Agent-to-Agent (A2A) Protocol#

The A2A protocol allows VEX agents to collaborate securely across different server instances. Every inter-agent task is cryptographically verified and recorded in the Merkle audit trail.

How it Works#

Discovery: Agents expose a capability card at /.well-known/agent.json.
Task Creation: A requester sends a TaskRequest via POST to /a2a/tasks.
Verification: The task response includes a Merkle hash of the execution trace.
Polling: The requester can query status via /a2a/tasks/{id}.

Endpoints#

Endpoint	Method	Description
`/.well-known/agent.json`	GET	Capability discovery
`/a2a/tasks`	POST	Submit inter-agent task
`/a2a/tasks/{id}`	GET	Check task status / result

Security#

The protocol uses:

JWT Authentication: Role-based access control for tenants.
Replay Protection: Nonce + timestamp validation.
Merkle Proofs: Ensuring the remote result hasn't been tampered with.

Enforce HTTPS

In production, the A2A protocol requires TLS. VEX will refuse to send task requests over unencrypted channels if VEX_ENV=production.

Example Request#

JSON

{
  "skill": "calculator",
  "input": {
    "expression": "24 * 42"
  },
  "nonce": "a1b2c3d4e5"
}