Skip to content
Get packs

Guide · Agent skill

How to use the Model Armor Template Creator agent skill

Updated · written with AI from the project's own documentation and checked against our catalogue data

The Model Armor Template Creator is an agent skill from the googleworkspace/cli project that teaches an AI agent to use the gws command-line tool to create a named Model Armor template in a Google Cloud project. Model Armor is a Google Cloud service that screens prompts and responses for things like jailbreak attempts. This guide explains what the skill does, how to install it, how to ask for it, and what to check before anything is written to your cloud account.

Model Armor Template Creatorgoogleworkspace/cli · Apache-2.0

Quick start

  1. 1Install the gws CLI. Make sure the gws command-line tool is installed and available on your system.
  2. 2Configure authentication. Set up auth using the gws-shared/SKILL.md instructions, or run gws generate-skills if that file is missing.
  3. 3Download and unzip the skill. Download the ZIP from this page and unzip it to ~/.claude/skills/ or a project's .claude/skills/ folder.
  4. 4Make the skill visible to your agent. Confirm the skill is in place, or upload it under Settings → Capabilities → Skills in the Claude apps.
  5. 5Make a natural request. Ask the agent to create a Model Armor template, naming the project ID, location and template ID.
  6. 6Review the command. When the agent proposes the write command, check the project, location and template ID before confirming.
  7. 7Use the template. Reference the resulting template name with +sanitize-prompt and +sanitize-response.

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

An agent skill is a folder containing a SKILL.md file that gives an AI agent instructions for a specific task. When the skill is installed and relevant, the agent reads those instructions and follows them. This skill covers one task: creating a new Model Armor template.

Once installed, the agent can set up a template with a chosen template ID in a given Google Cloud project and location. It can use the built-in jailbreak preset or a custom JSON configuration. The skill tells the agent to confirm with you before running the write command.

What you need before installing

The skill cannot work without its prerequisites. You need the gws command-line tool (the googleworkspace CLI), a Google Cloud project ID and a location with Model Armor access, and authentication configured through the gws shared skill at gws-shared/SKILL.md. If that shared file is missing, the SKILL.md says to run gws generate-skills to create it.

  • gws command-line tool installed
  • Google Cloud project ID with Model Armor access in your chosen location
  • Authentication set up via gws-shared/SKILL.md

Install the skill from the ZIP

Download the ZIP for this skill from this page and unzip it. Place the resulting skill folder in your personal skills directory at ~/.claude/skills/, or in a project's .claude/skills/ folder if you want it limited to one project.

Alternatively, in the Claude apps, open Settings, go to Capabilities, and upload the skill under Skills.

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

Trigger the skill with a natural request

You do not need to type a command yourself. Describe what you want in plain language and name the project, location and template ID. The agent should recognise a request to create a Model Armor template and use this skill. Be specific, because the agent needs the project ID, the location and the template ID to build the command.

The command the agent builds has this form, taken from the skill:

  • Create a Model Armor template called my-tmpl in project my-project at us-central1 using the jailbreak preset.
  • Make a Model Armor template with ID support-bot-screen in my GCP project in us-central1 from this JSON configuration: ...
gws modelarmor +create-template --project <PROJECT> --location <LOCATION> --template-id <ID>

Flags and options the agent uses

--project and --location and --template-id are required. --location takes a GCP location such as us-central1. --preset accepts one value in the source: jailbreak. --json takes a JSON body for the template configuration and overrides --preset when both are given.

If neither --preset nor --json is given, the skill defaults to the jailbreak preset.

  • --project: GCP project ID (required)
  • --location: GCP location, for example us-central1 (required)
  • --template-id: template ID to create (required)
  • --preset: use a preset template, currently jailbreak (optional)
  • --json: JSON body for the template configuration, overrides --preset (optional)
gws modelarmor +create-template --project P --location us-central1 --template-id my-tmpl --preset jailbreak

Safety notes: scripts, network and write access

This skill runs scripts and uses the network. Because it is a write command that creates real resources, the skill instructs the agent to confirm with you before executing it. Read the proposed command, check the project ID and location, and only approve it when they are correct.

Creating a template can incur cost or affect production settings in your project. If you only want to screen content without touching cloud resources, this skill is not the right tool. Use the template name afterwards with +sanitize-prompt and +sanitize-response, as the skill's tips note.

  • Always check the project ID and location before approving.
  • The template is created in a real Google Cloud project.
  • Do not run it against production unless you intend to.

Tips for smooth use

Give the agent the exact project ID, location and template ID in one request so it does not have to guess. Use a template ID you can recognise later, because you will need the template name when calling the sanitize commands.

If you need a custom configuration, supply it with --json. Otherwise the jailbreak preset is the default.

  • Pick a clear template ID such as my-tmpl so you can find it later.
  • Use the template name with +sanitize-prompt and +sanitize-response.
  • See the gws-modelarmor skill for all filter user-generated content for safety commands.

What people use it for

  • Create a jailbreak-screening template in a development Google Cloud project before wiring it into an LLM app.
  • Stand up a named template for a security team's prompt-screening setup in a specific GCP location.
  • Create a template from a custom JSON configuration you have already written.
  • Prepare a template ID that your application code later passes to the sanitize commands.

Frequently asked questions

How do I install the Model Armor Template Creator skill?

Download the ZIP from this page and unzip it into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder. In the Claude apps you can instead upload it under Settings → Capabilities → Skills.

What do I need to use this skill?

You need the gws command-line tool, a Google Cloud project ID and location with Model Armor access, and authentication configured through gws-shared/SKILL.md. The skill cannot work without these.

Is this skill free?

The skill is licensed under Apache-2.0. Creating a Model Armor template in Google Cloud may incur cost, so check your own project's billing before running it.

What is the difference between --preset and --json?

--preset uses a built-in template, and the only value listed in the source is jailbreak. --json takes a custom JSON body for the template configuration and overrides --preset if both are given.

What happens if I do not pass a preset or JSON?

The skill defaults to the jailbreak preset when neither --preset nor --json is given.

Does the agent create the template without asking me?

No. The skill marks this as a write command and tells the agent to confirm with you before executing it.

Can I use this skill to screen content without creating anything in my cloud project?

No. The skill creates a template in a Google Cloud project, so it is not suitable if you want to screen content without touching cloud resources.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

Use this skill if you build LLM applications on Google Cloud, or manage prompt-screening setup for a project, and already use the gws CLI. Pick something else if you do not have a Google Cloud project or gws installed, or if you only want to screen content without creating cloud resources.

Download the skill