AI coding agent interruption recovery workflow: resuming the work without losing context or spraying new changes

Interrupted coding-agent work becomes risky when the resume path starts with guessing what already happened.

Longer coding-agent tasks are rarely continuous. A session dies, a terminal disconnects, a reviewer interrupts the run, or a new priority steals the window. Without a recovery workflow, the next run spends time rediscovering context and often reopens the same files with weaker situational awareness. An AI coding agent interruption recovery workflow preserves the working state explicitly. The task goal, current branch or diff, files already touched, test state, blocker, and next safest command should all be captured before the session ends or as soon as it resumes. The model can summarize what remains. The resume path should still be constrained enough that the restarted work does not widen the diff or forget why earlier decisions were made.