| name | grafana-prd |
| description | Grafana production environment specialist for metrics, logs, traces, dashboards, and alerting. Use when asked to: query Prometheus metrics, search Loki logs, analyze Tempo traces, manage dashboards, configure alert rules, or investigate issues in the production environment. Trigger keywords: grafana-prd. Do NOT use for test environment (use grafana-tst instead).
|
| compatibility | Requires grafana-prd agent (.codex/agents/grafana-prd.agent.md) and grafana-prd MCP server. |
Grafana Test Environment Skill
Agent Delegation
Delegate ALL operations to the grafana-prd agent.
"Use the grafana-prd agent to complete this task."
The grafana-prd agent provisions the grafana-prd MCP server and has the required
tools (list_datasources, query_prometheus, query_loki_logs, search_traces, etc.).
Description
Access Grafana production environment for observability data. Covers metrics (Prometheus),
logs (Loki), traces (Tempo), profiling (Pyroscope), incident management, OnCall schedules,
Sift investigations, alerting, and dashboard management. Use for production workflows.
Capabilities
Metrics (Prometheus)
- Query PromQL against test Prometheus datasources
- List metric names, label names, label values, and metadata
- Query histograms and percentiles
Logs (Loki)
- Execute LogQL queries for test log retrieval
- List log label names and values
- Query log patterns and statistics
- Stream log data with filtering and parsing
Traces (Tempo)
- Find slow requests in test environment
- Analyze distributed traces
Profiling (Pyroscope)
- Fetch CPU, memory, goroutine profiles for test services
- List profile types and label values per service
Incidents & OnCall
- Create, list, and get incident details
- Add timeline notes to incidents
- List OnCall schedules, teams, users
- Get current on-call users for a schedule
Sift Investigations
- Create and retrieve automated investigations
- Find error patterns in logs
- Find slow requests across services
Alerting
- List, get, create, update, delete alert rules
- List contact points and notification policies
- List alert groups from OnCall
Dashboards
- Search, get, create, update dashboards
- Get panel queries and dashboard summaries
- Generate deeplinks to dashboards or panels
- Render panel/dashboard images (PNG)
- Create and manage annotations
Datasources
- List, get datasources by name or UID
Key Concepts
- Datasources: Named connections to backends (Prometheus, Loki, Tempo, Pyroscope). Always resolve UIDs via
list_datasources before querying.
- Prometheus / PromQL: Metrics backend. Use
list_prometheus_metric_names before writing PromQL to ensure valid metric names.
- Loki / LogQL: Log aggregation backend. Always call
list_loki_label_names / list_loki_label_values before building LogQL queries.
- Tempo: Distributed tracing backend. Use to find slow spans and trace request paths across services.
- Dashboards: JSON-defined panels; use
get_dashboard_summary for large dashboards to avoid token limits.
Workflow
- Connectivity check — call
list_datasources (with limit=1) as a lightweight probe. If it fails, stop and report: "grafana-prd MCP server is unavailable. Cannot proceed."
- Discover datasources — resolve correct datasource UIDs using
list_datasources.
- Discover schema — list metric names, label names/values, or log labels before querying.
- Query or manage — use
query_prometheus, query_loki_logs, search_traces, dashboard/alert tools as appropriate.
- Report — structure the response as: datasource → time range → results summary → anomalies/patterns → recommended actions.
Usage Examples
Test Metrics
grafana-prd Show CPU usage for test-api service in production
Log Investigation
grafana-prd Search Loki logs for errors in auth service last 1 hour
Dashboard Editing
grafana-prd Add a new panel to the production overview dashboard
Alert Testing
grafana-prd Create a test alert rule for high memory usage
Trace Analysis
grafana-prd Find slow requests for checkout service in test environment
Best Practices
- Always call
list_datasources first to get datasource UIDs before any query
- Always call
list_prometheus_metric_names before writing PromQL
- Always call
list_loki_label_names / list_loki_label_values before writing LogQL
- Use
query_loki_stats to check log volume before fetching raw entries
- Use
get_dashboard_summary instead of get_dashboard_by_uid for large dashboards
- Do NOT search the filesystem for
mcp-config.json — always route through the agent file
Limitations
- Test environment only — data does not reflect production
- Different data retention policies than production
- Write operations (create/update alerts, incidents) require appropriate permissions
- Image rendering requires Grafana Image Renderer service
Environment Isolation
CRITICAL: Never use grafana-prd and grafana-tst in the same request. Choose only one environment per query.
