Skip to main content

Wrap Google ADK agents as nodes

Stub

This How-to is a stub. Google ADK integration is planned as an adapter pattern; no first-class implementation is shipped yet.

Goal​

You will wrap a Google ADK agent as an ARP capability and execute it as a bounded, policy-gated step.

When to use this​

  • You have an existing ADK agent you trust and want to integrate.
  • You want ARP/JARVIS to remain the outer execution boundary.

Prerequisites​

  • A runnable ADK agent
  • A wrapper endpoint/service
  • Policy rules for external delegation

Steps​

  1. Define a NodeType that represents “delegate to ADK agent”.
  2. Implement the wrapper with strict timeouts and redaction.
  3. Register and test selection/execution under policy.

Verify​

  • Calls are bounded (timeouts/budgets).
  • Outputs are schema-valid and auditable.

Troubleshooting​

  • Unreliable output → add evaluators and multi-trial stability checks.
  • Too much authority → downscope tokens and enforce external egress policy.
  • Debugging hard → store intermediate evidence as artifacts.

Cleanup / Rollback​

  • Deprecate the node type and revoke credentials.

Next steps​