What problem does it solve? AI coding agents often generate code without following a spec-first, test-first discipline, leading to implementations that drift from approved designs and lack test evidence. This Skill enforces a structured implementation phase where tasks from a Spec-Driven Development (SDD) workflow are executed strictly against specs and design artifacts, with progress persisted and auditable. ## Core Features & Use Cases - Task-Based Implementation: Executes assigned tasks from tasks.md (or Engram artifacts) by reading specs as acceptance criteria and design decisions as constraints, then marking tasks complete as it goes. - Strict TDD Mode: When enabled and a test runner exists, loads a dedicated TDD module enforcing RED-GREEN-TRIANGULATE-REFACTOR cycles, assertion quality rules, and a mandatory TDD Cycle Evidence table. - Multi-Backend Persistence: Supports engram, openspec, hybrid, and none artifact store modes, with merge protocols that preserve progress across batched apply runs. - Workload Governance: Enforces review workload forecasts, chained/stacked PR strategies, and size-exception gates before writing code. - Use Case: An orchestrator agent delegates "Phase 1, tasks 1.1-1.3" of an auth change; this Skill reads the proposal, spec, and design, implements the JWT middleware following existing code patterns, marks tasks [x], persists apply-progress, and returns a structured summary ready for the verify phase. ## Quick Start Delegate the pending tasks for a specific change to the sdd-apply sub-agent so it implements them according to the specs and design artifacts.
