How to use the Service Mesh Observability agent skill

What the skill makes the agent do

Once installed, the agent follows a structured guide for observability in a service mesh. It explains the three pillars of observability: metrics (request rate, error rate, latency P50, saturation), traces (span context, latency, dependencies, bottlenecks) and logs (access logs, error details, debug info, audit trail).

It also gives the agent the golden signals to alert on: latency (P50 and P99 request duration, with an example alert threshold of P99 above 500ms), traffic (requests per second, with anomaly detection), errors (5xx error rate above 1%), and saturation (resource utilization above 80%).

• Best practices such as sampling rates: 100% in dev and 1-10% in prod.
• Avoiding high-cardinality labels by limiting label values.
• Propagating trace context headers consistently, correlating metrics and traces with exemplars, and using hot/cold storage tiers for retention.
• For detailed templates and worked examples, the agent reads an additional reference file, references/details.md, which includes an Istio setup with Prometheus and Grafana and Istio metrics queries.

Install the skill from the ZIP

Download the ZIP for this skill from this page and unzip it. You have two options for where the skill folder goes. For a personal setup, place it in ~/.claude/skills/. For a single project, place it in that project's .claude/skills/ folder.

You can also upload the skill in the Claude apps under Settings → Capabilities → Skills.

After installation, the folder should contain SKILL.md and references/details.md.

• Personal: ~/.claude/skills/
• Project: .claude/skills/ inside your project

Trigger the skill with a natural request

You do not need a special command. The skill's description tells the agent when to use it: setting up mesh monitoring, debugging latency issues, or implementing SLOs for service communication. Describe your goal in plain language and mention your mesh if you have one.

Requests that match the skill's purpose include setting up distributed tracing across services, building mesh metrics dashboards, finding out why calls between two services are slow or failing, defining SLOs for service communication, visualizing service dependencies, and troubleshooting mesh connectivity.

• "Help me set up Prometheus and Grafana monitoring for our Istio mesh."
• "Our checkout service calls to payments have high P99 latency. Walk me through how to find where the time goes."
• "Define SLOs for service-to-service calls in our Linkerd mesh and tell me what to alert on."
• "Which golden signals should I alert on for traffic between our microservices?"

What it needs and what it does not do

The skill lists no required dependencies. It does not run scripts and does not use the network. It is guidance and templates, not an executable setup tool, so it will not configure a live cluster for you.

The skill is conceptual and general. Its guidance targets mesh-specific tooling such as Istio and Linkerd, so it is less useful for a single application that is not running in a service mesh.

• Requires: nothing listed
• Runs scripts: no
• Uses network: no
• The templates are starting points you review and apply yourself.

Tips for getting useful results

Give the agent your mesh type, your monitoring stack, and the specific symptom. The reference file contains concrete templates, so ask for them by topic when you need YAML or queries.

Treat the thresholds in the skill as examples. Tune them to your own traffic and SLOs before you alert on them.

Follow the sampling advice: a high sample rate in production adds storage costs, and high-cardinality labels can make metrics expensive to store.

• Name your mesh (Istio or Linkerd) and your metrics tool in the request.
• Ask the agent to check the template in references/details.md for a worked example.
• Review every manifest before applying it to a cluster.