Post Deploy Monitor

Post-deploy Monitor

Status: scaffold (full implementation queued)

This skill is registered in pilot's routing table (Phase 8.5 Post-deploy) but its full content is queued. Until then, redirect per use case.

Redirect for now

• Manual log scrape → use diagnose to investigate any reported anomaly post-deploy.
• Error rate / latency check → if your team has a Grafana / Datadog / Sentry dashboard, open it manually; pilot doesn't yet wrap those APIs.
• Rollback decision → use diagnose to assess whether a rollback is warranted, then run your team's documented rollback procedure.

State to the user: "post-deploy-monitor is scaffold-only; using diagnose for any anomaly investigation."

When the full skill is implemented, it will cover

1. Pre-deploy baseline capture

• Before Ship completes, record:
  • 7-day p50 / p95 / p99 latency for the affected routes
  • 7-day error rate for the affected routes
  • 7-day log-volume baseline
• Store baseline in .pilot/post-deploy/<commit-sha>.json

2. Post-deploy delta watch

• For 15 / 30 / 60 minutes post-deploy, poll the metrics source
• Compute deltas against baseline
• Threshold alerts:
  • Error rate > baseline × 1.5 → CAUTION
  • Error rate > baseline × 3 → BLOCK (auto-suggest rollback)
  • p99 latency > baseline + 100ms → CAUTION
  • p99 latency > baseline × 2 → BLOCK

3. Log diff

• Sample logs from the affected services pre- and post-deploy
• Surface any new error patterns (ERROR / FATAL log lines not present in baseline)

4. Rollback recommendation

• Combine signals from sections 2 and 3
• Output one of: STABLE (no action), WATCH (continue monitoring), ROLLBACK (revert immediately)

Output artifact

Full skill writes POST-DEPLOY.md with the deltas, log diffs, and recommendation. Updated every 5 minutes for the first hour.

Acceptance criteria for the full skill

• Reads metrics from at least: Prometheus, Datadog, Sentry, CloudWatch, Grafana (configurable per-project via .pilot.json).
• Reads logs from at least: stdout via kubectl logs, CloudWatch Logs, Datadog Logs, Loki.
• Surfaces a recommendation within 15 minutes of deploy completion.
• Has fixture tests covering: baseline-vs-spike error rate, latency regression, new error pattern in logs.

Triggers (final list)

• Literal: "monitor", "after deploy", "did the deploy work", "rollback", "post-deploy", "did anything break"
• Phase ordering: fires automatically after Phase 8 Ship completes; runs for 60 minutes by default

Full content scheduled for a follow-up session.