monitor-experiment

name: monitor-experiment description: Monitor Beaker experiments until completion. Use when the user asks to monitor, check, or track a Beaker experiment. allowed-tools: Bash(beaker:*)

Monitor Beaker Experiment

Finding the experiment ID

If no experiment ID is provided, find the most recent experiment:

# List experiments in a workspace (e.g., ai2/open-instruct-dev)
beaker workspace experiments ai2/open-instruct-dev 2>&1 | head -10

Common workspaces:

• ai2/open-instruct-dev - Development experiments
• ai2/lm-eval - Evaluation experiments
• ai2/tulu-3-dev - Tulu development

Monitoring an experiment

1. Get the experiment status:

beaker experiment get <experiment-id> --format json | jq '.[0] | {id, name, status: .jobs[0].status}'

2. Check the status fields:

   • status.current: "scheduled", "starting", "running", "succeeded", "failed", "canceled"
   • status.exitCode: Exit code (0 = success, non-zero = failure)
   • status.started / status.exited: Timestamps

3. If still running, wait 30 seconds and check again

4. When complete:

   • If exitCode is 0: Report success
   • If exitCode is non-zero: Fetch and display logs

5. Continue monitoring until the experiment finishes or the user asks to stop

Useful commands

# Check status (compact)
beaker experiment get <id> --format json | jq '.[0].jobs[0].status.current'

# Get logs on failure
beaker experiment logs <id>

# Get logs with tail (last N lines)
beaker experiment logs <id> 2>&1 | tail -100

# Open in browser
echo "https://beaker.org/ex/<id>"