Google Agent Skills

The agent-platform-alert-configuration skill is used to configure best-practice alerting policies specifically for AI agents. It leverages OpenTelemetry (OTel) metrics to monitor key performance indicators such as agent latency, error rates, token usage, and quality metrics. The skill generates output in the form of Terraform (.tf) configuration files.

When to use this skill:

• When analyzing, writing, or deploying alerting policies for AI agents.
• When monitoring agent latency, error rates, token usage, and quality metrics.

When NOT to use this skill:

• For standard infrastructure monitoring that is unrelated to AI agents.
• If the agent is not instrumented with OpenTelemetry (specifically for Reliability, Cost, Safety, and Security alerts).

Alerting Types and Requirements:

• Reliability, Cost, Safety, and Security alerts: These use generic OTel metrics and are runtime-agnostic (working across platforms like Cloud Run or Vertex AI).
• Quality alerts: These are strictly bound to Vertex AI deployments and rely on Vertex AI Online Monitors.

This skill provides a structured workflow to design and implement a governed, secure pipeline for agentic analytics. It is specifically designed for workloads involving structured and unstructured data distributed across Google Cloud, on-premises, or other cloud providers (e.g., AWS, Azure).

Data residing outside Google Cloud (such as in Databricks, Snowflake, Salesforce, SAP, or Oracle) is accessed via federation mechanisms like Apache Iceberg, 'zero-copy ETL' methods, or remote query push-down.

The workflow follows four distinct phases:

1. Phase 1: Requirements discovery: Gathering detailed functional and non-functional requirements.
2. Phase 2: Solution architecture: Generating a detailed architecture based on Phase 1 requirements.
3. Phase 3: Solution validation: Creating and running validation plans and scripts.
4. Phase 4: Solution packing and presentation: Consolidating and presenting the final solution.

Workflow Constraints:

• Strict phase separation: During Phase 1, do not recommend or propose any architectural designs, technical decompositions, or cloud services. Only ask clarifying questions.
• Phase skipping: If a user indicates a phase is already complete (e.g., "architecture is confirmed"), skip that phase and proceed directly to the next requested task.

The google-cloud-solution-architecture skill follows a structured, four-phase workflow to design multi-product Google Cloud solutions. This skill is intended for complex, end-to-end architectural guidance and should not be used if specialized product-specific skills (e.g., google-cloud-recipe-*) already address the specific workload.

Workflow Phases

1. Phase 1: Requirements discovery: Gathering functional and non-functional requirements.
2. Phase 2: Solution architecture: Generating product recommendations, diagrams, descriptions, and deployment guidance.
3. Phase 3: Solution validation: Creating validation plans, instructions, and scripts.
4. Phase 4: Solution packing and presentation: Consolidating all generated content.

Core Operational Principles

• Strict phase separation: During Phase 1, no architectural designs or product recommendations are proposed. This prevents premature anchoring on specific tools before requirements are fully understood.
• Iterative approval: Every deliverable (technical decompositions, product recommendations, diagrams, etc.) must be presented to the user for explicit approval before moving to the next task or phase.
• No autonomous execution: The skill will not run generated scripts or code without unambiguous permission. Users are always offered the option to execute commands manually to avoid unintended costs or infrastructure mutations.
• Phase skipping: If a user indicates that a phase or task is already complete (e.g., "requirements discovery is completed"), the skill will skip directly to the requested task.

Google Cloud Storage (GCS) is a managed service for storing data as immutable objects in buckets. It provides strong global consistency and is designed for workloads ranging from static website hosting and backups to data lakes and AI/ML training.

Key Capabilities

• Object Management: Create, configure, upload, download, stream, and transfer data.
• Organization: Use folders and prefixes to organize objects.
• Access Control: Manage permissions via IAM and ACLs, including public access prevention and signed URLs.
• Storage Classes & Tiering: Choose between Standard, Nearline, Coldline, and Archive to balance cost and performance.
• Data Protection: Implement versioning, encryption (CMEK), retention policies (Bucket Lock), object holds, and soft delete.
• Automation & Integration: Trigger Pub/Sub notifications on object changes and mount buckets as a file system using gcsfuse.
• Performance & Scale: Optimize storage performance for high-throughput workloads and manage large-scale data via Storage Insights.

When NOT to use GCS

• Block Storage: Use Persistent Disk instead.
• Data Warehousing/Analytics: Use BigQuery instead.
• Databases: Use Cloud SQL, Spanner, Bigtable, or Firestore instead.

The google-cloud-recipe-foundation-builder skill deploys a baseline landing zone foundation for a Google Cloud Organization. It is designed for setting up a new Google Cloud Organization or establishing a secure, enterprise-grade landing zone foundation.

Note: This skill is currently in a preview state and does not include all advanced features. For more options, use Google Cloud Setup.

Core components provisioned:

• Security Guardrails: Enforces 17 baseline Organization Policies (13 Boolean, 4 List constraints).
• Resource Hierarchy: Establishes four folders (Common, Production, Non-Production, Development) and provisions corresponding projects using unique ID prefixes (logging-, prod-, non-prod-, dev-).
• Billing & API Enablement: Links projects to a billing account and activates logging/monitoring services.
• Centralized Logging & Monitoring: Deploys a global log bucket (30-day retention), an organization-wide audit log sink, and a cross-environment metrics scope.