Skip to main content

Build an atomic node in Python

Stub

This How-to is a stub. The JARVIS stack is still stabilizing the “author a new atomic node” developer workflow.

Goal​

You will create a new atomic capability in Python and make it executable via the ARP Standard Atomic Executor flow.

When to use this​

  • You want to add a new first-party capability to a JARVIS stack.
  • You want a bounded, testable “do one thing” unit you can evaluate and promote.

Prerequisites​

  • Python >=3.11
  • A running JARVIS stack (local dev is fine)
  • A clear decision on how new atomic nodes are delivered:
    • installed node pack (trusted in-process), vs
    • remote-only wrapper nodes.

Steps​

  1. Decide which authoring mode you want:

    • First-party pack: add a new node to arp-jarvis-atomic-nodes and install it alongside arp-jarvis-atomic-executor.
    • Remote-only wrapper: implement an atomic node whose handler makes a bounded outbound call (no RCE).

  2. Implement the node (code + metadata as one unit).

  3. Register the NodeType in Node Registry.

Verify​

  • Node Registry lists your new NodeType.
  • A composite run can select and execute your new node without violating constraints/policy.

Troubleshooting​

  • Node does not show up in registry → the registry seed/sync process is missing or disabled.
  • Execution fails at node.run.execute → policy or constraints are denying the node type.
  • Handler runs but outputs are missing → verify NodeRun.outputs/artifacts are emitted correctly.

Cleanup / Rollback​

  • Remove the node from the pack and re-seed the registry, or roll back the NodeType version.

Next steps​