subagent-driven-development

To dispatch subagents for task execution, provide a plan file and a sandboxed context. The system reads the plan upfront, extracts all tasks, and dispatches a fresh subagent for each task to ensure isolation and fast iteration without cross-task state leakage.

The two-stage review process evaluates tasks by first checking spec compliance and then assessing code quality. Tasks must pass both review stages before being marked complete, with automatic progression applying iterative fixes until all criteria are met.

Automate multi-task implementation by providing a plan file to extract per-task work items. The workflow uses a per-task toolset to guide deterministic execution, automatically applying iterative fixes until each task passes both spec compliance and code quality reviews.

Use fresh subagents for task delegation when coordinating complex implementation plans with largely independent tasks. This approach avoids cross-task state leakage, improves focus, and enables parallel work with strict review controls for deterministic execution and traceability.

Before dispatching subagents, you need a plan file containing the implementation tasks, a sandboxed context for isolation, and a per-task toolset. These inputs guide deterministic task execution and ensure each subagent operates within its required scope.