| name | observability-readiness |
| description | Verify that a deployment is observable before promotion. Use when DevOps must check logs, metrics, traces, dashboards, alerts, health endpoints, synthetic probes, SLO/error-budget signals, correlation IDs, runbook links, and incident triage evidence without exposing credentials or sensitive telemetry. |
| agents | ["devops"] |
| sdlc_phase | deploy |
| owner | Platform Eng |
| status | active |
Observability Readiness
When to use
Use this skill before staging/production deployment, before enabling traffic, and during release readiness for features whose failure needs operational detection.
Inputs and outputs
- Inputs: architecture/instrumentation contract, service list, health endpoints, expected logs/metrics/traces, dashboards/alerts, target environment, and incident runbook references.
- Outputs: observability readiness verdict, missing telemetry findings, dashboard/alert references, smoke probe evidence, and deploy-report observability fields.
Procedure
- Map critical signals. For each deployed service, identify health endpoint, readiness/liveness signal, key logs, metrics, traces, and business/user-impact indicators.
- Check deploy-time telemetry. Verify the new version emits startup logs, request logs, errors, latency, saturation/resource metrics, background job signals, and dependency health.
- Verify correlation. Confirm request IDs/correlation IDs connect frontend, backend, workers, and external calls where applicable.
- Confirm dashboards and alerts. Record dashboard names/links and alert names/status. Use names or URLs only; do not include tokens or private query credentials.
- Define smoke/synthetic probes. List post-deploy probes for user flows, API health, queues, scheduled jobs, and external integrations.
- Check runbook path. Ensure a human can move from alert to runbook to rollback/mitigation path.
- Emit verdict.
ready, ready-with-gaps, or blocked.
Hard rules
- Never promote a high-risk deployment with no health signal, no alert, and no rollback trigger.
- Never expose sensitive logs, PII, auth headers, trace baggage, API keys, or dashboard tokens.
- Never treat "container running" as observability readiness.
- Never invent SLOs or alert thresholds; record missing contracts as follow-up work.
- Never mark observability ready when incident responders cannot identify service/version/environment from telemetry.
References
