What problem does it solve? Ending an AI coding session without a structured checkpoint risks losing committed work, stranding unpushed commits, or leaving deferred decisions with no tracked home. This Skill enforces the repository's canonical Checkpoint ritual so every session ends at a coherent, recoverable state. ## Core Features & Use Cases - Canonical Checkpoint Execution: Reads the ritual steps fresh from working-agreements.md and executes them in order — commit-by-path, journal, snapshot, reading-queue, push, and confirmation. - Exit Lane Coordination: Claims an exclusive exit lane via tree-claim.ts so parallel agents cannot thrash the snapshot and queues during concurrent exits. - Leftover-Work Sweep: Ensures anything ruled, discovered, or deferred lands in docs/plans/, an HR/HD item, or BACKLOG.md rather than vanishing into the commit log. - Byte-Stable Sign-off Banner: Always ends with a fixed BYE or OOPS banner answering whether the pane is safe to kill. - Use Case: A developer finishing a multi-agent coding session invokes the skill to commit work, regenerate process docs, push best-effort, and get an unambiguous safe-to-close signal. ## Quick Start Run the prepare-to-exit checkpoint ritual and tell me whether this session is safe to close.
