managed-model-endpoints

npx skills add xuzhougeng/wisp-science --skill managed-model-endpoints --agent claude-code

Same command for any agent — swap --agent for codex, cursor, copilot.

# Managed model endpoint boundary Wisp does not currently expose an endpoint registry or a service-lifecycle backend. The Agent cannot allocate ports, configure tunnels, read secrets, register health checks, or start and stop a persistent inference service through Python. Do not model service startup as a normal `run_in_context` command: a Run tracks one process lifecycle, while a managed endpoint also needs a durable endpoint identity, health, routing, authentication, restart policy, and ownership. ## Supported path If the user already operates an endpoint outside Wisp and the selected local, WSL, or SSH context can reach it using credentials already configured in that execution environment, load `using-model-endpoint` to run a bounded inference client. Never request or print secret values merely to make the call. Otherwise explain that endpoint registration and service management are not available in this Wisp build. A future implementation should add a typed service or execution-context backend with keyring-backed secret binding, health checks, start/stop/recovery semantics, and auditable invocation Runs.