skill orchestrates a multi-agent workflow that performs parallel analysis, architecture synthesis, and decomposition to produce an implementation-ready plan. It validates each stage using LLM-as-Judge quality gates before promoting the task.

Parallel analysis in task planning is the concurrent evaluation of research, codebase, and business impacts for a draft specification. This Skill uses this parallelized analysis to gather comprehensive context before synthesizing an architectural overview and breaking down the task.

LLM-as-Judge verification for workflow orchestration acts as an automated quality gate to validate task refinement phases. This Skill applies LLM-as-Judge checks at each stage, ensuring the decomposed and parallelized steps meet target quality standards before promotion.

Yes, you can configure iteration limits and skip stages for task refinement via command-line arguments. This Skill supports customizing target quality thresholds, human-in-the-loop checkpoints, included or skipped stages, and incremental refinement modes to control the workflow.

You need multi-agent workflow orchestration for task planning when a high-level feature idea requires automated fleshing out of technical details, dependency identification, and sub-task breakdown. It transforms draft specifications into actionable steps with risk assessments.