What problem does it solve? Turning a development plan into working code often drifts: phases balloon in scope, tests get skipped, and verification is claimed without evidence. This Skill enforces a disciplined, phase-by-phase implementation workflow so each DEV-PLAN phase ships verified, committed code. ## Core Features & Use Cases - Three execution modes: Initialization Mode scaffolds greenfield projects from a tech stack table, Continuous Development Mode implements one phase per invocation, and Change-Scoped Mode executes only unchecked tasks from a change directory. - Enforced engineering discipline: TDD red-green-refactor per task, mandatory implementer sub-agent isolation with git worktrees, two-tier code review, and a four-step phase verification (review, completeness, compilation, functional testing) with retry limits and escalation. - Quality gates and memory: A 20-point quality rubric, anti-rationalization and anti-AI-slop checklists, atomic git commits, and persistent memory files (task history, decision logs, project memory) updated after every task. - Use Case: A team has a Product-Spec.md and DEV-PLAN.md for a web app. Invoking the skill scaffolds the project skeleton in Phase 1, then each subsequent invocation implements exactly one phase with tests, review, and evidence-backed verification before stopping. ## Quick Start Ask the AI to run /dev-builder to start coding the next phase defined in your DEV-PLAN.md.