Serving AHP

Use v0.9.0 or newer. The example path requires a v0.9.0 or newer repository checkout and your application running on port 9000. In one terminal run:

agenthealth serve examples/ahp-check/agenthealth.yaml

In a second terminal query:

curl http://127.0.0.1:8080/ready
curl http://127.0.0.1:8080/live

Set AHP_TOKEN to a private random token of at least 16 characters through your secret manager or environment. Bearer-authorized GET requests to /health/dependencies and /health/capabilities return detailed results. To enable authenticated endpoints, stop the server and restart it in the first terminal after setting AHP_TOKEN:

agenthealth serve examples/ahp-check/agenthealth.yaml --listen 127.0.0.1:8080 --token-env AHP_TOKEN

Without a token these endpoints remain inaccessible. Public /health and /ready disclose only aggregate status and observation time.

The initial health response is UNKNOWN/503 until the first run completes. Checks repeat 30 seconds after each run and expire after two minutes. Explicit active probes also repeat; use a passive configuration for routine monitoring. SIGINT/SIGTERM cancel checks and gracefully shut down HTTP within five seconds. Successful shutdown returns 0; invocation, binding and server failures return 6. A health failure does not exit the serving process.

Remote deployments should terminate HTTPS and enforce rate limits at a proxy. See the protocol and impact review. This experimental implementation is included in v0.9.0; v0.8.0 does not include it.

Kubernetes serving

Phase 14 scenarios use this server in a Deployment and conventional sidecar. /ready follows snapshot freshness and aggregate readiness; /live drives process liveness/startup. Unlike CLI exec probes, AHP readiness can accept DEGRADED. Refresh timing, protected endpoint authorization and graceful shutdown retain this guide's existing behavior. See the deployment guide for Secrets and network exposure.