| name | monitor |
| description | Watch Laboratory signals after a run, analysis change, publication update, CI
change, or repeated agent workflow. Uses local files, CI status, data
integrity checks, and experiment artifacts rather than deploy telemetry.
Trigger: /monitor.
|
| argument-hint | [experiment-path|report-path|--ci] |
/monitor
Laboratory has no production service to watch by default. Monitoring here means
watching the evidence surfaces that can drift after a change:
- CI failures
- flaky tests
- regenerated analysis outputs
- stale
*_latest.csv pointers
- simulated data accidentally feeding publishable reports
- missing trace/provenance artifacts
- broken Spellbook harness bridges
Procedure
- Identify the watched surface: experiment, analysis tool, report, docs, or
harness.
- Pick the signal:
- repo health:
make ci-smoke
- harness health: symlink/marker audit
- experiment health: simulate run plus artifact inspection
- publication health: data provenance and report consistency
- Poll or re-run only as long as useful. This is a bounded local watch, not a
daemon.
- Escalate to
/diagnose on a real regression.
Output
Return monitor.done or monitor.alert with:
- commands run
- artifacts inspected
- signal result
- recommended next step