What problem does it solve? Executing an epic that spans multiple phase docs, PRs, and repos often fails in predictable ways: invariants drift between repos, stacked branches conflict after squash merges, review findings get fixed instance-by-instance instead of by class, and merged code is reported as live when deploys are manual. This Skill owns the cross-phase state that no single-phase build process can see. ## Core Features & Use Cases - Sprint Ledger: Maintains a run log with kickoff gates, an invariant registry re-checked after every phase, a merge topology plan, phase results, and an owner-decision queue. - Per-Phase Protocol: Composes worktree isolation, single-phase build, simplify, verification, scope checks, deploy-reachability checks, and shipping into one ordered pipeline per phase. - Review Convergence Loop: Governs review rounds with a class-sweep rule (fix the pattern across all repos and branches, not the cited instance) and a stopping rule that escalates non-converging rounds. - Use Case: You have a nine-doc epic spanning two repos and eight PRs that must land in a declared order. Run this Skill to drive each phase in sequence while tracking invariants, merge topology, and deploy status in a single auditable ledger. ## Quick Start Ask the AI to execute all phase docs in your epic's planning directory as one ordered sprint using the build-all workflow.