sdd:plan

To turn draft feature files into ready-to-implement software plans, you refine draft specifications through multi-stage analysis, architecture synthesis, and task decomposition. This creates verified, implementation-ready steps organized for maximum parallel execution.

Multi-agent parallelization for software development tasks involves orchestrating multiple agents to run research, codebase analysis, and business analysis simultaneously. This merges findings into a coherent architectural overview and organizes work for maximum parallel execution.

To refine a draft task specification for a development team, you apply multi-stage analysis and judge evaluations to the draft file. This verifies the task specification and promotes it from the draft directory to the todo directory for implementation.

Task planning with LLM-as-judge checks works effectively for multi-stage analysis by verifying draft task specifications before implementation. It evaluates the synthesized architecture and decomposed steps, ensuring the refined task is fully planned and error-free.

The best way to decompose software architecture into parallel execution steps is by merging simultaneous research and codebase analysis findings. This synthesis breaks work down into implementation steps and reorganizes them for maximum parallel execution.