What is Grafana Skills

Skills, in Grafana’s Assistant feature set, are a way to capture your team’s operational know-how — “how we actually troubleshoot this service” — as a standardized, shareable document an AI agent can follow, instead of that knowledge living only in one senior engineer’s head.

---

What a Skill actually is

A concrete example: a “Service Troubleshooting” skill that lists the critical services and the canonical diagnostic steps for each — the same content that would otherwise live in a runbook, but written so an agent can execute it, not just a human read it.

Skill document
  ├── When to use this skill (trigger conditions)
  ├── Canonical steps (query X, check Y, escalate if Z)
  └── Optional: MCP actions to trigger in external systems
              (open a GitHub issue, page via Slack, create a Linear ticket)

Skills can be private (yours only) or shared with your team, and they can be configured to trigger What is Grafana MCP tool calls directly — with explicit instructions in the skill content about when and how the agent should use those tools, rather than leaving that judgment implicit.

How this differs from What is Grafana MCP

MCP is the transport — it’s what lets an agent call query_prometheus or create_alert_rule at all. Skills are the procedure layer on top — they tell the agent which sequence of MCP calls constitutes “how we troubleshoot a checkout latency spike” so every invocation follows the same playbook instead of the agent reinventing an investigation path each time.

Skill: "Checkout latency triage"
   │
   ▼
1. Query p99 latency for checkout-service (via grafana-mcp)
2. If p99 > SLO threshold → pull traces for the slowest 1% (Tempo)
3. Check for a recent deploy correlated with the spike
4. If found → open a GitHub issue with the trace links (MCP action)
5. Else → escalate via Slack (MCP action)

gcx: agentic-first Grafana Cloud CLI

Alongside Skills, Grafana shipped What is gcx — a CLI built specifically for agentic usage rather than interactive human use. It comes with 22 bundled skills for Claude Code and other .agents-compatible harnesses out of the box, covering:

• Querying PromQL, LogQL, and traces directly from a terminal/agent context
• Managing dashboards, SLOs, alerting, and Synthetic Monitoring without leaving the CLI

This effectively packages the “how a competent SRE uses Grafana” knowledge as install-and-go skills, rather than requiring every team to author their own from scratch. See What is gcx for the full command surface and how it compares to What is Grafana MCP.

Where it fits

Layer	Role
What is Grafana MCP	Transport — exposes Grafana as callable tools
Grafana Skills	Procedure — standardizes which tools, in what order, for a given investigation
What is gcx	Distribution — ships a starter set of skills pre-bundled, plus its own CLI transport

Why it’s on the backlog: this is the mechanism to encode existing runbooks/playbooks (the kind already tracked under g-reliability/) as agent-executable procedures, rather than leaving an SRE agent to reason from first principles on every incident.