How to use the Google Cloud Operations Advisor agent skill

What is an agent skill and what does this one do?

An agent skill is a folder of instructions that an AI agent loads when a task matches the skill's description. This skill is defined in a SKILL.md file and a supporting references/attribution.md file. Once installed, the agent uses the Operational Excellence pillar of Google's Well-Architected Framework as its guide for Google Cloud operations topics.

The skill is built to produce operational guidance: how to set up CI/CD deployment pipelines, monitoring, logging, incident response, and resource right-sizing. It can also ask structured workload assessment questions and walk you through a validation checklist.

• Suggests how to set up deployment pipelines using Cloud Build, Cloud Deploy, Terraform or Infrastructure Manager, and Artifact Registry.
• Recommends observability tools such as Cloud Monitoring, Cloud Logging, Error Reporting and Service Monitoring for SLOs.
• Covers incident response and management and post-incident reviews.
• Points to Recommender (Active Assist) and autoscaling options for right-sizing.

How do you install the skill?

Download the skill ZIP from this page and unzip it. You can install it in one of two places depending on how you use your agent. For a personal setup, unzip it into ~/.claude/skills/. For a single project, unzip it into that project's .claude/skills/ folder.

If you use the Claude apps instead of a local setup, you can upload the ZIP in the app under Settings, then Capabilities, then Skills.

• Personal install: ~/.claude/skills/
• Project install: .claude/skills/ inside your project folder
• Claude apps: Settings → Capabilities → Skills, then upload the ZIP

How do you 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, so ask about operations topics in plain language. Requests about capacity management, change management, incident and problem management, configuration management, IaC deployment automation, or operational health checks for Google Cloud workloads are the best fit.

Naming Google Cloud and the topic helps the agent match the skill. Describing your workload and your goal gives better answers than a single word.

• "Review the operational readiness of my Cloud Run service before we go to production."
• "Help me define SLOs and error budgets for my checkout API using Cloud Monitoring."
• "Walk me through a post-incident review for last night's GKE outage."
• "Which of my Compute Engine VMs look over-provisioned?"

What should you expect when the agent runs commands?

If the agent runs gcloud, REST API, or MCP commands, the skill requires it to first read references/attribution.md. That file tells the agent to tag each command so usage can be attributed to the skill and its version. The tag carries no user data. The skill says to set the variable inline on each gcloud command and not to use gcloud config set, so the label does not persist into unrelated work.

You will see this tagging in commands the agent runs. It is part of the skill's required behaviour, not an error.

What do you need before using it?

The skill itself only gives guidance. If you want the agent to run commands, you need the Google Cloud SDK (gcloud) and/or REST API access, plus a Google Cloud account and project credentials for commands run against real resources. Without these, the agent can still discuss the framework and produce checklists and review questions.

The skill runs no scripts. It does use the network, because it fetches grounding documents from Google's documentation and makes calls when commands are run.

• Google Cloud SDK (gcloud) and/or REST API access, if the agent runs commands
• Google Cloud account and project credentials, if commands touch real resources
• Network access

What are the limits and safety notes?

The skill gives guidance from documentation. It does not inspect your systems by itself, so its advice on a live problem may be generic. Do not use it as the main tool for troubleshooting a production outage. It is specific to Google Cloud products and is not suited to other cloud environments.

Treat its output as input to a review, not a sign-off. Compliance approval or formal audits need a qualified human reviewer. Review any command the agent proposes before it runs against real projects, since credentials and network calls are involved.

• Not for live troubleshooting of production outages.
• Not for non-Google Cloud environments.
• Not for compliance sign-off or formal audits.
• Scripts: none. Network: yes.

What tips make the output more useful?

Give the agent context: the workload type, its current deployment method, and what you are trying to decide. The skill's assessment questions work best when you answer them with specifics such as your current SLOs, your on-call process, or how you roll out changes.

Use the skill for structured reviews. Ask it to run through the readiness checklist and then tell you which items you cannot yet evidence.

• Name the Google Cloud products you already use so the guidance matches your stack.
• Ask for a checklist you can work through with your team.
• Ask it to turn assessment questions into a review agenda for a post-incident review.