wrap-continuelisted

# Wrap + Continue A checkpoint, not a retrospective. It exists because the heavy end-of-session flow burns the exact context the user is trying to reclaim — so proportion is a requirement, not a preference. ## Misroute check (one line, then move) If the signal is "done for the day" / "work is complete" / "switching to something else", this is the wrong flow. Say which flow applies and **then run that flow in the same turn** — do not stop and hand the user back a command to type; they already asked to wrap. A wrongly-run heavy flow costs context; a wrongly-run light flow loses capture. Name the signal that decided it in one sentence and get on with it. No pickup prompt in that case. Everything below assumes the task continues. Section 3 and Section 5 below both target 150–350 words with the structure rules given in each section. ## 1. Stop background writers first — ordering is load-bearing Nothing may be observed about repository state until everything that could still be writing has stopped. - List and stop background agents/shells (`TaskList` / `TaskStop`, plus background Bash this session launched). - **Exception: leave long-running processes the next thread will want** — dev servers, booted simulators, watchers. Note the PID, port/UDID and the command so the next thread reuses it instead of starting a second one. - **Then re-check `git status` in every location a stopped writer could have touched** — main tree *and* every worktree (`git worktree list`; worktrees l