name: long-running-tasks description: Handle tasks that may exceed tool timeouts (model training, large builds, data processing).

Long-Running Tasks

Use when

• Running tasks that may take more than 10 minutes (model quantization, training, large builds)
• Starting processes that should continue even if the connection drops
• Monitoring background jobs and reporting progress

Don't use when

• The task completes quickly and fits within the normal tool timeout window.
• The task must be interactive (requires continuous prompts or TTY input).

Outputs

• Log files (e.g., task.log, build.log) should be saved in artifacts/ when possible.
• Final outputs should be written to artifacts/ for easy retrieval.

Templates or Examples

• Use the command patterns in the “Commands” and “Examples” sections as templates.

Strategy

For any task that might exceed the Bash tool timeout (10 minutes):

1. Run in background with output logging
2. Return immediately to confirm the task started
3. Check periodically and report progress
4. Confirm completion when done

Commands

Start a long-running task

# Use nohup to persist after session ends, redirect all output to log
nohup <command> > task.log 2>&1 &
echo "Task started with PID $!"

Check if task is still running

# Check by process name
pgrep -f "<command_pattern>" && echo "Still running" || echo "Completed"

# Or check the log for completion indicators
tail -20 task.log

Monitor progress

# Watch log file for updates
tail -f task.log  # (use with timeout or Ctrl+C)

# Or get last N lines
tail -50 task.log

Examples

Model Quantization

# Start quantization in background
nohup python quantize.py --model GLM-4.7-flash --format NVFP4 > quantize.log 2>&1 &
echo "Quantization started. Check quantize.log for progress."

Then periodically:

tail -30 quantize.log
pgrep -f "quantize.py" && echo "Still running..." || echo "Process completed!"

Large Build

# Start build in background
nohup cargo build --release > build.log 2>&1 &
echo "Build started with PID $!"

Check progress:

tail -20 build.log

Data Processing Pipeline

# Start pipeline
nohup ./process_data.sh input/ output/ > pipeline.log 2>&1 &

# Check progress (if script outputs progress)
grep -E "Progress|Completed|Error" pipeline.log | tail -10

Best Practices

1. Always use nohup - Ensures task survives if connection drops
2. Redirect both stdout and stderr - Use > file.log 2>&1
3. Save the PID - echo $! right after starting
4. Check periodically - Every 5-10 minutes for long tasks
5. Look for completion markers - grep for "done", "error", "completed"
6. Clean up - Remove log files after confirming success

Reporting to User

When starting a long task, tell the user:

• What command was started
• Where logs are saved
• How to check progress manually
• Estimated completion time (if known)

When checking progress, report:

• Current status (running/completed/failed)
• Recent log output (last 10-20 lines)
• Any errors or warnings seen

When task completes:

• Confirm success or failure
• Summarize results
• Clean up temporary files if appropriate