Monitor
Perform one status cycle per invocation. Recurrence belongs to the host's scheduler.
- Resolve the experiment from the argument or
.pipeline-state.json. - Run
scripts/monitor_exp.sh <exp_id>when available and inspect the relevant log tail and process state. - Follow
../../references/agent-routing.mdfor theexp-managerrole when diagnosis is needed. - Report status, evidence, uncertainty, and the next safe action.
Do not kill, retry, or relaunch work without clear authorization and an exact target. For repeated checks, offer the current host's native scheduler when it has one; otherwise suggest another invocation.