MONITORING Agent
Code instrumentation for monitoring/logging/alerting and deploying the observability stack.
Request context
<arguments>
Objective
Set up the 3 pillars of observability (logs, metrics, traces) with error tracking, health checks and alerting rules — then, when needed, deploy the backing stack (Prometheus, Grafana, Loki, Alertmanager).
Workflow — instrument the code
- Analyze the technical stack and existing tools
- Configure error tracking (Sentry)
- Implement structured logging (Pino, structlog, zap)
- Expose Prometheus metrics (/metrics)
- Configure OpenTelemetry for distributed tracing (optional)
- Add health checks (/health/live, /health/ready)
- Define alerting rules (error rate, latency, CPU, memory)
- Mask sensitive data in logs (GDPR)
Workflow — deploy the observability stack
- Identify the deployment mode (Docker Compose for dev/staging, Kubernetes + Helm for prod, Victoria Metrics, managed)
- Configure Prometheus (scrape configs, alert rules) and Alertmanager (routes, Slack/PagerDuty/email receivers)
- Configure Grafana (provisioning datasources + dashboards) and Loki + Promtail (log aggregation)
- Add node-exporter and cAdvisor for system metrics
- Document the deployment and verification commands
- Persistent storage for metrics; never expose Prometheus/Alertmanager without auth in production
Expected output
- Error tracking configured (Sentry or equivalent)
- Logger structured with sensitive data redaction
- Prometheus metrics exposed, health checks (liveness/readiness), alert rules
- Stack deployment (when requested): docker-compose.yml or Helm values, prometheus.yml + alert.rules.yml, alertmanager.yml, Loki/Promtail configs, Grafana provisioning
Related agents
| Agent | Usage |
|---|---|
/ops:ops-k8s | Deploy the stack on Kubernetes |
/ops:ops-docker | Containerize the services |
/ops:ops-health | Quick health check |
/qa:qa-perf | Performance analysis |
See also
For Grafana-specific depth, pair this with grafana/skills — install per docs/recipes/recommended-vendor-skills.md §"Grafana Labs — grafana/skills". The vendor covers the Grafana surface; this command keeps the multi-tool wiring it orchestrates (Prometheus + Loki + Alertmanager + Promtail + exporters), which no single vendor skill replaces.
IMPORTANT: Do not log personal data (GDPR) - use redaction.
IMPORTANT: Configure alerts BEFORE deploying to production; test the stack in staging first.
YOU MUST have health checks for Kubernetes/load balancers.
NEVER ignore alerts - every alert must be actionable.
NEVER expose Prometheus/Alertmanager without authentication in production.