What problem does it solve? Working through a large engineering backlog item-by-item is slow and error-prone: parallel changes conflict on shared files, deploys collide, and board state drifts from what actually shipped. This Skill turns a NEXT.md backlog into an orchestrated pipeline of concurrent sub-agent waves that land on one batch branch and merge as a single PR. ## Core Features & Use Cases - Wave-based parallel execution: Groups unblocked backlog items by repository area and dispatches concurrent worktree sub-agents, stopping only when file contention requires it. - Batch branch and PR management: Lands each wave on one batch branch with one commit per task, pushes once per wave, and raises a single PR merged with --merge. - Coordinator discipline: Enforces briefing contracts for sub-agents, inbox checks between sessions, board resequencing, and red-run triage that fixes a whole run's failures together. - Use Case: An operator says "work the backlog"; the Skill renders the board, dispatches a wave of agents across infra, app, and web areas, merges their work into claude/b<n>-board, pushes, and hands off to /watch until the PR merges. ## Quick Start Say "do next" or "work the backlog" to have the coordinator read NEXT.md, dispatch a wave of worktree sub-agents, and drive the batch to a merged PR.
