How to use the Prometheus Monitoring Expert agent skill

What does the prometheus skill make the agent do?

Agent skills are folders of written instructions that an agent loads when a task matches them. This skill is a single file, SKILL.md, that tells the agent to act as an observability engineer. It is written guidance, not a tool: it does not connect to Prometheus or Grafana.

The guidance covers PromQL queries, alerting rules and Alertmanager routing, recording rules, controlling metric cardinality, Grafana dashboards with template variables, and service discovery. It also points to established methods: the four golden signals (latency, traffic, errors, saturation), the USE method for infrastructure and the RED method for services, and multi-window burn-rate alerts for SLOs.

How do you install the prometheus skill?

Download the ZIP of the skill from this page and unzip it. For a personal setup, place the unzipped folder in ~/.claude/skills/. For a single project, place it in the project's .claude/skills/ folder instead.

You can also upload the skill in the Claude apps: go to Settings, then Capabilities, then Skills, and upload the skill there.

• Personal install: ~/.claude/skills/prometheus/
• Project install: .claude/skills/prometheus/ inside your project
• Claude apps: Settings → Capabilities → Skills

How do you trigger the skill with natural requests?

You do not need a special command. Ask a monitoring question in plain language that mentions Prometheus, PromQL, alerts, Grafana or observability, and the agent will use the skill's guidance. Include your context: which service, which metrics you have, and what problem you are seeing.

Being specific helps. A request that names the metric, the symptom and the goal gives the agent something concrete to work with.

• Ask for a PromQL query for a latency percentile from a histogram metric
• Ask how to write an alert that does not fire on brief spikes
• Ask how to reduce cardinality for a metric with user IDs in its labels

What are good examples of requests?

The skill is most useful for design and review questions. Paste your existing rule or query and ask what is wrong with it. Because the file contains only guidance, the agent cannot read your live metrics, so include any values or config you want it to consider.

• Write a p99 latency query using histogram_quantile over a rate of the bucket metric
• Review this alerting rule and suggest a for: duration and a runbook link
• Propose recording rules to precompute per-job request rates for a dashboard
• Explain how to set group_by, group_wait and repeat_interval to batch related alerts
• Suggest template variables such as $job and $instance for a reusable Grafana panel

What do you need before using it?

The skill lists no requirements. It does not run scripts and does not use the network. You need an agent that supports skills and the metrics or configuration you want to discuss. The agent cannot inspect a running cluster unless you paste the relevant data into the conversation.

What are the safety notes for scripts and network access?

The skill contains only SKILL.md. It runs no scripts and makes no network calls, so it does not change your systems on its own. Any query, rule or configuration the agent writes is a suggestion: review it before you apply it to Prometheus, Alertmanager or Grafana, and test it in a non-production environment first.

What tips make the guidance work better?

Keep label values bounded. The skill warns that labels such as user IDs, request IDs, IP addresses and trace IDs destroy performance, so ask for logs or traces for that data instead. Use rate() rather than irate() in alerting rules, and avoid ranges shorter than two scrape intervals.

Give every alert a for: duration so brief spikes do not page on-call engineers, and make sure each alert has a runbook or remediation path. Finally, monitor the up metric so you know your alerting pipeline is working.

• Use rate() over a range of at least two scrape intervals
• Add a for: duration to every alert
• Do not put high-cardinality labels in metrics
• Watch the up metric to confirm your monitoring is running