Skip to main content

Prerequisites

The following are required for deploying the Agent Harness Security plug-in. Have this information available before deployment.
  • HiddenLayer API key and secret. See Create API Key.
    • API key permissions required: Interactions: read, write; Policies: read; Projects: read.
    • See AI Runtime Security, under API Permission Related to Products or Features table.
  • Create a HiddenLayer Project ID. See Create a Project.
    • After a project is created, you can copy the ID by clicking the copy button.

Deploy the Plug-in

  1. Obtain the plug-in from HiddenLayer. Contact your HiddenLayer sales representative or support.
  2. Deploy the plug-in to your endpoints.
    • For example:
      • Deploy using an MDM application, like Jamf or Microsoft Intune.
      • Manually copy the plug-in file to the endpoint.

Configure and Hook the Plug-in

Configuring the plug-in connects it to your tenant in the HiddenLayer Console. Hooking the plug-in secures your AI coding agents. You can unhook the plug-in when needed.
  1. Set the Client ID. This is the HiddenLayer API key. Replace my-client-id with the API key.
  2. Set the Client Secret. This is the API secret. Replace my-client-secret with the API secret. The secret is routed to the OS secret store. The value is not echoed.
  3. Set the Platform URL. This is the region for the Console that you log in to. Note: The following example is for the EU region. The US region is the default region.
  4. Optionally, set the Project ID if you are using a project other than the default. Replace my-project-id with the project ID you want to use.
  5. Check the status of the plug-in to verify the current configuration.
  6. Verify the configuration authentication.
  7. Install the hooks for all supported harnesses.
System-Level InstallationFor system-level installation, run with sudo. This is supported for Copilot and Cursor.

Unhook the Coding Agents

  1. To remove the hook configurations, run the following command.

Validation

After setup, send a test command. Traffic should appear in the console filterable by requester / project ID, showing ping, tool use, write interactions, and all agent activity.