How to use the Google Cloud Reliability Advisor agent skill

What does the Google Cloud Reliability Advisor skill make the agent do?

Once installed, the agent answers reliability and resilience questions about Google Cloud systems using the Reliability pillar of Google's Well-Architected Framework. It can explain principles such as defining SLOs, building redundancy across zones and regions, graceful degradation, backup and restore testing, and blameless postmortems.

It also asks structured questions about your workload and targets, such as RTO and RPO. The skill is a guidance and assessment aid. It is not a tool that changes cloud resources.

• Explains reliability principles, including user-experience-based reliability, SLO targets, observability, graceful degradation and postmortems.
• Asks questions about how your organisation measures reliability, sets targets and tests recovery.
• Checks an architecture against a validation checklist of reliability recommendations.

How do you install the skill?

You can install the skill for yourself or for a single project. Download the ZIP file from this page, then unzip it into one of these locations:

For a personal install, unzip it into ~/.claude/skills/. For a project install, unzip it into the project's .claude/skills/ folder. In the Claude apps, you can also upload the skill under Settings → Capabilities → Skills.

• Personal: ~/.claude/skills/
• Project: .claude/skills/ inside your project
• Claude apps: Settings → Capabilities → Skills (upload)

What files come with the skill?

The package contains a SKILL.md file, which holds the skill's instructions, and a references/attribution.md file. The attribution file sets out how commands the agent runs are tagged. Keep both files together when you unzip.

• SKILL.md
• references/attribution.md

How do you trigger the skill with natural requests?

You do not need a special command. Ask about reliability in plain language, and mention Google Cloud, your workload and any targets you have. The skill's description covers topics such as uptime, high availability, disaster recovery, RTO/RPO targets, SRE practices, backup strategies, multi-region failover and fault-tolerant architecture.

If you ask about cost, operational excellence or security, the skill is not designed to answer. It also does not handle hands-on deployment or live incident troubleshooting.

• "Review the high availability design of our Cloud Run and Cloud SQL setup against the Reliability pillar."
• "Help me set SLOs and an error budget for our checkout service."
• "Our RTO is 4 hours and RPO is 15 minutes. What questions should we answer before choosing a backup strategy?"
• "Which parts of our multi-region failover design need testing before a reliability review?"

What can you expect from the agent's answers?

The skill uses a set of workload assessment questions and a validation checklist. Expect the agent to ask what your organisation does for each area, such as how it defines reliability from user experience, how it uses redundancy and how it tests recovery from failures and data loss.

Relevant Google Cloud products the skill may mention include Compute Engine Managed Instance Groups, GKE, Cloud Run, Cloud Load Balancing, Cloud Storage multi-region, Cloud SQL High Availability, Spanner, Cloud Monitoring and Backup and DR Service.

• Expect questions before recommendations when your inputs are incomplete.
• Expect links to Google's reliability guidance pages as grounding documents.
• Expect a recommendation to test recovery, but not an actual test run.

What does the skill need, and what should you check for safety?

The listing states that the skill requires nothing beyond the agent itself, runs no scripts and uses no network access. Its own instructions do say that any gcloud, REST API or MCP command the agent runs must be tagged with a usage attribution variable, set inline on each command rather than with gcloud config set.

Because the skill is guidance-only, review any command the agent proposes before you run it against a live project, especially commands that change resources.

• Requires: nothing listed.
• Runs scripts: no. Uses network: no.
• Attribution tagging applies only when the agent runs gcloud, REST or MCP commands.