What problem does it solve? Turning an approved backend implementation plan into working code often loses track of what was actually built: checkboxes drift from reality, decisions get re-litigated, and feedback lands in the wrong place. This Skill executes the plan task-by-task or via multi-agent orchestration while keeping the plan, PRD, and decisions log truthful as code lands. ## Core Features & Use Cases - Plan execution in two modes: implement task-by-task with per-task developer review, or orchestrate parallel subagents with dependency-aware, file-disjoint scheduling and a single final review. - Direct implementation without ceremony: for small, well-described changes it implements directly with no artifacts, escalating to planning only when the change outgrows "trivial". - Change management on finished work: triages follow-up feedback (QA findings, PR review comments) into code fixes, plan amendments, superseding decision entries, or PRD patches. - Use Case: A developer says "implement the plan for LO-58" — the Skill reads .ai/<TICKET>/backend-plan.md, proposes the next ready task, implements it against the real dependency code, checks off steps as they land, and hands the diff to the developer for review before committing. ## Quick Start Ask the assistant to implement the approved backend plan for your ticket, for example by saying "implement the plan for LO-58" or "orchestrate the rest of the ticket".
