Experiments reads per-task run telemetry under ~/.agents/sentinel/experiments/<task>/ (runs, success rate, sparklines, optional synthesis.md). Use it to spot flaky or regressing scheduled tasks.

If another browser or Cursor’s preview is missing tabs like this one, compare the build stamp next to the freshness dot in the header, or open /api/meta on the same host and port.

Shows live output from the Sentinel anomaly detectors (sentry-anomaly, shopify-anomaly, ga4-anomaly-check). Health status, baseline window, and any current WARN/CRITICAL findings are pulled straight from the per-detector state files in ~/.agents/sentinel/.

The shared findings bus (~/.agents/sentinel/shared-findings.jsonl) is how tasks broadcast cross-cutting observations — API changes, anomalies, auth issues, rate-limit discoveries — so downstream tasks see them on their next run.

Entries auto-expire after their ttl_hours. Use "Prune" to compact expired rows immediately.

Per-task learned memory at ~/.agents/sentinel/experiments/<task>/memory.md. Tasks can read and append to their own memory to build up long-term learned context across runs.

Pick a task below to see its memory.md, recent quality scores, and synthesis notes.