How to use the Agent Platform Model Registry agent skill

What does the skill make the agent do?

An agent skill is a set of instructions that an AI agent loads so it can handle a specific kind of task. Once this skill is installed, the agent knows how to work with models stored in the Agent Platform Model Registry on Google Cloud.

The skill covers listing models, looking up model details, uploading new models or new versions, updating metadata, and deleting models. It runs its work through gcloud commands and Python SDK code.

• List models and find their numeric IDs
• Describe a model or a specific version
• Upload a new model or version
• Update model metadata
• Delete a model

How do you install the skill?

You can install the skill for a single user or for one project. Download the ZIP file from the skill's page on this site and unzip it. Then place the unzipped folder in one of these locations:

• Your personal skills folder: ~/.claude/skills/
• A project's skills folder: .claude/skills/ inside the project
• Or upload the skill in the Claude apps under Settings, then Capabilities, then Skills

How do you trigger the skill with a request?

You do not need a special command. Describe the registry task in plain language and the agent will use the skill when the request matches it. The skill is meant for uploading, listing, describing, updating or deleting models and their versions in the Agent Platform Model Registry.

Include the details the agent will need, such as the project and region, or the model ID. If you leave out the model ID for an update or describe request, the agent is instructed to ask you for it or offer to list models first.

• Show me the newest models in my registry.
• Describe model 1234567890 in project my-project, region us-central1.
• Upload a new version of my fraud model and show me the command before running it.
• Delete the model with this ID, after I confirm.

What happens with the safety confirmations?

The skill sorts actions into three tiers. Read-only actions such as list and describe run immediately. Uploads and updates are reversible but change the registry, so the agent must first show the exact gcloud command and wait for you to reply 'Yes' in a later turn. Deletions are irreversible, so the agent asks you to type a confirmation, such as 'I confirm', in a separate turn before it deletes anything.

If you reject a prompt or change a parameter such as the display name, the agent must not run the old command. It should present a new confirmation with the updated command.

What do you need before you start?

The skill can only run if your environment is ready. You need the following:

• The gcloud CLI installed and authenticated to a Google Cloud account that can access the target project
• A Google Cloud project ID and region, either set in your session or available to the agent
• Python 3 with the google-cloud-aiplatform library, used by the display-name search script
• Cloud Storage access for stored model artifacts. Standard storage charges apply

How does the agent list and describe models?

Listing always uses a limit so that large registries do not return thousands of models. Results come back most recently updated first, so the default list shows the 50 newest models. To find a model by display name, the skill uses a Python search that filters on the server.

To describe a model, the agent runs a gcloud command with the model ID. To target a specific version, it uses the model ID followed by the version ID.

• Model Registry operations use the region and project you give, and the agent passes both on every gcloud ai models command
• Registry operations do not provision serving compute, so the skill states there are no serving compute charges. Standard Cloud Storage pricing still applies to model artifacts

What does the skill not do?

This skill is not for model training or for deploying models to endpoints. Use a different skill for those tasks. It also does not manage models outside the Agent Platform Model Registry.