Troubleshooting
No Console Output in UI
If the "Console Output" box in the UI is empty but the agent seems to be running:
- Check Backend Terminal: Look at the terminal where you started
run_ui.sh. Do you see logs there?- AgentCommander uses Dual Logging: Logs are printed to the terminal and sent to the UI.
- Browser Refresh: If the terminal has logs but the UI doesn't, the WebSocket connection likely dropped. Refresh the page to reconnect.
- Check "Console" Visibility: Ensure you are on the Control tab. The console is located at the bottom of the Control Panel.
"Run Task" Error: Unexpected keyword argument 'no_exec_list'
This error indicates a mismatch between the running ui_server.py and the installed pylib.
- Fix: Restart the UI Server (
Ctrl+Cthenbash run_ui.sh). Python code changes in the library require a restart to be loaded.
Metric Not Found / Infinite Score
If the agent keeps failing with "Metric is INF":
- Check Evaluator Output: Use the "White-box Debugging" method to inspect
eval_out.txtin the experiment directory. - Check Print Format: Ensure your
evaluator.pyprints exactlyBest metric: 0.123(Case sensitive: lowercase 'm' in metric). - Check Leakage: If the Anti-Cheating check fails, the score is forced to Infinity. Check if your model is modifying
y_testin place.
Windows WSL2 Issues
- Accessing UI: If
localhost:8080doesn't work, try the WSL IP address (runwsl hostname -I). - Permissions: Ensure your project folder is inside the WSL filesystem (e.g.,
~/projects/), not on the mounted Windows drive (/mnt/c/), for better performance and permission handling.