Guide · Research

Watch a long run from your Lock Screen

Have your agent launch or watch a training run, simulation or pipeline, and keep a live card with the progress and the metric you care about.

1. The card only moves when the agent sends an update, so the agent session has to outlive your laptop lid. Start it inside tmux or screen on the machine running the job:
   tmux new -s run

   Copy
   Then start Claude Code (or your agent) inside it. Detach with Ctrl-b then d; the session keeps running.

1. Paste this, with your script and metric:
   Run python train.py and watch its log. Keep a Startr card updated each epoch with the epoch count, validation loss and an ETA. Mark it warn if validation loss rises three epochs in a row, alert if the job crashes, and end it with the best result.

   Copy
2. Already running? Point the agent at the log instead:
   My training job is already running and writing to logs/run3.log. Watch that file and keep a Startr card updated each epoch with the epoch, validation loss and an ETA. End the card when the log says training finished.

   Copy

1. Each card lasts up to eight hours. Add this line so the agent hands off cleanly:
   If the run will take longer than seven hours, end the card with the current status before seven hours and start a new card to continue.

   Copy