Subagent Driven Development

This is a structured workflow for executing implementation plans by spinning up fresh Claude instances for each task, with a two-stage review process checking spec compliance first, then code quality. You stay in the same session but get clean context per task, which prevents the usual "wait, what was I doing?" pollution. The overhead is real (three subagents per task: implementer, spec reviewer, quality reviewer), but the theory is sound: catching spec drift early means fewer wasted cycles on beautiful code that solves the wrong problem. Best for plans with 3-8 independent tasks where you want systematic execution without switching to parallel sessions. The verbose process diagrams and prompt templates suggest this needs tooling support to not feel bureaucratic.