subagent-driven-development

Subagent-driven development dispatches a fresh subagent per task to prevent context pollution, enabling rapid parallel iteration within a single session. Each subagent operates with its own context and review loop to balance speed and quality.

A two-stage review evaluates each task sequentially: first checking spec compliance against requirements, then assessing code quality. If issues are found, the corresponding reviewer subagent is re-dispatched until the task passes both stages.

Dispatching fresh subagents for each independent task prevents context pollution. This approach gives every task its own isolated context window and review loop, maintaining execution speed without degrading output quality.

Yes, you can enforce spec compliance by dispatching a dedicated reviewer subagent after implementation. It checks the completed work against the original plan's task text before passing it to the code-quality review stage.

Use subagent-driven development for planning-driven work where tasks are mostly independent within a single session. It is ideal when you need rapid iteration and explicit plan parsing with strict per-task spec compliance and code quality enforcement.

Yes, subagent-driven development operates without external dependencies. It relies entirely on templated prompts for implementer, spec reviewer, and code-quality reviewer subagents to orchestrate task execution natively.