monitor

name	monitor
description	Check experiment status and diagnose failures via exp-manager. Triggers on "check status", "看看实验", "how's the run", "any failures", "is it done", "monitor experiment", "experiment status".
disable-model-invocation	true

Monitor

One-command experiment status check via @exp-manager.

Instructions

When this skill is invoked with optional <exp_id>:

Step 1: Determine target experiment

• If argument provided → use as exp_id
• If no argument → read .pipeline-state.json for current_exp
• If current_exp is null and no argument → ask user which experiment

Step 2: Verify experiment exists

• Check exp/{exp_id}/ directory exists
• Check exp/{exp_id}/results/runs.log exists
• If not → tell user: "Experiment {exp_id} not found. Run /labmate:new-experiment first."

Step 3: Delegate to @exp-manager

Check experiment {exp_id} status. Read exp/{exp_id}/results/runs.log and .pipeline-state.json. Diagnose any failures, auto-retry if appropriate, report status.

Step 4: Suggest continuous monitoring

After the check completes, tell user:

For continuous monitoring, run: /loop 5m /monitor {exp_id}