One interruption should not send the whole job back to the start.
Your computer restarts. An agent fails. A tool is unavailable. Or you simply come back tomorrow. Bellfly keeps the work history so you can see what is finished and continue from a known point.
Come back to the work, not another explanation.
Suppose an agent was processing a folder of documents when the computer stopped. Some outputs exist, others do not. Bellfly starts by finding out which is which.
Read the saved work and identify what has already finished.
Check the actual files, available tools and current permissions.
Keep the usable results; repair the affected part or continue the remaining work.
Check the combined output and bring it back for review.
Decide what should happen next.
Pause when you need to think. Stop future work when the goal no longer fits. Add an instruction or a new file when the project changes.
Continue from completed work
Bellfly checks whether the saved progress still fits the current files and tools before using it.
Try an affected step again
A failed step can be retried without hiding its earlier failure or rerunning unrelated finished work.
Check before repeating an outside action
If a message may already have been sent or a change published, Bellfly checks the external state before doing it again.
Stopping future work keeps the history intact.
Stop does not undo files already changed or external actions already performed. The record shows what happened, what remains uncertain and what needs your decision before continuing.