What problem does it solve? Turning an approved implementation plan into actual code changes often loses scope control, approval discipline, and traceability. This Skill executes a formalized SDD task plan one approved batch at a time, keeping tasks.md, apply-progress.md, and state.yaml consistent so work stays resumable and auditable. ## Core Features & Use Cases - Batch-scoped execution: Runs exactly one approved batch per invocation, requiring explicit user approval before any code-touching batch. - Safety triggers and dirty-tree policy: Blocks on dirty working trees, code-plan desync, out-of-scope blast radius, or ambiguous instructions, and records blockers in state.yaml. - Traceable progress logging: Appends per-batch evidence, changed files, and decisions to apply-progress.md while keeping state.yaml operational. - Use Case: After planning a feature change with SDD v2, invoke this Skill to implement batch 1, have it stop when a referenced function was renamed, and resume later from the recorded checkpoint. ## Quick Start Ask the assistant to execute the next approved batch of the current SDD change using sdd-apply and update the progress files.