agent-platform-deploy

What problem does it solve?

This Skill eliminates the manual effort and risk of human error when managing model deployments on Google's Agent Platform, including setting up endpoints, verifying that deployed models serve correctly, and cleaning up resources to avoid unexpected compute charges.

Core Features & Use Cases

• Deploy Models from Multiple Sources: Deploy open models from Model Garden, custom weights stored in GCS, or copy and deploy first-party tuned models to new regions or projects.
• Verify Deployment Health: Test deployed endpoints with sample prompts to confirm they are serving correctly before use.
• Safe Resource Cleanup: Undeploy models and delete endpoints to stop incurring charges, with strict safety confirmation tiers to prevent accidental destructive actions.
• Troubleshoot Deployment Errors: Get guidance for resolving common issues like quota limits or resource exhaustion during deployment.
• Use Case: For example, if you need to test a Gemma 3 model for your application, this skill handles the full deployment, verification, and cleanup workflow without requiring you to manually run and track gcloud commands.

Quick Start

Use the agent-platform-deploy skill to deploy the google/gemma3@gemma-3-27b-it model from Model Garden to a new endpoint in the us-central1 region, verify it works with a test prompt, then undeploy and delete the endpoint when finished.