What problem does it solve? Before ending a conversation with an AI agent, it is easy to forget loose ends: unpushed commits, PRs with pending CI, promised follow-ups never executed, or background tasks still running. This Skill performs a session-closeout audit so nothing silently falls through the cracks. ## Core Features & Use Cases - Session-scoped audit: Checks in-flight subagents and background tasks, git status and unpushed commits, PRs touched this session, and promised-but-undone work — strictly limited to what this conversation created or committed to. - Command-verified findings: Verifies state with real commands (git status, git log) rather than recalling from memory, keeping pre-existing repo noise out of scope. - Strict verdict format: Returns either a one-line "SAFE TO CLOSE" verdict or a "NOT DONE" table capped at 5 rows with per-item state (now / safe / you) and one next action. - Use Case: After a long debugging and PR session, ask whether it is safe to close the chat and receive a table showing one unpushed commit and a PR awaiting CI, plus the single next action to take. ## Quick Start Ask the agent to wrap up and confirm whether anything is left outstanding before closing this session.