← All documentation

Agent, peer and communication-path checks

This Phase 7 configuration is a template for your running services. Phase 7 requires v0.5.0 or newer; see installation.

Before running agenthealth.yaml:

• Expose the first agent's HEAD/GET health resource at port 8080, implementing the agent health document.
• Expose the A2A peer at port 8081 with an agent card advertising skill answer.
• Implement a safe POST probe handler in the first agent. When it receives downstream: peer, it must contact that peer using its configured direct or A2A integration and report completed success/failure with the same peer ID.
• Replace the optional model-health HTTP URL with your supporting service's health endpoint. This checks HTTP health, not model inference.

AgentHealth implements sending and evaluating probes. This example does not start servers or install the receiving handler. Peer endpoint health alone cannot prove the first agent can communicate with the peer.

go run ./cmd/agenthealth check examples/agent-check/agenthealth.yaml --format json
go run ./cmd/agenthealth doctor examples/agent-check/agenthealth.yaml

The result tree separates first, peer, first-to-peer, and model-health. The peer and path are critical; the model-health dependency is optional. The path probe is explicitly active and executes once. Configure credentials independently for each target when your services require authentication.

Phase 10 compatibility

This nested example remains valid. To share a backend across paths, give that node an explicit id and use ref edges. Each communication path keeps a distinct node and its own checks; edge policy remains independent. See the graph example. Graph fields require v0.8.0 or newer; v0.7.0 binaries reject them.

Kubernetes compatibility

This example includes an active first-to-peer probe; do not copy it unchanged into periodic readiness checks. The passive Kubernetes fixture keeps peer health and supporting path evidence distinct. See deployment scenarios and the deployment guide.

version: v1
targets:
  - name: first
    type: agent
    endpoint: http://localhost:8080/health
    agent:
      required_capabilities: [answer]
    dependencies:
      - name: peer
        type: a2a
        endpoint: http://localhost:8081
        a2a:
          required_skills: [answer]
      - name: first-to-peer
        type: agent
        endpoint: http://localhost:8080/health
        checks: [functional]
        agent:
          functional:
            safe: true
            text: Return a fixed health acknowledgement without tools or writes.
            downstream: peer
      - name: model-health
        type: http
        endpoint: http://localhost:8082/health
        critical: false