Pause long coding-agent runs and resume after limits reset
A developer built Steward, which stops unattended Claude Code runs at 90% of the usage window and resumes them after the reset with the conversation intact.
Done with Claude Code
What they did
A developer wrote Steward to manage overnight Claude Code runs. Every run goes through one manager that saves the session and a checkpoint: done, in progress, next, files touched, how to verify. At 90% of the five-hour usage window it stops the run and queues a resume for just after the reset. Crashes write a report with the exact resume command.
How it went
The author shows a demo of a run pausing at 91% usage and resuming with its conversation. They say a real Codex task passed, while timings in the launch film are illustrative.
Worth knowing
Automatic context rotation and model fallback only work with Claude. Switching providers starts a fresh chat from the checkpoint rather than the full conversation.
Try it yourself with Claude Code
Work on [task description] in [repo path]. Before you start, write a checkpoint file with five sections: Done, In progress, Next, Files touched, How to verify. Update it after each meaningful step so a future session with no memory could continue. If you are stopped partway, I will resume you and point you at this file. Do not push, deploy or delete anything without showing me the plan first.
Swap the highlighted [parts] for your own details.
Source: X · @CMills_Dev · Oct 18, 2026
Written with AI assistance from the original post and checked against our editorial guidelines. Results are as reported by the source, not independently verified. Spotted a mistake? Tell us.
