Cursor Agent stops early: what to check

Read Cursor’s stop status first

Cursor reports completed, aborted, or error in its stop payload. Completed with open boxes keeps the clock-out gate active. The Stop button reports aborted and releases that turn while preserving the punch list. A user close can mark clean closure. An error is not silently converted into a successful completion; inspect Doctor and shift-log.md before restarting.

• A question or command approval wait needs an attended answer or an approved permissions policy.
• A model or service limit cannot be bypassed by Nightshift.
• A sleeping computer cannot run a local recovery worker.

Make Cursor Agent keep working until the list is done

Install the released plugin in Cursor, open the persistent project, run Setup, review rules and the punch list, then Start a small attended shift. Cursor-native skills and hooks guard an IDE shift. For CLI recovery, Cursor currently requires project .cursor/hooks.json; Setup offers to copy the shipped hook file only with your approval.

If the Cursor IDE dies: the CLI handoff

If a Cursor IDE-owned shift dies with enough evidence, Nightshift can start a CLI worker on the same on-disk contract. The IDE conversation ID and CLI session store are different: the CLI worker is not the same chat. Later wakeups can resume the recorded worker; the old tab receives an attach-or-stop instruction. For example, two finished boxes and one open box remain exactly that state across the handoff, but the owner must review the worker's output and checks.