manage-agent

Run amctl agent create with the required flags for your subtype, build type, and provisioning mode, which auto-builds and auto-deploys to the default environment. Always run amctl agent create --help first since required flags vary by combination.

Do not rely on amctl agent get, since its status field is usually empty and not a liveness indicator. Instead check agent logs for app-level output past the init container, or confirm at least one memoryUsage sample via agent metrics.

Builds are addressed by buildName (e.g. hotel-booking-agent-1778760196718), not the buildId UUID. Pull the buildName from amctl agent build list --json and pass it to build get or build logs.

context show only reflects the instance URL and looks identical for unauthenticated contexts. Treat amctl project list --json as the real login canary; auth failure surfaces as error.code NO_ORG.

Delete is asynchronous, so agent list may show the agent for 5-15 seconds after data.deleted returns true. Poll agent list until the name disappears before assuming deletion completed.

The CLI has no pod describe, events, or readiness visibility, so ImagePullBackoff and CrashLoopBackoff can look like a fresh deploy. After about 3 minutes without metrics or app logs, escalate to kubectl describe pod and kubectl get events.